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
Epithelial fusion is a crucial process in embryonic development , and its failure underlies several clinically important birth defects . For example , failure of neural fold fusion during neurulation leads to open neural tube defects including spina bifida . Using mouse embryos , we show that cell protrusions emanating from the apposed neural fold tips , at the interface between the neuroepithelium and the surface ectoderm , are required for completion of neural tube closure . By genetically ablating the cytoskeletal regulators Rac1 or Cdc42 in the dorsal neuroepithelium , or in the surface ectoderm , we show that these protrusions originate from surface ectodermal cells and that Rac1 is necessary for the formation of membrane ruffles which typify late closure stages , whereas Cdc42 is required for the predominance of filopodia in early neurulation . This study provides evidence for the essential role and molecular regulation of membrane protrusions prior to fusion of a key organ primordium in mammalian development . The fusion of apposed epithelial sheets is an essential process in the completion of many morphogenetic events including closure of the neural tube , optic fissure , palatal shelves , and cardiac septa . Failure of these fusion events leads to clinically important congenital malformations including neural tube defects ( NTDs: anencephaly and open spina bifida ) , coloboma , cleft palate , and cardiac septal defects , respectively ( Pai et al . , 2012; Ray and Niswander , 2012 ) . NTDs are among the commonest human birth defects , affecting 0 . 5–2 per 1000 pregnancies worldwide ( Copp et al . , 2013 ) . Understanding the mechanisms by which the vertebrate neural plate folds up and fuses to form a closed neural tube is thus of paramount importance for gaining insight into the embryonic pathogenesis of NTDs , and for developing improved methods for their prevention . In recent years , some of the molecular mechanisms underlying different morphogenetic aspects of neural tube closure have been unravelled . For example , the initial convergence and extension movements that narrow and elongate the neural plate were found to be regulated by the non-canonical Wnt-planar cell polarity pathway ( Wallingford and Harland , 2002; Williams et al . , 2014; Ybot-Gonzalez et al . , 2007b ) , whereas the subsequent bending of the mammalian neural plate at discrete medial and dorsolateral hinge points was shown to be regulated by Shh and BMP signalling ( Ybot-Gonzalez et al . , 2002; Ybot-Gonzalez et al . , 2007a ) . Much less is known , however , about the final steps of neurulation , involving fusion and remodelling of the neural folds at the dorsal midline . During epithelial ‘fusion’ , individual cells do not actually fuse with one another , but rather cells at the leading edges of apposed tissues form de novo adhesions to create a continuous epithelium . Neural tube closure involves a particular kind of epithelial fusion , in which two distinct tissues need to fuse and remodel: the pseudostratified neuroepithelium ( NE ) and the squamous surface ectoderm ( SE ) . Initially , these two tissues form a continuous ectodermal layer; however , during neural fold fusion , the continuity of this epithelium is disrupted at the bilateral NE/SE junctions , and new adhesions form between concurring epithelia from each side . Remodelling then generates the closed neural tube covered by the future epidermis ( Figure 1 ) . Cellular protrusions are often observed prior to apposition at the onset of epithelial fusion events ( Pai et al . , 2012 ) . It has long been known that membrane ruffles are present at the edges of apposed neural folds during neural tube closure in amphibians ( Mak , 1978 ) , birds ( Bancroft and Bellairs , 1975; Schoenwolf , 1979 ) , and mammals ( Geelen and Langman , 1979; Waterman , 1976 ) ( Figure 1 ) . Recently , filopodia have been observed at the neural fold tips in mice ( Massarwa and Niswander , 2013; Pyrgaki et al . , 2010 ) and in ascidians ( Ogura et al . , 2011 ) , together with F-actin enrichment along the NE/SE boundary ( Hashimoto et al . , 2015; Ogura et al . , 2011 ) . Early morphological studies in mice found that the initial contact between neural folds in the midbrain and hindbrain regions is made by SE cells , from which cellular protrusions emanate , whereas at the forebrain level initial contact is made by NE cells ( Geelen and Langman , 1977; 1979 ) . In chick , during cranial neurulation the SE and NE layers contact at the same time , but in the spinal region this first contact is made by SE cells ( Schoenwolf , 1979 ) , and in frog neurulation the SE closes first , and this closure is actually uncoupled from NE closure , which occurs later ( Davidson and Keller , 1999 ) . In mouse spinal neural tube closure , however , the cell type of origin of the protrusive cells and initial contact point have not been previously identified . Moreover , whether these protrusions are required for vertebrate neural tube closure at any level of the body axis is unknown . 10 . 7554/eLife . 13273 . 003Figure 1 . Schematic representation of the final events of neurulation in the spinal region of the mouse embryo . The apposing neural folds exhibit cell protrusions from their tips ( left ) , the neural folds then undergo fusion ( middle ) , and the two epithelia remodel to generate a closed neural tube covered by SE ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 003 Small GTPases of the Rho family are ubiquitously expressed molecular switches that cycle between active ( GTP-bound ) and inactive ( GDP-bound ) states and have a pivotal role in linking extracellular signals with several specific downstream effectors . In particular , Rac1 and Cdc42 are well-known regulators of the actin cytoskeleton that drives cellular protrusion . Rac1 induces the formation of the branched actin networks that underlie lamellipodia and membrane ruffles , and Cdc42 drives the assembly of unbranched actin bundles that form filopodia ( Heasman and Ridley , 2008; Ridley , 2011 ) . Even though the specific roles of the different Rho-GTPases were initially described using constitutively active and/or dominant negative forms , as well as pharmacological approaches , these techniques were later recognised to create limitations , owing to issues with specificity and dosage control . Conditional gene targeting has subsequently become the preferred method for studying Rho GTPase function in vivo in mammals , particularly mice ( Wang and Zheng , 2007 ) . Here , we describe the formation of different types of protrusions at the edges of the mouse spinal neural folds immediately prior to fusion . Using conditional targeting of Rac1 or Cdc42 in the NE and/or the SE , we show that these protrusions originate from SE , rather than NE cells . Furthermore , we show that Rac1 regulates the formation of ruffles , without which neurulation fails leading to open spina bifida , whereas Cdc42 is implicated in the formation of filopodia during earlier stages of neurulation . Membrane protrusions at the tips of the mouse neural folds have been described using both transmission electronic microscopy ( TEM ) ( Geelen and Langman , 1979 ) and scanning electronic microscopy ( SEM ) ( Waterman , 1976 ) . TEM provides very detailed imaging of both the cellular protrusions and the cells they emanate from . However , the sectional views obtained with TEM do not allow for a three-dimensional analysis of protrusive morphology . We therefore initially chose to describe the protrusive activity in the mouse spinal neural folds using SEM . We observed elaborate membrane protrusions at the point of fold apposition throughout spinal neurulation , and these protrusions were found to vary qualitatively as neurulation progressed ( Figure 2 ) . At the onset of neural tube closure ( somite-stage ( ss ) 7 ) , protrusions consisted mainly of long , finger-like filopodia ( Figure 2A ) . As neurulation progressed ( e . g . ss12 ) , we observed a mixture of filopodia and ruffles at the spinal fusion point ( Figure 2B ) . By late spinal neurulation stages ( from ss24 onwards ) , membrane ruffles only were observed , devoid of filopodial extensions ( Figure 2C ) . In some cases , cell protrusions were seen not only at the fold apposition point , but also along the edges of both open neural folds ( Figure 2C and Figure 2—figure supplement 1 ) . For consistency in the analysis between different embryos , we have focussed our analysis of protrusion types solely on the activity seen at the fold apposition point ( for further details of protrusive analysis , see 'Materials and methods' section and Figure 2—figure supplements 2 and 3 ) . 10 . 7554/eLife . 13273 . 004Figure 2 . Cell protrusions emanate from the interface between the NE and the SE of apposing neural folds during spinal neurulation . ( A–C ) SEMs of ss7 ( A ) , ss12 ( B ) , and ss24 ( C ) embryos . The point of spinal neural fold apposition exhibits filopodia at ss7 , filopodia and ruffles at ss12 and membrane ruffles at ss24 . Note the presence of ruffles at intervals along the edges of the PNP neural folds ( Ci , arrows ) . At least 10 different embryos were analysed . Scale bars: 100 µm ( A , B , C ) , 20 µm ( Ai , Ci ) , 10 µm ( Aii , Bi , Cii ) , and 5 µm ( Bii ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 00410 . 7554/eLife . 13273 . 005Figure 2—figure supplement 1 . Protrusions are present on the tips of both neural folds . Dorsal view of the PNP of the same embryo depicted in Figure 2C , showing membrane ruffles ( arrows ) on the edges of both neural folds . Anterior is top left , posterior is bottom right . Note also membrane blebs present on the apical surface of the NE , as described ( Waterman , 1976 ) . Scale bar: 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 00510 . 7554/eLife . 13273 . 006Figure 2—figure supplement 2 . Examples of different types of protrusions visualized by SEM at the PNP point of fusion . Ruffles , in which membrane ruffles without any filopodia protruding from them are the main or predominant type of protrusion observed . Ruffles and Filopodia , in which a mixture of the two types of protrusions are observed , sometimes with filopodia or microspikes emanating from the edges of ruffles . Filopodia , in which filopodia are the main or predominant type of protrusion observed . Absent , in which no or very few incipient protrusions are observed . Scale bar: 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 00610 . 7554/eLife . 13273 . 007Figure 2—figure supplement 3 . Quantification of the types of protrusion observed in all control embryos used in the different crosses in this study . There is a significant difference between embryos at the onset of spinal neurulation ( ss6-10 ) , with filopodia being predominant here , and in later stages embryos being replaced by ruffles only or ruffles and filopodia ( p=0 . 00011 and p=0 . 00000 for comparison with ss15-22 and ss23-30 , respectively ) . The percentage of embryos displaying ruffles only increases further at the end of spinal neurulation ( ss23-30 ) , but this difference is not statistically significant when compared to ss15-22 embryos ( p=0 . 21169 ) . A – Absent or incipient protrusions , F – Filopodia only ( or predominantly ) , RF – mixture of Ruffles and Filopodia ( or filopodia emanating from ruffles ) , R – Ruffles only ( or predominantly ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 007 Rac1 and Cdc42 are small Rho GTPases that regulate the cytoskeleton , particularly the actin networks that underlie the formation of lamellipodia/ruffles and filopodia , respectively ( Heasman and Ridley , 2008; Ridley , 2011 ) . Knock-outs of both Rac1 and Cdc42 are embryonic lethal before neurulation ( Chen et al . , 2000; Sugihara et al . , 1998 ) , and therefore , analysing their role in neural tube closure required the generation of conditional knock-out mice . We initially chose to conditionally ablate these GTPases by recombining floxed alleles of either Rac1 or Cdc42 with Cre recombinase expressed under the control of the Pax3 promoter . Pax3 is a transcription factor expressed in the dorsal-most cells of the developing neural plate and neural tube from early neurulation stages ( embryonic day ( E ) 8 . 5 ) ( Goulding et al . , 1991 and Figure 3—figure supplement 1 ) . To confirm effective Cre-driven recombination at the appropriate tissues and stages , we crossed Pax3Cre/+ mice ( Engleka et al . , 2005 ) with a homozygous ROSA26-EYFP reporter line ( Srinivas et al . , 2001 ) . As expected , YFP was expressed in the dorsal NE from E8 . 5 onwards ( Figure 3A , B ) , with some YFP-expressing cells also detected ventral to the Pax3 expression domain , consistent with recent findings ( Moore et al . , 2013 ) . Surprisingly , however , at neurulation stages later than ss20 , we also detected YFP expression in cells of the dorsal SE , mainly those directly in contact with the NE of the open neural folds ( Figure 3B ) . In confirmation of their SE identity , we found that these cells robustly express E-cadherin , whereas Pax3 was expressed only at very low intensity , or not at all ( Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 13273 . 008Figure 3 . Pax3Cre-Rac1 mutants display late failure of PNP closure , with absence of ruffles . ( A , B ) Pax3Cre-driven recombination in the dorsal neural folds and neural tube , detected from E8 . 5 by direct YFP-reporter expression ( A ) , and by immunofluorescence in transverse sections of the closing neural tube at E9 . 5 ( B ) . After ss20 , recombination is also detected in the dorsal SE ( red arrows ) , but not at earlier stages ( red crosses ) . Note also recombination in cells of the ventral NE ( red arrowheads; see also Figure 3—figure supplement 1 ) . At least three different embryos were analysed for each stage . ( C , D ) Pax3Cre-Rac1 mutants display open spina bifida at E11 . 5 ( C , white arrowheads and inset , quantified in Table 1 ) and delayed PNP closure from ss24-27 onwards ( D , p<0 . 001 – see Figure 3—source data 1 for raw values and statistical details ) . ( E , F ) SEMs of the PNP fusion point of control embryos show predominantly ruffles and filopodia at ss15-22 and ruffles at ss23-30 , whereas Pax3Cre-Rac1 mutants show ruffles and filopodia at ss15-22 and absent protrusions at ss23-30 ( E , quantified in F , p=0 . 29604 for ss15-22 and p=0 . 0002 for ss23-30 ) . A – Absent or incipient protrusions , F – Filopodia only ( or predominantly ) , RF – mixture of Ruffles and Filopodia ( or filopodia emanating from ruffles ) , R – Ruffles only ( or predominantly ) . Scale bars: 100 µm ( A and B ) , 1 mm ( C ) and 10 µm ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 00810 . 7554/eLife . 13273 . 009Figure 3—source data 1 . Source data and statistical analysis for Figure 3D . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 00910 . 7554/eLife . 13273 . 010Figure 3—figure supplement 1 . Pax3Cre drives recombination in a domain of cells that includes the dorsal SE , in addition to dorsal NE . ( A , B ) Transverse sections through the E9 . 5 PNP ( >ss20 ) of Pax3Cre-YFP embryos showing immunolocalisation of YFP and E-cadherin ( A ) and Pax3 protein ( B ) . Note co-localisation of E-cadherin and YFP in ( A ) . The YFP expression domain , which includes SE , dorsal NE , and scattered ventral NE cells , appears more extensive than the Pax3 expression domain , which is confined to dorsal NE ( B ) . Minimum of three embryos analysed . See also Figure 3B . Scale bars: 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 01010 . 7554/eLife . 13273 . 011Figure 3—figure supplement 2 . Pax3Cre-Rac1 conditional mutants show tissue-targeted deletion of Rac1 . ( A ) Whole-mount in situ hybridisation with sense and antisense RNA probes against mouse Rac1 exons 4 and 5 show specificity of the antisense probe used . ( B ) Pax3Cre-Rac1 mutants show Rac1 depletion in the dorsal NE ( asterisks ) and SE ( arrows ) , consistent with the findings of reporter expression using the Pax3Cre line ( Figure 3B ) . Scale bars: 500 µm ( A ) and 200 µm ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 01110 . 7554/eLife . 13273 . 012Figure 3—figure supplement 3 . Pax3Cre-Rac1 mutants show normal bending of the neural plate . Transverse sections , stained with haematoxylin and eosin , through the PNP of E9 . 5 embryos . Pax3Cre-Rac1 conditional mutants form dorsolateral hinge points ( arrowheads ) during spinal neurulation , similar to Pax3Cre-Con embryos ( n = 3 for each group ) , suggesting that faulty neural fold bending is unlikely to account for the neural tube defects in these mutant embryos . Scale bar: 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 01210 . 7554/eLife . 13273 . 013Figure 3—figure supplement 4 . Pax3Cre-Rac1 mutants show normal F-actin and adherens junction components distribution . Transverse sections through the recently closed neural tube of E9 . 5 embryos showing actin ( phalloidin staining ) , and β-catenin and E-cadherin immunolocalisation . Pax3Cre-Rac1 conditional mutants show a closely similar distribution of all these proteins ( see insets ) in both the targeted areas ( dorsal NE and dorsal midline SE; above dashed lines ) and non-targeted areas ( below dashed lines ) . This includes apical actin accumulation in the NE ( top and bottom panels; n = 3 for phalloidin ) β-catenin in both NE and SE ( top panel; n = 3 ) , and E-cadherin in SE ( bottom panel; n = 2 ) . Scale bar: 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 01310 . 7554/eLife . 13273 . 014Figure 3—figure supplement 5 . Pax3Cre-Rac1 mutants show defects in neural crest-derived structures . ( A ) Pax3Cre-Rac1 mutant embryo at E12 . 5 showing split face ( arrows ) , pools of blood indicating circulation defects , spina bifida ( arrowheads ) , and general embryo discolouration indicating imminent death ( quantified in Table 1 ) . Lines in ( A ) show the approximate levels of the sections shown in ( B ) ( i ) and ( C ) ( ii ) . ( B ) Histological transverse sections through the trunk at heart level of Pax3Cre-Con and Pax3Cre-Rac1 mutant embryos at E12 . 5 . The mutant embryo has formed a common arterial trunk ( CAT ) rather than displaying normal outflow tract septation with separate aortic ( a ) and pulmonary ( p ) trunks , and displays dorsal root ganglia ( DRG ) of reduced size ( 2 embryos analysed per group ) . ( C ) Histological transverse sections through the spinal cord at hindlimb bud level of Pax3Cre-Con and Pax3Cre-Rac1 mutant embryos at E12 . 5 . The mutant embryo exhibits dorsal root ganglia ( DRG ) of reduced size , and an open spinal cord at this level ( arrowheads ) ( two embryos analysed per group ) . Scale bars: 1 mm ( A ) , 500 µm ( B ) , and 200 µm ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 014 When Rac1 was ablated in the Pax3 lineage ( confirmed by mRNA in situ hybridisation , see Figure 3—figure supplement 2 ) , 76% of embryos displayed spinal NTDs consisting of either open spina bifida or a curled tail ( Figure 3C; Table 1 ) . These defects occurred at a similar frequency in both Pax3Cre/+; Rac1flox/- ( 21/27 ) and Pax3Cre/+; Rac1flox/flox ( 16/22 ) embryos ( p=0 . 947 ) , and hence these genotypes were combined for further analysis ( denoted Pax3Cre-Rac1 ) . In contrast , Pax3Cre/+; Rac1flox/+ and Pax3Cre/+; Rac1+/- control embryos ( denoted Pax3Cre-Con ) , which had conditional or constitutional heterozygous Rac1 loss of function , exhibited only 8% spina bifida , a significantly lower frequency than in Pax3Cre-Rac1 embryos ( Figure 3C; Table 1 ) . The third genotype group comprised embryos lacking the Pax3Cre allele , which were either wild-type ( floxed ) or heterozygous at the Rac1 locus ( denoted Non-Cre ) . Only 1/80 ( 1% ) of these embryos exhibited spina bifida , not significantly different from the frequency in Pax3Cre-Con embryos ( Table 1 ) . 10 . 7554/eLife . 13273 . 015Table 1 . Conditional genetic analysis of the roles of Rac1 and Cdc42 . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 015CrossGenotypeAbbreviated genotypeEmbryonic DayPax3Cre/+;Rac1f/+ X Rac1f/-TotalExencephalySpina bifida and/or curly tailPax3+/+;Rac1f/f , f/+ . f/- or +/-Non-Cre10 . 5 – 13 . 5801 ( 1% ) 1 ( 1% ) Pax3Cre/+;Rac1f/+ or +/-Pax3Cre-Con10 . 5 – 13 . 5393 ( 8% ) 2 ( 5% ) Pax3Cre/+;Rac1f/f or f/-Pax3Cre-Rac110 . 5 – 13 . 5494 ( 8% ) 37 ( 76% ) TotalDead or dyingSplit facePax3+/+;Rac1f/f , f/+ . f/- or +/-Non-Cre13 . 54300Pax3Cre/+;Rac1f/+ or +/-Pax3Cre-Con13 . 51700Pax3Cre/+;Rac1f/f or f/-Pax3Cre-Rac113 . 52118 ( 86% ) 21 ( 100% ) Pax3Cre/+;Cdc42f/+ X Cdc42f/-TotalExencephalySpina bifida and/or curly tailPax3+/+;Cdc42f/f , f/+ . f/- or +/-Non-Cre10 . 5 – 13 . 55200Pax3Cre/+;Cdc42f/+ or +/-Pax3Cre-Con10 . 5 – 13 . 51700Pax3Cre/+;Cdc42f/f or f/-Pax3Cre-Cdc4210 . 5 – 13 . 52300TotalDead or dyingSplit facePax3+/+;Cdc42f/f , f/+ . f/- or +/-Non-Cre13 . 5261 ( 4% ) 0Pax3Cre/+;Cdc42f/+ or +/-Pax3Cre-Con13 . 5600Pax3Cre/+;Cdc42f/f or f/-Pax3Cre-Cdc4213 . 51110 ( 91% ) 11 ( 100% ) Grhl3Cre/+;Rac1f/+ X Rac1f/f or f/-TotalExencephalySpina bifida and/or curly tailGrhl3+/+;Rac1f/f , f/+ . f/- or +/-Non-Cre10 . 5 – 13 . 514101 ( <1% ) Grhl3Cre/+;Rac1f/+ or +/-Grhl3Cre-Con10 . 5 – 13 . 5731 ( 1% ) 0Grhl3Cre/+;Rac1f/f or f/-Grhl3Cre-Rac110 . 5 – 13 . 54411 ( 25% ) 39 ( 89% ) TotalUnattached allantoisGrhl3+/+;Rac1f/f , f/+ . f/- or +/-Non-Cre9 . 51341 ( <1% ) Grhl3Cre/+;Rac1f/+ or +/-Grhl3Cre-Con9 . 5860Grhl3Cre/+;Rac1f/f or f/-Grhl3Cre-Rac19 . 56921 ( 30% ) Grhl3Cre/+;Cdc42f/+ X Cdc42f/fTotalDead or underdevelopedGrhl3+/+;Cdc42f/+Non-Cre9 . 5 – 10 . 5160Grhl3Cre/+;Cdc42f/f or f/-Grhl3Cre-Cdc429 . 5 – 10 . 51212 ( 100% ) Nkx1-2CreERT2/+;Rac1f/+ X Rac1f/-TotalExencephalySpina bifida and/or curly tailNkx1-2+/+;Rac1f/f , f/+ . f/- or +/-Non-Cre10 . 5 – 13 . 55100Nkx1-2CreERT2/+;Rac1f/+ or +/-Nkx1-2Cre-Con10 . 5 – 13 . 51600Nkx1-2CreERT2/+;Rac1f/f or f/-Nkx1-2Cre-Rac110 . 5 – 13 . 51700*p<0 . 001 when compared to either Non-Cre or DriverCre-Con . A small percentage of Pax3Cre-Rac1 embryos also developed the cranial NTD exencephaly , but at the same low frequency ( 8% ) as was observed in Pax3Cre-Con embryos ( Table 1 ) . This could reflect the predisposition of Pax3 heterozygotes to exencephaly ( Dempsey and Trasler , 1983 ) , although this exencephaly frequency was not significantly different from the 1% observed in Non-Cre controls . In any event , the finding of exencephaly in this study is unlikely to be related to the conditional ablation of Rac1 . The open spina bifida lesions in Pax3Cre-Rac1 embryos tended to be small and never extended further anterior than the level of the hindlimb bud ( Figure 3C ) . Other embryos displayed a curled tail but no open lesion ( not shown ) which , in other mouse mutants , can result from delayed spinal neural tube closure ( Copp , 1985 ) . To assess closure directly , we measured the length of the posterior neuropore ( PNP ) , the region of open spinal neural folds , in embryos between ss16 and ss31 . PNP length diminished progressively in control embryos as neurulation proceeded along the spinal region ( Figure 3D ) . Comparing Pax3Cre-Rac1 and Pax3Cre-Con embryos , before ss23 there was no detectable difference in PNP length but , from ss24 onwards , the PNP lengths of Pax3Cre-Rac1 embryos were significantly greater than those of Pax3Cre-Con embryos ( Figure 3D ) , consistent with a late failure of PNP closure and consequent relatively mild spinal neurulation defects . At ss20-23 , we also detected a significant difference in PNP length between Non-Cre and Pax3Cre-Con embryos ( Figure 3D ) . This is consistent with a delay in PNP closure in Pax3 heterozygotes ( Auerbach , 1954; Dempsey and Trasler , 1983 ) and the small percentage of spina bifida observed in these embryos ( Table 1 ) . To investigate whether the failure of PNP closure in Pax3Cre-Rac1 embryos was accompanied by a defect in the formation of protrusions at the fold apposition point , we analysed this region by SEM at mid ( ss15-22 ) and late ( ss23-30 ) spinal neurulation . Comparing the types of protrusions formed by Non-Cre and Pax3Cre-Con embryos did not reveal a difference between the two groups ( p=0 . 44 ) . Nonetheless , because Pax3 is known to regulate the cytoskeleton in osteogenic cells , and Rac1 activity is required for this function ( Wiggan et al . , 2002; Wiggan and Hamel , 2002; Wiggan et al . , 2006 ) , we also investigated protrusions in the Pax3 mutant mouse Sp2H , which contains a 32bp deletion in the Pax3 gene ( Epstein et al . , 1991 ) . No differences were observed between the protrusions of wild-type , heterozygous and Sp2H mutant embryos ( data available in doi:10 . 5061/dryad . rm660 ) . We conclude that Pax3 heterozygosity does not affect protrusion formation; from here onwards , the protrusion analysis utilised a single category of Controls , comprising pooled Non-Cre and Pax3Cre-Con embryos . Comparing Pax3Cre-Rac1 and control embryos , there was no significant difference between the proportions of protrusion types formed at ss15-22 , whereas at ss23-30 these proportions differed significantly between the two groups . In contrast to control embryos at these late neurulation stages , Pax3Cre-Rac1 embryos never formed ruffles only: they either exhibited both ruffles and filopodia ( 60% of cases ) , or had absent or incipient protrusions ( 40%; Figure 3F ) . These observations are consistent with the failure or delay in PNP closure seen in Pax3Cre-Rac1 embryos resulting from a defect in cell protrusive activity at late neurulation stages . In particular , Pax3Cre-Rac1 embryos are impaired in the formation of membrane ruffles devoid of filopodia . While it seemed most likely that the neural tube closure defect in Pax3Cre-Rac1 resulted from faulty formation of membrane ruffles , we considered whether conditional deletion of Rac1 might also lead to an earlier defect in neural tube closure , such as NE bending . Failure of dorsolateral bending can cause spina bifida in mice ( Ybot-Gonzalez et al . , 2007a ) . Transverse sections through the PNP of these embryos at E9 . 5 confirmed the presence of normal-appearing hinge points at both midline and dorsolateral positions ( Figure 3—figure supplement 3 ) , arguing against a mechanism of spina bifida in Pax3Cre-Rac1 embryos based on NE bending defects . Nevertheless , because Rac1 is required for the formation of adherens junctions ( Ehrlich et al . , 2002; Kovacs et al . , 2002; Yamada and Nelson , 2007 ) , we analysed the localisation of F-actin and β-catenin in the NE of Pax3Cre-Rac1 mutants , as well as E-cadherin in the SE . The distribution of these proteins was found to be closely similar in control and mutant embryos , as well as in both the targeted and non-targeted regions of the NE in mutant embryos ( Figure 3—figure supplement 4 ) , thus ruling out an effect of Rac1 knock-out on epithelial stability , which could impair neural tube closure . To address the question of whether the protrusion defects observed in Pax3Cre-Rac1 embryos are indeed a cause of failure of PNP closure , rather than a consequence of that failure , we chose to analyse the cell protrusive activity in a different mouse mutant with spina bifida . The curly tail ( ct ) mutant is homozygous for a hypomorphic allele of the transcription factor grainyhead-like-3 ( Grhl3 ) and exhibits spina bifida with 15–20% penetrance ( Gustavsson et al . , 2007; Ting et al . , 2003; van Straaten and Copp , 2001 ) . The size of the spina bifida lesions and the timing of failure of PNP closure in ct/ct embryos are similar to those in Pax3Cre-Rac1 embryos . Furthermore , the PNP defect in ct/ct embryos is known to be caused by a defect of cell proliferation in the hindgut , leading to excessive curvature of the body axis ( Brook et al . , 1991; Copp et al . , 1988 ) , and hence is unlikely to be related to protrusion and fusion events at the neural fold tips . We collected ct/ct embryos at ss24-30 and measured their PNP lengths ( Figure 4A and Figure 4—figure supplement 1 ) . For analysis of protrusions , embryos were grouped into two categories: those with PNP lengths above 0 . 6 mm ( large PNP ) , which are destined to develop spina bifida ( Copp , 1985 ) and those with PNP lengths below 0 . 4 mm ( small PNP ) that undergo normal PNP closure . Embryos with intermediate sized PNPs were not included in the analysis ( Figure 4—figure supplement 1 ) . The relative proportions of the different types of protrusions did not differ significantly between embryos with large and small PNPs ( Figure 4B , C ) , showing that delay in PNP closure does not cause defects in protrusive activity of cells at the dorsal fusion point . 10 . 7554/eLife . 13273 . 016Figure 4 . Failure of PNP closure does not cause defective protrusive activity . ( A ) SEMs of E9 . 5 curly tail embryos showing examples of small and large PNPs ( double arrows ) . ( B , C ) SEMs of the PNP fusion point of curly tail embryos show either membrane ruffles or ruffles and filopodia at ss23-30 ( B , quantified in C ) . There is no difference in protrusion type or frequency between embryos with small and large PNPs ( p=0 . 71782 ) . Definition of protrusion types as in Figure 3 . Scale bars: 100 µm ( A ) and 10 µm ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 01610 . 7554/eLife . 13273 . 017Figure 4—figure supplement 1 . Size range of PNPs of curly tail embryos collected at E9 . 5 and their grouping according to PNP length . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 017 The requirement for Rac1 solely at later stages of spinal neural tube closure raises two possibilities . Rac1 might be required for the formation of ruffles that are devoid of filopodia , but not for the combined appearance of ruffles and filopodia . These ruffles without filopodia might be needed only for late spinal NT closure . Alternatively , it is possible that Rac1 is required for neural fold protrusions along the whole spinal axis throughout all NT closure stages , but these protrusions arise on SE cells , and so Rac1-dependency is seen only at late neurulation stages , when the Pax3 lineage becomes targeted in the SE as well as the NE ( Figure 3B and Figure 3—figure supplement 1 ) . To address the first hypothesis , and to test a possible candidate for the formation of filopodia , we generated conditional Cdc42 mutants in the Pax3 cell lineage ( Pax3Cre-Cdc42 ) . None of the Pax3Cre-Cdc42 embryos showed NTDs ( Table 1 ) . Moreover , Pax3Cre-Cdc42 embryos collected during neurulation between ss16 and ss27 resembled control litter mates in PNP length ( Figure 5A ) . The proportions of protrusion types at mid ( ss15-22 ) and late ( ss23-30 ) neurulation did not differ between Pax3Cre-Cdc42 and control embryos ( Figure 5B , C ) . These results suggest that the protrusions seen at earlier neurulation stages are either regulated independently of both Rac1 and Cdc42 , or may emanate from SE rather than NE cells . 10 . 7554/eLife . 13273 . 018Figure 5 . Pax3Cre-Cdc42 mutants do not show defects in neural tube closure or protrusive activity . ( A ) Pax3Cre-Cdc42 mutants show a normal rate of PNP closure from ss16-27 ( see Figure 5—source data 1 for raw values and statistical details ) . ( B , C ) SEMs of the PNP fusion point of control and Pax3Cre-Cdc42 embryos show no difference in the types of protrusions formed at ss15-22 and ss23-30 ( B , quantified in C –p=0 . 0 . 1533 and p=0 . 36722 for ss15-22 and ss23-30 , respectively ) . This is consistent with the lack of spina bifida seen in Pax3Cre-Cdc42 mutants ( see Table 1 ) . Definition of protrusion types as in Figure 3 . Scale bar: 10 µm ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 01810 . 7554/eLife . 13273 . 019Figure 5—source data 1 . Source data and statistical analysis for Figure 5A . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 01910 . 7554/eLife . 13273 . 020Figure 5—figure supplement 1 . Pax3Cre-dc42 mutants show defects in neural crest-derived structures . ( A ) Pax3Cre-Cdc42 mutant embryo at E12 . 5 showing split face ( arrows ) , pools of blood indicating circulation defects , and general embryo discoloration indicating imminent death ( quantified in Table 1 ) . Lines in ( A ) show the approximate levels of the sections shown in ( B ) ( i ) and ( C ) ( ii ) . ( B ) Histological transverse sections through the trunk at heart level of Pax3Cre-Con and Pax3Cre-Cdc42 mutant embryos at E12 . 5 . The mutant embryo has formed a common arterial trunk ( CAT ) rather than displaying normal outflow tract septation with separate aortic ( a ) and pulmonary ( p ) trunks , exhibits dorsal root ganglia ( DRG ) of reduced size and occlusion of the dorsal part of the neural tube lumen ( ) ( two embryos analysed per group ) . ( C ) Histological transverse sections through the spinal cord at hindlimb bud level of Pax3Cre-Con and Pax3Cre-Cdc42 embryos at E12 . 5 . The mutant embryo exhibits dorsal root ganglia ( DRG ) of reduced size and occlusion of the dorsal part of the neural tube lumen ( * ) . Pax3Cre-Rac1 mutants show an open spinal cord at this level ( arrowheads ) ( two embryos analysed per group ) . Scale bars: 1 mm ( A ) , 500 µm ( B ) , and 200 µm ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 020 In view of the different effects on neural fold protrusions that we observed when deleting Rac1 and Cdc42 in the Pax3 cell lineage , it was important to confirm that Pax3Cre-Rac1 and Pax3Cre-Cdc42 embryos both developed expected phenotypes later in development . These small GTPases are required in the neural crest for proper development of structures including craniofacial primordia , dorsal root ganglia , and the cardiac outflow tract septum ( Fuchs et al . , 2009; Thomas et al . , 2010 ) . The Pax3 lineage encompasses the neural crest population ( Engleka et al . , 2005 ) , and therefore defects in neural crest derivatives were expected in both Pax3Cre-Rac1 and Pax3Cre-Cdc42 embryos . Embryos of both genotypes were dead by E13 . 5 ( Table 1 ) , as expected in cases of defective outflow tract septation ( Conway et al . , 1997 ) . Moreover , both displayed a split face phenotype from E11 . 5 onwards , consistent with a defect in neural crest-related craniofacial development ( Table 1 , Figure 3—figure supplement 5A and Figure 5—figure supplement 1A ) . Transverse sections at the heart level of E12 . 5 embryos confirmed the absence of an outflow tract septum in these embryos , as well as dorsal root ganglia that were severely reduced in size ( Figure 3—figure supplement 5B and Figure 5—figure supplement 1B ) . Transverse sections further down the spinal cord at the hindlimb level , where spina bifida occurs in Pax3Cre-Rac1 embryos , also showed severely reduced dorsal root ganglia ( Figure 3—figure supplement 5C and Figure 5—figure supplement 1C ) , thus confirming effective knockdown of Rac1 and Cdc42 protein throughout the spinal cord . Additionally , Pax3Cre-Cdc42 embryos displayed an apparent disorganisation of the dorsal neural tube , including occlusion of the lumen ( Figure 5—figure supplement 1C ) , consistent with a role described for Cdc42 in the polarity and organisation of the layers in the mouse brain ( Cappello et al . , 2006 ) . To test the hypothesis that the protrusions observed at the neural fold fusion point emanate from SE cells , we generated a conditional knock-out for Rac1 in this tissue by crossing Rac1 floxed mice with mice expressing Cre under the control of Grhl3 ( see Materials and Methods ) . Grhl3 is expressed predominantly in the SE during early neurulation ( Gustavsson et al . , 2007; Ting et al . , 2003 ) , and Grhl3Cre has been used as an early SE driver ( Camerer et al . , 2010; Massarwa and Niswander , 2013; Ray and Niswander , 2016 ) . Grhl3Cre-Rac1 mice were generated previously and found to have NTDs , including highly penetrant spina bifida , but cell protrusion analysis was not performed ( Camerer et al . , 2010 ) . We confirmed that Grhl3Cre drives recombination throughout the SE from E8 . 5 , and continuing at all stages of neurulation . We also detected recombination in a small proportion of scattered neuroepithelial cells ( Figure 6A , B ) , consistent with reported expression of Grhl3 in the neuroepithelium at E9 ( Gustavsson et al . , 2007 ) . Rac1 ablation in Grhl3Cre-Rac1 embryos was confirmed by mRNA in situ hybridisation ( Figure 6—figure supplement 1 ) . A proportion ( 30% ) of Grhl3Cre-Rac1 embryos displayed failure of chorioallantoic fusion ( Table 1; data not shown ) , a phenotype not previously described in these mice , which was accompanied by defects of growth and axial rotation that can result from such failure ( Morin-Kensicki et al . , 2006; Saunders et al . , 2004; Stumpo et al . , 2004 ) . These embryos were excluded from further analysis . Most of the remaining Grhl3Cre-Rac1 embryos had cranial and/or spinal NTDs: 25% exhibited exencephaly and 89% had spina bifida ( Table 1 ) . The size of the spina bifida lesions in these embryos was larger than that observed in Pax3Cre-Rac1 embryos , usually starting rostral to the hindlimb bud ( Figure 6C ) , which suggested an earlier failure of PNP closure . To examine this possibility , we measured the PNP of embryos at ss16-27 and detected a significant delay in PNP closure in Grhl3Cre-Rac1 embryos from ss20-23 , compared with Grhl3Cre-Con and NonCre-Con embryos ( Figure 6D ) . This confirmed that Grhl3Cre-Rac1 embryos fail to close their PNP earlier than Pax3Cre-Rac1 mutants . 10 . 7554/eLife . 13273 . 021Figure 6 . Grhl3Cre-Rac1 mutants show failure of PNP closure from ss20-23 , accompanied by abnormal protrusive activity . ( A , B ) Grhl3Cre-driven recombination in the SE is detected from E8 . 5 by direct YFP-reporter expression ( A ) , as well as by immunofluorescence in transverse sections of the closing neural tube ( B , red arrows ) . Note additional recombination in randomly scattered cells in the NE ( B , red arrowheads ) and other tissues . At least three different embryos were analysed for each stage . ( C , D ) Grhl3Cre-Rac1 mutants display open spina bifida at E11 . 5 ( C , between white arrowheads and inset , quantified in Table 1 ) and delayed PNP closure from ss20-23 ( D , *p<0 . 001 – see Figure 6—source data 1 for raw values and statistical details ) . ( E–H ) SEMs of the PNP fusion point of Grhl3Cre-Rac1 embryos show protrusive activity that differs from control embryos . Filopodia without ruffles are observed in Grhl3Cre-Rac1 embryos , especially at ss15-22 , and membrane ruffles without filopodia are never detected ( E , quantified in F , p=0 . 27024 for ss15-22 and p=0 . 02735 for ss23-30 ) . Definition of protrusion types as in Figure 3 . In the cases where filopodia where present ( RF and F categories ) , these were present in higher number ( G ) and were longer ( H ) in Grhl3Cre-Rac1 embryos when compared to Controls ( p<0 . 05 , p<0 . 001 , see Figure 6—source data 2 for raw values and statistical details ) . Scale bars: 100 µm ( A and B ) , 1 mm ( C ) and 10 µm ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 02110 . 7554/eLife . 13273 . 022Figure 6—source data 1 . Source data and statistical analysis for Figure 6D . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 02210 . 7554/eLife . 13273 . 023Figure 6—source data 2 . Source data and statistical analysis for Figure 6G , H . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 02310 . 7554/eLife . 13273 . 024Figure 6—figure supplement 1 . Grhl3Cre-Rac1 conditional mutants show tissue-targeted deletion of Rac1 . Grhl3Cre-Rac1 mutants show Rac1 depletion in the SE ( arrows ) , as well as a generalised reduction in signal , consistent with the findings of reporter expression using these Cre lines ( Figure 6A , B ) . Scale bar: 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 02410 . 7554/eLife . 13273 . 025Figure 6—figure supplement 2 . Grhl3Cre-Rac1 mutants show normal bending of the neural plate . Transverse sections , stained with haematoxylin and eosin , through the PNP of E9 . 5 embryos . Grhl3Cre-Rac1 conditional mutants form dorsolateral hinge points ( arrowheads ) during spinal neurulation , similar to Grhl3Cre-Con embryos ( n = 3 for each group ) , suggesting that faulty neural fold bending is unlikely to account for the neural tube defects in these mutant embryos . Scale bars: 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 025 Analysis of transverse sections through the PNP at E9 . 5 revealed normal-appearing dorsolateral hinge points in Grhl3Cre-Rac1 embryos , thus arguing against a neural plate bending defect as the underlying cause for the spina bifida phenotype ( Figure 6—figure supplement 2 ) . We next analysed the protrusive activity at the PNP fusion point in Grhl3Cre-Rac1 embryos . Despite only reaching statistically significant difference from controls at ss23-30 , Grhl3Cre-Rac1 embryos never showed membrane ruffles at any stage analysed , and also showed an increased occurrence of filopodia without associated ruffles ( Figure 6E , F ) . Furthermore , the density and length of filopodia in mutants appeared increased ( Figure 4E ) , and this was confirmed by measuring the number of filopodia present around the point of fusion , as well as their length ( see 'Materials and methods' for details ) ( Figure 6G , H ) . These results indicate that , like in late neurulation , cell protrusions emanate from SE cells during earlier neurulation and are at least partly regulated by Rac1 . The analysis of Pax3Cre-Rac1 and Grhl3Cre-Rac1 embryos suggested that Rac1 is required in the SE for the regulation of membrane protrusions: the implication being that the protrusions emanate from SE not NE cells . However , since Cre recombination also occurred in dorsal NE cells in Pax3Cre lines , and in scattered NE cells in Grhl3Cre lines , we could not rule out a role of Rac1 in the NE . To resolve this issue , we performed a further experiment to test specifically whether Rac1 function in NE cells may mediate cell protrusions and neural tube closure . We used Nkx1-2Cre-ERT2 ( Rodrigo Albors et al . , 2016 ) to generate Nkx1-2Cre-Rac1 embryos in which Rac1 was conditionally inactivated solely in NE cells . Nkx1-2 , also known as Sax1 , is a homeobox gene expressed in the posterior neural tube ( Schubert et al . , 1995 ) . In situ hybridisation confirmed that Nkx1-2 is expressed solely in the closing PNP of E9 . 5 embryos ( Figure 7A ) , with transcripts detectable only in the NE ( Figure 7B ) . Using the ROSA26-EYFP reporter , we confirmed that Nkx1-2Cre drives recombination in the NE of the PNP and previously closed neural tube ( Figure 7C ) , whereas there is no recombination in the SE ( Figure 7D ) . Nkx1-2Cre-Rac1 embryos developed entirely normally , and did not display NTDs ( Table 1 ) . Moreover , they formed normal ruffles typical of late neurulation ( Figure 7E , F ) . This experiment demonstrates unequivocally that Rac1-dependent cellular protrusions do not emanate from the NE during spinal closure . 10 . 7554/eLife . 13273 . 026Figure 7 . Nkx1-2Cre is expressed in NE and Nkx1-2Cre-Rac1 mutants display normal cellular protrusions . ( A , B ) In situ hybridisation for Nkx1-2 in whole mount E9 . 5 embryos . ( A ) Left: lateral view; right: dorsal view of the PNP . Nkx1-2 transcripts are confined to the neural plate and very recently closed neural tube . A transverse vibratome section at the level of the closing PNP ( B ) reveals Nkx1-2 expression solely in the NE , and not in adjacent mesoderm ( asterisks ) nor overlying SE ( arrows ) . ( C , D ) Nkx1-2Cre-driven recombination in the closing neural tube detected by immuno-fluorescence of YFP-reporter expression at E9 . 5 . Note the presence of YFP in the NE of the closing PNP ( red double-arrow ) and previously-closed neural tube ( white double-arrow ) ( C ) , but not in the SE lateral to the NE . A transverse section through the closing neural tube at E9 . 5 ( D ) shows the complete absence of YFP from the SE ( red crosses; 10 embryos analysed ) . ( E , F ) Nkx1-2Cre Rac1 mutants have no neurulation defects ( see Table 1 ) and SEMs of their PNP fusion point at ss24-30 show predominantly membrane ruffles , similar to control embryos ( E , quantified in F , p=0 . 75 ) . Definition of protrusion types as in Figure 3 . Scale bars: 500 µm ( A ) , 50 µm ( B ) , 100 µm ( C , D ) , and 10 µm ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 026 Our finding of increased density and length of filopodia in Grhl3Cre-Rac1 mutants ( Figure 6E , F ) suggested that , rather than being required for the formation of filopodia , Rac1 may in fact be needed to suppress them and maintain a balance between the formation of the different kinds of protrusions . To test the role of Cdc42 in the formation of filopodia in early neurulation , we conditionally ablated Cdc42 predominantly in the SE lineage , using Grhl3Cre . The Grhl3 and Cdc42 genes are located less than 2 Mb apart on mouse chromosome 4 ( www . ensembl . org ) , and therefore have less than 2% chance of recombination . In initial crosses , we obtained a rare recombinant in which the Grhl3Cre and Cdc42f alleles were in coupling , allowing the necessary Grhl3Cre/+; Cdc42f/+ x Grhl3+/+; Cdc42f/f mating to be performed , but precluding the generation of Grhl3Cre/+; Cdc42f/+ controls for this cross ( Table 1; see 'Materials and methods' for details ) . Remarkably , Grhl3Cre-Cdc42 embryos were dead by E10 . 5 ( Table 1 ) , and at E9 . 5 were already severely growth retarded , having failed to undergo axial rotation ( Figure 8A ) . These embryos underwent normal chorioallantoic fusion ( not shown ) , unlike some Grhl3Cre-Rac1 mutants . Due to their early lethality , we analysed the protrusive activity of Grhl3Cre-Cdc42 embryos at E8 . 5 . The initial closure event ( Closure 1 ) occurred normally at ss6-7 , and Grhl3Cre-Cdc42 embryos at ss6-10 were indistinguishable from control littermates ( Figure 8B ) . At the spinal closure point , cellular protrusions differed significantly from control embryos: they were predominantly ruffles , in sharp contrast with the predominance of filopodia seen in controls at these early neurulation stages ( Figure 8C , D ) . This finding suggests that Cdc42 is required for filopodia formation during early spinal closure . 10 . 7554/eLife . 13273 . 027Figure 8 . Grhl3Cre-Cdc42 embryos show altered protrusive activity . ( A ) Grhl3Cre-Cdc42 mutants have an embryonic lethal phenotype , with E9 . 5 embryos displaying reduced size and failure of axial rotation ( quantified in Table 1 ) . ( B ) SEMs of E8 . 5 embryos with fewer than 11 somites . At this stage , Grhl3Cre-Cdc42 mutants are indistinguishable in overall morphology from control littermates . ( C , D ) SEMs of the PNP fusion point of Grhl3Cre-Cdc42 mutants at ss6-10 show a predominance of membrane ruffles , in contrast to the filopodia seen in control embryos at this stage ( C , quantified in D , p<0 . 05 ) . Definition of protrusion types as in Figure 3 . Scale bars: 1 mm ( A ) , 100 µm ( B ) and 10 µm ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 027 Our genetic experiments indicated that the cell protrusions required for neural tube closure emanate from SE cells . To examine this question by a different methodology , we performed serial block-face SEM of the PNP fusion point ( Figure 9A–C ) . This technique allows high-resolution imaging of cells coupled with the ability to perform three-dimensional reconstructions which allow analysis of the entire cell shape , thus combining the advantages of TEM and SEM ( Hughes et al . , 2014 ) . In the case of the closing spinal neural folds , we were able to identify the membrane protrusions and their cell of origin in each section , and thus trace them in all the sections to finally obtain a reconstruction of the entire protrusive cell shape ( Figure 9 and Video 1 ) . We also traced non-protrusive SE and NE cells for comparison . 10 . 7554/eLife . 13273 . 028Figure 9 . Protrusive cells have a SE-like morphology . ( A–F ) Still images from Video 1 . ( A–C ) A series of transverse section images obtained through serial block-face SEM imaging of the closure point of the PNP at E9 . 5 . Protrusions are visible at the tips of the neural folds ( orange arrow in B ) . Black object in ( A ) is an artefact . ( D–F ) Three-dimensional reconstructions of different cell types from the same section-stack . Examples are shown of typical morphologies of pseudostratified NE cells ( dark blue: spindle shaped cell; cyan: wedge shaped cell ) and of a squamous SE cell ( yellow ) . A single pair of cells are extending protrusions , one from each neural fold ( green and red ) , and these have a squamous-type cell morphology , similar to SE cells . Relative to the sections in A–C , the reconstructed cell volumes are shown in the same orientation ( D ) , rotated 90° forward ( E ) or rotated 90° forward with zoom ( F ) . ( G , H ) Orthoslices from the analysed stack with superimposed three-dimensional reconstructions of the cells described above . Three different embryos were analysed at ss20-26 , with similar results . Scale bar: 100 µm ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 02810 . 7554/eLife . 13273 . 029Video 1 . Serial block-face SEM reveals that protrusive cells have a SE-like morphology . Animation showing a series of transverse images of the spinal neural tube closure point at E9 . 5 , with superimposed three-dimensional reconstructions of NE cells ( dark blue and cyan ) , a SE cell ( yellow ) , and protrusive cells at the tips of the neural folds ( green and red ) . See Figure 9 legend . DOI: http://dx . doi . org/10 . 7554/eLife . 13273 . 029 SE cells have a simple squamous epithelial morphology , with a short apico-basal dimension ( yellow cell , Figure 9D , E , G , and Video 1 ) , whereas NE cells have the typical shapes of a pseudo-stratified epithelium , with a long apico-basal axis , and wedge , spindle or inverted wedge morphology , depending on nuclear position ( dark blue and cyan cells , Figure 9D , E , G , and Video 1 ) , as described previously in the neural plate ( Schoenwolf , 1985; Smith et al . , 1994 ) . We confirmed that , in each transverse section , membrane protrusions emanated from only a single pair of cells bilaterally , one on each neural fold tip . Moreover , these cells were positioned precisely at the junction between the SE and NE . Their cell bodies were flat in the apico-basal direction and elongated in the plane of the tissue , closely similar to SE cells , except for the elaborate membrane protrusive activity on their free ends ( red and green cells , Figure 9D–F , H , and Video 1 ) . We conclude that the protrusive cells are most likely of SE origin , although we cannot rule out the hypothesis that they might be highly modified NE cells . Besides neural tube closure , other morphogenetic events involve epithelial tissue fusions accompanied by cellular protrusive activity . For example , during palatal shelf fusion in mice , filopodia are present at the medial edge epithelial cells just before fusion , and TGF-β3 knock-out mice that lack such protrusions display cleft palate ( Taya et al . , 1999 ) . Similarly , during eyelid closure , filopodia extend from the leading edge epithelia and are reduced in number and length in c-jun mutants that display defective eyelid closure ( Zenz et al . , 2003 ) . But perhaps the best studied case of cell protrusive activity in epithelial fusion during development is the process of dorsal closure in Drosophila . In this system , the leading edge cells of the lateral epidermis extend cell protrusions as they advance over the amnioserosa layer . When Rac or Cdc42 functions are perturbed , using dominant-negative proteins or loss-of-function mutations , this leads to defects of lamellipodia or filopodia , respectively , accompanied by failed dorsal closure , as well as misalignment of any segments that manage to close ( Harden et al . , 1995; 1999; Jacinto et al . , 2000; Woolner et al . , 2005; Hakeda-Suzuki et al . , 2002 ) . This suggests that in Drosophila , these protrusions may have a dual role , both in terms of mechanically participating in the closure process , and as exploratory structures that assure proper matching of fusing segments . However , because Rac is also involved in contraction of the underlying amnioserosa cells ( Harden et al . , 1995; 2002 ) , it is possible that its role in the leading edge cells is mainly exploratory . Unlike the lateral epidermis leading edge cells in Drosophila , which are moving over the amnioserosa , the protrusions that emanate from the mouse neural folds are not crawling on top of other cells or on extracellular matrix; rather , they extend into a fluid-filled space , and therefore could not be exerting any type of traction force to drive fusion . Moreover , in the segmented epidermis of the early Drosophila embryo , exact alignment of A-P segments appears vital and is sub-served by the epithelial protrusions ( Millard and Martin , 2008 ) . In contrast , the mouse developing spinal cord is not overtly segmented and it would appear that the role of neural fold protrusions is more likely to involve cell-cell recognition and/or signalling across the midline , to initiate epithelial fusion and ensure robust closure of the neural tube . Cell-cell recognition is likely a key step in triggering fusion , and many epithelial fusions involve cell recognition through Eph-ephrin interactions , including palatal shelf development ( Compagni et al . , 2003; Davy et al . , 2004; Risley et al . , 2009 ) , optic fissure closure ( Noh et al . , 2016 ) , and neurulation ( Abdul-Aziz et al . , 2009; Holmberg et al . , 2000 ) . The EphA2 receptor is present on the mouse spinal neural folds just before fusion , and its expression can be detected by TEM on the protrusions themselves ( Abdul-Aziz et al . , 2009 ) , raising the possibility that Eph-ephrin signalling may initiate upon contact between cell protrusions from apposing cells . Epithelial cell protrusions may also be involved in initiation of de novo cell adhesions . In cultured MDCK cells , E-cadherin accumulation is induced by contacts between Rac1-driven exploratory lamellipodia from opposing cells . The initial contact then spreads , driven by actomyosin contraction , while Rac1 activity and lamellipodial extension cease , and new junctions are formed between the two cells ( Yamada and Nelson , 2007 ) . What drives the formation of protrusions in the first place ? Recent studies in Drosophila suggest that during dorsal closure the epithelial leading edge cells undergo an incomplete epithelial-to-mesenchymal transition , caused by loss of apico-basal polarity ( Bahri et al . , 2010; Pickering et al . , 2013 ) . Loss of polarity in these cells results in a reduction of PTEN phosphatase , which in turn causes an increase in PIP3 ( Pickering et al . , 2013 ) , a known activator of Rac . As the cells meet in the midline , they switch back to ‘full’ epithelial character and restore cell-cell adhesion , an event mediated by Pak , an effector of both Rac and Cdc42 ( Bahri et al . , 2010 ) . Our results with Pax3Cre-Rac1 mutants show a clear requirement for Rac1 and ruffle formation on SE cells during the final stages of primary neurulation , from ss24 onwards . However , Rac1 is required in the SE earlier on , as Grhl3Cre-Rac1 mutants show delayed PNP closure from ss20 , and a lack of ruffles alone in these mutants is detected at ss15-22 , despite no overall statistical significance . This argues for a requirement of ruffles from at least mid-neurulation , and perhaps a gradual transition between protrusive types , with a balance of different protrusions needed . Despite the lack of ruffles , the shift towards excessive filopodia formation in Grhl3Cre-Rac1 mutants was unexpected given that , in cultured fibroblasts , expression of a dominant-negative form of Rac1 leads to inhibition of filopodia ( Johnston et al . , 2008 ) . Fibroblasts genetically deficient for Rac1 , on the other hand , are able to spread and move by extending filopodia , possibly through an Arp2/3-independent process ( Steffen et al . , 2013 ) . Filopodial initiation can be driven by either Arp2/3 ( branched nucleation ) or by formins ( unbranched nucleation ) ( Yang and Svitkina , 2011 ) . Moreover , Rac1 can both activate and inhibit the Arp2/3 complex , through either the Scar/WAVE complex or the Arpin protein , respectively ( Dang et al . , 2013 ) . If filopodium formation on the leading edge SE cells occurs through an Arp2/3-dependent process , then in the absence of Rac1 perhaps initiation of filopodial extension can be driven by other activators of the Arp2/3 complex . The occurrence of filopodia initiation would then be enhanced if Rac1 activates an Arp2/3 inhibitor such as Arpin in these cells . On the other hand , if initiation of these filopodia is driven independently of Arp2/3 , then the absence of Rac1 might shift the balance from branched to unbranched actin nucleation , resulting in more actin filaments being incorporated into filopodium-forming cross linked bundles . Our results show that Cdc42 is not needed in mid and late spinal neurulation , as the Pax3Cre-Cdc42 mutants have no defects in PNP closure or protrusive activity . But Cdc42 does play a role in early neurulation , as Grhl3Cre-Cdc42 mutants show a shift towards the extension of ruffles at the expense of filopodia . Whether or not this would impair progression of neural tube closure past ss10 could not be determined , given the early lethality of these embryos . It is possible that , in the absence/reduction of filopodia , membrane ruffles could take over their role and closure would progress . In fact , mutant mice for Ena/VASP proteins ( actin regulators involved in filopodium formation ) successfully close their PNP despite having cranial neural tube closure defects ( Kwiatkowski et al . , 2007; Lanier et al . , 1999; Menzies et al . , 2004 ) , arguing that filopodia are dispensable for spinal neurulation . Animal studies were performed according to the UK Animals ( Scientific Procedures ) Act 1986 and the Medical Research Council’s Responsibility in the Use of Animals for Medical Research ( July 1993 ) . Non-mutant embryos were from random-bred CD1 mice for standard SEM analysis , and BALB/c for serial block-face imaging SEM . Curly tail mice were maintained as a random-bred homozygous colony ( Gustavsson et al . , 2007 ) . Cre-driver lines were Pax3Cre/+ ( Engleka et al . , 2005 ) , Grhl3Cre/+ ( Camerer et al . , 2010 ) and Nkx1-2CreERT2/+ ( Rodrigo Albors et al . , 2016 ) . Floxed lines were Rac1f/f ( Glogauer et al . , 2003 ) , Cdc42f/f ( Wu et al . , 2006 ) , and ROSA26-EYFP ( Srinivas et al . , 2001 ) , all maintained on a C57BL/6 background . For the generation of conditional mutants , the following general scheme was followed ( where ‘Driver’ refers to either Pax3 , Grhl3 or Nkx1-2 , and ‘GTPase’ refers to Rac1 or Cdc42 ) : heterozygous floxed lines were initially crossed with mice carrying the ubiquitously expressed transgene Actb-Cre ( Lewandoski and Martin , 1997 ) to generate heterozygous Actb-Cretg; GTPase+/- mice , which were then back-crossed to GTPasef/f to generate heterozygous GTPasef/- ( with removal of the Actb-Cre ) . Doubly heterozygous DriverCre/+; GTPasef/+ mice were generated and crossed with GTPasef/- mice to obtain conditional mutants . This scheme was altered when the Grhl3Cre/+ line was found to drive recombination in the germ line of about 50% of the progeny ( not shown ) , and in that case the crosses were Grhl3Cre/+; GTPasef or -/+ X GTPasef/f . For the crosses with Nkx1-2CreERT2/+ , Cre activation was induced by intraperitoneal injection of pregnant mothers with a mixture of 20 mg/ml Tamoxifen ( Sigma ) and 10 mg/ml Progesterone ( Sigma ) , total volume 75 µl , at both E7 . 5 and E8 . 5 . Embryos were dissected in DMEM ( Invitrogen ) containing 10% fetal bovine serum ( Sigma ) and rinsed in PBS prior to fixation . Yolk sacs were used for embryo genotyping . Embryos were fixed for at least 2 hr in 4% paraformaldehyde in PBS , pH 7 . 4 , at 4°C , and dehydrated in a methanol series , except for the embryos stained for F-actin . Immunofluorescence was performed on 12-µm-thick cryosections of gelatine-embedded embryos ( 7 . 5% gelatine [Sigma] in 15% sucrose ) . F-actin was detected using Alexa-Fluor-568–phalloidin ( Life Technologies A12380 ) . β-catenin was detected using a rabbit polyclonal antibody ( Abcam ab16051 ) . YFP was detected using anti-GFP rabbit polyclonal Alexa488-conjugated antibody ( Life Technologies A21311 ) at 1:1500 dilution ( for single-label detection ) or anti-GFP chicken polyclonal antibody ( Abcam ab13970 ) at 1:500 dilution ( for double-labelling with E-cadherin ) . E-cadherin was detected using a rabbit monoclonal antibody ( Cell Signaling Technology 23E10 ) at 1:100 dilution . Pax3 was detected using a 1:50 dilution of mouse anti-Pax3 monoclonal antibody concentrate ( Developmental Studies Hybridoma Bank , created by the NICHD of the NIH and maintained at The University of Iowa , Department of Biology , Iowa City , IA 52242 ) . For E-cadherin and Pax3 , epitopes were unmasked by boiling three times for 3 min in citrate buffer . Secondary antibodies were goat anti-rabbit Alexa568 ( A21069 ) , goat anti-rabbit Alexa488 ( A11070 ) , goat anti-mouse Alexa568 ( A11004 ) , and goat anti-chicken Alexa488 ( A11039 ) ( all Life Technologies ) , all at 1:500 dilution . Images were captured on an Olympus IXZ1 inverted microscope or on an LSM710 confocal system mounted on an Axio Observer Z1 microscope ( Carl Zeiss Ltd , UK ) , and linear adjustments made using Fiji software . Specific primers ( 5’- ACGTGTTCTTAATTTGCTTTTCCCT-3’ and 5’- CCCCTGCGGGTAGGTGAT-3’ ) were designed to amplify exons 4 and 5 of mouse Rac1 cDNA ( the exons deleted in the conditional mutants used [Glogauer et al . , 2003] ) , generating a 200 bp fragment . Nkx1-2 probe was a kind gift from Dr F . Schubert ( Schubert et al . , 1995 ) . Whole-mount in situ hybridisations were performed using digoxigenin-labelled sense and anti-sense RNA probes , followed by preparation of 40 µm vibratome sections . Embryos were fixed overnight in Bouin’s solution ( Sigma ) , dehydrated in an ethanol series and embedded in paraffin-wax . Seven micron sections were stained using Harris’ haematoxylin solution and 2% Eosin Y ( both Sigma ) . Images were captured on an Axiophot2 upright microscope . Embryos were fixed overnight in 2% glutaraldehyde , 2% paraformaldehyde in 0 . 1 M phosphate buffer , pH7 . 4 , at 4°C , post-fixed in 1% OsO4/1 . 5% K4Fe ( CN ) 6 in 0 . 1 M phosphate buffer at 3°C for 1 . 5 hr and then rinsed in 0 . 1 M phosphate buffer . After rinsing with distilled water , specimens were dehydrated in a graded ethanol-water series to 100% ethanol , followed by one acetone wash . The samples were then critical point dried using CO2 and mounted on aluminium stubs using sticky carbon taps . The mounted samples were then coated with a thin layer of Au/Pd ( approximately 2 nm thick ) using a Gatan ion beam coater and imaged with a JEOL 7401 FEGSEM . Embryos were fixed for 12–36 hr in 3% glutaraldehyde and 1% paraformaldehyde in 0 . 08 M sodium cacodylate buffer , pH 7 . 4 , and then en bloc stained with osmium ferricyanide-thiocarbohydrazide-osmium , uranyl acetate , and Walton’s lead citrate as described ( West et al . , 2010 ) with two modifications . First , the osmium concentration was reduced to 1% and , second , graded alcohols ( 50 , 70 , 90 , 3 x 100% ) and propylene oxide were used instead of acetone to dehydrate specimens for infiltration and curing overnight at 60°C in Durcupan ACM resin . Specimens were then superglued to aluminium pins and trimmed to place the region of interest within a 0 . 5 x 0 . 5 x 0 . 4 mm mesa and sputter coated with 5 nm gold palladium . Stacks of backscatter electron micrographs were automatically acquired using a Gatan 3 view system in conjunction with a Zeiss Sigma field emission scanning electron microscope working in variable pressure mode at a chamber pressure of 9 Pa and 4 kV . At a standard magnification of x1000 and a pixel resolution of 4096 x 4096 , the total area sampled measured 255 . 4 µm2 on x and y and , depending on the number of 100-nm-thick sections sampled , between 67 and 150 µm on z . The resulting stacks were normalised for contrast and brightness and then converted to TIFF images in Digital Micrograph prior to importation into Amira 5 . 3 . 3 software for semi-automated segmentation and presentation . Protrusions were scored based on SEM images of the PNP fusion point taken at 2000x magnification , and categorised in four different classes: Ruffles ( comprised predominantly or solely of membrane ruffles ) , Ruffles and Filopodia ( either a mixture of both types of protrusions , filopodia that emanate from ruffles , or ruffles with microspikes ) , Filopodia ( comprised predominantly or solely of filopodial protrusions ) , Absent ( total absence of protrusions , or just one or two incipient protrusions ) . Examples of these types of protrusions can be found in Figure 2—figure supplement 2 , and the full dataset of protrusion images can be found in the Dryad Digital Repository ( doi:10 . 5061/dryad . rm660 ) . Scoring was done blind to embryonic stage and genotype by two different persons . In the minority of cases where the two scorings did not concur , a final decision was made by consensus . Where analysed , filopodial density was measured by counting the number of individual filopodia in an area of 2000 µm2 around the point of neural fold fusion , and filopodial length was measured in the same area using Fiji software . Only filopodia that measured above 1 µm were considered for these analyses . Kruskal-Wallis ANOVA on ranks was used for comparison of PNP size between different groups within each stage range . Fisher exact test was used for comparison of proportions of different types of protrusions and Chi-square test was used for comparison of phenotype frequencies in Table 1; when more than two groups were compared in multiple tests , the alpha-level was protected manually . Mann-Whitney Rank Sum Test was used to compare filopodial number and filopodial length ( Figure 6G , H ) .	The neural tube is an embryonic structure that gives rise to the brain and spinal cord . It originates from a flat sheet of cells – the neural plate – that rolls up and fuses to form a tube during development . If this closure fails , it leads to birth defects such as spina bifida , a condition that causes severe disability because babies are born with an exposed and damaged spinal cord . As the edges of the neural plate meet , they need to fuse together to produce a closed tube . It was known that cells at these edges extend protrusions . However , it was unclear how these protrusions are regulated , whether they arise from neural or non-neural cells and whether or not they are required for the neural tube to close fully . By studying mutant mouse embryos , Rolo et al . found that cellular protrusions are indeed required for the neural tube to close completely . These protrusions proved to be regulated by proteins called Rac1 and Cdc42 , which control the filaments inside the cell that are responsible for cell shape and movement . Rolo et al . also found that the cells that give rise to the protrusions are not part of the neural plate itself . Instead , these cells are neighboring cells from the layer that later forms the epidermis of the skin ( the surface ectoderm ) . Future studies will need to investigate which signals instruct those precise cells to make protrusions and to discover what happens to the protrusions after contact is made with cells on the opposite side . It will also be important to determine whether spina bifida may arise in humans if the protrusions are defective or absent .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2016	Regulation of cell protrusions by small GTPases during fusion of the neural folds
Convergence and extension movements elongate tissues during development . Drosophila germ-band extension ( GBE ) is one example , which requires active cell rearrangements driven by Myosin II planar polarisation . Here , we develop novel computational methods to analyse the spatiotemporal dynamics of Myosin II during GBE , at the scale of the tissue . We show that initial Myosin II bipolar cell polarization gives way to unipolar enrichment at parasegmental boundaries and two further boundaries within each parasegment , concomitant with a doubling of cell number as the tissue elongates . These boundaries are the primary sites of cell intercalation , behaving as mechanical barriers and providing a mechanism for how cells remain ordered during GBE . Enrichment at parasegment boundaries during GBE is independent of Wingless signaling , suggesting pair-rule gene control . Our results are consistent with recent work showing that a combinatorial code of Toll-like receptors downstream of pair-rule genes contributes to Myosin II polarization via local cell-cell interactions . We propose an updated cell-cell interaction model for Myosin II polarization that we tested in a vertex-based simulation . Polarised cell rearrangements drive the simultaneous elongation and narrowing of cell sheets ( convergence and extension ) during development . These collective cell behaviours have been mostly studied in the context of axis elongation that accompanies gastrulation in bilaterian animals , but are also found in organogenesis , for example underlying kidney tubule elongation ( Keller , 2002; Tada and Heisenberg , 2012 ) . Understanding convergence and extension movements is important as their failure is associated with congenital diseases , including neural tube defects ( Wallingford et al . , 2013 ) . The first molecular mechanism for convergence and extension was found in Drosophila , where planar polarisation of actomyosin was shown to underlie the polarised cell rearrangements of germband extension ( GBE ) ( Zallen and Wieschaus , 2004; Bertet et al . , 2004 ) . This discovery paved the way for in-depth studies of how the planar polarisation of actomyosin and other components such as Bazooka ( Par-3 ) and E-Cadherin drives the selective shortening of cell-cell junctions during active intercalation of epithelial cells ( Zallen and Wieschaus , 2004; Rauzi et al . , 2008; 2010; Levayer et al . , 2011; Levayer and Lecuit , 2013; Blankenship et al . , 2006; Fernandez-Gonzalez et al . , 2009; Simões et al . , 2010; 2014; Tamada et al . , 2012 ) . Recently , actomyosin planar polarisation was also found to be required during convergence and extension in vertebrate tissues ( Rozbicki et al . , 2015; Nishimura et al . , 2012; Lienkamp et al . , 2012; Shindo and Wallingford , 2014 ) . The upstream signals that pattern these polarities in the plane of the converging and extending tissues are starting to be deciphered . In vertebrates , the conserved planar cell polarity ( PCP ) pathway controls planar cell rearrangements during axis extension ( Wallingford , 2012 ) . In the Xenopus model , this pathway was recently shown to do so by biasing the polarisation of actomyosin ( Shindo and Wallingford , 2014 ) . In Drosophila , the PCP pathway is not required for polarisation of the actomyosin cytoskeleton ( Zallen and Wieschaus , 2004 ) , which instead depends on the segmentation cascade , the most downstream cues being the striped expression of pair-rule transcription factors such as Eve or Runt ( Zallen and Wieschaus , 2004; Bertet et al . , 2004 ) . Misexpression of these pair-rule transcription factors causes a local reorientation of polarity , which led to the hypothesis that local cell-cell interactions generate planar polarity in the Drosophila germband , rather than more global cues ( Zallen and Wieschaus , 2004 ) . Recent work has provided molecular evidence for this; three Toll-like receptors are expressed in overlapping stripes in the early embryo under the control of the pair-rule genes eve and runt ( Paré et al . , 2014 ) . Genetic disruption of these receptors leads to defects in GBE and a corresponding loss of the planar polarisation of Myosin II and Bazooka in the tissue . A model was proposed in which the germband is planar polarised through the preferential enrichment of Myosin II at sites of heterophilic Toll-like receptor interactions ( Paré et al . , 2014 ) . The overlapping expression domains of Toll-like receptors would therefore establish a combinatorial code where every cell along the antero-posterior ( AP ) axis has a different 'identity' , resulting in the bipolar distribution of Myosin II in every cell . These findings open new questions . One is what becomes of the combinatorial code and the planar polarisation of Myosin II once the cells have started intercalating and the number of cells increases along AP ? Specifically , if the cell identity stripes defined by the Toll-like receptor code are one cell wide to start with as hypothesised ( Paré et al . , 2014 ) , then these would increase to two cells wide on average after one round of cell intercalation . Heterophilic interactions between Toll receptors would no longer be expected at the interfaces between pairs of cells of the same 'identity' . Therefore one possibility is that these interfaces are not enriched in Myosin II at later stages of GBE . Alternatively , a secondary mechanism might be required to polarise the germband in later GBE , for example relying on a global polarising signal , more akin to PCP pathway-reliant polarisation in vertebrates ( Devenport , 2014; Goodrich and Strutt , 2011 ) . Another unsolved question is how the AP patterns established early in development are maintained during the cell movements of convergent extension ( Dahmann et al . , 2011; Vroomans et al . , 2015 ) . Cell rearrangements by intercalation are sufficient to cause mixing of adjacent cell populations ( Umetsu et al . , 2014 ) , therefore it is likely that a mechanism exists to maintain order along the AP axis of the germband . At later stages of embryonic development in Drosophila , an enrichment of actomyosin at parasegmental boundary ( PSB ) cell-cell interfaces is required to prevent cell intermingling caused by cell proliferation ( Monier et al . , 2010; 2011 ) . The actomyosin enrichment in this case is thought to act as a mechanical barrier , since the enriched PSB cell-cell interfaces align , indicating line tension . Supporting this notion , laser ablation experiments have demonstrated an increase in interfacial tension at compartmental boundaries in the wing disc and abdomen ( Umetsu et al . , 2014; Aliee et al . , 2012; Landsberg et al . , 2009 ) . Since parasegmental boundaries are defined genetically by pair-rule gene expression before gastrulation starts ( Lawrence and Johnston , 1989 ) , an unexplored possibility is that actomyosin enrichments at PSBs could form early , during GBE , and limit intermingling of cells during the large-scale cell rearrangements of convergence and extension . Here we take a systems biology approach to answer these questions by investigating the relationship between segmentation , the planar distribution of the motor Myosin II and the cell behaviours contributing to axis extension . We aimed to develop an analysis of these at the scale of the tissue , in living wild-type embryos . In particular , we asked what the relationship is between the described bipolar distribution of Myosin II at AP interfaces early in GBE and the later formation of parasegmental boundaries that stop mixing between anterior and posterior compartments . We show that Myosin II has a bipolar distribution in early embryos , which then transitions to a unipolar distribution as a direct consequence of polarised cell intercalation in the germband . Such an observation strongly supports that a cell identity mechanism polarises Myosin II throughout the whole of GBE . We show that the boundaries defined by the unipolar patterns , which include the PSBs , are the sites of the cell intercalation events driving GBE . We demonstrate that the PSB is a distinct mechanical structure from very early in GBE . These findings suggest that the boundaries we identify have a dual role , driving axis extension while ensuring that cell mixing remains limited . Finally , we propose an updated differential cell identity model . We reasoned that analysing the spatiotemporal modulations of actomyosin enrichment during GBE might answer the above questions by revealing undiscovered patterns . To quantify changes in Myosin II polarisation during GBE , we imaged the ventral surface of Drosophila embryos co-expressing the fluorescent fusion proteins GAP43-mCherry ( Martin et al . , 2010 ) , to label the cell membranes , and sqh-GFP ( Royou et al . , 2004 ) , to label Myosin II ( Figure 1A , Video 1 ) . Because sqh-GFP was expressed in a sqhAX3 null mutant background , all Myosin II molecules were tagged with GFP ( Royou et al . , 2004 ) . Images were acquired every 30 s , from before the start of extension , until the enrichment of Myosin II at parasegmental boundaries ( PSBs ) ( Monier et al . , 2010 ) was clearly detectable at the end of extension ( Video 1 , Figure 2A ) . The GAP43-mCherry signal was used to segment and track apical cell membranes over time ( Blanchard et al . , 2009; Butler et al . , 2009; Lye et al . , 2015 ) , while the sqh-GFP signal was used to quantify Myosin II fluorescence intensities for each cell-cell interface identified by cell tracking ( Figure 1A , Video 1 ) . We synchronised movies from 6 embryos to the start of GBE , using our previously described measure of tissue strain rate in the anteroposterior ( AP ) axis ( Butler et al . , 2009 ) ( Figure 1—figure supplement 1 ) . This allowed us to average the Myosin II fluorescence intensities associated with apical cell-cell junctions ( interfaces ) across embryos , which increased from the start of GBE as expected ( Figure 1B ) . 10 . 7554/eLife . 12094 . 003Figure 1 . Quantifying Myosin II polarisation over time during Drosophila axis extension . ( A ) sqhAX3; sqh-GFP; GAP43-mCherry embryos are imaged ventrally by confocal microscopy with a 196 x 173 μm field of view , with cell membranes visualised in the red channel and Myosin II in the green channel . Apical cell-cell interfaces are tracked over time based on the cell membrane signal . Next , Myosin II fluorescence intensities associated with the tracked cell-cell interfaces are quantified . Six movies were collected . VML: ventral midline . SEM image on the left from Flybase ( dos Santos et al . , 2015 ) . ( B ) Total fluorescence intensities for Myosin II at apical cell-cell interfaces over time . Data shown in B , C’ , C” , D’ and D” is averaged for the 6 movies . ( C–C” ) Quantification of Myosin II bidirectional polarity . ( C ) Fourier quantification of Myosin II bipolarity , depicted here on a movie frame 8 . 5 min after GBE onset . The length of the bipolar green vector represents the amplitude of polarity and its angle , the orientation of the polarity relative to the AP embryonic axis . Because the polarity is essentially aligned along the AP embryonic axis ( rosette in C’ ) , the polarity amplitude can be projected onto the AP axis and quantified using a Gaussian fit which allows a better separation between bidirectional and unidirectional polarity signals , compared to the Fourier method ( Figure 1—figure supplement 2 ) . ( C” ) Amplitude of Myosin II bidirectional polarity along the AP axis and over time , calculated using the Gaussian method . ( D ) Fourier quantification of Myosin II unidirectional polarity , shown on a movie frame 39 min after GBE onset . The length of the unipolar green vector represents the amplitude of polarity and its slope , the orientation of the polarity relative to the AP embryonic axis ( see corresponding rosette in D’ ) . The vector either points towards the anterior or the posterior , depending which side of a given cell is enriched in Myosin II . ( D” ) Absolute amplitude of Myosin II unidirectional polarity along the AP axis and over time , calculated using the Gaussian method ( Figure 1—figure supplement 2 ) . ( E ) Spatio-temporal map showing Myosin II bidirectional polarity for one representative movie ( SG_4 , Figure 1—figure supplement 1 and 3 ) , as a function of the AP position in the field of view ( x-axis , in µm ) and time relative to the start of GBE ( y-axis , in min ) . Movie frames corresponding to 0 and 50 min are shown on the left . At time 0 , the mesoderm is invaginating through the ventral furrow ( VF , white streak in the middle of the image ) . Mesoderm and mesectoderm cells on either side of the VF are not included in the analysis , nor are the germband cells at the posterior , because these move out of the field of view with the convergence extension of the tissue . Germband cells included in the analysis are labelled in magenta on both frames . At 50 min , most of the cells in the field of view are included in the analysis , except the mesectoderm cells at the midline ( VML ) and very dorsal germband cells coming in the field of view ( bottom ) . The amplitude of Myosin II bipolarity is expressed as a proportion ( Abbreviated as pp in all figures ) of the mean Myosin II intensity around the perimeter of each cell . Scale shows highest bidirectional polarity in bright green and no polarity in black . White lines on the plot follow the displacement of AP coordinates over time , which move posteriorly as the tissue undergoes extension . ( F ) Spatio-temporal map showing Myosin II unidirectional polarity for the same representative movie . The amplitude of unipolarity is expressed as a proportion ( pp ) of mean Myosin II intensities at cell-cell interfaces . Scale shows enrichment towards anterior cell-cell interfaces as green ( negative values ) and towards posterior as magenta ( positive values ) . Input data and statistics are in Figure 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 00310 . 7554/eLife . 12094 . 004Figure 1—source data 1 . Source data for Figure 1 , including statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 00410 . 7554/eLife . 12094 . 005Figure 1—source data 2 . Source data for Figure 1—figure supplement 1 , including statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 00510 . 7554/eLife . 12094 . 006Figure 1—figure supplement 1 . Synchronisation of sqhAX3; sqh-GFP; GAP43-mCherry movies . ( A ) Rates of tissue strain in the AP direction ( in proportion per hour ) comparing axis extension for the 6 movies sqhAX3; sqh-GFP; GAP43-mCherry ( named SG_1 to 6 ) . The movies have been synchronised to the start of GBE ( time zero ) ( B ) To check that the 6 movies are appropriately synchronised , we plotted when polarised cell intercalation starts and ends ( determined by our calculated intercalation strain rate , see Materials and methods ) and when the first ectodermal cell divisions occur ( determined by eye ) . ( C ) Frequency in the y-axis gives the number of tracked germband cells used in subsequent polarity analyses , per 30 s time bins from the start of GBE ( x-axis ) , pooled from the 6 movies . Input data and statistics are in Figure 1—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 00610 . 7554/eLife . 12094 . 007Figure 1—figure supplement 2 . Methods for calculating bidirectional and unidirectional Myosin II polarity . ( A , A’ ) To quantify bi- and uni- directional polarisation of Myosin II , we aim at identifying either two peaks ( A ) or one peak ( A’ ) around the perimeter of a given cell . Myosin II fluorescence intensities ( f . i . , y-axis ) are plotted along the unwrapped cell perimeter from 0 to 360 degrees ( x-axis ) , starting from the East side of the cell and going anti-clockwise . ( B-D’ ) We used two different methods , 'Fourier' and 'Gaussian' , to identify one ( unidirectional polarity ) or two ( bidirectional polarity ) peaks in the Myosin II fluorescence signal around the perimeter of each cell . Graphs B-D’ show simulated data for a cell with either bidirectional polarity in Myosin II ( B , B’ ) or unidirectional polarity ( C , C’ ) or a mixture of the two ( D , D’ ) . In the first method ( B , C , D ) , Fourier decomposition gave period 1 ( unidirectional polarity ) and period 2 ( bidirectional polarity ) amplitude estimates . In the second method ( B’ , C’ , D’ ) , two Gaussians curves with their means 180 degrees apart were fitted to Myosin II fluorescence intensities through minimisation , varying the amplitude of each Gaussian curve and a standard deviation common to both . ( B , B’ ) Example cell with bidirectional polarity . Both Fourier ( B ) and Gaussian ( B’ ) methods succeed well at fitting the two peaks of Myosin II intensities ( fluorescence intensities , f . i . , represented as a black castellated curve as in A , A’ ) . ( C , C’ ) Example cell with unidirectional polarity . The Fourier method fits a unidirectional peak at the correct position , but because the Myosin II signal is discrete , or castellated , there is also a strong period 2 harmonic in phase with the unidirectional signal . Thus with the Fourier method , the quantification of uni- and bi- directional polarities is not completely independent: specifically the period 2 estimate used for estimating bidirectional polarity will also describe some of the unidirectional polarity signal ( C ) . Gaussian fitting finds the correct unidirectional peak ( C’ ) . ( D , D’ ) Example cell with a combination of uni- and bidirectional polarity . Similar to ( C ) , the period 2 amplitude is augmented by the castellated signal in the Fourier method ( D ) , whereas the Gaussian fitting correctly identifies two independent peaks of different amplitude: the smallest amplitude identified by the West peak ( blue ) corresponds to the bidirectional polarity , whereas the higher East peak correspond to the unidirectional polarity ( D’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 00710 . 7554/eLife . 12094 . 008Figure 1—figure supplement 3 . Spatiotemporal maps for all sqhAX3; sqh-GFP; GAP43-mCherry movies . ( A ) Spatiotemporal maps for the representative sqhAX3; sqh-GFP; GAP43-mCherry movie SG_4 shown in Figure 1E , F . Cell number ( left panel ) , bidirectional ( middle ) and unidirectional ( right ) polarity contour maps calculated using the Fourier component method . Note the more persistent bidirectional polarity signal from 20 min onwards compared to Figure 1F ( which shows the Gaussian quantification ) , due to contamination from the unidirectional signal . ( B-F ) Cell number ( left panel ) , bidirectional ( middle ) and unidirectional ( right ) polarity contour maps for the 5 other movies , using the Gaussian fitting method as in Figure 1E , F . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 00810 . 7554/eLife . 12094 . 009Video 1 . Representative sqhAX3; sqh-GFP; gap43-Cherry movie ( SG_4 ) , showing the red ( top left ) and green ( top right ) fluorescence channels as well as the tracked cell shapes ( bottom left ) and the quantification of Myosin II fluorescence at tracked interfaces ( bottom right ) . See also Figure 1A . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 00910 . 7554/eLife . 12094 . 010Figure 2 . Parasegmental boundaries become mechanically active early during axis extension . ( A , B , D ) Frames of a representative sqhAX3; sqh-GFP; GAP43-mCherry movie ( SG_6 ) at 60 min after the start of GBE . ( A ) PSBs are identified at the end of the movie by strong enrichments in actomyosin ( arrows ) . ( B ) These are used to manually identify each parasegment ( differently coloured cell centroids ) . Note that the mesectodermal cells ( ME , highlighted in yellow ) present at the midline are not included in our analyses . ( C , D ) Using parasegment identification , we define 3 classes of linked columns of interfaces , the PSB interfaces ( green ) and those one cell anterior ( named ‘-1’ , in blue ) and posterior ( named ‘+1’ , in red ) to the PSB , shown in a schematic ( C ) and on the representative movie frame ( D ) . ( E ) Myosin II fluorescence intensities ( y-axis ) found at the three different classes of interfaces over time ( x-axis ) for the six sqhAX3; sqh-GFP; GAP43-mCherry embryos . Solid lines represent means . Ribbons ( error bands ) show an indicative confidence interval of the mean , calculated as a sum of the variance of the embryo means and the mean of the within-embryo variances . Blue and red bars at the top of the panel show time intervals over which -1 and +1 interfaces , respectively , differ from the PSB . Significance is calculated for each one-minute bin using a mixed model ( ‘lmer4’ package in ‘R’ ) using variation between embryos as the random effect . We use p<0 . 0005 as the significance threshold , which corresponds to a 0 . 05 threshold ( * ) modified by a Bonferroni correction to take into account the 81 one-minutes bins . The same conventions for displaying confidence intervals and statistical significance are used in all subsequent ribbon plots . ( F-I ) Comparison of junctional tension at PSB and +1 cell-cell interfaces using laser ablation . ( F ) Overlay of a PSB junction immediately prior to ( -1 time point , green ) and after ablation ( +5 time point , magenta ) . The rectangle shows the ablated region . Green arrows show the position of the vertices flanking the junction just prior to ablation . White dashes indicate the line used to produce the kymograph in ( G ) . Scale bar , 3μm . The kymograph shows the vertices recoil after ablation ( black frame indicated by yellow arrow at time zero ) . Time corresponds to -3 . 65 to 29 . 95 s relative to ablation . The changes in distance between vertices as measured on similar kymographs for each ablation are plotted in ( H ) . The graph shows the mean change in vertex distance over time for ablations at PSB ( black ) and +1 ( red ) interfaces ( N=19 ablated junctions for each ) . Error bars represent the 95% confidence interval of the mean . ( I ) Graph showing linear regression ( solid lines ) for the first 5 time points after ablation . The 95% confidence interval of the regressed line is also shown ( dotted lines ) . The data did not significantly deviate from linearity . Slopes were significantly different , with gradients of 0 . 2245 ( +-0 . 02665 ) for PSBs and 0 . 1084 ( +-0 . 0201 ) for +1s , so a ratio of 2 . 07 between the two . ( J ) Immunostaining of an eve-EGFP embryo at stage 8 using α-GFP and α-Engrailed antibodies , showing that the odd-numbered stripes of Engrailed-expressing cells are faithfully labelled by Eve-EGFP . Scale bar=25 μm . ( K ) -1 , PSB and +1 interfaces were identified in the three eve-EGFP , GAP43-mCherry movies and their orientation relative to the AP embryonic axis measured . The graph shows the proportion ( pp ) of interfaces oriented between 60 and 90 degrees relative to the AP axis , as a function of time . A LOWESS curve with a smoothing window of 10 points has been fitted to the data , for this graph and all other interface alignment graphs . Statistical comparisons are shown for the time point 40 min ( Cumulative interface orientation distribution for all interfaces at 40 min are shown in Figure 2—figure supplement 1K ) . The convention for P values for this graph and all subsequent similar graphs are: NS: p>0 . 05; p<0 . 05; p<0 . 01; **p<0 . 001 . ( L ) Same analysis for 3 wgCX4; eve-EGFP , GAP43-mCherry movies ( See also Figure 2–supplement 1L ) . This shows that in wingless mutants , PSB interfaces are more DV-oriented than -1 or +1 , as in wildtype ( K ) . ( M ) Graph comparing Myosin II enrichment at PSBs relative to +1 interfaces in fixed embryos labelled with α-Sqh1P antibodies , during GBE ( stage 8 ) and at later stages ( stage 9 , 10 , 11 ) , in wildtype and wingless mutants . Input data and statistics are in Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 01010 . 7554/eLife . 12094 . 011Figure 2—source data 1 . Source data for Figure 2 , including statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 01110 . 7554/eLife . 12094 . 012Figure 2—source data 2 . Source data for Figure 2—figure supplement 1 , including statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 01210 . 7554/eLife . 12094 . 013Figure 2—figure supplement 1 . Identification and characterisation of parasegmental boundaries properties during axis extension . ( A ) -1 , PSB and +1 interfaces were identified in 6 sqhAX3; sqh-GFP; GAP43-mCherry movies , and their orientations relative to the AP axis measured . The graph shows the proportion ( pp ) of interfaces between 60 and 90 degrees ( y-axis ) , as a function of time ( x-axis ) . Statistical comparisons are shown for the 40 min timepoint ( see also the cumulative distribution of all interface orientations for this timepoint in J ) . ( B , C ) , Separate Engrailed and GFP channel images for the stage 8 embryo shown in Figure 2J . Scale bars = 25 μm . ( D-F ) Immunostainings of stage 6 , 7 and 9 eve-EGFP , GAP43-mCherry embryos using anti-GFP and anti-En antibodies . The Eve-EGFP reporter labels the anterior edge of odd-numbered engrailed-expressing stripes faithfully , therefore marking the PSBs in every other parasegment . Scale bars = 25 μm . ( G-I ) Further analyses for the ablations shown in Figure 2 F-I . For all graphs N=19 for both PSB and +1 interfaces . ( G ) Graph comparing lengths of ablated PSB and +1 interfaces ( 95% confidence interval of the mean is shown ) . Interface lengths were not significantly different . ( H ) Graph comparing Sqh-GFP fluorescence intensities at ablated PSB and +1 interfaces ( 95% confidence interval of the mean in shown ) . Myosin II intensities were significantly higher in PSB junctions . ( I ) Cumulative distribution of the orientation of ablated PSB and +1 interfaces relative to the AP axis . AP=0 degrees; DV=90 degrees; pp= proportion . PSB interfaces were significantly more aligned to the DV axis than +1 interfaces . ( J-L ) Cumulative distributions of all interface orientations at the 40 min timepoint for the 3 types of movies analysed . ( M-M”’ ) Example of images used for the quantification of Myosin II in fixed embryos ( Figure 2M ) , here in a WT stage 9 embryo . Embryos are triple stained with Sqh-1P , DE-CAD and Engrailed antibodies . The En and DE-CAD channel is used to identify PSB interfaces ( arrowheads ) and +1 interfaces . The Sqh-1P channel is used for Myosin II quantification at the corresponding interfaces . Scale bar = 20 μm . Input data and statistics are in Figure 2—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 013 We further extracted independent measures of bidirectional and unidirectional Myosin II planar polarities in the orientation of the AP axis ( Figure 1—figure supplement 2 ) . Bidirectional polarity of Myosin II ( an enrichment at both anterior and posterior cell-cell interfaces for a given cell , Figure 1C , C’ ) was detectable just before the onset of extension and then peaked very early ( at 10 min ) before declining gradually ( Figure 1C” ) , consistent with previous studies ( Kasza et al . , 2014 ) . In contrast , unidirectional polarity ( an enrichment in Myosin II at either anterior or posterior cell-cell interfaces , Figure 1D , D’ ) increased progressively for most of GBE ( Figure 1D” ) . This suggests that there is a transition from bidirectional to unidirectional Myosin II polarisation over the course of GBE . To ask whether actomyosin polarities are patterned across the AP axis , we generated spatiotemporal heat maps for both types of polarity for the 6 embryos , as a function of time and position along the AP axis ( maps for a representative embryo in Figure 1E , F; see other embryos in Figure 1—figure supplement 3 ) . Note that in these maps , the data for all cells along the dorsoventral ( DV ) axis , within an AP bin of defined width , are averaged . Myosin II bidirectional enrichment is strong across the whole AP axis from just before the start of extension until about 20 min , then fades away ( green signal in Figure 1E ) . In the unidirectional polarity maps , cells with posterior interfaces enriched in Myosin II ( positive values , magenta signal ) are distinguished from those where the enrichment is at anterior interfaces ( negative values , green signal ) ( Figure 1F ) . A juxtaposition of opposing unidirectional polarities along the AP axis ( magenta next to green signal ) thus indicates that shared interfaces between neighbouring cells are enriched in Myosin II . Although the signal is noisy for single embryos , many such juxtapositions are found ( Figure 1F ) . These motifs follow the movement of the tissue as it extends towards the posterior ( white guide lines in Figure 1F ) . The most prominent ones occur at a regular spacing ( arrows in Figure 1F ) . We hypothesised that those correspond to early cable-like enrichments of Myosin II at parasegmental boundaries ( PSBs ) ( Monier et al . , 2010 ) . To test this , we tracked PSBs using two different approaches . First , we identified PSBs from clear cable-like enrichments of Myosin II at the end of the 6 movies analysed above , 60 min after the start of GBE ( arrows in Figure 2A ) . Using these boundaries , we manually assigned a parasegment identity to each tracked cell ( Figure 2B ) , which could be followed back to the beginning of each movie . This identified PSB cell-cell interfaces at each time point ( Video 2 ) . We also identified the cell-cell interfaces one-cell diameter anterior and posterior to each PSB ( named ‘-1’ and ‘+1’ interfaces , respectively ) over time ( Figure 2C , D ) . We then quantified the amount of Myosin II found at these three columns of interfaces over time for 3 to 4 parasegments per embryo , for all 6 embryos ( Figure 2E ) . We found that the enrichment in Myosin II at PSB interfaces becomes stronger than in the flanking columns of interfaces by 10–15 min of extension . If these cell-cell interfaces enriched in Myosin II were interconnected , they would be expected to straighten , a signature of line tension as shown for other tissue boundaries ( Umetsu et al . , 2014; Monier et al . , 2010; Aliee et al . , 2012; Landsberg et al . , 2009; Fagotto , 2014; Calzolari et al . , 2014 ) . To test this , we quantified the proportion of interfaces oriented between 60 and 90 degrees relative to the AP axis ( thus DV-oriented ) , for each class ( Figure 2—figure supplement 1A , J ) . We find that PSB interfaces are more DV-oriented compared to flanking -1 and +1 interfaces , throughout most of GBE ( note that all interfaces become briefly very DV-oriented at the beginning of GBE , which is caused by mesoderm invagination transiently stretching the germband cells along DV , see Lye et al . , 2015 ) . We interpret this as evidence that PSB interfaces align more than flanking interfaces . Together with the preferential enrichment in Myosin II at PSBs ( Figure 2E ) , this suggests that PSB interfaces are under higher tension than flanking interfaces during GBE . 10 . 7554/eLife . 12094 . 014Video 2 . Representative sqhAX3; sqh-GFP; gap43-Cherry movie ( SG_4 ) showing the green channel ( sqh-GFP ) with identification of the different parasegments and the parasegmental boundary interfaces . See also Figure 2A , B , D . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 014 To confirm this , we performed laser ablations to probe tension at specific cell-cell interfaces ( Figure 2F–I ) ( Rauzi et al . , 2008; Farhadifar et al . , 2007 ) . We ablated interfaces located at the PSBs at 40 min ( identified by their enrichment in Myosin II , see Materials and methods ) and compared them with the ablation of +1 interfaces ( one cell diameter posterior to PSB ) . We checked that PSB and +1 interfaces selected for ablation did not have significantly different lengths ( Figure 2—figure supplement 1G ) . PSB interfaces had more Myosin II than +1 interfaces , as expected ( Figure 2—figure supplement 1H ) . PSB interfaces are also more DV-oriented than +1 interfaces ( Figure 2—figure supplement 1I ) as expected from our interface alignment analysis . We found that PSB vertices recoiled significantly faster than +1 interfaces and to a greater extent ( Figure 2H ) . We estimated the difference in recoil velocities to be a factor of 2 ( Figure 2I ) . This confirms that PSB interfaces are under higher tension than flanking interfaces and validates our interface alignment analysis . To further confirm that PSBs are mechanically active during axis extension , we used a second approach to identify these boundaries , using eve-EGFP ( Venken et al . , 2009 ) to directly label the PSBs in embryos expressing GAP43-mCherry ( the latter to track cell interface behaviours as before ) . We found that eve-EGFP reliably marks the anterior edge of odd-numbered parasegments throughout GBE ( Figure 2J , Figure 2—figure supplement 1B–F ) . This allowed us as before to assign parasegment identities to cells and to track the PSB and flanking -1 and +1 interfaces at odd-numbered PSBs through time , for 3 eve-EGFP , GAP43-mCherry movies . We confirmed that interface orientation differences between PSB and flanking interfaces were replicated in these movies , where PSBs are labelled without relying on their enrichment in Myosin II ( Figure 2K and Figure 2—figure supplement 1K ) . Together , these results show that PSB interfaces are mechanically active by 15–20 min at the latest after GBE onset , much earlier than their previously known role at stage 10 when they segregate dividing boundary cells ( Monier et al . , 2010 ) . Since cell division in the germ-band ectoderm does not commence until 40 min after GBE onset in our movies ( Figure 1—figure supplement 1B ) , this suggested that PSBs have an early mechanical role during polarised cell intercalation . Because later in development , Myosin II enrichment at PSBs depends upon Wingless ( Wnt-1 homologue , expressed in one row of cells immediately anterior to the PSB interfaces; Monier et al . , 2010; Sanson , 2001 ) , we asked if this signalling pathway was also required for the mechanical activity of the PSBs during GBE . To test this , we generated 3 movies expressing eve-GFP and GAP43-mCherry in a wingless null mutant background ( wgCX4; eve-EGFP , GAP43-mCherry embryos ) . We performed the same interface orientation analysis as before , and found that the PSBs straightened in wingless mutant embryos as in wildtype ( compare Figure 2K and L and Figure 2—figure supplement 1K and L ) . We also quantified Myosin II enrichment at PSBs ( relative to +1 interfaces ) in fixed embryos at stages 8 to 11 ( Figure 2M and Figure 2—figure supplement 1M–M”’ ) . Although Myosin II is significantly decreased in wingless mutants at PSBs once the germband has finished extending ( stages 9 , 10 and 11 ) , confirming our previous findings ( Monier et al . , 2010; 2011 ) , we found no difference during GBE ( stage 8 ) . We conclude that the selective enrichment in Myosin II at PSB interfaces and their straightening during GBE is not controlled by Wingless , suggesting that it is under pair-rule gene control . There were more unidirectional polarity patterns in our spatiotemporal maps than just those corresponding to PSBs ( Figure 1F and Figure 1—figure supplement 3 ) . To characterise those , we increased the resolution of our maps by averaging the data collected for each of the 6 sqhAX3; sqh-GFP; GAP43-mCherry movies . We used our identification of PSB interfaces to attribute a within-parasegment coordinate value to each cell from 0 ( anterior-most ) to 1 ( posterior-most ) over time ( Figure 3A and Video 3 ) . Using this coordinate system , we averaged data from 3 to 4 parasegments per movie for our 6 movies . We replotted bidirectional and unidirectional polarity patterns at this parasegmental scale ( Figure 3B , C ) . Confirming individual movie maps ( Figure 1E and Figure 1—figure supplement 3 ) , we found that AP-oriented bidirectional Myosin II polarisation is strong across parasegmental domains until about 15 min after extension , decreasing thereafter ( bright to dark green signal in Figure 3B; statistics in Figure 3—figure supplement 1A , B , D ) . In contrast , unidirectional polarity emerges gradually from the start of GBE ( Figure 3C; Figure 1F; statistics in Figure 3—figure supplement 1A , C , E ) . First , as expected , anterior and posterior interfaces at PSBs have strong Myosin II enrichments of opposite sign , from as early as 10 min after GBE onset ( green and magenta respectively at each edge of the plot in Figure 3C ) . Second , the increase in resolution reveals two more positions along the AP axis where anterior and posterior unidirectional polarities alternate ( magenta/green boundaries highlighted with arrows in Figure 3C; statistics in Figure 3—figure supplement 1C ) . This suggests that there are columns of interfaces in at least two stereotypical locations within each parasegment that become enriched in Myosin II . This gradual transition from global bidirectional polarities to precisely located unidirectional polarities suggested that new DV-oriented junctions not enriched in Myosin II form as a consequence of cell rearrangements . To monitor the progress of cell rearrangements , we quantified the number of cells across the parasegmental domains over time . The average number of cells per parasegment width ( along AP ) almost exactly doubles , from 3 . 6 cells at the start of axis extension to 7 . 3 after 60 min ( Figure 3D and Figure 3—figure supplement 1F ) . This shows that the emergence of unidirectional polarity is concurrent with the progress of polarised cell intercalation . 10 . 7554/eLife . 12094 . 015Figure 3 . Within-parasegmental patterns reveal two further myosin-enriched boundaries at stereotypical AP locations . ( A ) Schematic of Drosophila embryo showing the parasegment domains along AP ( VML: ventral midline ) . Cells expressing Engrailed ( En ) and Wingless ( Wg ) abut the posterior and anterior edge , respectively , of each parasegmental boundary ( PSB ) . The identification of PSBs in movies ( Figure 2 ) was used to allocate an AP coordinate to each cell within each parasegment domain . The anterior-most position is recorded as 0 ( red in the heat scale ) and the posterior-most position is recorded as 1 ( blue ) . This coordinate system is used to pool cell information from all the different parasegments present throughout each movie , in order to look for stereotypical within-parasegment patterns . AP coordinates for a representative movie are shown at a late ( 38 . 5 min ) and an early ( 4 min ) timepoint , for 3 parasegments tracked ( PS4 , PS5 and PS6 ) . ( B ) Spatio-temporal map showing Myosin II bidirectional polarity for all averaged PS domains , as a function of the within-parasegmental AP coordinate ( x-axis ) and time relative to the start of GBE ( y-axis , in min ) . Heat scale shows highest bidirectional polarity in bright green and no polarity in black ( See statistics in Figure 3—figure supplement 1A , B ) . ( C ) Spatio-temporal map showing Myosin II unidirectional polarity for all averaged PS domains , as a function of the within-parasegmental AP coordinate ( x-axis ) and time relative to the start of GBE ( y-axis , in min ) . Heat scale shows enrichment of posterior cell-cell interfaces as magenta ( positive values ) and of anterior ones as green ( negative values ) ( See statistics in Figure 3—figure supplement 1A , C ) . ( D ) Quantification of average cell number per parasegment domain as a function of time relative to the start of GBE ( y-axis , in min ) . Cell numbers are obtained by dividing the average parasegment width ( psw ) by the average cell width ( cw ) . ( E ) Diagram showing the proposed model: at the start of GBE , 3 to 4 cells of distinct identity per parasegment enrich Myosin II at their shared interfaces . After cell rearrangement , stripes of cells of the same identity become adjacent . Myosin II is enriched preferentially at interfaces shared between cells of different identity ( PSBs , S1/2Bs and S2/3Bs , also marked on panel ( C ) . There is more Myosin II enrichment at PSBs compared to other boundaries , indicated as thicker green lines . We postulate that the third stripe , S3 as defined by unidirectional polarity data above ( panel C ) , is composed of a mixture of two identities , named 3 and 4 here , whose boundary is more variable . In support of this , S3 is wider than S1 and S2 , but not wide enough for 4 cells across ( 2+2 ) ( see also cell numbers per stripe in Figure 4D ) . Input data and statistics are in Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 01510 . 7554/eLife . 12094 . 016Figure 3—source data 1 . Source data for Figure 3 , including statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 01610 . 7554/eLife . 12094 . 017Figure 3—source data 2 . Source data for Figure 3—figure supplement 1 , including statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 01710 . 7554/eLife . 12094 . 018Figure 3—figure supplement 1 . Within-parasegmental patterns of Myosin II cell polarity . ( A ) Number of cells ( N ) per grid square for panels B , C in this figure and for the spatiotemporal maps in Figure 3B , C . Data pooled from 6 sqhAX3; sqh-GFP; GAP43-mCherry embryos . The total number of cell instances sampled was 124 , 241 . ( B , C ) Statistical information for spatiotemporal maps in Figure 3B , C . White squares indicate where Myosin II polarity is not significantly different from zero . Embryo variances are assumed to be equal . ( B ) Data significance for bidirectional Myosin II polarity . Coloured grid squares show where the data is significant , defined as where the mean value for each spatio-temporal grid square is greater than the 99 . 9% confidence interval of its distribution . Bipolarity strength is shown with the same scale as in Figure 3B . ( C ) Data significance for unidirectional Myosin II polarity . Coloured grid squares show where the data is significant , defined as where the absolute mean value per grid square is greater than the 95% confidence interval of its distribution . Unipolarity is colour-coded by the direction of polarity ( anterior or posterior ) , without the strength of polarity shown in Figure 3C . ( D , E ) Comparison of within-parasegment bidirectional ( D ) and unidirectional ( E ) polarity patterns prior to GBE ( -20 to -5 min ) and at three time periods during GBE; 0 to 20 min , 20 to 40 min and 40 to 60 min , from bottom to top panels . Y-axes show the strength of either bidirectional ( D ) or unidirectional ( E ) polarity . X-axes show the within-parasegment coordinates , 0 anterior-most , and 1 , posterior-most ( See Figure 3 ) . Grey ribbons close to the x-axes show where along AP the data is significantly different . ( F ) Expected number of cells as a function of within-parasegment AP coordinates ( x-axis ) versus time ( y-axis ) , averaged for all 6 sqhAX3; sqh-GFP; GAP43-mCherry movies . The number of cells per parasegment increases from 3 . 6 at the onset of GBE , to 7 . 3 on average at 60 , as shown in Figure 3D . The total number of cell instances sampled was 124 , 886 . Input data and statistics are in Figure 3—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 01810 . 7554/eLife . 12094 . 019Video 3 . Representative sqhAX3; sqh-GFP; gap43-Cherry movie ( SG_4 ) showing the tracked cell contours with within-parasegment coordinate colour-coded as shown in Figure 3A . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 019 To explain these patterns , we propose the following model . Because of the precision of the segmentation cascade ( Dubuis et al . , 2013; Tkačik et al . , 2015 ) , it is conceivable that there are as many cell identities as there are cells per parasegment width ( 3–4 on average , see Figure 3D and model in Figure 3E ) . At the start of GBE , actomyosin enrichment would occur at each cell-cell interface based on these differences in identity along the AP axis . When cells intercalate and make new contacts , this would bring cells of the same identity adjacent to each other along AP . Because their identities are the same , their new shared interfaces would not enrich in Myosin II ( Figure 3E ) . In contrast , interfaces between stripes of cells of different identity would continue to enrich in Myosin II , driving the emergence of persistent unidirectional polarity . A corollary of this model is that Myosin II polarisation is a consequence of local cell-cell interactions rather than global signals . If a global mechanism was at play , actomyosin would be expected to be enriched at all new DV-oriented interfaces , maintaining bidirectional polarisation , which is not what we find ( Figure 3B , C ) . This model generates specific predictions that we can test . In particular , the two new columns of interfaces identified as having strong unidirectional polarity within each parasegment should have more Myosin II and straighten more than the intervening cell-cell interfaces , after they emerge through cell intercalation . We tracked these , by manually identifying junctions enriched in Myosin II at the end of each movie ( as previously done for the PSBs ) , at the AP locations mapped in our spatiotemporal plots ( Figure 3C ) . This initial analysis defined 3 stripes per parasegment ( S1 , S2 and S3 , Figure 4A and Video 4 ) and identified cell-cell interfaces separating stripes 1 and 2 ( S1/2B ) and stripes 2 and 3 ( S2/3B ) ( boundary interfaces ) , from cell-cell interfaces within each stripe ( non-boundary interfaces ) ( Figure 4B , see also Figure 3C , E ) . We checked that the S1/2B and S2/3B interfaces identified at the end of GBE had AP positions consistent with their expected boundary positions throughout the movies ( Figure 4C ) . Next , we checked that cell numbers for each stripe matched those expected from the model , with S1 and S2 increasing approximately from 1 to 2 cells wide , and S3 from 1 . 5 to 3 cells wide , from start to end of GBE ( Figure 4D ) . The larger width of S3 is explained in our model: S3 would be composed of a mixture of cell identities 3 and 4 , because there are not enough cells in a parasegment ( 3 . 6 cells on average rather than 4 at the onset of GBE , Figure 3D ) to make a two-cell stripe for either identities 3 or 4 at the end of GBE ( Figure 3E ) . 10 . 7554/eLife . 12094 . 020Figure 4 . Behaviour of S1/2 and S2/3 boundaries . ( A ) The image is taken from a sqhAX3; sqh-GFP; GAP43-mCherry movie 60 min after the start of GBE , where cells have been manually allocated to putative within-parasegment stripes S1 ( red centroids ) , S2 ( green centroids ) and S3 ( blue centroids ) , based on Myosin II enrichment and position along AP . In this movie , allocation was done for 3 parasegments ( magenta centroids highlight cells belonging to other parasegments and yellow centroids belong to midline cells , ME ) . ( B ) Same movie frame where interfaces are classified as belonging to boundaries between stripes ( PSB interfaces in magenta , S1/2B interfaces in yellow , S2/3B interfaces in cyan ) or not belonging to any boundaries ( red interfaces in S1 , green interfaces in S2 and blue interfaces in S3 ) . ( C ) Spatiotemporal plot ( time in y-axis and within-parasegment coordinates in x-axis ) to check that the locations of manually identified within-parasegment boundaries correspond to the location of the S1/2B and S2/3B given by the unidirectional polarity map ( arrows , see Figure 3C ) . The proportion ( pp ) of non-boundary interfaces is colour-coded so that 1 is green ( only non-boundary interfaces ) and 0 is magenta ( only boundary interfaces ) . There is high concordance between the locations of S1/2B and S2/3B interfaces in both plots ( compare with Figure 3C ) . ( D-J ) Once stripe and interface identities are allocated , analyses can be performed on all tracked parasegments throughout GBE . ( D ) Average cell number per stripe in AP ( y-axis ) as a function of time from the start of GBE ( x-axis ) . At the bottom of the panel , red bar indicates the time intervals where S1 differs from S2 , and blue bar where S1 and S2 differ from S3 . ( E ) Average Myosin II intensity at boundary interfaces between stripe 1 and 2 ( S1/2B ) compared to interfaces immediately anterior ( -1 ) or immediately posterior ( +1 ) . Mean for PSB interfaces is shown for reference ( dashed line ) . Blue and red bars at the top of the panel show time intervals where -1 and +1 interfaces , respectively , differ from S1/2B interfaces . ( F ) Same quantifications as in E but for S2/3B . ( G ) Proportion of interfaces with orientation between 60 and 90 degrees relative to the AP axis ( y-axis ) , as a function of time ( x-axis ) , for S1/2B interfaces compared to -1 or +1 interfaces . The same measure for PSB interfaces is shown for reference ( grey curve ) . A statistical comparison is shown at 40 min ( see also I ) . ( H ) Same quantifications as in G , but for S2/3B . ( I , J ) show the cumulative distributions of interface orientation for S1/2B and S2/3B and control interfaces at 40 min . Input data and statistics are in Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 02010 . 7554/eLife . 12094 . 021Figure 4—source data 1 . Source data for Figure 4 , including statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 02110 . 7554/eLife . 12094 . 022Video 4 . Representative sqhAX3; sqh-GFP; gap43-Cherry movie ( SG_4 ) showing the within-parasegment stripes colour-coded as in Figure 4A . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 022 Next , we quantified Myosin II enrichment at the within-parasegment boundaries . As predicted , interfaces belonging to the boundaries S1/2B and S2/3B become more enriched in Myosin II than interfaces immediately anterior ( -1 ) or posterior ( +1 ) ( Figure 4E , F ) . We then examined the orientation of the different classes of interfaces over time . As predicted , S1/2B and S2/3B are more DV-oriented than +1 or -1 control interfaces ( Figure 4G–J ) . Note that as expected from the unipolarity maps ( Figure 3C ) , S1/2B and S2/3B are less enriched in Myosin II and less DV-oriented than the PSBs ( see PSB curves shown for comparison in Figure 4E–H ) , but overall these three boundaries have comparable behaviours . Based on our analysis of cell number , Myosin II enrichment and interface orientation , we conclude that we have identified two new columns of interfaces enriched in Myosin II within parasegments , with the behaviour predicted by our model ( Figure 3E ) . A further prediction of our model is that Myosin II enrichment should respond to the juxtaposition of different cell identities rather than to the orientation of the cell-cell interfaces relative to the main embryonic axes . To test this prediction , we examined Myosin II enrichment at boundary interfaces ( PSBs , S1/2B and S2/3B ) relative to non-boundary interfaces , as a function of interface orientation relative to the AP axis . Before 25 min of GBE , boundary interfaces have more Myosin II than non-boundary interfaces for all orientations except those parallel to AP ( 0 to about 20 degrees ) ( left panel in Figure 5A and Figure 5—figure supplement 1A–C’ ) . For both types of interface , there is some dependency upon orientation , with higher enrichment for interfaces closer to 90 degrees relative to AP ( DV-oriented interfaces ) , consistent with previous studies ( see for example Figure 4D in Kasza et al . , 2014 ) . This dependency upon orientation is lost after 25 min , with boundary interfaces strongly enriched compared to non-boundary interfaces , irrespective of orientation ( right panel in Figure 5A and Figure 5—figure supplement 1A–C’ ) . We conclude that although some more global mechanism might contribute to Myosin II enrichment at the beginning of GBE , cell-cell interactions dominate overall . 10 . 7554/eLife . 12094 . 023Figure 5 . Characterisation of the behaviours of boundary and non-boundary interfaces . ( A ) Average Myosin II intensity in boundary versus non-boundary interfaces for two time periods of GBE ( 0–25 and 25–50 min ) , as a function of their orientation relative to the AP embryonic axis . 0 degrees is parallel to AP , 90 degrees parallel to DV . ( B , C ) Analysis of cell neighbour exchanges . ( B ) Example of a T1 transition where the interface between cells C and D shortens to a single vertex , followed by the growth of a new interface between cells A and B . The graph gives the interface length ( y-axis ) as a function of time after the start of GBE ( x-axis ) . In this particular example , the T1 transition starts at 5 min and finishes at 15 min after the start of GBE . ( C ) Aligning all interfaces in time so that the T1 transitions are at zero min , this plot shows how the shortening of interfaces ( black curves ) correlates with the increase in Myosin II fluorescence intensity ( magenta curves ) during neighbour exchange . ( D-F ) Analysis of cell geometries . ( D ) We compared interface lengths predicted by a Voronoi tessellation ( black on the left , dotted grey on the right ) with real interface lengths ( magenta ) to extract a length deviation from the Voronoi tessellation , a geometric proxy for local stress . ( E ) Graph showing the average deviation in length from a Voronoi prediction ( y-axis ) , for all interfaces ( black line ) , for boundary interfaces ( magenta curve ) and for non-boundary interfaces ( green curve ) , as a function of GBE time ( x-axis ) . Non-boundary interfaces are on average longer and boundary interfaces shorter than the average length deviation for all interfaces . ( F ) On average , boundary interfaces become increasingly geometrically stressed ( shorter than Voronoi prediction ) over a period of 15 min just prior to T1 transitions . ( G-K ) Fate of boundary ( abbreviated to B ) and non-boundary ( abbreviated to nonB ) interfaces during GBE , for stripes S1 and S2 ( Data pooled from 6 embryos , N=96 , 343 interface instances ) . ( G ) S1 and S2 interfaces behaviours fall into four main types: interfaces that remain boundary throughout GBE and do not go through a T1 transition ( black ) ; interfaces that remain boundary throughout but go through a T1 transition ( grey ) ; boundary interfaces that go through a T1 transition and become non-boundary ( orange ) ; interfaces that remain non-boundary interfaces throughout ( purple ) . The percentage of each interface type is shown . Within each type , interfaces are sorted according to the time of T1 transition ( white lines ) . Black arrows indicate two infrequent subtypes . In the orange class , a subtype of boundary interfaces corresponds to interfaces between either cell identities 1 and 3 ( cell identity 2 is missing ) or 3 and 2 ( cell identity 1 is missing ) . We call these interfaces 'super-boundaries' ( abbreviated to superB ) ( see main text ) . We have inferred that identity 1 or 2 are skipped because for this subtype the tracking data shows that either stripe S1 or stripe S2 has a local width of zero . The other subtype is in the purple class ( arrow ) and corresponds to rare non-boundary interfaces that do go through a T1 transition . ( G’ ) Comparison of the timings of T1 transitions in the different interface types . The two infrequent subtypes have opposite behaviours: the super-boundary T1 transitions ( dashed orange curve ) are earliest , while non-boundary T1 transitions are latest ( purple curve ) compared to boundary T1 transitions ( orange and grey ) . ( G’’ ) Comparison of the distributions of the orientations of interfaces 2 . 5 min prior to T1 transition for the different types of interface . 'Super-boundary' interfaces are the most DV-oriented ( dashed orange ) . ( H ) Cartoon showing the expected location of the four types of interfaces relative to the position of the stripe boundaries ( dashed black lines ) . Green shows Myosin II enrichment . ( I ) Graph giving the frequency ( y-axis ) of each type of interface as a function of the AP position within a S1 or S2 stripe ( x-axis ) . Each AP location ( bin ) within a stripe is attributed a within-stripe coordinate from 0 ( anterior-most ) to 1 ( posterior-most ) . ( J ) Plot showing the fates of each type of interface during GBE . The mean interface orientation ( y-axis ) and length ( x-axis ) is plotted for each type over time . Dashed arrows show the direction of time . Dashed lines connect interfaces before and after T1 transitions . See Figure 5—figure supplement 1E for individual curves for each sqhAX3; sqh-GFP; GAP43-mCherry movie . ( K ) Cartoon summarising the behaviour of each type of interface during GBE . Changes in length and orientation of interfaces are depicted as well as the transition between boundary and non-boundary class . Direction of time is indicated by arrows . The dashed part of the grey arrow depicts the situation where a boundary interface remains a boundary after a T1 swap . Input data and statistics are in Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 02310 . 7554/eLife . 12094 . 024Figure 5—source data 1 . Source data for Figure 5 , including statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 02410 . 7554/eLife . 12094 . 025Figure 5—source data 2 . Source data for Figure 5—figure supplement 1 , including statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 02510 . 7554/eLife . 12094 . 026Figure 5—figure supplement 1 . Analysis of cell-cell interface behaviour . ( A-B’ ) Relationship between interface orientation and Myosin II fluorescence over time . Spatiotemporal plots show interface orientation relative to the AP axis ( x-axis; deg . , degree ) as a function of time during GBE ( y-axis ) . Plots A and B show the number of cells sampled per square grid , for boundary ( A ) and non-boundary interfaces ( B ) . Plots A’ and B’ show Myosin II fluorescence intensities associated with boundary ( A’ ) and non-boundary ( B’ ) interfaces . Boundary interfaces ( A’ ) have higher levels of Myosin II fluorescence ( green ) compared to non-boundary interfaces ( B’ ) throughout GBE , irrespective of interface orientation . ( C , C’ ) Comparisons of Myosin II fluorescence intensities ( f . i . ) over time at boundary ( C ) and non-boundary ( C’ ) interfaces for three groups of interface orientations; 0–30˚ ( parallel to AP ) , 30–60˚ and 60–90˚ ( parallel to DV ) . Blue and red bars show time intervals in which DV and AP-oriented interfaces , respectively , differ from 30–60 degree interfaces . Although there are subtle differences in Myosin II enrichment depending on interface orientations , the main difference in Myosin II is between boundary and non-boundary interfaces ( Compare graphs C and C’ ) . ( D ) Interface length deviation from topologies predicted from a Voronoi tessellation . Interfaces shorter than expected from a Voronoi topology ( negative values; yellow to red ) are strongly DV-oriented and heading towards T1 transitions . ( E ) Fate of each interface type as a function of time in the course of GBE , with a curve for each movie ( legend in Figure 5J ) . ( F-H ) Evidence for an additional boundary within stripe S3 . ( F ) Number of T1 transitions ( y-axis ) as a function of AP position , attributing coordinates 0 ( anterior-most ) to 1 ( posterior-most ) for each stripe S1 , S2 and S3 ( x-axis ) . The number of T1 transitions is highest close to 0 and 1 , corresponding to the PSB , S1/2B and S2/3B boundaries . Another boundary is suggested by a peak in the middle of S3 , which according to our model would arise between cell identities 3 and 4 ( Figure 3E ) . ( F’ ) Cumulative frequency distributions of the number of T1 transitions ( y-axis ) as a function of the distance from the stripe boundaries ( x-axis ) . S3 has significantly more T1 transitions away from the flanking boundaries , as expected from an additional peak within S3 . ( G ) Comparison of interface Myosin II fluorescence across the three stripes . Dashed lines show confidence intervals of data pooled from the 6 sqhAX3; sqh-GFP; GAP43-mCherry movies . Stripe 3 has a noticeable peak mid-stripe that is missing in stripes S1 and S2 . ( H ) Rate of intercalation convergence in the DV-axis for the first 25 min of germband extension ( y-axis ) as a function of within-parasegment coordinates ( x-axis ) . The rate of intercalation increase from anterior to posterior , being highest in S3 . 95% confidence intervals are shown for data pooled as in G . Input data and statistics are in Figure 5—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 026 Another prediction from our model is that the boundary interfaces should drive convergence extension , in other words they should shorten actively , since they are more enriched in Myosin II than non-boundary interfaces . We have already shown that PSBs , S1/2Bs and S2/3Bs become straighter than intervening interfaces , which is evidence that they are more contractile . To ask if they participate more in cell rearrangements , we developed a method to capture the cell neighbour exchanges called T1 transitions ( see Materials and methods ) . T1 transitions are identified by following the shrinkage of a given interface and linking it to the growth of a new interface ( Figure 5B ) . Using this method , we identified every T1 transition occurring in stripes S1 and S2 for all tracked parasegments in our 6 sqhAX3; sqh-GFP; GAP43-mCherry movies . We did not analyse stripe S3 as we cannot unambiguously identify boundary interfaces separating the putative cell identities 3 and 4 in that stripe . For each T1 transition identified , we have information on how much Myosin II is found at shortening and elongating interfaces . Pooling all the T1 transitions in S1 and S2 together , we find that Myosin II increases with interface shortening prior to the interface swap ( Figure 5C ) , consistent with prior studies ( see for example Figure 1g , i in Rauzi et al . , 2010 ) . To distinguish between the interfaces that are shortening actively from those that may shorten passively , we developed another method to probe geometric stress ( see Materials and methods ) . We assume that a Voronoi tessellation based on cell centroid locations represents a mechanically neutral configuration for the cell-cell interfaces . We measured the deviation in interface length from this tessellation for boundary and non-boundary interfaces ( Figure 5D , Figure 5—figure supplement 1D and Materials and methods ) . We find that boundary interfaces are shorter than predicted by a Voronoi tessellation ( Figure 5E ) , particularly so in the 15 min prior to a T1 transition ( Figure 5F ) , indicating that they actively shorten during GBE . We conclude that the boundary interfaces that we have identified drive convergence of the germ-band . Next , we examined the behaviour of all interfaces during GBE for S1 and S2 ( Figure 5G–K ) . We identify four main interface behaviours . About a quarter of interfaces are boundary interfaces which are not involved in any T1 transitions and remain boundary interfaces throughout GBE ( black in Figure 5G–K ) . At the start of GBE , these interfaces are oriented on average about 50 degrees relative to the AP axis , then rotate to become oriented closer to DV , around 70 degrees ( Figure 5J , K and Figure 5—figure supplement 1E ) . Another quarter of interfaces are boundary interfaces involved in T1 transitions , with two distinct behaviours: some remain boundary interfaces after the T1 swap , while others become non-boundary interfaces ( grey and orange , respectively , in Figure 5G–K and Figure 5—figure supplement 1E ) . Finally , the rest of the interfaces are non-boundary interfaces which , for their large majority , are not involved in T1 transitions as expected ( purple , Figure 5G–K and Figure 5—figure supplement 1E ) . This confirms that boundary interfaces are those involved in cell neighbour exchange . Each interface behaviour occurs at the expected AP locations within each stripe , giving further support to our model ( Figure 5H , I ) . Furthermore , examining now the whole data set ( considering all three stripes S1 , S2 and S3 ) , we find that the number of T1 transitions consistently peaks at the expected locations for PSBs , S1/2B and S2/3B ( Figure 5—figure supplement 1F ) . Interestingly , we also find a T1 transition peak in the middle of S3 ( blue curves in Figure 5—figure supplement 1F , F’ ) , which would correspond in our model to an incomplete or variably located boundary between cell identities 3 and 4 ( Figure 3E ) . This is corroborated by a peak in Myosin II in the middle of stripe S3 ( Figure 5—figure supplement 1G ) . Using an independent measure of cell intercalation ( intercalation strain rate , see ( Butler et al . , 2009; Blanchard et al . , 2009 ) and Materials and methods ) , we find that the rate of intercalation is higher in stripe S3 , compared to stripe S1 and S2 ( Figure 5—figure supplement 1H ) . We think that this higher rate of intercalation in stripe S3 is caused by missing cells of identity 3 or 4 in this stripe . We postulate that when a cell identity is missing in the AP parasegmental sequence , such as cell identity 3 or 4 , the resulting interface enriches more Myosin II and consequently intercalates faster and earlier that other interfaces . We expect these 'superboundary' interfaces ( behaving as 'superintercalators' ) to be most prevalent in stripe S3 because of insufficient cells there , but our data suggest that these can be found also ( but rarely ) in stripe S1 and S2 ( SuperB subtype in Figure 5G–G” ) . We conclude that the variable number of cells per parasegment along AP causes a faster intercalation rate in the posterior part of the parasegment compared to the anterior part ( Figure 5—figure supplement 1H ) . The current molecular explanation for the planar polarization of Myosin II during GBE is that pair-rule genes control the expression in stripes of three Toll-like receptors that provide a heterotypic code for the enrichment of Myosin II at AP cell-cell interfaces ( Paré et al . , 2014 ) . The code is thought to be incomplete because it currently does not explain interface enrichment at PSBs ( Paré et al . , 2014 ) . Here we asked what is the minimum number of receptors that could explain all of the Myosin II patterns that we have uncovered in this study . We first considered a scenario recapitulating as closely as possible the expression of the three Toll-like receptors ( Toll-2 , Toll-6 and Toll-8 ) identified in Paré et al . ( 2014 ) . We noted that Toll-6 and Toll-8 were largely interchangeable ( Paré et al . , 2014 ) . Therefore our first scenario has a receptor A and a receptor B respectively expressed in pair-rule patterns broadly similar to Toll-2 and Toll-6/8 ( Figure 6A ) . Assuming initially 4 cells per parasegment , we counted by how many receptors adjacent cells differed along the AP axis . For example , if a cell expresses a receptor and the adjacent cell does not , then we recorded a difference of 1 for the corresponding AP interface ( Figure 6A ) . We postulate that a difference of one receptor or more triggers Myosin II enrichment at the corresponding interface . In this first scenario , all interfaces along AP differ by one receptor , except at the PSBs where there are no differences , consistent with the conclusion that the Toll-like receptor patterns currently do not explain Myosin II enrichment at PSBs ( Paré et al . , 2014 ) . 10 . 7554/eLife . 12094 . 027Figure 6 . Finding the smallest number of receptors explaining Myosin II planar polarization during axis extension . ( A ) Expression patterns of two putative receptors A and B repeated every double parasegment ( corresponding to the expression patterns of , respectively , Toll-2 and Toll-8 as described in Figure 1p in Paré et al . , 2014 ) . PSBs are shown as solid black lines . Summing the number of receptor differences at each boundary , this combination lacks a difference at the PSBs . For cell pairs brought together when single cells are missing ( second line ) , the number of cell receptor differences increases only when cell identity 2 or cell identity 3 is missing ( grey boxes highlight an increase in receptor differences ) . There is no increase in receptor differences , hence no robustness , built in if two contiguous cells are missing at any location ( third line ) . We calculate a 'robustness' score by adding the number of instances , for two parasegments , where there is an increase in receptor differences in the event of 1 or 2 cells missing: the score for this scenario is 10 ( number of grey boxes for a double parasegment unit ) ( see Figure 6—figure supplement 1C ) . ( B ) When considering 0 , 1 or 2 missing cells , the most robust solution with three receptors is achieved with an additional receptor C spanning one parasegment out of two ( either odd or even ) . This provides a receptor difference at the PSBs ( grey boxes in first line ) and systematically increases receptor differences when one cell is missing at a given location ( grey boxes in second line ) . When 2 cells are missing , the number of receptor differences increases at a subset of locations , notably in the case where cell identities 3 and 4 are missing ( grey boxes in third line ) . The robustness score for this solution is 20 ( arrow in Figure 6—figure supplement 1C ) . Code for receptor permutations is in Source code 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 02710 . 7554/eLife . 12094 . 028Figure 6—source data 1 . Source data for Figure 6—figure supplement 1 , including statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 02810 . 7554/eLife . 12094 . 029Figure 6—figure supplement 1 . Combinatorial receptor patterns . ( A , B ) Receptor differences when cells are missing . Permutations of receptors and receptor differences when S3 ( A ) or both S3 and S4 cells ( B ) are missing , for the scenario in Figure 6B . In locations where S3 cells are missing alone ( A ) , receptor differences increase to 2 . In locations where both S3 and S4 cells are missing ( B ) , receptor differences increase to 3 . ( C ) Scoring robustness with a three-receptor combination . We randomised the location of three receptors , each expressed in 4 cells per double parasegment , one million times . Each permutation was scored as the sum of i ) the number of receptor differences between neighbouring cells ( grey boxes in first line of example in Figure 6B ) , ii ) the increase in number of receptor differences when one cell is missing at a given location ( grey boxes in second line in Figure 6B ) and iii ) the increase in number of receptor differences when two cells are missing ( grey boxes in third line in Figure 6B ) . The total score for the solution in Figure 6B is 20 , shown by the red arrow on the frequency histogram . Note that receptor C can be expressed in either even or odd parasegments , making no difference to the robustness score . ( D ) We also randomized the locations of four receptors , each expressed in 4 stripes per double-parasegment . Considering again events where 0 , 1 or 2 cells are missing , the only permutation of 4 receptors that has a higher robustness score , 24 , than the solution shown in Figure 6B , 20 , is shown in ( D ) . Remarkably , the number of receptor differences increases progressively when 1 or 2 cells are missing at any given location ( grey boxes ) . Note that this is a very similar permutation to the solution in Figure 6B , but the expression pattern for previously named receptor A is split into two pair-rule domains , now named receptors A and B , which confer additional robustness when 2 cell identities are missing . ( E ) Frequency histogram of the robustness score of one million randomisations for the expression of 4 receptors . The best score , 24 ( red arrow ) , corresponds to the solution shown in ( D ) . Input data and statistics are in Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 029 We then considered what happens when one cell is missing in the sequence of four cell identities along AP in each parasegment . We know this has to be frequently the case since we find an average of 3 . 6 cells per parasegment at the start of GBE ( Figure 3D ) . We counted again the number of receptors at interfaces , when a cell is missing at a given position . For example , if cell identity 2 is missing , cell identities 1 and 3 become adjacent; since cell identity 1 is expressing receptor A and cell identity 3 , receptor B , we scored a difference of two receptors for this particular interface ( Figure 6A ) . Remarkably , we find that the number of receptor differences increases by one in many locations when a cell identity is missing ( Figure 6A ) . We predict that the number of receptor differences is likely to be proportional to the amount of Myosin II recruited . In other words , we propose that the receptor identity system is quantitative . If the amount of Myosin II enriched is indeed proportional to the number of receptor differences , then more rapid cell intercalation would be expected to occur where cells are missing in the AP sequence ( see 'superboundaries' and 'superintercalators' introduced earlier ) . Increased cell intercalation would fill the gaps in cell identity during GBE and maintain the cell order along AP . From our data , the cell identities that are most likely to be missing are 3 and 4 , since we find that there are not enough cells to make two columns of two cells at the end of GBE in stripe 3 ( Figure 3D , E and 4D ) . For example , according to the scenario in Figure 6A , if cell identity 3 is missing , the receptor numbers at adjacent interfaces 2/4 increases from 1 to 2 ( Figure 6—figure supplement 1A ) . This in turn should translate into an increase in Myosin II at those interfaces , which then would increase the rate of intercalation . This notion is supported by our data , since we find that the cell intercalation rate is higher in stripe 3 containing identities 3 and 4 , than in stripes 1 or 2 ( Figure 5—figure supplement 1H ) . From the above , we propose that the receptor system is robust to missing cells because gaps in the pattern will be 'repaired' by speeding up intercalation at cell-cell interfaces most different in their receptor composition . Building on this hypothesis , we looked for the most likely expression pattern for a third putative receptor that would both explain the enrichment at the PSBs but also confer enhanced robustness to missing cells . To do this , we explored all possible permutations of three receptors , where each is expressed in a putative pair-rule pattern ( four cell-stripes out of eight , in a given double parasegment unit ) ( Figure 6B ) . We scored each permutation by summing both the number of immediate neighbour receptor differences , and also the increase in receptor differences at each interface if one or two cell identities are missing in a given row of cells . The permutation that scored highest ( 20 , see Figure 6—figure supplement 1C ) expresses two receptors in the same exact pattern as our first scenario ( Figure 6A ) and a third receptor in every other parasegment ( Figure 6B ) . With this solution , the number of receptors at cell-cell interfaces increases systematically from one to two when a cell identity is missing anywhere in a double parasegment unit , therefore showing robustness . When two cells are missing , this number increases to three receptors at a subset of locations . Interestingly , one of these locations corresponds to the case where cell identities 3 and 4 are both missing ( Figure 6B; Figure 6—figure supplement 1B ) . Since our data suggest that identities 3 and/or 4 are those most likely to be absent ( 1 and 2 being more systematically specified ) , this solution confers adequate robustness , taking into account the observed polarity of the parasegment . Note that a solution with 4 receptors instead of 3 does show a better robustness throughout the double parasegment unit ( Figure 6—figure supplement 1D , E ) , but since cell identities 1 and 2 are less likely to be missing according to our data , we conclude that the three-receptor solution shown in Figure 6B is the most parsimonious . To test our cell identity model more formally , we implemented a vertex model with a starting configuration of 20 rows and 14 columns of regular hexagonal cells , organised into 4 parasegments , with each parasegment comprising 3 to 4 cell identities along AP ( Video 5 ) . In vertex models , the movement of junctional vertices is governed by the strength of cell-cell adhesion , the contractility of the actomyosin cortex and cell elasticity ( Farhadifar et al . , 2007; Fletcher et al . , 2014; Honda and Eguchi , 1980 ) . These contributions are encoded in a ‘free energy’ function , whose gradient determines the velocity of each vertex . In addition , cell neighbour exchanges ( T1 transitions ) occur whenever a cell-cell interface’s length falls below a threshold value . We use a free energy function based on ( Farhadifar et al . , 2007 ) ( Figure 7A ) , keeping the ‘cell elasticity’ and ‘cortical contractility’ terms the same throughout , but varying the ‘line tension energy’ term in successive simulations to model different features of interface contractility inferred from the real data ( Figure 7—figure supplement 1A ) . 10 . 7554/eLife . 12094 . 030Video 5 . Movie of simulation 4 shown in Figure 7G . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 03010 . 7554/eLife . 12094 . 031Figure 7 . A vertex model based on cell-cell interactions replicates the interface behaviours during axis extension . ( A ) Summary of vertex model of germband extension . Cells are considered as two-dimensional polygons representing cell-cell interfaces , with vertices forming where three polygons meet . An ‘energy function’ is calculated and used to update the position of every vertex i over time . This energy function encodes mechanical contributions associated with cell elasticity , cortical contractility and interfacial ‘line tension energy’ . We consider a number of simulations ( see main text ) , which differ in the hypothesised dependence of the line tension f on interface lengths . ( B ) The initial configuration for each simulation comprises regular hexagonal cells organised into parasegments , each comprising cells of 4 stripe identities ( S1-S4 ) . Note that the initial configuration starts with 14x20 cells ( Video 5 ) and smaller snapshots are shown in this Figure and Figure 7—figure supplement 1 . ( C ) In simulation 2 , the line tension energy f varies linearly with interface length for non-boundary interfaces , but we specify a nonlinear dependence for boundary interfaces to represent a positive feedback between interface shortening and Myosin II enrichment . In this simulation , cells undergo neighbour exchanges but become stuck locally in four-cell junctions and hence convergent extension cannot proceed ( D , D’ ) . In Simulation 3 , we apply our non-linear dependence of line tension to the total length of contiguous boundary interfaces for a given cell ( length Li , m ) ( E ) , rather than to individual boundary interfaces ( length li , j ) ( D , D’ ) . This allows vertices to slide independently on either side of a column of interfaces that makes a boundary ( E’ , E” ) and the simulated tissue now undergoes convergent extension ( F ) , but identities S3 and S4 clump together . ( G ) Simulation 4 resolves this issue by incorporating ‘supercontractility’ , where boundary interfaces between cells of non-adjacent identities ( ‘skipped boundary’ ) are more contractile . Code for vertex model is in Source code 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 03110 . 7554/eLife . 12094 . 032Figure 7—source data 1 . Source data for Figure 7—figure supplement 1 , including statistical analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 03210 . 7554/eLife . 12094 . 033Figure 7—figure supplement 1 . Further details of vertex model simulations for interface behaviours during axis extension . ( A ) The model simulations 1–4 that we consider ( see main text ) differ in their hypothesised dependence of the ‘line tension’ energy term f ( see Figure 7A ) on interface lengths . ( B ) In simulation 1 , the line tension energy associated with each cell-cell interface varies linearly with its length , but with a higher constant of proportionality for boundary interfaces than for non-boundary interfaces . In this simulation , cells fail to undergo neighbour exchange . ( C ) Analysis of the interface behaviours in simulation 4 as in real data in Figure 5I ( legends therein ) . Input data and statistics are in Figure 7—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 033 In simulation 1 , the line tension energy associated with each cell-cell interface varies linearly with its length , but the boundary interfaces are twice as contractile as the non-boundary interfaces ( Figure 7—figure supplement 1B ) . Cells fail to undergo neighbour exchange in this simulation . In simulation 2 , the linearity of line tension energy is replaced at boundary interfaces by a non-linear relationship , where the line tension energy decreases at an ever-faster rate as the interface shortens . This models a positive feedback between interface shortening and Myosin II enrichment , supported by our data ( Figure 5C ) . In this simulation 2 , cells do now undergo neighbour exchanges , but become stuck in a four-cell junction topology ( Figure 7C–D’ ) . In simulation 3 , we allow vertices to slide independently on either side of a column of interfaces that makes a boundary ( Figure 7E–F ) . We implement this ( for boundary interfaces only ) by applying our non-linear dependence of line tension to all interfaces present at a given boundary for a given cell ( combined length L ) ( Figure 7E–F ) , rather than to individual interfaces ( length l ) ( Figure 7C–D’ ) . The cells are now able to intercalate and the simulated tissue undergoes convergent extension , elongating in AP while shortening in DV ( Figure 7F ) . As a consequence , single columns of cell identities 1 and 2 become double columns of cells at the end of the simulation , as predicted in Figure 3E . However , because of their insufficient number , cells of identities 3 and 4 end up clumping together according to their identity , thereby disrupting the AP order of the starting pattern ( Figure 7F ) . To address this , we implemented a fourth simulation ( Figure 7G ) that incorporates our hypothesised ‘supercontractility’ , where interactions between cells of non-adjacent identities in the parasegmental sequence generate more contractile interfaces than cells of adjacent identities . For example , contractility would be higher at interfaces between identities 2 and 4 , than between 2 and 3 or 3 and 4 . Implementing this , simulation 4 solves the clumping problem and maintains the AP order of cell identities throughout axis extension ( Figure 7G ) , as postulated in Figure 3E . So simulation 4 recapitulates the intercalary cell behaviours that we hypothesise based on our data . Finally , we analysed interface behaviours as for the real data ( Figure 5I ) . We find that boundary and non-boundary interfaces in simulation 4 have behaviours qualitatively similar to real data ( Compare Figure 7—figure supplement 1C with Figure 5I ) , demonstrating that this simulation successfully models the cell interface behaviours of GBE . We have developed new computational methods to quantify and analyse patterns of Myosin II planar polarisation and cell behaviours in the extending Drosophila germband in both time and space . In previous studies , the analysis of Myosin II planar polarity has focused on bipolarity , often comparing the enrichment in Myosin II at the DV-oriented sides ( also called vertical sides ) of germband cells relative to their AP-oriented sides ( also called horizontal sides ) ( for example , see Simões et al . , 2014; Paré et al . , 2014; Kasza et al . , 2014 ) . Here , in addition to using a measure of bidirectional polarity , we have developed a measure of unidirectional polarity , to identify when one side of a cell is enriched relative to all other sides . By distinguishing between bi- and unidirectional polarities , we have been able to identify novel patterns that inform how Myosin II planar polarisation arises and drives cell and tissue behaviours . Furthermore , by taking a live-imaging approach , we have been able to observe how these polarities evolve with unprecedented temporal resolution . Our study provides further experimental evidence that differential cell identity generates the planar polarity of Myosin II in the germband and extends existing models . A long-standing hypothesis in Drosophila segmentation is that the cascade of genes from maternal determinants , gap genes and then pair-rule genes is able to establish differential 'identities' with single-cell precision along the AP axis ( Dubuis et al . , 2013; Tkačik et al . , 2015 ) . The discovery of a role for Toll-like receptors , under the control of pair-rule genes in GBE , has provided compelling molecular evidence for this model ( Paré et al . , 2014 ) . One question arising from this work is what happens to Myosin II planar bipolarity once polarized cell intercalation proceeds . Indeed , polarized cell intercalation will increase the cell number along AP , thus bringing cells with the same identity next to each other along this axis . If differential cell identity via heterotypic interactions drives Myosin II polarisation throughout GBE , then some cells should find themselves in homotypic interaction with either an anterior or posterior neighbour , which would not lead to Myosin II enrichment ( See Figure 3E ) . The unipolarity patterns that we find are consistent with this hypothesis , identifying alternating domains of enriched and not enriched cell-cell interfaces along AP , which emerge during the course of axis extension . These correspond to Myosin II-enriched boundaries between parasegmental domains ( PSBs ) and to at least two more locations within each parasegment from early in GBE . The AP position of these enrichments is consistent with these being the consequence of the doubling of cell numbers along AP via polarized cell intercalation . Therefore the differential identity model predicts a transition between bidirectional and unidirectional polarities over the course of GBE , which is validated by our data . Another prediction of the differential identity model is that Myosin II enrichment should be dependent upon the type of cell-cell interface ( homotypic versus heterotypic ) rather than interface orientation ( DV versus AP-oriented ) . We were able to test this by comparing the orientation of enriched boundary interfaces ( heterotypic in our model ) versus non-boundary interfaces ( homotypic ) . Early GBE ( 0–25 min ) is characterised by two features . First , as predicted by a cell-cell interaction model , boundary interfaces are significantly more enriched than non-boundary interfaces for most orientations . However , overlaid on this , DV-oriented interfaces are also more enriched in Myosin II , irrespective of their boundary/non-boundary identity . This relationship between interface orientation and Myosin II enrichment in early GBE is at odds with a model based solely on cell-cell interactions . It is unclear what the cause of this relationship might be . Some planar polarity and cell intercalary behaviours remain in mutants for all three Toll-like receptors identified ( Paré et al . , 2014 ) . A possibility is that the remaining polarity is due to a more distant polarising signal operating in early embryos , which would direct Myosin II to all DV-oriented interfaces . Later in GBE ( 25–50 min ) , our analysis shows that Myosin II enrichment becomes independent of interface orientation , indicating that distant polarising signals are not acting on the germband at this stage and that local cell-cell interactions dominate . The Toll receptor model proposed in Paré et al . ( 2014 ) relies on each parasegment being four cells wide . Our quantification shows that parasegments are in fact on average only 3 . 6 cells wide at the onset of GBE ( sampling parasegments 4 to 7 , see Materials and methods ) . The widths of the stripes containing cell identities 1 and 2 are consistent with single-cell wide columns increasing to two-cell wide columns and therefore behave as expected from the differential cell identity model . However , the distinction between the stripes containing cell identities 3 and 4 as predicted by Paré et al . ( 2014 ) was less clear . Instead we observe a third stripe , which is 1 . 5 cells wide in AP on average at the start of GBE , increasing to 3 after 60 min . We think it likely that the cell types 3 and 4 do exist as postulated by Paré et al . ( 2014 ) , since there are detectable peaks of Myosin II and neighbour exchanges in the middle of our third stripe ( Figure 5—figure supplement 1F , G ) . But because parasegments are less than 4 cells across at GBE onset , some rows would have only cell types 1 , 2 , 3 or 1 , 2 , 4 , while others have the full complement of cell types 1 , 2 , 3 , 4 ( see Figure 3E ) . After 60 min of GBE , stripes 3 and 4 would then give a mixture of arrangements , such as 3 , 3 , 4 and 3 , 4 , 4 . As a result , the expected enrichment of Myosin II at heterotypic interfaces between cells of identity 3 and 4 would not align well , explaining why we cannot resolve a stripe 3/4 boundary in our data . If our reasoning is correct , this implies that there is an inherent polarity within each parasegment , with the anterior half made of cell types 1 and 2 being robustly specified , while in the posterior half , specification of cell identities 3 and 4 is more variable . This polarity might be important for the tissue to cope with the variation of cell number across parasegments and to repair the AP patterns during cell intercalation . Indeed , at the start of axis extension , although parasegments are usually 3 or 4 cells across , they occasionally have rows that are fewer or more cells across ( Lawrence and Johnston , 1989; Busturia and Lawrence , 1994 ) . We conclude that the mechanism of active convergence of the germ-band must be robust to variable cell number within each parasegmental unit . Our modeling suggests a mechanism by which the embryo copes with this variable cell number during axis extension . We postulate that the cell-cell interaction mechanism that triggers Myosin II enrichment at interfaces along AP is quantitative . It has been proposed that the stripy expression of Toll-2 , 6 and 8 receptors generate heterotypic interactions that result in Myosin II enrichment ( Paré et al . , 2014 ) . We further propose that these receptors , in addition to at least another receptor at the PSB , produce Myosin II enrichment which is proportional to the strength of the heterotypy . In other words , the more adjacent cells differ in the number of receptors they express , the more Myosin II will accumulate at their shared interfaces . We find that three receptors expressed in a pair-rule pattern is sufficient in theory to explain the planar polarization of Myosin II at every interface along AP in the germband , including the PSB interfaces which were not accounted for by the Toll-2 , 6 , 8 combinatorial code ( Paré et al . , 2014 ) . Two of the receptor patterns we identify correspond to the patterns of Toll-2 and Toll-6/8 ( Toll-6 has a pattern similar to Toll-8 ) and the third provides heterotypy at the PSB . The remarkable finding with this minimal combination of receptors is that heterotypy increases when one cell is missing in any position along AP . Moreover , heterotypy increases further when two cells are missing at half of the positions along AP . This is true in particular when identities 3 and 4 are both missing , which are the identities we think are most likely to be absent , based on our data . So when cells are missing , heterotypy would increase , triggering more Myosin II enrichment . This would increase the intercalation rate at the most mismatched interfaces and lead to pattern repair . In support of this , we do find an increased rate of cell intercalation in the posterior part of the parasegment ( Figure 5—figure supplement 1H ) , where we predict more mismatches because of too few cells of identities 3 and 4 . We tested these hypotheses in a vertex model and recapitulated qualitatively the tissue-scale behaviours in the data . We had to implement specific interface behaviours in the model to have successful convergence-extension of the in silico tissue . These are based on plausible behaviours in vivo . In particular , one limitation of vertex models is that apposed cortices are modeled as a single interface . The changes between Simulation 2 and 3 attempt to go round this limitation: what we tried to model is a situation where cells behave independently on either side of a boundary . For example , junctions could slide independently of each other on either side of the boundary . This is possible in vivo because a boundary is made of two cell cortices , and each cell cortex at the boundary interface could elongate or shorten independently . This could conceivably happen if the two cell cortices on either side of a boundary have different contractile forces . In addition to junctional sliding , cell-cell sliding could occur along the boundary , for example if adhesion is decreased there . Further work is required to determine if these processes are happening during GBE . Another point of note , we have implemented ratios of 1:2:8 for the line tension energies between non-boundary , boundary and 'supercontractile' boundary interfaces in Simulation 4 . The 1:2 ratio is quantitatively consistent with observed ratios of tension between PSB boundary and non-boundary interfaces obtained by laser ablation ( Figure 2I ) . We do not know what to expect as a ratio between boundary and supercontractile boundary interfaces , but 8 seems high . A discrepancy between the ratios of tension needed for a successful simulation of boundary behaviour and the ratios estimated in vivo by laser ablation has been noted by Landsberg et al . ( 2009 ) and so the relationship between line tension energies in vertex models and tension measured by laser ablation might not be simple/linear . For this paper , the key point is that the model qualitatively supports the idea that some boundary interfaces are more contractile than others . In combination with the two other receptor patterns which would correspond to those of Toll-2 ( receptor A in our model ) and Toll-6/8 ( receptor B ) , our parsimonious three-receptor combination is in theory sufficient to explain all of the Myosin II polarity patterns we identify in our study . By identifying PSB interfaces at late stages by their strong myosin enrichment and backtracking to earlier in development , we have further demonstrated that the PSB dominates over the two intra-parasegmental boundaries in terms of myosin enrichment . The predominance of the PSB is detectable from very close to the start of GBE . At the onset of GBE , PSBs are already demarcated genetically by the expression of the pair-rule genes such as eve and ftz and the gradually increasing expression of segment polarity genes wg and en ( Jaynes and Fujioka , 2004 ) . However , this is the first time that a cellular ( rather than genetic ) characteristic has been identified for PSBs this early . After the end of germband extension , later in development when epidermal cells are actively dividing , the movement of dividing cells across PSBs is prevented because the boundary interfaces enrich in Myosin II relative to non-boundary interfaces ( Monier et al . , 2010 ) , as for other compartmental boundaries in Drosophila ( Umetsu et al . , 2014; Aliee et al . , 2012; Landsberg et al . , 2009; Major and Irvine , 2006 ) and for tissue boundaries in zebrafish ( Calzolari et al . , 2014 ) and Xenopus ( Fagotto , 2014; Fagotto et al . , 2013 ) . In all these cases , the enrichment in Myosin II has been proposed to increase interfacial tension and promote tissue segregation . A possibility is that the PSBs fulfill a similar role during GBE , to prevent mixing between adjacent parasegments that cell intercalation might cause otherwise . Our interface orientation analyses and our laser ablation experiments demonstrate that there is indeed an increase in interfacial tension at PSBs early in GBE . We propose that elevated line tension at PSBs and also , to a lesser extent , at the two intra-parasegmental boundaries that we have identified , contribute to maintain the AP sequence of cell identities while cell rearrangements are occurring . It is unclear why the PSB boundaries are enriching Myosin II more than the other two intraparasegmental boundaries we have identified . This could be explained if the heterotypy between cell identity 1 on the posterior side of the PSB is strongest in combination with cell identities 3 or 4 on the anterior side of the PSB . We predict that a not yet identified receptor , with a pattern of expression corresponding to receptor C in our most parsimonious model ( Figure 6B ) , directs myosin II recruitment at the PSB interfaces . It could be that this putative receptor triggers a stronger response at the PSBs compared to the Toll-like 2 , 6 , 8 receptors at the other boundaries . Alternatively , more than one receptor might be contributing heterotypy at the PSBs . Our data suggest that we can rule out an early role for Wingless signaling in contributing to a PSB-specific response . Indeed , while Wingless is required to maintain Myosin II enrichment at the PSB later in development ( Monier et al . , 2010; 2011 ) , it is not required for the enrichment during germ-band extension ( Figure 2M ) , which is corroborated by the fact that PSBs straighten in wingless mutants as in wildtype ( Figure 2L ) . Thus it is likely that the pathway directing strong enrichment of Myosin II specifically at PSBs is under pair-rule control . Finally , our analysis shows that cell interface behaviour associated with active intercalation predominantly occurs at the boundary interfaces that we identify . Thus in Drosophila GBE , intraparasegmental boundaries and PSBs enriched in actomyosin appear to drive GBE . Supracellular Myosin II cables are already known to drive tissue elongation through the formation of multicellular rosettes ( Blankenship et al . , 2006 ) . These have not been linked to specific positions along the AP axis , but it is likely that rosettes form exclusively at PSBs or intraparasegmental boundaries , where our analysis suggests that Myosin II is enriched in continuous cable-like structures . In conclusion , we think that we have identified segmentally repeated boundaries , which enrich in Myosin II and simultaneously drive cell intercalation while keeping cells ordered along the AP axis . Our findings contribute to the growing evidence that cell fate heterogeneities are translated into differential interface contractility to govern morphogenesis ( Paré et al . , 2014; Bielmeier et al . , 2016; Bosveld et al . , 2016 ) . Extending the work of Paré et al . ( 2014 ) , we propose an updated differential cell identity model that is robust to missing cells , postulating a third receptor expressed in every other parasegment as the most parsimonious solution . In Xenopus , the antero-posterior patterning of the mesoderm also drives convergent extension ( Ninomiya et al . , 2004 ) , and thus similar ordering mechanisms might operate in vertebrate systems . As a whole , this system is reminiscent of the 'self' versus 'non-self' recognition mechanisms thought to play a role during neuronal wiring in the nervous system ( Zipursky and Grueber , 2013; He et al . , 2014 ) , and might represent a more ancient and primitive 'non-self' avoidance system , co-opted here by AP patterning to control cell behaviours . The logic and rules that we have uncovered for Drosophila axis extension provides a paradigm for more complex structures such as the brain ( Hassan and Hiesinger , 2015 ) . We used the null mutants sqhAX3 ( Jordan and Karess , 1997 ) and wgCX4 ( Baker , 1987 ) and the transgenes en-lacZ ( on II ) ( Busturia and Morata , 1988 ) , sqh-GFP42 ( on II ) ( Royou et al . , 2004 ) , GAP43-mCherry ( on III ) ( Martin et al . , 2010 ) and eve-EGFP ( on III ) ( Venken et al . , 2009 ) to construct the stocks sqhAX3; sqh-GFP42; GAP43-mCherry/TM6B , yw;;eve-EGFP , GAP43-mCherry/TM6B and w; wgCX4 , en-lacZ/CTG . yw67embryos were used as WT . The CTG balancer chromosome was CyO , twi-GAL4 , UAS-GFP ( Halfon et al . , 2002 ) . We followed standard methods for fixing and staining Drosophila embryos , using the primary antibodies goat anti-GFP ( ab6662 , Abcam , 1:200 ) , rabbit anti-Engrailed ( d300 , Santa Cruz Biotechnology; 1:50 ) , rabbit anti-β-gal ( ECK0341 , MP Biomedicals; 1:2500 ) , rat anti-DE-CAD ( DCAD2 , DSHB; 1:50 ) , guinea pig anti-Sqh-1P ( Zhang and Ward , 2011; 1:100 ) . We used the following secondary antibodies: goat anti-rabbit-Alexa-594 , goat anti-rat-Alexa-594 and goat anti-guinea pig-Alexa-488 ( Life Technologies , 1:500 ) . To improve immunostaining against Sqh-1P , embryos were post-fixed in 4% formaldehyde for 15 min before secondary antibody staining . Embryos were mounted individually on slides in VECTASHIELD ( Vector Labs ) under a coverslip suspended by a one-layer thick magic tape ( Scotch ) bridge on either side . This flattened the embryos sufficiently so that all cells were roughly in the same z-plane . Prior to placing the coverslip , embryos were rolled so that their ventral surfaces were facing upwards towards the coverslip . Embryos were imaged on a Nikon Eclipse TE2000 inverted microscope incorporating a C1 Plus confocal system ( Nikon ) . Images were captured using Nikon EZ-C1 software . Optical z-stacks were acquired with a depth of 0 . 25 µm between successive optical z-slices and with a total optical z-stack depth sufficient to capture both the top of the embryo and any more basal markers of the parasegment boundary ( PSB ) . All embryos were imaged using a violet corrected 60x oil objective lens ( NA of 1 . 4 ) . Laser illumination at 488 nm wavelength was used for Alexa-488 fluorophores and 543 nm for Alexa-594 . Neutral density ( ND ) 4 filters were applied to all lasers . Recursive averaging of 4 was used . The gain and offset were optimized for each embryo . For each stage and each genotype quantified ( yw67 or w; wgCX4 , en-lacZ ) , 9–10 embryos ( 1–4 boundaries per embryo ) were analysed . Embryos were immunostained for Sqh-1P ( as a marker of Myo II ) , DE-CAD ( as a marker of cell membranes ) and a PSB marker ( En or βgal , depending on the embryo genotype ) . Quantification was performed on PSB interfaces and +1 interfaces , which could all be identified relative to the position of PSB marker staining . Connected interfaces , in which Myo II was to be quantified , were traced using the FIJI plugin Simple Neurite Tracer ( Longair et al . , 2011 ) based on DE-CAD staining . Where possible , a line of interconnected interfaces was traced between the ventral midline and the amnioserosa . If a region of dividing cells was encountered along one of these lines of interconnected interfaces , the tracing was stopped and restarted the other side of the dividing cells . The traced lines were then increased in width by one pixel each side , giving a total line width of three pixels . Quantification was performed in the Sqh-1P channel . Fluorescence values lower than the modal pixel intensity were subtracted as background fluorescence . Average fluorescence intensity was calculated for each 3-pixel wide line trace using ImageJ . PSB interface fluorescence intensity was then normalised to +1 interface fluorescence intensity on a per PSB basis . Statistics were performed in Prism ( GraphPad ) . Embryos were dechorionated in commercial bleach before being rinsed thoroughly in water . An oxygen permeable membrane was pulled tightly over a custom-made metal imaging insert . Nine stage 5 embryos were mounted , ventral-side towards the objective , on the membrane in halocarbon oil ( Voltalef PCTFE , Atofina , France ) in a 5 mm spaced 3x3 array . A coverslip was placed over the embryos , supported by a bridge of a single coverslip on each side . Embryos were imaged under a 40x oil objective lens ( NA of 1 . 3 ) on a Nikon Eclipse E1000 microscope with a Yokogawa CSU10 spinning disk head and a Hamamatsu EM-CCD camera . Embryos were illuminated using a Spectral Applied Research LMM2 laser module ( 491 nm and 561 nm excitation ) . Images were captured using Volocity Acquisition Software ( PerkinElmer ) . Embryos were positioned under the objective lens so that the field of view was slightly posterior to the point at which embryos were widest in their DV axis . Optical z-stacks of a thickness of 28 µm were captured , with 14 µm above the top of the embryo and 14 µm into the embryos at the beginning of acquisition ( to allow for movement of the embryo in the z-axis ) . Consecutive optical z-slices were separated by 1 µm . Embryos were imaged every 30 s from late stage 5 for 100 min . Movies were recorded at 20 . 5 ± 1°C , measured with a high-resolution thermometer ( Checktemp1 ) . To check that embryos survived the imaging process to the end of embryogenesis . sqhAX3; sqh-GFP; GAP43-mCherry and eve-EGFP , GAP43-mCherry embryos were allowed to develop on the imaging insert to hatching in a humidified box . wgCX4; eve-EGFP , GAP43-mCherry embryos were treated similarly , but because wingless mutants are embryonic lethal , the cuticle of embryos was prepared using standard methods to check their phenotype . Occasional movies acquired for embryos that did not hatch or did not make a cuticle at the end of embryogenesis were discarded . The confocal z-stacks were converted into stacks of curved quasi-two-dimensional representations , the outermost of which followed the surface of the embryo with deeper layers shrinking progressively in 0 . 5 µm steps towards the centre of the embryo . The section giving the clearest view of cell apices was selected for tracking . Bespoke tracking software identifies cells and links them in an iterative process using an adaptive watershedding algorithm ( Blanchard et al . , 2009; Butler et al . , 2009 ) . For each cell at each time point , coordinates of cell centroids , perimeter shapes , cell-cell interfaces , and links forwards and backwards in time for both cells and interfaces ( even through neighbour exchange ) are stored . No statistical method was used to predetermine embryo number . We previously tracked cells in 5 embryos per treatment in Butler et al . ( 2009 ) , which was sufficient to show treatment differences . WT morphogenesis is remarkably reproducible ( see Figure 1—figure supplement 1A , B ) so we considered 6 sqhAX3; sqhGFP; GAP43-mCherry embryos would be sufficient to show robust patterns . For eve-EGFP , GAP43-mCherry and wgCX4; eve-EGFP , GAP43-mCherry embryos , we performed manual correction of segmented cell outlines at all time points . This improved the tracking in the embryos , hence we required only 3 embryos per treatment ( summarized in Table 1 ) . 10 . 7554/eLife . 12094 . 034Table 1 . Summary of embryos analysed per genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 034Embryo Genotype# Movies analysedMode of TrackingsqhAX3; sqh-GFP; GAP43-mCherry6Automatedeve-EGFP , GAP43-mCherry3Automated , manual correctionwgCX4; eve-EGFP , GAP43-mCherry3Automated , manual correction Movie x and y pixel coordinate axes were rotated and transformed into embryonic AP and DV coordinates in µm . The origin of embryonic coordinates was set at the start of GBE as the intersection between the anterior of the field of view and the ventral mid-line , with positive AP aligned towards the embryonic posterior . The origin of this coordinate system moved with the location of the intersection point , for example if there was any lateral movement of the embryo in AP or if the embryo rolled in DV . Using the relative movements of cell centroids , local tissue 2D strain ( deformation ) rates were calculated for small spatio-temporal domains ( see Figure 8 below and Blanchard et al . , 2009; Butler et al . , 2009; Lye et al . , 2015 ) , composed of a focal cell and one corona of neighbouring cells over a 2 min interval ( contained within five movie frames ) . A separate direct measure of 2D cell shape change was calculated by first approximating each cell with its best-fit ellipse , then finding the strain rate tensor that best mapped a cell’s elliptical shape to its shape in the subsequent time point . The difference between the local tissue strain rates ( calculated above ) and the average cell shape strain rate of cells in the same spatio-temporal domain was attributed to cell intercalation . All strain rates were then projected onto the embryonic axes , AP and DV . 10 . 7554/eLife . 12094 . 035Figure 8 . Methodology for quantifying tissue strain rates . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 035 WT movies were synchronized in time ( Figure 1—figure supplement 1A ) with zero min defined as the last frame in which there was no extension at the posterior edge of the image . This was further refined to the frame before which tissue extension in the AP axis exceeded a proportional rate of 0 . 01 / min . We confirmed that ectodermal cell division and the timing of the cessation of cell intercalation in different embryo movies were clustered in time as a result ( Figure 1—figure supplement 1B ) . For all analyses , we included only neurectoderm cells , having classified and excluded all head , mesoderm , mesectoderm , non-neural ectoderm and amnioserosa cells . We subtracted the modal pixel intensity as background from raw images in the Myosin II channel at each time point . We set the width of cell-cell interfaces at 3 pixels , a compromise between being wide enough to encompass all interface Myosin II fluorescence , and narrow enough to minimise the inclusion of medial Myosin II . The fluorescence intensity for each cell-cell interface in each movie frame ( every 30 s ) was calculated as the average intensity of interface pixels . We measured apical cell membrane Myosin II polarity using the Myosin II fluorescence intensities of each cell-cell interface calculated above . We first expressed interface fluorescence intensity around each cell perimeter as a function of angle , from the embryonic posterior ( zero ) anti-clockwise ( Figure 1—figure supplement 2A ) . Treating this intensity signal from 0 – 360 degrees as a periodic repeating signal , we calculated its Fourier decomposition , extracting the amplitude of the period 2 component as the strength of Myosin II bipolarity ( equivalent to planar cell polarity ) , with its phase representing the orientation of cell bipolarity ( Figure 1—figure supplement 2B–D , red lines ) . We also extracted the period 1 component as a Myosin unipolarity measure ( Figure 1—figure supplement 2B–D , cyan lines ) . The orientations of both uni- and bipolarity distributions for our dataset were strongly and consistently biased towards the AP-axis ( Figure 1C’ , D’ ) . However , there was some pollution of the unipolarity signal in the bipolarity signal , with the latter enhanced because of the castellated ( discontinuous ) nature of the average interface intensity signal ( Figure 1—figure supplement 2A–D , black lines ) . We therefore explored further methods to calculate independent uni- and bipolarity quantities . Based on the consistent AP bias to both kinds of polarity , we measured the polarity in the AP axis only . We found that fitting two Gaussians independently , centred on the anterior and posterior sides of each cell works well , and is able to separate combinations of uni- and bipolarity ( Figure 1—figure supplement 2B’-D’ ) . We fitted the amplitudes and variances of anterior and posterior Gaussians through minimising the discrepancy between the combined Gaussian signal and the Myosin II signal . The bipolarity signal was taken as two peaks of the amplitude of the smaller of the two Gaussians . Subtracting the bipolarity signal from the combined Gaussians , the remainder is the unipolarity signal . Because overall Myosin II intensity differed between embryos , we normalised the strength of both polarities by dividing the allocated Gaussian amplitude area by the cell’s mean perimeter Myosin II signal , so that they would be consistent across embryos . Finally , we made an adjustment to account for an imaging artefact that results in a domed intensity of Myosin II in all images , with corners less bright than the image centres . The differences in brightness are not an issue per se , since we express polarity amplitudes as a proportion of mean cell perimeter fluorescence , but an artefactual gradient in intensity across a cell will introduce a unipolarity signal . We therefore fitted a smooth to the Myosin II intensity across each image separately ( with a kernel size of 1/20th of the image width ) , calculated the local gradient of this smooth for each cell , and rebalanced the local gradient effect while keeping the mean cell perimeter fluorescence the same . Using the above methods we produced uni- and bipolarity measures projected along the AP axis for each cell at each time point , that are independent of each other and normalised to control for variation in Myosin II fluorescence . Heat maps show time on the y-axis plotted against some measure of AP location on the x-axis , with heat colour representing a third variable . Variation of the third variable was averaged over the DV axis . Heat maps show the mean values of the third variable for each grid square of the plot , the size of which is shown in ‘N’ heat maps . For example , for Figure 3B , C , the ‘N’ heat map is Figure 3—figure supplement 1A , with 80 time bins and 60 AP coordinate bins . White guidelines drawn over contoured heat maps are the average cell trajectories , showing the gross extension of the tissue in the AP axis over time . Tissue domains were defined in individual tracked movies using two different techniques , depending on the embryo’s genotype . For movies of sqhAX3; sqh-GFP; GAP43-mCherry embryos , strong PSB enrichments of Sqh-GFP were identified at the end of movies . Groups of cells in between strong Sqh-GFP enrichments were manually selected ( each group corresponding to a single parasegment ) in a single time point at the end of each movie . Because cells were tracked over time , these classifications of parasegmental group identity could be automatically backtracked to define the same groups of cells at all earlier time points . For movies of eve-EGFP , GAP43-mCherry embryos , the anterior boundaries between parasegments were identified by clear anterior margins of Eve-EGFP positive nuclei . Groups of cells in between successive clear anterior margins of Eve-EGFP positive nuclei were manually selected ( each group corresponding to two parasegments ) in a single time point at the end of each movie . Groups of cells were again classified at earlier time points by backtracking through movies . We only used parasegments that were seen throughout each movie , excluding , for example , posterior parasegments that flowed out of the field of view as a result of axis extension . Data used in subsequent analyses were from parasegments 4–7 ( summarised in Table 2 ) , as calculated from the distance along the AP axis of the embryo , and from the timing and location of cell division nests in abdominal parasegments ( Foe , 1989 ) . 10 . 7554/eLife . 12094 . 036Table 2 . Summary of parasegments analysed for each sqhAX3; sqh-GFP; GAP43-mCherry movie . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 036Movie IdentifierPS4PS5PS6PS7SG_1✗✔✔✔SG_2✗✔✔✔SG_3✔✔✔✗SG_4✔✔✔✗SG_5✔✔✔✗SG_6✔✔✔✔ Interface orientations , relative to the embryonic axes , were calculated for PSB , -1 and +1 interfaces at all time points in movies from eve-EGFP , GAP43-mCherry or wgCX4; eve-EGFP , GAP43-mCherry and for PSB , S1/2B , S2/3B , +1 and -1 interfaces at all times points in movies from sqhAX3; sqh-GFP , GAP43-mCherry embryos . All distributions of interface orientations ( from 0 , parallel to the AP embryonic axis , to 180° ) were reflected around 90° , producing distributions from 0° , AP-aligned , to 90° , DV-aligned . As a measure of co-alignment , the proportion of interfaces oriented between 60 and 90° relative to the AP axis was plotted over time , from -20 to 60 min . Cumulative frequencies were calculated for each reflected distribution of interface orientations at 40 min ( corresponding to when Myosin II levels were significantly different ) . Two-sample Kolmogorov-Smirnov tests on the cumulative frequency distributions of interface orientation were used to compare treatments ( Prism , GraphPad ) . We repeated the analysis , treating the distribution of interface angles from 0° to 180° as a circular distribution , and calculating the parameter of concentration ( κ ) of the von Mises probability density function . Both the plots of κ versus time and the comparison of distributions at 40 min gave very similar results to the above methods ( data not shown ) . Junctional laser ablation experiments were carried out in sqhAX3; sqh-GFP; GAP43-mCherry embryos . PSBs were located by eye by finding i ) connected junctions that had the strongest Sqh-GFP intensity and ii ) had mirror image Sqh-GFP enrichments the other side of the embryonic midline . We confirmed that the PSB interfaces we selected were significantly more strongly enriched in Myo II than + 1 interfaces by quantifying the mean Sqh-GFP intensity in a line section drawn over the junction at the time point prior to ablation ( Figure 2—figure supplement 1H ) . We also confirmed that +1 interface orientations relative to the embryonic midline were more broadly distributed than PSB interface orientations ( Figure 2—figure supplement 1I ) . Further quantification showed that interface types did not differ in mean length ( Figure 2—figure supplement 1G ) . Ablations were carried out as described in Lye et al . ( 2015 ) . 2 to 4 ablations were performed in each embryo and a total of 15 embryos were used . 19 ablations were carried out for both PSB and +1 junctions . A single ablation was performed in each parasegment and all ablations were confined to the Vnd and Ind domains along the DV axis of the embryo . The region of interest selected for ablation was placed over the middle of the chosen junction . 5 images were collected prior to ablation ( any longer and the junction would move away from the region of interest due to axis extension movements ) and up to at least 30s after ablation . Line sections were then manually drawn over ablated junctions and the Dynamic Reslice tool in ImageJ was used to produce kymographs . The distances between the two vertices at either end of a junction were measured from 5 time points before ablation until 30s after ablation . Linear regression was performed on the first 5 time points after ablation . The slope of the regressed line was used as a measure of the vertex recoil velocity . The 'equal slopes' test function in Prism ( GraphPad ) was used to test for significant differences between slopes and thus difference in recoil velocities . Pooling the normalised polarity data from 6 sqhAX3; sqh-GFP; GAP43-mCherry embryos , each grid square of the heat maps has a distribution , with number of data points per grid square ( Figure 3—figure supplement 1A ) , mean ( Figure 3B , C ) and a confidence interval that we can calculate . We tested whether the mean value in each grid square of contoured heat maps ( averaged over the 6 embryos ) is significantly different from zero using t-tests . Figure 3—figure supplement 1B , C show squares in white that are not different from zero at the 95% two-tailed confidence level . In Figure 3—figure supplement 1C , where unipolarity is significantly different from zero , the direction rather than the strength of unipolarity ( see Figure 3C ) is shown . For each parasegment , we calculated the average width of the parasegment in AP ( psw ) and the average width of each cell in that parasegment , also in AP ( cw ) . To give the average number of cells per parasegment width , we divided psw by cw ( Figure 3D ) . For the number of cells per stripe width , the numerator was the width of the stripe ( Figure 4D ) . We manually defined within-parasegment stripe boundaries , looking for Myosin II accumulation along DV interfaces linked in cable-like structures parallel to PSBs and classifying cells as being in stripe S1 , S2 or S3 within each parasegment . We checked our stripe classifications by plotting the locations of stripe boundary ( scoring 0 ) and non-boundary ( scoring 1 ) interfaces against within-parasegment coordinate ( Figure 4C ) . The peaks in location of boundary interfaces align very well with the mean location of within-parasegment boundaries S1/2B and S2/3B ( black arrows ) taken from Figure 3C and Figure 3—figure supplement 1C . We registered neighbour exchange events when a cell-cell interface swapped ownership from one pair of neighbours to an orthogonal pair of neighbours . Most neighbour exchange events were straightforward , with the reducing interface swapping cleanly into a new growing interface . However , some swapped repeatedly before resolving , or did not resolve , or reverted to the original cell connectivity . We therefore set a threshold time window of 5 min over which we ignored repeated neighbour swaps . We expected that polarised Myosin II ( uni- or bidirectional ) at cell junctions would lead to cell shapes that differed from relaxed geometries of a kind that would be expected if , for example , Myosin II was either absent or uniformly distributed . We therefore constructed a measure to quantify the degree of difference from a putative relaxed geometry , both at the scale of cell perimeters and of individual cell-cell interfaces . We first defined relaxed geometries . We chose a Voronoi tessellation , based on cell centroid locations , as a simple first approximation to relaxed geometries . A Voronoi tessellation identifies cell-cell interfaces as the set of points equidistant from two neighbouring cell centroids . Vertices are located where these interfaces from local pairs of cell centroids intersect ( Figure 5D ) . The tessellation will stretch with tissue ( cell centroid ) stretch , so we expected our comparisons to be robust to cell elongation per se . Using existing cell centroids ( centres of mass ) , we used a Voronoi tessellation to obtain expected vertex locations , cell-cell interface lengths and cell perimeters . We quantified the difference between actual cell perimeters and those based on Voronoi predictions . By definition , as cell shapes become geometrically stressed , cell perimeters will on average become longer than those predicted by the Voronoi tessellation . We subtracted tessellated interface lengths from observed interface lengths to get a measure of geometric stress . A value near zero indicated a relaxed geometry , with increasing deviation from zero indicating increasingly stressed geometries ( Figure 5—figure supplement 1D ) . We expected relaxed cell geometries twenty min before the start of GBE , when cells have finished cellularisation but before gastrulation and before polarised Myosin II expression . Indeed , perimeter stress was low and stable until the start of GBE , when it rose sharply then remained high throughout GBE ( Figure 5E , black line ) . Mesoderm invagination no doubt introduces some stress from -5 to 5 min ( Lye et al . , 2015 ) , but the fact that the geometric stress index remained high thereafter shows that this stress is likely to be actively maintained in the germ-band . Boundary interfaces behaved differently from non-boundary interfaces , with the latter longer than expected ( Figure 5E ) . We investigated further , aligning interfaces in time to zero at the point of neighbour exchange . The deviation of boundary interface length increased prior to exchange events , coinciding with a similarly increase in interfacial myosin prior to exchange as interfaces shortened ( Figure 5F ) . Upon neighbour exchange , these interfaces became non-boundary interfaces and showed an elongated signature . Overall , these data suggest that the active contraction of boundary interfaces is driving convergence in DV , and that as soon as they become non-boundary interfaces they take on a passive signature . The scoring for each permutation is explained in Figure 6 and Figure 6—figure supplement 1 . The code for generating the permutations is given in Source code 1 . We used mathematical modelling to investigate the mechanical implications of actomyosin planar polarisation during Drosophila axis extension . Vertex models are a particularly successful description of epithelial mechanics that model the polygonal tessellation that cells’ adherens junctions form in two dimensions ( Farhadifar et al . , 2007; Fletcher et al . , 2014; Honda and Eguchi , 1980 ) . In such models , the movement of junctional vertices and the rearrangement of cells are governed by the strength of cell-cell adhesion , the contractility of the actomyosin cortex and cell elasticity . We describe the epithelial sheet by a set of connected vertices in two dimensions . Assuming that the motion of these vertices is overdamped , the position ri ( t ) of vertex i evolves according to the first-order equation of motion ( 1 ) ηdri ( t ) dt=Fi ( t ) , where Fi ( t ) denotes the total force acting on vertex i at time t and η denotes the common drag coefficient . We specify the forces acting on vertices through a ‘free energy’ function U , for which ( 2 ) Fi= − ∂U∂ri . Our choice of U is based on that proposed in Farhadifar et al . ( 2007 ) and is given by ( see Figure 7A ) : ( 3 ) U= ∑αK2 ( Aα− A0 ) 2+∑αΓ2Pα2+∑ ( i , j ) f ( lij ) . The first term in this free energy function describes an area elasticity with common elastic coefficient K , for which Aα is the area of cell α and A0 is a common ‘target’ area , and the sum runs over all cells at time t . The second term describes the contractility of the cell perimeter Pα by a common coefficient Γ , with the sum again running over all cells at time t . The third term represents ‘line tensions’ at cell-cell interfaces , where lij denotes the length of the edge shared by vertices i and j and the sum runs over the set of cell-cell interfaces at time t . Line tensions can be reduced by increasing cell-cell adhesion or reducing actin-myosin contractility . The precise functional form of this line tension energy term varies across our simulations . In addition to these equations of motion for cell vertices , we need to ensure that cells are always non-intersecting and to allow cells to form and break bonds . This is achieved through an elementary operation called edge rearrangement ( a T1 transition or swap ) , which corresponds biologically to cell intercalation . Mathematically , such arrangements are necessary in the vertex model due to the finite forces acting on a cell's vertices arbitrarily far from equilibrium . We implement a T1 swap whenever two vertices i and j are located less than a minimum threshold distance dmin apart ( taken to be much smaller than a typical cell diameter ) . In this case , the two vertices are moved orthogonally to a distance pdmin apart and the local topology of the cell sheet is modified such that they no longer share an edge . The configuration of the cell sheet is updated using the following algorithm . Prior to numerical solution , we non-dimensionalize the model , following previous implementations ( Farhadifar et al . , 2007; Kursawe et al . , 2015 ) by rescaling all lengths with A0 and all times with η/KL2; thus , all presented model results are non-dimensional . Starting from an initial configuration ri ( 0 ) , we update the state of the system until time T over discrete time steps Δt . At each time step we: implement any required T1 swaps; compute the forces Fi on each vertex from the free energy U; solve the equation of motion for each vertex over the time step numerically , using an explicit Euler method; and finally update the positions of all vertices simultaneously . We implement this model in Chaste ( Fletcher et al . , 2013 ) , an open source C++ library that allows for the simulation of vertex models . The code is given in the file Source code 1 . We consider several alternative model simulations of axis extension , which differ only in the hypothesised dependence of the line-tension energy described above on the length and type of cell-cell interfaces . In each simulation , we model the movement , shape change and neighbour exchange of a small tissue that is initially comprised of 20 rows and 14 columns of hexagonal cells . Prior to the start of each simulation , we simulate the evolution of the tissue to mechanical equilibrium under the assumption that the line-tension energy varies linearly with interface length , f ( lij ) = Λijlij , with the same ( constant ) coefficient for every interface , Λij=Λint . This avoids compounding the later dynamics by artefacts associated with starting the tissue from a non-equilibrium cell size . The value of Λint and all other parameters used in the simulations described below are provided in Table 3 . We then simulate the tissue until time T under a different hypothesised dependence of the line-tension energy , as described below . In each simulation , we introduce four distinct stripes of cell identities within each parasegment ( Figure 7B ) . Note that both stripes 3 and 4 are initially discontinuous , reflecting our in vivo finding that these two stripes have a combined initial AP width of 1 . 5 cells ( S3 in Figure 4D ) . 10 . 7554/eLife . 12094 . 037Table 3 . List of parameters and their values used in simulations . DOI: http://dx . doi . org/10 . 7554/eLife . 12094 . 037ParameterDescriptionValueSimulationsηDrag coefficient1 . 0AllTSimulation end time500AllΔtTime step0 . 001AlldminT1 swap threshold0 . 01AllpT1 swap distance multiplier1 . 5AllKElastic coefficient1AllA0Cell target area1AllΓContractility coefficient0 . 04AllΛintLine-tension coefficient for non-boundary ( or tissue-boundary ) interfaces0 . 051-3ΛbdyLine-tension coefficient for ( stripe- ) boundary interfaces2Λint2-3ΛsupLine-tension coefficient for super-contractile ( stripe- ) boundary interfaces8Λint4 Here , we follow ( Farhadifar et al . , 2007 ) in setting the line tension energy to vary linearly with the length of a cell-cell interface: ( 4 ) f ( lij ) = Λijlij . In our model , the line-tension coefficient Λij takes one of two values , depending on the type of interface . If the interface is shared by two cells of the same stripe identity ( a non-boundary interface ) , or it is contained in a single cell ( a tissue-boundary interface ) , then we set Λij=Λint . If the interface is shared by two cells of different stripes identities ( a stripe-boundary interface ) , then we set Λij=Λbdy , where Λbdy>Λint and thus boundary interfaces are more contractile than non-boundary interfaces ( Figure 7—figure supplement 1A ) : ( 5 ) f ( lij ) ={Λintlij , for a non-boundary or tissue-boundary interface , Λbdylij , for a stripe-boundary interface . In this simulation , we find that cells are unable to execute neighbour exchanges and hence axis extension is not achieved ( Figure 7—figure supplement 1B ) . In our next model , we consider a nonlinear dependence of the line-tension energy f ( lij ) on the cell-cell interface length . Here , we wish to study the effect of including a feedback or runaway component , in which shorter stripe-boundary interfaces become enriched in Myosin II and thus more contractile , on the axis extension process . To this end , we choose the functional form ( see Figure 7—figure supplement 1A ) : ( 6 ) f ( lij ) ={Λintlij , for a non-boundary or tissue-boundary interface , Λbdyloglij , for a stripe-boundary interface . In this simulation , we find that while some cells exchange neighbours , most 4-way junctions do not resolve ( Figure 7C ) . To address the resolution of 4-way junctions encountered in Simulation 2 , we next consider a more complex model of line tension , where now the value of the coefficient Λij is computed as follows for boundary interfaces . For each of the two cells sharing the boundary interface , we sum the lengths of the ( contiguous ) boundary interfaces shared by the cell , including the boundary interface of interest . Having computed this number for each of the two cells , we then compute the smaller of these two numbers , which we denote by Li , m , the indices reflecting the variable number of contiguous boundary interfaces ( Figure 7E ) . The line-tension coefficient then takes the form: ( 7 ) f ( lij ) ={Λintlij , for a non-boundary or tissue-boundary interface , ΛbdylogLi , m , for a stripe-boundary interface . This resolves the 4-way junction issue encountered in simulation 2 because cells are able to shear along boundaries ( Figure 7F ) . However , the initially discontinuous stripes 3 and 4 remain in clumps , unable to repair their stripe continuity . Our next model builds on Simulation 3 to include super-contractility , in which the line-tension coefficient Λij now takes a different value for boundary interfaces between stripes whose identities differ by 2 ( for example , an interface between a cell belonging to stripe 2 and a cell belonging to stripe 4 ) to those between stripes whose identities differ by 1 . We denote these values by Λsup and Λbdy , respectively , where Λsup> Λbdy to reflect our hypothesis that mismatched or skipped identity boundary interfaces are more contractile than other boundary interfaces: ( 8 ) f ( lij ) ={Λint lij , for a non-boundary or tissue-boundary interface , ΛbdylogLi , m , for a stripe-boundary interface , ΛsuplogLi , m , for a skipped stripe-boundary interface . In this simulation stripes 3 and 4 repair their continuity and the patterns of cell and interface behaviours qualitatively mimic in vivo data ( Figure 7G and compare Figure 7—figure supplement 1C with Figure 5I ) .	Early in development , a growing embryo elongates to form its main body ( head–tail ) axis . This elongation is driven by a process called cell intercalation – when cells insert between each other . The mechanism that controls this coordinated cell movement is well understood on a small scale . However , it is not known how hundreds of cells rapidly intercalate across a whole tissue without deforming a tissue or inappropriately mixing . During fruit fly development , an embryo divides into repeated segments of tissue while elongating . While this happens , cells redistribute an essential structure called the actomyosin cytoskeleton so that it is found more commonly along certain sides of the cell . This structure , which can be thought of as the cell’s “muscle” , is a contractile web made of proteins called actin and myosin . It is closely associated with the cell’s membrane and causes cells to contract and push past each other . The enrichment of the actomyosin cytoskeleton on certain sides of a cell is determined by signaling systems , which are controlled by the segmentation genes in the fruit fly and by the so-called planar cell polarity pathway in vertebrates . Tetley , Blanchard et al . have now investigated cell intercalation across a whole tissue by filming live fruit fly embryos in which both actomyosin and cell membranes were made visible with fluorescent markers . Computational tools were then used to quantify how much actomyosin is enriched in the sides of thousands of cells in the embryo at particular points in time while the embryos elongated . This revealed reproducible patterns of actomyosin enrichment . As embryos elongated , the actomyosin cytoskeleton redistributed itself inside the cells: whereas at the start two opposite sides of each cell were enriched in actomyosin ( a bipolar distribution ) , at later times the enrichment occurred on just one side ( a unipolar distribution ) . Incorporating these patterns into a model of tissue-wide cell intercalation showed that cells along the head–tail axis acquire a specific identity depending on their position . Interactions between the cells then allow the cells to compare their identities with each other and modify their pattern of actomyosin enrichment accordingly . Where the identities of neighbouring cells are different , the cells enrich actomyosin along their shared sides , creating boundaries between stripes of cells that share the same identity . These findings show that actomyosin-rich boundaries drive the elongation of the head–tail axis while limiting cell intermixing . Future work will investigate how the patterns of actomyosin enrichment are altered in fly mutants in which the identities of the cells along the head–tail axis are disrupted .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "computational", "and", "systems", "biology" ]	2016	Unipolar distributions of junctional Myosin II identify cell stripe boundaries that drive cell intercalation throughout Drosophila axis extension
RNA editing by adenosine deamination alters genetic information from the genomic blueprint . When it recodes mRNAs , it gives organisms the option to express diverse , functionally distinct , protein isoforms . All eumetazoans , from cnidarians to humans , express RNA editing enzymes . However , transcriptome-wide screens have only uncovered about 25 transcripts harboring conserved recoding RNA editing sites in mammals and several hundred recoding sites in Drosophila . These studies on few established models have led to the general assumption that recoding by RNA editing is extremely rare . Here we employ a novel bioinformatic approach with extensive validation to show that the squid Doryteuthis pealeii recodes proteins by RNA editing to an unprecedented extent . We identify 57 , 108 recoding sites in the nervous system , affecting the majority of the proteins studied . Recoding is tissue-dependent , and enriched in genes with neuronal and cytoskeletal functions , suggesting it plays an important role in brain physiology . The central dogma of biology maintains that genetic information passes faithfully from DNA to RNA to proteins; however , with the help of tools such as alternative splicing , organisms use RNA as a canvas to modify and enrich this flow of information . RNA editing by deamination of adenosine to inosine ( A-to-I ) is another process used to alter genetic information ( Nishikura , 2010 ) . Unlike alternative splicing , which shuffles relatively large regions of RNA , editing targets single bases in order to fine-tune protein function . Because inosine is interpreted as guanosine by the cellular machinery , this process can recode codons ( Basilio et al . , 1962 ) . A-to-I RNA editing is catalyzed by the ADAR ( adenosine deaminase that acts on RNA ) family of enzymes . All eumetazoans , from cnidarians to mammals , express ADARs but the extent to which they use them to recode has been explored in few representatives ( Nishikura , 2010 ) . Recent advances in DNA sequencing and the supporting computational analyses have permitted transcriptome-wide screens for RNA editing events . So far , such studies have been limited to organisms with a sequenced genome ( Ramaswami et al . , 2012 , 2013 ) . In general , these screens have looked for variation in RNA at positions that are invariant in the genome . In humans , inosine is abundant in RNA ( Paul and Bass , 1998; Bazak et al . , 2014 ) , but almost all of it lies within transcribed repetitive elements in untranslated regions or introns ( Nishikura , 2010 ) . A compilation of recoding sites in human transcriptomes revealed 1183 events ( Xu and Zhang , 2014 ) , but most were observed in only a single sample . Individual searches ( Danecek et al . , 2012; Ramaswami et al . , 2013 ) uncovered only 115 ( non-repetitive ) recoding events , and 53 in mice; 34 recoding sites are conserved across mammals ( Pinto et al . , 2014 ) . In Drosophila , an order of magnitude more recoding sites have been identified , residing in about 3% of all messages ( St Laurent et al . , 2013 ) . Although individual editing sites are clearly essential ( Brusa et al . , 1995 ) , these data suggest that RNA editing is not a broadly used mechanism for proteome diversification . However , anecdotal data suggest this assumption might not apply across the animal kingdom . For example , using traditional cloning methods , scores of recoding sites have been uncovered in a small number of squid and octopus transcripts encoding potassium channels , ADARs , and ion pumps ( Garrett and Rosenthal , 2012a ) . As for most organisms , there are no genomes available for cephalopods . Here we apply a novel approach for editing site detection in the absence of a sequenced genome . We use it to comprehensively identify editing sites in the squid giant axon system and other areas of the nervous system . Surprisingly , almost 60% of all mRNAs studied harbor recoding events , and most at multiple sites . These data show orders of magnitude more recoding in the squid proteome than in any other species studied to date . In squid , editing is so pervasive that the central dogma should be modified to include this process . Our results open the possibility that extensive recoding is common in many organisms , rivaling alternative splicing as a means of creating functional diversity . To detect RNA editing sites in the squid nervous system , we generated millions of RNA and genomic DNA reads from an individual squid . Our method differed from previous approaches by using a de novo transcriptome as the point of reference instead of a genome ( Figure 1A ) . The transcriptome was assembled from RNA-seq reads , and each nucleotide within it represents the consensus of many reads . If the majority of RNA reads were edited ( ‘strong’ editing sites ) , the transcriptome would differ from the genomic DNA and read ‘G’ where gDNA reads would show ‘A’ ( the sequencing process identifies inosines as guanosines ) . We detected such sites by aligning DNA-seq reads to the transcriptome ( Figure 1B ) . At positions where editing occurred in the minority of RNA-seq reads ( ‘weak’ editing sites ) , however , the transcriptome and the genomic DNA would be identical . These sites were detected by identifying variability in RNA-seq , but not DNA-seq , reads ( Figure 1B ) . This general approach is applicable to all organisms that lack a sequenced genome . 10 . 7554/eLife . 05198 . 003Figure 1 . A general approach to detect RNA editing sites in organisms that lack a sequenced genome . ( A ) Squid RNA-seq data is used to create a de novo transcriptome followed by the detection of conserved ORFs . ( B ) ‘Weak’ and ‘strong’ editing sites are detected by comparing RNA and DNA reads from the same animal to the ORFs from the transcriptome . ‘Weak’ editing sites were detected by observing the minority of the RNA reads to differ from the consensus transcriptome nucleotide . ‘Strong’ editing sites , where the consensus transcriptome includes the edited nucleotide , were detected by observing all DNA reads to differ from the transcriptome nucleotide . DOI: http://dx . doi . org/10 . 7554/eLife . 05198 . 003 We sequenced cDNA from the giant axon system ( giant fiber lobe: GFL ) , the optic lobes ( OL ) and matching germline gDNA isolated from the same animal . cDNA was also sequenced from the vertical lobe ( VL ) , buccal ganglion ( BG ) and the Stellate Ganglion ( SG ) from another animal ( The SG and GFL are parts of the peripheral nervous system; all the rest are from the central nervous system ) . The GFL and OL RNA-seq reads were used to construct a transcriptome model ( Grabherr et al . , 2011 ) . To focus on editing sites inside bona fide coding regions , we retained only transcript-fragments with open reading frames ( ORFs ) homologous to known proteins ( UniProt Consortium , 2014 ) ( Figure 1A ) and used the editing detection procedures outlined in Figure 1B . Surprisingly , our pipeline identified 81 , 930 weak sites , and 5644 strong sites , due to A-to-G transitions ( Figure 2A ) . Only 12 , 403 weak sites and 219 strong sites were identified for the other 11 possible types of modifications . These numbers suggest false-positive rates of 15% and 4% respectively , mainly due to transcriptome assembly problems , SNPs , somatic mutations and systematic mis-alignments . Note that these false-positive rates are considerably lower than those for similar searches for editing within human coding sequences , where a genome reference was employed ( Ramaswami et al . , 2013 ) . 10 . 7554/eLife . 05198 . 004Figure 2 . High number of RNA editing sites in squid translates into an extraordinary number of recoding events . ( A ) The number of nucleotide modifications observed in the squid nervous system for each possible substitution type ( in blue , 87% of all detected modifications were A-to-G ) . A similar analysis of human and Rhesus macaque sequencing data ( green and brown , respectively ) shows low levels , and no enrichment , of A-to-I editing in coding regions , as reported previously . In the inset , the distribution of nucleotide modifications observed in squid mitochondria-encoded genes , used here as a negative control . The ADAR enzymes have no reported activity in the mitochondria and , accordingly , no A-to-G overrepresentation is observed . Also see Figure 2—figure supplement 1 . ( B ) Scope of recoding due to RNA editing in squid , both in the total number of recoding events and the total number of genes affected , is orders of magnitude higher than human , mouse , and fly ( numbers for other organisms are based on recent publications using RNA-seq datasets comparable to the one used here [Danecek et al . , 2012; Ramaswami et al . , 2013; St Laurent et al . , 2013] ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05198 . 00410 . 7554/eLife . 05198 . 005Figure 2—figure supplement 1 . A-to-G modifications appear in clusters of consecutive identical mismatches and show distinctive 5′ and 3′ neighbor preferences . ( A ) The number of modifications observed for each possible modification type , considering only modifications that appear in clusters . About half of the A-to-G modifications appear in clusters of at least three consecutive same-type mismatches , in accordance with the expected properties of A-to-I editing sites , found in other organisms ( Morse et al . , 2002; Levanon et al . , 2004 ) . ( B ) The number of reads with 3 , 4 , and 5 consecutive identical mismatches for each possible modification type . Most of these reads contained A-to-G modifications . ( C ) The sequence surrounding of the observed A-to-G modifications , compared with that surrounding random adenosines in our model transcriptome . The sequence surrounding the ‘weak’ sites and the ‘strong’ sites ( Figure 1B ) are similar to each other and to what is known for other species ( Kleinberger and Eisenberg , 2010 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05198 . 00510 . 7554/eLife . 05198 . 006Figure 2—figure supplement 2 . Hierarchical clustering reveals tissue selectivity in the modification levels of the A-to-G sites , but not in the non A-to-G sites . ( A ) For the A-to-G sites , different tissues show varying levels of editing , globally , as well as site-specific tissue-dependent regulation . For example , higher modification levels ( red ) are observed in the GFL tissue whereas low levels ( green ) are observed in the VL tissue . Yet , some sites ( top rows ) are edited more strongly in VL . ( B ) Hierarchical clustering of the non A-to-G modification levels in the five different neuronal tissues . Consistently with the modifications being due to genomic polymorphisms , data cluster according to the animal it was taken from: modification levels in the GFL and OL tissues , which were taken from one individual animal , form one cluster , as do the VL , BG and SG tissues , taken from another individual animal . Modification levels are , by and large , uniform across tissues coming from the same individual animal . Note that in both panels only sites with significantly variable modification levels are presented ( binomial analysis was performed with Bonferroni-corrected p-value of 0 . 05 as a cutoff ) , each row represents one modification site . Abbreviations: Giant fiber lobe ( GFL ) , Optic lobe ( OL ) , Vertical lobe ( VL ) , Buccal ganglia ( BG ) , and Stellate ganglion ( SG ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05198 . 00610 . 7554/eLife . 05198 . 007Figure 2—figure supplement 3 . Validation of editing using Sanger sequencing . ( A ) An example of editing sites verified using Sanger sequencing in the squid protein piccolo . Arrowheads mark the locations of the editing sites . ( B ) Editing levels measured by Sanger sequencing for the 40 sites correlate with RNA sequencing results . DOI: http://dx . doi . org/10 . 7554/eLife . 05198 . 00710 . 7554/eLife . 05198 . 008Figure 2—figure supplement 4 . Quality controls for the A-to-G modifications and the non A-to-G modifications . ( A ) The distribution of the quality scores for all the sites used ( all the positions inside all the analyzed reads ) , A-to-G modifications , and non A-to-G modifications . No difference is observed between these three groups . Note that sites with Q < 30 were excluded . ( B ) The number of mismatches detected as a function of the position inside the read . Non A-to-G mismatches tend to occur at reads' ends , suggesting alignment artifacts ( which tend to affect reads' ends ) are responsible to some of these mismatches ( Ramaswami et al . , 2012 ) . A-to-G mismatches do not show such tendency . ( C ) The distribution of modification levels for A-to-G and non A-to-G sites , for the GFL and OL tissues . The increased number of non A-to-G sites with ∼50% modification level hint at some genomic polymorphisms ( SNPs ) , that were not represented in our DNA reads due to the limited coverage , are included among the non A-to-G mismatches . Consistently , 51% of the sites with non A-to-G modification levels between 40–60% recur in both tissues ( coming from the same individual animal ) , compared to only 22% of the A-to-G modifications in the same range . Similarly , 50% of non A-to-G modification levels higher than 90% recur in both tissues ( coming from the same individual animal ) , compared to only 21% for A-to-G modifications in the same range . These two ranges are the only ones in which such difference is observed . Abbreviations: Giant fiber lobe ( GFL ) , Optic lobe ( OL ) , quality score ( Q ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05198 . 00810 . 7554/eLife . 05198 . 009Figure 2—figure supplement 5 . Number of modification sites detected as a function of the amount of DNA and RNA reads . ( RNA-seq from the GFL and OL tissues only , originating from the animal whose DNA was sequenced ) . ( A ) The number of A-to-G sites detected increases with the number of RNA reads , with no sign for saturation . Thus , we expect the number of editing sites to be much larger than the one reported here . ( B ) The number of A-to-G sites detected in each gene correlates with the gene's RNA coverage , demonstrating that with much larger RNA-seq data , the number of detected editing sites could be as high as ∼200 , 000 ( expected number of 17 sites per protein , on average , for each of the ∼12K ORFs in our model transcriptome ) . ( C ) The number of modification sites detected as a function of the number of DNA reads . Detection of modification sites is based on mismatches between cDNA reads and the consensus . However , one of the main sources for such mismatches , which masks the signal due to RNA editing , is heterozygosity of the genome . The more DNA reads available , the better one can identify and exclude genomically heterozygous sites ( SNPs ) and improve signal-to-noise ratio . ( i ) ‘weak’ sites detection ( ii ) ‘strong’ sites detection - here exclusion of SNPs is part of the detection scheme itself ( see Methods ) and thus the number of detected sites ( and not only the signal-to-noise ratio ) increases with gDNA coverage ( iii ) ‘Weak’ and ‘strong’ sites , combined . Abbreviations: Giant fiber lobe ( GFL ) , Optic lobe ( OL ) , Standard Error ( SE ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05198 . 009 Although the number of A-to-G discrepancies was unexpectedly large , subsequent analyses support the idea that they are caused by RNA editing rather than other sources of error . First , we applied our pipeline to similarly sized data sets from a human blood sample and from the rhesus macaque brain , each containing matching RNA and DNA sequence reads . As expected for mammals , the quantity of AG mismatches in coding regions were similar to those from non-AG mismatches , and both were quantitatively indistinguishable from the noise determined from the squid data ( Figure 2A and Supplementary file 1A , B ) . These controls demonstrate that the enormous number of AG mismatches in the squid data is not an artefact of our analysis pipeline . Other features point to the biological origin of our AG mismatches . Similar to A-to-I editing sites in other organisms ( Morse et al . , 2002; Levanon et al . , 2004; Kleinberger and Eisenberg , 2010 ) , those identified here tend to cluster and show distinctive 5′ and 3′ neighbor preferences ( Figure 2—figure supplement 1 ) . In addition , hierarchical clustering of results from five tissues reveals that A-to-G modifications , but not other types , exhibit clear tissue-specificity , suggesting they do not result from genomic polymorphisms and mapping artifacts ( Figure 2—figure supplement 2 ) . No A-to-G overrepresentation is observed in mitochondria-encoded genes ( Figure 2A ) , in agreement with the absence of ADARs , and by extension A-to-I editing , in the mitochondria . Finally , direct Sanger sequencing from a second individual confirmed editing at 40/40 A-to-G sites , and deep-sequencing validated 120/143 A-to-G sites but none of the 12 non A-to-G sites tested ( Figure 2—figure supplement 3 , Supplementary file 1C–G ) . Taken together , the overrepresentation of A-to-G modifications over all other types , the motifs surrounding the A-to-G sites , the tissue-specific modification levels , and the validation experiments , provide evidence that the majority of the A-to-G modifications are true editing events , while most non A-to-G modifications are likely technical artifacts or genomic variations ( Zaranek et al . , 2010; Ramaswami et al . , 2012 ) ( Figure 2—figure supplement 4 ) . Unlike with humans , the large number of A-to-I editing events translates into a large number of recoding events: Overall , 57 , 108 recoding events were detected in 6991/12 , 039 ORFs . These numbers are orders of magnitude higher than any other species studied ( Danecek et al . , 2012; Ramaswami et al . , 2013; St Laurent et al . , 2013; Pinto et al . , 2014 ) ( Figure 2B ) . Moreover , a large fraction of the proteins are recoded at multiple sites ( Figure 3A ) : about 1/3 harbor ≥3 sites and 10% harbor ≥10 sites . Even when focusing only on recoding sites with editing levels >10% , about 10% of the squid proteins harbor ≥5 sites ( Figure 3A ) . On the extreme end of the spectrum , the ORFs encoding α Spectrin and Piccolo have 247 and 182 recoding sites , respectively ( Figure 3B and Figure 3—figure supplement 1 ) . It should be noted that only annotated ORFs were examined in our pipeline , and the number of editing sites did not saturate with respect to the number of sequence reads ( Figure 2—figure supplement 5 ) . Moreover , incompleteness of the de novo transcriptome , as well as incorrect assembly of paralogs and splice variants , may cause our pipeline to miss many additional sites ( Supplementary file 1B ) . Therefore there are probably many more recoding sites in the squid transcriptome . 10 . 7554/eLife . 05198 . 010Figure 3 . Editing often recodes multiple amino acids in the same protein . ( A ) The fraction of the squid genes that harbor multiple recoding events . About a third of the squid proteins harbor three or more recoding sites and more than 10% harbor 10 or more recoding sites . ( B ) Homology-modelling of the α Spectrin protein in which 10% of the amino acids ( 247/2412 ) are recoded by editing . Amino acids 1602 to 1918 of the squid α Spectrin protein are included in the 3-D model . Recoding sites are highlighted in green . Recoding sites with tissue-dependent levels are highlighted in red and the corresponding editing levels are indicated in the table . Also see Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05198 . 01010 . 7554/eLife . 05198 . 011Figure 3—figure supplement 1 . Homology-modelling of the squid Piccolo protein in which 9% of the amino acids ( 182/2098 ) are recoded by editing . Amino acids 1830 to 1966 of the squid Piccolo protein are included in the 3-D model . Recoding sites with tissue-dependent and -independent editing levels are highlighted in red and green , respectively . Aspartate residues involved in Ca2+ binding are highlighted in yellow . Abbreviations: Giant fiber lobe ( GFL ) , Optic lobe ( OL ) , Vertical lobe ( VL ) , Buccal ganglia ( BG ) , and Stellate ganglion ( SG ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05198 . 011 Consistent with other organisms ( Stapleton et al . , 2006 ) , recoding events are enriched in genes with neuronal and cytoskeletal functions ( Figure 4—figure supplement 1A and Supplementary file 1H ) . To gain insight on the effected pathways , squid ORFs were mapped to all human KEGG pathways ( Kanehisa and Goto , 2000 ) . Editing has a global effect on most pathways ( Supplementary file 1I ) , and those related to the nervous system are even more affected . For example , of the 27 proteins in the ‘Synaptic vesicle cycle’ pathway , 22 are edited and 14 heavily so ( Figure 4A ) . Similarly , of the 39 proteins in the ‘Axon guidance’ pathway , 33 are edited and 19 heavily so . Other notable pathways are ‘Regulation of actin cytoskeleton’ and ‘Circadian rhythm’ ( Figure 4—figure supplement 1B ) . By contrast , proteins in the pathways ‘Ribosome’ and ‘RNA polymerase’ are edited less than average ( Supplementary file 1I ) , demonstrating that some pathways may be protected from editing . Consistently , editing levels observed in non-nervous system tissues are considerably lower ( Alon et al . , 2015 ) . 10 . 7554/eLife . 05198 . 012Figure 4 . Recoding due to RNA editing affects complete molecular pathways and is likely to be more advantageous in sites with high editing levels . ( A ) All the squid proteins present in the KEGG ‘Synaptic vesicle cycle’ pathway are edited , and most are heavily edited . We define ‘heavily edited proteins’ as those for which the cumulative recoding level , that is the editing level summed over all recoding sites , exceeds unity . These are marked red , other edited proteins yellow , and proteins not identifiable in the squid transcriptome are shown in green . Also see Figure 4—figure supplement 1 . ( B ) The fraction of nonsynonymous codon changes as a function of the editing levels , using data from the GFL and OL tissues combined . The higher the editing level , the higher the fraction of nonsynonymous codon changes . The fraction expected by chance is shown in red . A similar relationship is also true for every tissue separately ( Figure 4—figure supplement 2A ) . Asterisks mark p-value <0 . 001 estimated using 1000 bootstrap runs . DOI: http://dx . doi . org/10 . 7554/eLife . 05198 . 01210 . 7554/eLife . 05198 . 013Figure 4—figure supplement 1 . Recoding events are enriched in genes with neuronal and cytoskeletal functions and globally affect molecular pathways . ( A ) The top-scoring Gene Ontology ( GO ) terms ( rated by false discovery rate , FDR ) , enriched in a list of squid ORFs ranked by the cumulative recoding level , that is the editing level summed over all recoding sites ( Eden et al . , 2009 ) . ( B ) All of the identifiable squid proteins present in the KEGG pathway ‘Circadian rhythm’ are edited , and many are heavily edited . We define ‘heavily edited proteins’ as those for which the cumulative recoding level exceeds unity ( i . e . , each copy of the protein is expected to have at least one modified amino acid , on average ) . These are marked red , other edited proteins in magenta , and proteins not identifiable in the squid transcriptome in green . This figure was created using the KEGG ( Kanehisa and Goto , 2000 ) pathway database website ( http://www . genome . jp/kegg/pathway . html ) . Editing levels were calculated using data from the Giant fiber lobe ( GFL ) and Optic lobe ( OL ) tissues combined . DOI: http://dx . doi . org/10 . 7554/eLife . 05198 . 01310 . 7554/eLife . 05198 . 014Figure 4—figure supplement 2 . The fraction of nonsynonymous codon changes as a function of editing levels and the amino acid modifications due to editing . ( A ) For high editing levels , the fraction of nonsynonymous codon changes is significantly different from the fraction expected by chance for all the neuronal tissues examined . The fraction expected by chance ( see ‘Materials and methods’ ) is shown in red . ( B ) , The observed distribution of recoding types for sites with editing levels lower than 10% , between 10–50% , and higher than 50% . Highly edited sites favor the creation of glycine and arginine , mainly at the expense of lysine , in a statistically significant manner . Expected values and error bars were calculated by using the mean values and standard deviation of 100 bootstrap runs , respectively , generated by randomly modifying adenosine in a way that preserves the editing sequence preference and the number of events . ( C ) and ( D ) : Amino acid targeted by the editing and created by the editing in both squid and Drosophila ( St Laurent et al . , 2013 ) . The most frequent target for removal is lysine , and glycine and arginine are frequently created due to editing . Editing levels were calculated using data from the GFL and OL tissues combined . Abbreviations: Giant fiber lobe ( GFL ) , Optic lobe ( OL ) , Vertical lobe ( VL ) , Buccal ganglia ( BG ) , and Stellate ganglion ( SG ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05198 . 01410 . 7554/eLife . 05198 . 015Figure 4—figure supplement 3 . Editing tends to avoid potentially deleterious recoding events . Each squid ORF was aligned against the conserved domains in the Conserved Domain Database ( CDD ) ( Marchler-Bauer et al . , 2013 ) , and the score for substituting each amino acid by all other types of amino acids was calculated ( Boratyn et al . , 2012 ) . The substitution score is a positive or negative integer , reflecting amino acid substitution which , compared to chance , occur frequently or infrequently in the alignment of the conserved domains , respectively . ( A ) The average editing levels , using data from the GFL and OL tissues combined , as a function of the amino acid substitution score . The average editing levels for negative substitution scores is significantly lower compared to what is expected by chance . ( B ) The distribution of the recoding sites as a function of the amino acid substitution score . Recoding sites tend to avoid large negative substitution scores compared with random changes . ( C ) The average substitution score as a function of the editing levels , using data from the GFL and OL tissues combined . The higher the editing levels , the higher the average substitution score , indicating that highly edited sites are more likely to recode to amino acids that occur frequently in other species . Expected values and error bars were calculated by using the mean values and standard deviation of 10 , 000 bootstrap runs , respectively . For A and C , the editing levels in all the sites with the same recoding type were randomly shuffled . For B , adenosines were randomly modified in a way that preserves the sequence preference and the total number of editing events . One asterisk mark p-value <0 . 05 , two mark p-value<1e-4 . Abbreviations: Giant fiber lobe ( GFL ) , Optic lobe ( OL ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05198 . 015 Recently , it was suggested that RNA editing is generally not advantageous in humans ( Xu and Zhang , 2014 ) , as nonsynonymous events are less frequent than expected by chance ( Xu and Zhang , 2014 ) . Strikingly , for sites with high editing levels in squid , the opposite is true ( Figure 4B and Figure 4—figure supplement 2A ) . Recoding events favor creation of glycine and arginine , mainly at the expense of lysine ( Figure 4—figure supplement 2B–D ) . Moreover , highly edited sites within conserved domains tend to recode to amino acids that occur frequently in other species at the same position ( Figure 4—figure supplement 3 ) , suggesting selection towards functional substitutions and against deleterious ones . The squid giant axon has been one of the most important models for neurophysiology . Studies using this preparation serve as a foundation for our current understanding of excitability , ion homeostasis , and axonal transport . Accordingly , we examined the extent to which RNA editing might affect these processes ( Supplementary file 1J ) . In total , 87 GFL ORFs that encode voltage and neurotransmitter gated ion channels , ion transporters , synaptic release machinery and molecular motors were identified in our transcriptome . In agreement with the overall editing frequency in squid , 70% harbored editing sites . Unexpectedly , however , 54% were heavily edited , many more than the 24% expected . Thus even in a background of hyper RNA editing , squid , like other organisms , preferentially edit nervous system targets . These data suggest that editing in squid has fundamentally different underpinnings and consequences . Have squid ADARs evolved novel structures that account for the high-level editing ? A past study showed that a squid ADAR2 ortholog can be expressed as two isoforms due to alternative splicing ( Palavicini et al . , 2009 ) : one , having two double-stranded RNA ( dsRNA ) binding motifs , resembles vertebrate ADAR2s . A second , however , contains an ‘extra’ dsRNA binding motif at its N-terminus . This non-canonical isoform edits RNA more efficiently , edits more sites , and binds dsRNA with a far higher affinity ( Palavicini et al . , 2009 , 2012 ) . Further Squid ADAR2 messages themselves contain many editing sites , leading to many subtly different isoforms . An ADAR1 isoform is also present in our transcriptome and is the focus of an ongoing study . An equally intriguing question is why squid edit to this extent ? The process clearly creates tremendous protein diversity , and this may in part explain the behavioral sophistication of these complex invertebrates . A recent study showed that editing can be used for temperature adaptation in octopus ( Garrett and Rosenthal , 2012b ) and this makes sense based on the codon changes that it catalyzes ( Garrett and Rosenthal , 2012a ) ( Figure 4—figure supplement 2C–D ) . In Drosophila , editing can respond to acute temperature changes ( Savva et al . , 2012 ) . The large number of sites in squid suggests that editing is well positioned to respond to environmental variation . Most model organisms studied so far are mammals which are homeotherms . Future studies of more diverse species are needed to reveal the extent to which cold-blooded organisms might utilize extensive editing to respond to temperature changes and other environmental variables . Specimens of the squid Doryteuthis pealeii were collected by trawl in the Vineyard Sound by the animal collection department of the Marine Biological Laboratory in Woods Hole , Massachusetts during the month of July . The giant fiber lobe ( GFL ) tissue of the Stellate ganglion , the optic lobe ( OL ) tissue and a portion of the sperm sack were manually dissected from a single adult male immersed in chilled , filtered seawater . The Buccal ganglia ( BG ) , Stellate ganglion with the giant fiber lobe removed ( SG ) and Vertical lobe ( VL ) tissues were dissected from a second adult male individual . Tissues were also dissected from non-neuronal regions: the branchial heart , the Gill , the ventral epithelial layer on the pen , the marginal epithelial layer on the pen , the iridophore layer of the skin , and the chromatophore layer of the skin . Each of these six tissues originated from a different animal . RNA from all tissues was extracted with the RNAqueous kit ( Life Technologies , Carlsbad , CA ) , and genomic DNA was extracted from the sperm sack using Genomic Tip Columns ( Qiagen , Venlo , Limburg , The Netherlands ) . The genomic DNA sequencing library was prepared using the TruSeq DNA Sample Prep kit , as described by the manufacturer ( Illumina , San Diego , CA ) , and sequenced using three lanes of Illumina HiSeq 2000 instrument . The RNA-Seq libraries for all the samples were prepared using the TruSeq Stranded mRNA Sample Prep Kit , as described by the manufacturer ( Illumina ) , and were sequenced using Illumina HiSeq 2000 instrument . The GFL and OL libraries were sequenced together on a single lane , the same for the VL and BG libraries ( with one unrelated library ) and the SG library was sequenced on a single lane ( with one unrelated library ) . Illumina sequencing was utilized to generate ∼87 million paired-end , 100 nt reads , using RNA from GFL tissue , the same number of reads using RNA from OL tissue , and ∼440 million paired-end , 100 nt reads , using germline DNA . The Trinity transcript assembly tool ( Grabherr et al . , 2011 ) was used with the default parameters ( except for ‘min_kmer_cov’ which was set to 2 instead of 1 ) to construct the genes sequences from the GFL and the OL sequencing data combined; giving 99 , 226 putative gene-fragments ( termed ‘components’ ) . Most of the generated components were short ( median of 379 bases ) . As Trinity attempts to identify the different transcripts ( isoforms ) of the components , 14 , 643 components were represented by more than one sequence ( putative transcript-fragment ) . For these components , the longest sequence was selected and further used . The average length of the representing sequences for all components was 589 bases , bringing the total size of the recovered squid transcriptome to about 58 million bases . DNA reads and RNA reads were separately aligned against the RNA components using Bowtie2 with local alignment configuration and default parameters ( Langmead and Salzberg , 2012 ) . Only uniquely aligned reads were used ( taking only reads with the maximal ‘mapping quality’ ) . As most of the components were short , we didn't demand that both mate pairs be aligned to the same component . Instead , each read was separately analyzed . Overall , 77% ( ∼268 million out of ∼350 million ) of the GFL and OL RNA reads combined were uniquely aligned to the components . However , as expected , only 3 . 5% of the DNA reads ( ∼30 million out of ∼880 million ) were uniquely aligned to the components . To focus on editing sites inside coding regions , and avoid repetitive elements that are prone to assembly and alignment errors , we retained only those components that were found to be significant similar ( Blastx E-value<1e-6 ) to the Swiss-Prot proteins dataset ( UniProt Consortium , 2014 ) . In these components , the alignment to the homolog Swiss-Prot protein covered most of the squid sequence ( 63% on average ) . The detected ORFs were extended in both directions until a stop codon or the end of a component was reached . Overall , 12 , 039 ORFs , with average length of 1370 bases , were detected . These ORFs represent 8276 Swiss-Prot proteins with distinct names ( two or more different ORFs may be two squid paralogs of the same Swiss-Prot protein , or fragments of the same squid gene aligned to different regions of the Swiss-Prot protein ) . The total size of the detected coding sequences in the squid was ∼16 million bases , 28% of the total transcriptome length recovered using Trinity ( the constructed squid coding sequences are given in a text file in a fasta format , Alon et al . ( 2015 ) , available via Dryad digital repository ) . The coverage of these ORFs was high; in the GFL and OL data combined , ∼199 million RNA reads were aligned to these Swiss-Prot ORFs , with an average RNA coverage of 1206× . The average DNA coverage was 106× as ∼21 million DNA reads , with average length of 83 bases were aligned to the coding sequences . The above alignment data and the ORFs information were used to find the locations of all DNA-RNA mismatches inside the coding regions . In the following , bases called with quality score Q < 30 were discarded . Note , however , that Trinity consensus sequence does take into account these bases , as well as reads that might have not been uniquely aligned to the transcriptome . It is often customary to filter out reads' ends when analyzing RNA-DNA mismatches . A main reason for that is the common mismatches at reads' ends due to alignment artifacts when a splicing junction occurs near the ends . In our case , as alignment is done to the transcriptome , we did not observe any increase in AG mismatch rate near reads' ends ( Figure 2—figure supplement 4 ) , and thus no such filter was used . Two procedures were used to detect editing sites: ( A ) ‘weak’ editing sites procedure: a binomial test was applied to find the significant modifications between RNA reads and the Swiss-Prot ORFs . The binomial statistics uses the number of successes ( the number of reads with a mismatch of a given type in a given position ) , the number of trials ( the total number of RNA reads aligned to the given position ) and the error probability . The probability of having a sequencing error ( the error probability ) was estimated using the sequence quality score . We counted only mismatches with >=30 quality score , and therefore the expected error probability was set to 0 . 1% . The binomial test was applied to every position inside the Swiss-Prot ORFs . The p-value for each location was corrected for multiple testing using a Benjamini-Hochberg false-discovery rate of 10% . Furthermore , in order to exclude RNA variability due to genomic polymorphisms , we filtered out all modification sites in which any of the DNA reads aligned to the site does not agree with the transcriptome . This procedure assumes the RNA consensus in the site is identical to the gDNA reads , and thus is not suited to detect sites in which the editing appear in most of the RNA reads and therefore also in the Trinity-generated ORFs ( ‘strong’ editing sites ) . ( B ) ‘strong’ editing sites procedure: the locations in which all DNA reads showed a different base than the ORF . The probability of such sites to be not a result of editing but rather a single nucleotide polymorphism ( SNP ) was estimated by ( 1/2 ) ( #DNA reads ) ( no allele-specific expression ) multiplied by the prior probability of a SNP which was taken to be 0 . 001 . Here too , the p-value for each location was corrected for multiple testing using a Benjamini-Hochberg false-discovery rate of 10% . Overall , 81 , 930 weak and 5649 strong A-to-G modification sites were detected in 7776 ORFs , and only 12 , 403 weak and 254 strong non A-to-G modifications sites ( the modification sites and their number of supporting reads in all the tissues studied are tabulated in Alon et al . ( 2015 ) available via Dryad digital repository ) . Interestingly , 2905 of the A-to-G modification sites reside in 268 out of the 475 ORFs with only non-metazoans homologs . The number of weak A-to-G sites detected ( but not strong ones ) is likely to increase with RNA coverage , as we demonstrated by sampling parts of the sequencing data and re-calculating the number of A-to-G sites ( Figure 2—figure supplement 5A ) . Another way to look at the dependence between the RNA coverage and the detected number of weak A-to-G sites is by recording the number of A-to-G sites detected in each ORF as a function of the ORF RNA coverage . The sorted RNA coverage was divided into ten equal bins and the mean number of A-to-G sites in each bin was calculated . As expected , higher RNA coverage is correlated with high number of A-to-G sites ( Figure 2—figure supplement 5B ) . As with the RNA reads , the dependence of the detection procedures on the DNA reads was examined by sampling parts of the DNA sequencing data . Increasing the DNA coverage increased the precision of the ‘weak’ site detection procedure ( as more SNPs are removed from the observed modifications ) and strongly increased the number of detected ‘strong’ sites ( Figure 2—figure supplement 5C ) . We did not apply additional read-number filters , as these seem to only marginally increase accuracy while reducing the number of detected sites . However , we did exclude five strong sites for which there were no uniquely-aligned , high-quality , supporting RNA reads . This brings the number of strong A-to-G and non A-to-G modifications to 5644 and 219 , respectively . The full list of weak and strong sites ( Alon et al . ( 2015 ) available via Dryad data repository ) provides the number of DNA and RNA reads per site . We examined clusters of mismatches of the same type ( several consecutive identical mismatches ) , revealing a high number of ORFs with A-to-G mismatch clusters ( Figure 2—figure supplement 1A ) . For example , examining only ORFs with three ( and above ) consecutive identical mismatches , 45 , 199 A-to-G sites in 4265 ORFs were detected , and only 470 non A-to-G sites . Interestingly , the clusters of A-to-G sites also appear at the level of the individual reads; for example , examining only reads with four and above consecutive identical mismatches , 85% of the reads contained A-to-G modifications ( Figure 2—figure supplement 1B ) . This data implies that single RNA molecules contain A-to-G clusters . We note that similar results were obtained using an algorithm which analyzes only reads with several consecutive mismatches , including reads that cannot be mapped by standard alignment tools ( Porath et al . , 2014 ) . Using this algorithm 23 , 737 A-to-G modification sites were found in the squid coding regions , compared to only 2933 non A-to-G modifications . Importantly , only 728 A-to-G modifications were found in the coding regions of human using this algorithm ( Porath et al . , 2014 ) , even though a much larger RNA-seq dataset was used ( ∼5 billion reads of Illumina Human BodyMap 2 . 0 ) , thus supporting the dramatic difference in editing between squid and other organisms . The sequence surrounding the A-to-G modifications sites is similar for ‘weak’ sites and ‘strong’ sites , and both differ from what is expected by chance ( Figure 2—figure supplement 1C ) . As expected for bona fide A-to-I editing sites ( Kleinberger and Eisenberg , 2010 ) , G is significantly underrepresented in the nucleotide before the editing site ( 6922 out of 87 , 574 sites detected ) and over represented in the nucleotide that follows the editing site ( 38 , 564 out of 87 , 574 sites ) . A and T are over represented in the nucleotide before the editing site ( 42 , 475 and 25 , 061 , respectively , out of 87 , 574 sites ) . We characterized the differences between the A-to-G modifications and all the other types of modifications ( non A-to-G ) . One possible source of apparent modifications is sequencing errors . We thus compared the quality score between A-to-G and non A-to-G modification sites . However , no significant difference in the distribution of the quality scores was observed ( Figure 2—figure supplement 4A ) . Another technical explanation for the non A-to-G modifications can be read-end artifacts: the ends of the reads are more likely to be misaligned due to splicing , or to contain errors generated in the RNA sequencing protocol ( Ramaswami et al . , 2012 ) . Indeed , non A-to-G modifications ( unlike A-to-G ones ) tend to be located in the reads-end ( Figure 2—figure supplement 4B ) , suggesting a larger fraction of these sites is likely to be a result of technical artifacts . In addition , we studied the modification level distribution ( the fraction of cDNA reads exhibiting the modification in a given position ) for both types of modifications ( Figure 2—figure supplement 4C ) . A higher fraction of non A-to-G sites show ∼50% modification levels ( compared to the A-to-G sites ) , indicating a higher fraction of missed genomic polymorphisms . Consistently , 50% of the sites with non A-to-G modification levels between 40–60% or between 90–100% recur in both the GFL and the OL tissue ( coming from the same individual animal ) , compared to only 21% of the A-to-G modifications in the same ranges . To find statistically significant differences in the modification levels between the GFL and the OL tissues , a binomial analysis was performed with Bonferroni-corrected p-value of 0 . 05 as a cutoff . Overall , 19% ( 16 , 425 out of 87 , 574 ) of the detected A-to-G modifications differ significantly between the GFL and OL tissues . Most of these sites ( 84% , 13 , 731 out of 16 , 425 ) have higher modification levels in the GFL tissue . In contrast , only 6% ( 783 out of 12 , 614 ) of the non A-to-G modifications are significantly different for the same two tissues , with no clear tissue preference: 45% and 55% of these sites have higher modification levels in the GFL and the OL tissue , respectively . As the same technical artifacts and genomic polymorphisms are expected to replicate in both tissues , these data increase our confidence that most of the A-to-G sites are bona fide editing sites . To further characterize the tissue-dependence of modifications levels , Illumina sequencing was again utilized to generate ∼54 , ∼62 and ∼42 million paired-end , 100 nt reads , using RNA from Buccal ganglia ( BG ) , Stellate ganglion with the giant fiber lobe removed ( SG ) , and Vertical lobe ( VL ) , respectively , collected from a different animal . The same alignment procedure was applied to quantify the modification level at the previously described sites for each of the additional samples . Statistically significant differences in the modification levels between all the five neuronal tissues were detected using a binomial analysis with Bonferroni-corrected p-value of 0 . 05 as a cutoff . The sites with significantly variable A-to-G modification levels were subjected for hierarchical clustering , revealing clear tissue selectivity with higher modification levels in the GFL tissue and low levels in the VL tissue ( Figure 2—figure supplement 2A ) . The same analysis for the non A-to-G modification levels demonstrates consistency between the individual animals as the modification levels in the GFL and OL tissues form one cluster , and the VL , BG and SG tissues form a second cluster , again suggesting that many of the non A-to-G modifications are due to genomic polymorphisms ( Figure 2—figure supplement 2B ) . To characterize the modification levels in non-neuronal tissues , Illumina sequencing was again utilized to generate ∼23 , ∼23 , ∼19 , ∼26 , ∼19 and ∼14 million paired-end , 150 nt reads , using RNA from the branchial heart , the Gill , the ventral epithelial layer on the pen , the marginal epithelial layer on the pen , the iridophore layer of the skin , and the chromatophore layer of the skin . The same alignment procedure was applied to quantify the modification level at the previously described sites for each of the additional samples , revealing considerably lower editing levels in the non-neuronal tissues ( Alon et al . ( 2015 ) , available via Dryad data repository ) . Direct validation of editing was performed on a subset of the detected A-to-G sites using Sanger sequencing and deep sequencing . To reduce the chance for RNA contamination in the DNA or vice versa , primers were designed to differentially amplify the gDNA , by residing in introns , or cDNA , by spanning exon–exon boundaries . To identify intronic sequence and exon–exon junctions the following steps were performed: ( a ) we recorded all the cases in which the beginning or the end of the DNA read was trimmed during the local alignment against the Trinity sequences . ( b ) If the flanking region could be aligned against the Trinity sequences , even if aligned separately without the rest of the read , it was discarded . ( c ) If at least three DNA reads showed the same flanking sequence starting from the same position inside the ORFs , it was considered to be an intron fragment . This procedure revealed the positions of part of the exon–exon junctions and part of the intron sequences that correspond to the junctions . Overall , this procedure resulted in about 2100 regions , containing ∼5% of all the detected editing sites , in which the gDNA and the cDNA could potentially be differentially amplified . For the Sanger sequencing , primers were designed to differentially amplify the gDNA and cDNA of 19 ORFs ( Supplementary file 1C ) . To allow better detection of editing , all the sites tested using Sanger sequencing were chosen to have >=20% modification levels ( in our original GFL analysis ) . For the Sanger validation experiment the GFL tissue from a single animal was used ( different from the animals used for the HiSeq sequencing experiments ) . All the sites tested ( 40 out of 40 ) were validated using Sanger sequencing ( Figure 2—figure supplement 3 and Supplementary file 1D ) . For the deep sequencing validations , three groups of targets were tested ( Supplementary file 1E , F ) : ( a ) twenty ‘interesting’ genes , that is , squid components with homology to genes known to be implicated in human diseases or in other important pathways , selected from the dataset of ∼2100 regions described above . ( b ) 20 regions randomly-selected from the above dataset of ∼2100 regions . ( c ) 20 regions randomly-selected from the set of putatively edited regions for which we could not design unique gDNA primers , and thus gDNA and the cDNA could not be differentially amplified . For this group the same primers were used to amplify the gDNA and the cDNA . As with the Sanger validation experiment , the deep sequencing validations were done using GFL tissue from a single animal ( different from the animals used for the HiSeq and Sanger sequencing experiments ) . All primer sets were designed with an overhang so that sequencing and indexing primers could be added to the amplicons in a nested PCR reaction . Samples from gDNA and cDNA were distinguished by unique sequencing indexes . After nested PCR , amplicons were pooled , purified by Ampure XP ( Beckman Coulter , Danvers , MA ) , and sequenced on an Illumina MiSeq instrument . All the cDNA targets were amplified , but only 49 of the 60 gDNA targets amplified well enough for analysis ( Supplementary file 1E , F ) . For the other eleven targets , the presence of undetected introns could have disrupted amplification . The DNA and RNA reads were analyzed using the same detection procedure described above ( Figure 1B ) , with the sole exception that sequence variations below 0 . 1% ( the expected sequencing error rate ) were allowed in the DNA reads ( as mandated by the much larger DNA coverage in this validation study ) . Altogether , 84% ( 120 out of 143 ) of the A-to-G sites examined were validated ( Supplementary file 1F , G ) . In contrast , none of the 12 non A-to-G sites examined were validated . Moreover , 170 additional A-to-G sites ( 86% of all novel detected sites , a similar fraction to the HiSeq data used in the original detection , Figure 2A ) were identified , more than doubling the number of A-to-G sites detected . Similar results were observed for the three groups examined . Finally , for the validated A-to-G sites , high correlation was obtained between the editing levels ( the fraction of cDNA reads exhibiting the editing in a given position ) measured using the HiSeq data and the MiSeq data ( Pearson's r of 0 . 86 , p-value = 1e-35 ) ( Supplementary file 1G ) . To demonstrate the extent of massive recoding on two examples , homology-modelling of the squid proteins α Spectrin and Piccolo was performed with SWISS-MODEL using default parameters ( Biasini et al . , 2014 ) and visualized using UCSF Chimera package http://www . cgl . ucsf . edu/chimera ( Figure 3B and Figure 3—figure supplement 1 ) . To find if the recoding events are enriched in genes with specific functions , we have calculated the cumulative recoding level , that is , the editing level summed over all recoding sites within each squid ORF . This gives a single score representing the extent of recoding in the whole protein . The squid ORFs list was ranked using the cumulative recoding level and all the Swiss-Prot annotations were converted to human Swiss-Prot annotations ( when possible ) for consistency . The GO analysis tool GOrilla was used to find enriched GO annotations in the ranked list ( Eden et al . , 2009 ) . In order to control for possible detection bias in highly expressed genes , the same list was ranked using the gene expression level ( FPKM ) and was also analyzed using GOrilla . As expected , the enriched GO annotations in the control list ( that is , genes ranked by expression levels ) were mainly connected to the ribosome ( translational elongation , structural constituent of ribosome , RNA binding and so on ) . In contrast , the list ranked by the cumulative recoding level gave enriched GO annotations which are mainly connected to neuronal and cytoskeleton functions ( Figure 4—figure supplement 1A and Supplementary file 1H ) . Trying to detect enriched GO annotations at the bottom of the list ( ranked by the cumulative recoding level ) did not produce any significant results . The extensive recoding due to RNA editing can affect many molecular pathways . In order to appreciate the extent of this phenomenon , squid ORFs were mapped to all human KEGG pathways ( Kanehisa and Goto , 2000 ) , revealing that RNA editing has a global effect on the majority of the squid pathways ( Supplementary file 1I ) . In this analysis , the homologs to the human proteins can be: ( a ) ‘heavily edited’ , defined as proteins for which the cumulative recoding level , that is the editing level summed over all recoding sites , exceeds unity , ( b ) ‘edited’ , if the protein has at least one recoding site , ( c ) not edited , or ( d ) not identifiable in the squid transcriptome . Editing levels were calculated using data from the GFL and OL tissues combined . On average , 74% and 22% of the identifiable proteins in each pathway are edited or heavily edited , respectively . Pathways related to the nervous system are even more extensively edited: 75% and 35% of the identifiable proteins in these pathways are edited and heavily edited , respectively . On the flip side , in the pathways ‘Ribosome’ and ‘RNA polymerase’ only 50% and 33% of the identifiable proteins are edited and 0% and 4% of the identifiable proteins are heavily edited , respectively ( Supplementary file 1I ) , demonstrating that some crucial pathways are protected from editing . We have also specifically examined the extent of recoding in squid ORFs that encode voltage and neurotransmitter gated ion channels , ion transporters , synaptic release machinery and molecular motors . Overall , we examined 87 ORFs that are homologous to the following proteins: Voltage-gated potassium channel alpha subunit , Voltage-dependent sodium channel alpha subunit , Voltage-dependent calcium channels , Ionotropic glutamate receptors , Synaptotagmin , Synaptobrevin , Syntaxin , Synapsin , Snap 25 , Sodium/potassium-transporting ATPase alpha subunit , Sodium/calcium exchanger ( SLC8 ) , Sodium bicarbonate exchanger ( SLC4 ) , Sodium/hydrogen exchanger ( SLC9 ) , Sodium/potassium/calcium exchanger ( SLC24 ) , Dynein , and Kinesin ( Supplementary file 1J ) . In the GFL tissue , 47 out of the 87 proteins ( 54% ) are heavily edited , more than twofold higher than expected by chance ( p-value<1e-6 ) . In fact , in all the examined neuronal tissues , this group of 87 proteins is heavily edited , significantly higher than expected by chance ( Supplementary file 1J ) . An important question is whether editing in the squid ORFs tends to avoid recoding by preferring synonymous modifications or , alternatively , tends to create more recoding sites than expected by chance . In squid , the expected fraction of nonsynonymous changes is 0 . 65 , estimated by random changes of adenosines in the ORFs , accounting for the observed local sequence preference of the editing sites ( Figure 2—figure supplement 1C ) , as follows: ( a ) we examined the sequence surrounding sites detected as edited ( the base preceding and the following the site ) , and counted the number of times each one of the 16 possible combinations of upstream and downstream nucleotide appears . These numbers were normalized by the number of times each combination appears for all the A bases in all the ORFs , to produce the observed probability of being targeted by editing given a certain combination of 5′ and 3′ nucleotides . ( b ) all the locations inside the ORFs were screened in a random order until an A base was encountered . ( c ) a random number was generated , if it was below the normalized probability corresponding to the sequence surrounding the site in question , this site was selected for further use . ( d ) the potential effect on the codon due to changing the A base to G was examined , in particular , whether the change is nonsynonymous or synonymous . ( e ) steps ( b–d ) were repeated until the number of A bases changed matched the observed number of editing sites . The described randomization procedure was also used to calculate the expected codon changes due to editing ( Figure 4—figure supplement 2B ) . We found that the higher the editing levels , the higher the fraction of nonsynonymous changes , and for editing levels >20% the nonsynonymous fraction is significantly higher than the expected fraction of 0 . 65 ( Figure 4B and Figure 4—figure supplement 2A ) . A recoding event which creates an amino acid rarely found in homologous proteins may indicate that the editing is deleterious . Therefore , each squid ORF was aligned against the conserved domains in the Conserved Domain Database ( CDD ) using DELTA-BLAST ( Boratyn et al . , 2012; Marchler-Bauer et al . , 2013 ) . DELTA-BLAST calculates the position-specific score matrices ( PSSM ) for each possible amino acid substitution . Positive scores indicate that a certain amino acid substitution occurs more frequently in the alignment against the conserved domains than expected by chance , while negative scores indicate that the substitution occurs less frequently than expected . The substitution score of each editing event was recorded as well as the editing level in the same position , using data from the GFL and OL tissues combined . As a control , the editing levels in all the sites with the same recoding type were randomly shuffled . This accounts for the fact that editing creates specific amino acid changes ( Figure 4—figure supplement 2B–D ) , and in turn , these specific changes may be correlated with average editing levels and with substitution scores . Thus , the shuffled dataset preserves both the distribution of recoding types and the distribution of editing levels for each recoding type . This analysis revealed that the average editing levels in sites with large negative substitution scores , which may be deleterious , are significantly lower than what is expected by chance ( Figure 4—figure supplement 3A ) . The distribution of recoding sites for each substitution score was calculated and compared with random changes that preserve the sequence preference and the total number of editing events ( Figure 2—figure supplement 1C ) . Consistent with the above finding , there are significantly less recoding sites with large negative substitution score than expected by chance ( Figure 4—figure supplement 3B ) . Finally , on average , the higher the editing level in a given site , the higher the average substitution score , above what is expected by chance ( Figure 4—figure supplement 3C ) . Thus , this analysis indicates that recoding by editing tends to avoid potentially deleterious sites and that editing sites with high editing levels might be more important functionally than editing sites with low editing levels . We searched publicly available datasets and found two datasets of matched DNA- and RNA-seq data which are comparable in size and read-lengths to our squid data: ( i ) Human RNA-seq and DNA-seq data of blood samples ( Chen et al . , 2012 ) ( ii ) Macaque RNA-seq and DNA-seq brain data ( Chen et al . , 2014 ) . The same pipeline as described above was applied to the data , with one single exception: while comparing the transcriptome model with SwissProt we used only non-vertebrate SwissProt sequences , in order to mimic the situation for squid in which there are no SwissProt entries from closely related species . The analysis of the human and macaque data was done for two purposes: ( A ) to show that the enormous number of AG mismatches in the squid data is real and not an artefact of the analysis pipeline , and ( B ) to demonstrate that our pipeline could identify RNA editing sites established by previous studies . The results of the analysis are summarized in Figure 2A and Supplementary file 1A . For non-AG mismatches , we obtained roughly the same numbers as those for squid . However , as expected , the AG mismatches in the human and macaque samples did not show any enrichment , and their abundance ( in coding regions ) was similar to those for other mismatches . In other words , the sensitivity of our method allows the detection of the super-strong editing signal of squid , but cannot separate the rare editing sites in mammals ( in coding regions ) from noise . Note that the task of detecting recoding sites in mammals is highly non-trivial even when one takes advantage of all information available , including the accurate genome sequence . The best efforts , so far , have yielded enrichment of AG mismatches , but still most detected sites are non-AG ( i . e . , most-probably , false-positives ) ( Ramaswami et al . , 2013 ) . In order to assess the precision of our method in recalling true editing sites we have looked at the validated editing sites found in the two studies above , and checked whether they have been picked up by our algorithm as well . The full data is presented in Supplementary file 1B . Overall , about half of the sites were picked up by our pipeline . The sites that were not identified , mostly resided in regions poorly described by our Trinity transcriptome assembly ( some were just very weakly edited ) . Thus , one may conclude that the recall rate of our method is lower than 0 . 5 , and the true extent of squid recoding is even much larger than we report . The data reported in this paper was deposited to the Sequence Read Archive ( SRA ) , under accession SRP044717 . The constructed squid coding sequences and all the A-to-G modification sites detected in the coding regions of the squid are available via Dryad digital repository ( Alon et al . , 2015; http://dx . doi . org/10 . 5061/dryad . 2hv7d ) .	For living cells to create a protein , a genetic code found in its DNA must first be ‘transcribed’ to create a corresponding molecule of messenger RNA ( mRNA ) . DNA and RNA are both made from smaller molecules called nucleotides that are linked together into long chains; the information in both DNA and RNA is contained in the sequence of these molecules . The mRNA nucleotides coding for proteins are ‘translated’ in groups of three , and most of these nucleotide triplets instruct for a specific amino acid to be added to the newly forming protein . DNA sequences were thought to exactly correspond with the sequence of amino acids in the resulting protein . However , it is now known that processes called RNA editing can change the nucleotide sequence of the mRNA molecules after they have been transcribed from the DNA . One such editing process , called A-to-I editing , alters the ‘A’ nucleotide so that the translation machinery reads it as a ‘G’ nucleotide instead . In some—but not all—cases , this event will change , or ‘recode’ , the amino acid encoded by this stretch of mRNA , which may change how the protein behaves . This ability to create a range of proteins from a single DNA sequence could help organisms to evolve new traits . Evidence of amino acid recoding has only been found to a very limited extent in the few species investigated so far . There has been some evidence that suggests that recoding might occur more often , and alter more proteins , in squids and octopuses . However , this could not be confirmed as the genomes of these species have not been sequenced , and these sequences were required to investigate RNA recoding using existing techniques . Alon et al . have now developed a new approach that allows the recoding sites to be identified in organisms whose genomes have not been sequenced . Using this technique—which compares mRNA sequences with the DNA sequence they have been transcribed from—to examine the squid nervous system revealed over 57 , 000 recoding sites where an ‘A’ nucleotide had been modified to ‘G’ and thereby changed the coded amino acid . Many of the identified mRNA molecules had been recoded in more than one place , and many more of these than expected changed the amino acid sequence of the protein translated from them . Alon et al . therefore suggest that RNA editing may have been crucial in the evolution of the squid's nervous system , and suggest that recoding should be considered a normal part of the process used by squids to make proteins .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "short", "report", "neuroscience", "genetics", "and", "genomics" ]	2015	The majority of transcripts in the squid nervous system are extensively recoded by A-to-I RNA editing
Autophagy is a catabolic process for bulk degradation of cytosolic materials mediated by double-membraned autophagosomes . The membrane determinant to initiate the formation of autophagosomes remains elusive . Here , we establish a cell-free assay based on LC3 lipidation to define the organelle membrane supporting early autophagosome formation . In vitro LC3 lipidation requires energy and is subject to regulation by the pathways modulating autophagy in vivo . We developed a systematic membrane isolation scheme to identify the endoplasmic reticulum–Golgi intermediate compartment ( ERGIC ) as a primary membrane source both necessary and sufficient to trigger LC3 lipidation in vitro . Functional studies demonstrate that the ERGIC is required for autophagosome biogenesis in vivo . Moreover , we find that the ERGIC acts by recruiting the early autophagosome marker ATG14 , a critical step for the generation of preautophagosomal membranes . Autophagy is a conserved catabolic process underlying the self-digestion of cytoplasmic components through the formation of double-membraned vesicles termed autophagosomes . One basic role of autophagy is to turn over damaged proteins and organelles to maintain cellular homeostasis . Autophagy also allows cells to cope with stresses such as starvation , hypoxia , and pathogen infection ( Mizushima et al . , 2008; Burman and Ktistakis , 2010; Levine , 2005; Yang and Klionsky , 2010; Weidberg et al . , 2011 ) . In the process of starvation-induced autophagy , several upstream signals are triggered , including inhibition of the mechanistic target of rapamycin ( MTOR ) , and activation of the Jun N-terminal kinase ( JNK ) and AMP kinase ( AMPK ) ( Noda and Ohsumi , 1998; Wei et al . , 2008; Kim et al . , 2011; Rubinsztein et al . , 2012; Kim et al . , 2013 ) . These signals are conveyed to activate the serine/threonine-protein kinase complex containing the Atg1 homologs ULK1/2 , ATG13 , FIP200 ( RB1CC1 ) and ATG101 ( C12orf44 ) ( Mizushima , 2010 ) . Together with upstream signals , this complex promotes the formation and membrane docking of the class III phosphatidylinositol 3 ( PtdIns3 ) -kinase ( PI3K ) complex consisting of ATG14 ( ATG14L/Barkor ) , the Atg6 homologue BECN1 ( Beclin1 ) , VPS34 ( PIK3C3 ) and VPS15 ( p150 ) for phosphatidylinositol 3-phosphate ( PI3P ) production ( Obara and Ohsumi , 2011 ) . Subsequently , DFCP1 ( ZFYVE1 ) , an endoplasmic reticulum ( ER ) -associated PI3P binding protein , is recruited to the site of newly-generated PI3P to form omegasomes ( Axe et al . , 2008 ) . This is followed by two ubiquitin-like conjugation systems to produce the ATG12–ATG5 conjugate and phosphatidylethanolamine ( PE ) -lipidated ATG8/LC3 , which initiates the formation of a preautophagosomal organelle termed the phagophore or isolation membrane ( Mizushima et al . , 1998a , 1998b; Ichimura et al . , 2000; Geng and Klionsky , 2008 ) . The membrane then expands and engulfs cytoplasmic components . Finally , the crescent-shaped tubular membrane seals to form a double-membraned autophagosome with cytoplasmic components enclosed within the inner membrane . Fusion of the autophagosome with the lysosome leads to the breakdown of the inner membrane together with the trapped cytosolic material ( Mizushima et al . , 1998a; Burman and Ktistakis , 2010; Yang and Klionsky , 2010; Weidberg et al . , 2011; Rubinsztein et al . , 2012 ) . A long-standing quest in the autophagy field has been to define the origin of the autophagosomal membrane . Recent data suggest a multi-membrane source model for autophagosome biogenesis . The endoplasmic reticulum ( ER ) supports PI3P-dependent formation of the omegasome , a cradle for phagophore generation and elongation ( Axe et al . , 2008; Hayashi-Nishino et al . , 2009; Yla-Anttila et al . , 2009 ) . The outer membrane of the mitochondrion may also supply lipids for the phagophore and autophagosome ( Hailey et al . , 2010 ) . Recently , a study by Hamasaki et al . ( Hamasaki et al . , 2013 ) indicates the ER–mitochondrial junction as being required for autophagosome biogenesis , possibly reconciling these two origins . In addition , clathrin-coated vesicles from the plasma membrane have been shown to promote phagophore expansion through the SNARE protein VAMP7 and its partner SNAREs ( Ravikumar et al . , 2010; Moreau et al . , 2011 ) . Moreover , ATG9-positive vesicles cycle between distinct cytoplasmic compartments to deliver membrane to a developing autophagosome or , in yeast , to phagophore assembly sites ( PAS ) ( Young et al . , 2006; Sekito et al . , 2009; Mari et al . , 2010; Nair et al . , 2011; Orsi et al . , 2012; Yamamoto et al . , 2012 ) . Autophagosomes may also acquire membrane from other sources including Golgi ( Geng et al . , 2010; Ohashi and Munro , 2010; Yen et al . , 2010; van der Vaart et al . , 2010 ) , early endosomes ( Longatti et al . , 2012 ) and vesicles budding from the ER and Golgi ( Hamasaki et al . , 2003; Zoppino et al . , 2010; Guo et al . , 2012 ) . Although tremendous progress has been made , a direct functional link between a membrane source and autophagosome biogenesis has not been established . Furthermore , the identity of the membrane determinant that responds to a stress signal to initiate autophagosome formation is unknown . A variety of visual techniques have been developed to define the origin of the autophagosome membrane . Here , we developed a functional approach relying on the lipidation of LC3 to assay an early stage in autophagosome biogenesis . We establish a cell-free system that reflects many of the physiological and biochemical landmarks of early events in the autophagic pathway and define the ER–Golgi intermediate compartment ( ERGIC ) , a membrane compartment between ER and Golgi for cargo sorting and recycling ( Appenzeller-Herzog and Hauri , 2006 ) , as a key membrane determinant for autophagosome biogenesis . A key step in autophagosome biogenesis is the generation of PE-lipidated LC3 by a ubiquitin-like conjugation system ( Ichimura et al . , 2000; Kabeya et al . , 2000 ) . The level of LC3 lipidation has long been a reliable measure of autophagy activity in vivo ( Klionsky et al . , 2012 ) . In vitro LC3 lipidation has recently been reconstituted with synthetic liposomes , recombinant LC3 and other components including the ATG12–ATG5 conjugate , ATG7 and ATG3 ( Sou et al . , 2006; Hanada et al . , 2007; Oh-oka et al . , 2008; Shao et al . , 2007; Gao et al . , 2010 ) . We sought to capture this modification in a more physiological context by relying on native membranes and cytosol to provide the core components of LC3 lipidation as well as any regulatory proteins that may be required for early autophagosome formation . To establish such an assay , we mixed autophagosome precursor-deficient membranes with cytosol from normal and starved cells . Cells lacking ATG5 are deficient in starvation-induced autophagy and phagophore formation ( Mizushima et al . , 2001 ) . Hence they only contain unmodified LC3 ( LC3-I ) in both cytosol and membrane fractions ( Figure 1A ) . Cytosols derived from WT cells ( including WT MEF [mouse embryonic fibroblast] , COS-7 [Cercopithecus aethiops fibroblast-like kidney cells] , and HEK293T [human embryonic kidney 293T cells] ) were highly enriched in LC3-I whereas the lipidated form of LC3 ( LC3-II ) sedimented with membranes ( Kabeya et al . , 2000 and Figure 1A ) . We incubated membranes from Atg5 KO MEFs with cytosol from WT MEFs in the presence of GTP and an ATP regeneration system ( Figure 1B ) and observed the formation of LC3-II in a time- ( Figure 1B ) and ATP-dependent manner ( Figure 1C ) . 10 . 7554/eLife . 00947 . 003Figure 1 . In vitro reconstitution of endogenous LC3 lipidation . ( A ) The distribution of LC3-I and LC3-II between the cytosol ( C ) and membrane ( M ) fractions from indicated cells . Cytosol and membranes from indicated cells were separated and evaluated by immunoblot ( IB ) with indicated antibodies . TFR , transferrin receptor ( B ) cell-free reconstitution of LC3 lipidation . Membranes from Atg5 knockout ( KO ) MEFs were incubated with cytosol from wild type ( WT ) cells plus GTP and an ATP regeneration system ( ATPR ) for the indicated times . Then SDS-PAGE and immunoblot were performed to detect the generation of lipidated LC3 ( LC3-II ) . ( C ) ATP dependence of in vitro LC3 lipidation . Reactions similar to ( B ) were performed in the absence or presence of indicated reagents followed by SDS-PAGE and immunoblot . ( D ) Delipidation of LC3 by ATG4B . A reaction similar to ( B ) was performed and the 16 , 000×g membranes were sedimented and solubilized with 1% TritonX-100 . The indicated concentrations of ATG4B were incubated with the samples for 30 min followed by SDS-PAGE and immunoblot . Asterisk , non-specific band . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 00310 . 7554/eLife . 00947 . 004Figure 1—figure supplement 1 . Characterization of the in vitro-lipidated LC3 . ( A ) LC3-II distributes in the 16 , 000×g membrane pellet fraction . Reactions similar to those of Figure 1B were performed . After the indicated times , the post-reaction mixtures were centrifuged at 16 , 000×g for 10 min . The pellet fractions ( 16k P ) were collected and the supernatant fractions were further centrifuged at 100 , 000×g yielding the pellet ( 100k P ) and supernatant ( 100k S ) fractions . SDS-PAGE and immunoblot were performed with indicated antibodies . RPN1 , Ribophorin 1 ( B ) LC3-II is tightly anchored to membranes . A reaction similar to that in ( A ) was performed and 16 , 000×g membranes were collected . The pellet was resuspended , divided into aliquots and incubated with indicated reagents followed by another 16 , 000×g centrifugation to separate into pellet ( P ) and supernatant ( S ) fractions , which were examined by SDS-PAGE and immunoblot . TFR , transferrin receptor . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 004 We compared the fractionation and biochemical properties of the in vitro-generated LC3-II to its in vivo counterpart . In a crude fractionation study , we found that the in vitro-generated LC3-II partitioned in the 16 , 000×g membrane fraction ( Figure 1—figure supplement 1A ) . Moreover , the in vitro product resisted extraction with urea or Na2CO3 ( Figure 1—figure supplement 1B ) and was delipidated to LC3-I by ATG4B ( Figure 1D ) , a cysteine protease that cleaves the C-terminal tail of LC3 and removes PE from LC3-II ( Tanida et al . , 2004 ) . These properties are shared with LC3-II generated in vivo ( Kabeya et al . , 2000; Tanida et al . , 2004 ) . Starvation-induced lipidation of LC3 requires the ATG12–ATG5 conjugate ( Mizushima et al . , 2001 ) . To test the ATG5 dependence and starvation effect on in vitro LC3 lipidation , we incubated cytosols from either untreated or starved WT cells or Atg5 KO MEFs with the corresponding membranes from Atg5 KO MEFs ( Figure 2A ) . LC3-II formation was stimulated about threefold in incubations containing cytosol from starved WT MEFs and membranes from starved Atg5 KO MEFs , compared to incubations containing cytosol and membranes from non-starved MEFs ( Figure 2A ) . Cytosol from Atg5 KO MEFs did not generate LC3-II when combined with membranes from Atg5 KO MEFs ( Figure 2A ) . In addition , cytosols from COS-7 and HEK293T cells also reconstituted starvation-induced lipidation of LC3 ( Figure 2—figure supplement 1 ) . These data suggest that the cell-free LC3 lipidation is regulated by starvation-induced components in cells and is dependent on ATG5 . 10 . 7554/eLife . 00947 . 005Figure 2 . The in vitro lipidation of LC3 is regulated by ATG5 , starvation and PI3K . ( A ) Starvation-promoted and ATG5-dependent lipidation of LC3 . Indicated cells were either untreated or starved for 30 min . The in vitro lipidation reaction with the indicated combination of cytosols and membranes was performed . The formation of LC3-II was analyzed by SDS-PAGE and immunoblot . Asterisk , non-specific band ( B ) PI3K inhibitors 3-methyladenine ( 3-MA ) and wortmannin ( Wtm ) inhibit LC3 lipidation . The in vitro lipidation reaction , with cytosol from starved WT MEFs and membrane from Atg5 KO MEFs , was performed in the absence or presence of the indicated concentrations of 3-MA and wortmannin for 60 min . LC3 lipidation was analyzed by SDS-PAGE and immunblot . ( C ) PI3P dependence of in vitro LC3 lipidation . The in vitro lipidation reaction similar to ( B ) was performed in the absence or presence of increasing concentrations of GST-FYVE or FYVE ( C/S ) proteins for the indicated times . SDS-PAGE and immunoblot were performed to analyze the level of LC3-II . Quantification of lipidation activity was shown as the ratio of LC3-II to LC3-I ( II/I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 00510 . 7554/eLife . 00947 . 006Figure 2—figure supplement 1 . Starvation-promoted lipidation of LC3 by COS-7 or HEK293T cytosol . COS-7 , HEK293T and Atg5 KO MEF cells were either untreated or starved for 60 min . The in vitro lipidation reaction with indicated combination of cytosols and membranes was performed followed by SDS-PAGE and immunoblot to examine the formation of LC3-II . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 00610 . 7554/eLife . 00947 . 007Figure 2—figure supplement 2 . Purification and verification of GST-FYVEs . ( A ) Purification of GST-fusion PI3P binding FYVE domains and mutants ( C/S ) . ( B ) PIP Strip blot with 10 µg/ml GST-FYVE . ( C ) PIP Strip blot with 10 µg/ml GST-FYVE ( C/S ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 007 To test the physiological relevance of the cell-free reaction , we examined the effect of inhibitors of autophagy on the lipidation of LC3 in vitro . Starvation-induced autophagosome biogenesis requires the class III PI3K complex which contains ATG14 , BECN1 , VPS15 , and the PI3K subunit VPS34 ( Burman and Ktistakis , 2010; Obara and Ohsumi , 2011 ) . Inhibition of the PI3K activity prevents autophagy . LC3 lipidation was inhibited in a dose-dependent manner by 3-methyladenine ( 3-MA ) and wortmannin , two PI3K inhibitors of different potency but which act in the same concentration ranges to block autophagy in intact cells ( Figure 2B and Klionsky et al . , 2012 ) . In starved cells , downstream effector proteins recognize the PI3P generated by the autophagy-specific VPS34 PI3 kinase . The FYVE domain binds to PI3P ( Stenmark and Aasland , 1999 ) and when expressed in excess blocks autophagy in the cell by sequestering PI3P ( Axe et al . , 2008 ) . To study the role of PI3P in the in vitro reaction , we isolated a FYVE domain derived from FENS-1 ( WDFY1 ) , an endosomal protein ( Ridley et al . , 2001; Axe et al . , 2008 ) , and included the peptide in a lipidation reaction mixture ( Figure 2—figure supplement 2A , B; Figure 2C ) . As reported in intact cells , the FYVE domain peptide inhibited LC3 lipidation in a dose-dependent manner whereas a cysteine to serine ( C/S ) mutation , which abolishes the ability of FYVE to bind PI3P ( Figure 2—figure supplement 2A , C and Axe et al . , 2008 ) , had no effect on lipidation ( Figure 2C ) . One technical limitation is that the lipidation reaction relies on the conversion of endogenous LC3-I to LC3-II . In order to control the level of substrate , we isolated tagged recombinant LC3 expressed in Escherichia coli . LC3 is synthesized as a precursor that is processed by ATG4 cleavage to expose a glycine at position 120 , the site of PE attachment ( Tanida et al . , 2004 ) . Cell-free lipidation of recombinant T7-LC3 ( aa1-120 ) required the glycine at position 120 and responded in a cytosol- and membrane concentration-dependent manner ( Figure 3—figure supplement 1A , B and C ) . Lipidated T7-LC3 sedimented along with membranes at 16 , 000×g , a property shared with the endogenous LC3-II generated in vitro ( Figure 3—figure supplement 1D and Figure1—figure supplement 1A ) . We next evaluated the physiological requirements for lipidation using the T7-LC3 substrate . Cytosol and membranes isolated from starved cells stimulated T7-LC3 lipidation in vitro ( Figure 3A ) , just as we observed with endogenous LC3 ( Figure 2A ) . Likewise , PI3K inhibitors , 3-MA and wortmannin , and the PI3P blocking peptide , FYVE , blocked in vitro T7-LC3 lipidation in a dose-dependent manner ( Figure 3B , C ) . Furthermore , cytosol deficient in ATG7 , ATG3 or ATG5 , key factors in the ubiquitin-like pathway for LC3 lipidation , failed to generate lipidated T7-LC3 in vitro ( Figure 3D , E and F ) . 10 . 7554/eLife . 00947 . 008Figure 3 . Recapitulation of the major regulatory pathways for autophagy by in vitro lipidation of T7-LC3 . ( A ) Starvation-induced lipidation of T7-LC3 . HEK293T and Atg5 KO MEF cells were either untreated or starved for 90 min . The in vitro lipidation reaction was performed by incubating T7-LC3 with HEK293T cytosols and Atg5 KO MEF membranes with indicated treatments for the indicated times followed by SDS-PAGE and immunoblot . ( B ) 3-methyladenine and wortmannin inhibit T7-LC3 lipidation . The in vitro lipidation reaction was performed by incubating T7-LC3 with cytosol from starved HEK293T and Atg5 KO MEF membranes in the absence or presence of the indicated drugs for 60 min followed by SDS-PAGE and immunoblot . ( C ) PI3P dependence of in vitro T7-LC3 lipidation . In vitro lipidation reactions similar to ( B ) were performed in the absence or presence of the indicated concentrations of GST-FYVEs for 60 min followed by SDS-PAGE and immunoblot . ( D ) Dependence on ATG5 for T7-LC3 lipidation . The in vitro lipidation reaction was performed by incubating T7-LC3 with starved cytosols as indicated and Atg5 KO MEF membranes for 60 min followed by SDS-PAGE and immunoblot to analyze LC3-II in the membrane fraction . ( E ) Dependence on ATG3 for T7-LC3 lipidation . A similar experiment was performed using cytosols from Atg3 KO and WT MEFs , and membrane from Atg3 KO MEFs . ( F ) Dependence on ATG7 for T7-LC3 lipidation . A similar experiment was performed using cytosols from Atg7 KO and WT MEFs , and membrane from Atg7 KO MEFs . ( G ) Dependence on ULK1 for starvation-induced T7-LC3 lipidation . The in vitro lipidation reaction was performed by incubating T7-LC3 with untreated or starved cytosols as indicated and Ulk1 KO MEF membranes for 60 min followed by SDS-PAGE and immunoblot of the membrane fraction . ( H ) Rapamycin-induced lipidation of T7-LC3 . Cells were treated with 1 µM rapamycin or a control solution for 2 hr and cytosol was incubated with membranes as in ( A ) . ( I ) Torin 1-induced lipidation of T7-LC3 . Cells were treated with 200 nM Torin 1 or a control solution for 90 min and incubated with membranes as above . Quantification of lipidation activity was shown as the ratio of LC3-II to LC3-I ( II/I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 00810 . 7554/eLife . 00947 . 009Figure 3—figure supplement 1 . Purification of the T7-tagged LC3 and characterization of the lipidation . ( A ) Purification of HisT7-LC3 and the G/A mutant . ( B ) Lipidation of HisT7-LC3 . The in vitro lipidation reaction was performed by incubating HisT7-LC3 or G/A mutant with indicated cytosols and Atg5 KO MEF membranes for 60 min followed by SDS-PAGE and immunoblot . ( C ) Cytosol and membrane dependence of HisT7-LC3 lipidation . The in vitro lipidation reaction was performed by incubating HisT7-LC3 with increasing concentrations of cytosol or membrane for 60 min followed by SDS-PAGE and immunoblot . ( D ) T7-LC3-II distributes in the 16 , 000×g membrane pellet fraction . T7-LC3 protein was generated by thrombin digestion of the HisT7-LC3 protein to remove the N-terminal His tag . The in vitro lipidation reaction was performed by incubating T7-LC3 with cytosol from starved HEK293T cells and Atg5 KO MEF membranes for 60 min . A centrifugation procedure similar to the experiment in Figure 1—figure supplement 1A was employed and the reaction products were evaluated by SDS-PAGE and immunoblot . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 009 Starvation induces autophagy through an MTORC1-ULKI protein kinase regulatory scheme ( Kim et al . , 2011 ) . We found that cytosol from starved or untreated Ulk1 KO MEFs reduced lipidation two to threefold relative to cytosol from WT cells ( Figure 3G ) . Furthermore , T7-LC3 lipidation was stimulated two to threefold by two MTOR inhibitors , rapamycin ( Heitman et al . , 1991 ) and Torin 1 ( Liu et al . , 2010 ) , known to induce autophagy ( Figure 3H , I ) . Thus , for endogenous and recombinant LC3 , the cell-free reaction reflects and responds to the major regulatory pathways of autophagy . We employed the cell-free reaction as an assay to isolate the membrane responsible for LC3 lipidation . For this purpose , we devised a three-step membrane fractionation procedure and monitored enrichment of the lipidation activity with respect to a variety of membrane marker proteins and the lipid donor PE , in relation to a bulk membrane marker , phosphatidylcholine ( PC ) ( Figure 4 ) . 10 . 7554/eLife . 00947 . 010Figure 4 . Membrane fractionation scheme . Briefly , Atg5 KO MEFs were homogenized and the lysates were subjected to differential centrifugations with indicated g forces . The ability of each fraction to trigger T7-LC3 lipidation was examined . The 25 , 000×g ( 25k ) pellet , which had the most activity , was selected and a sucrose gradient ultracentrifugation was performed to separate the 25k pellet to L ( light ) and P ( pellet ) fractions . The L fraction , which contained the majority of the activity to promote T7-LC3 lipidation , was further resolved on an OptiPrep gradient after which ten fractions from the top were collected and the lipidation activity was examined in each . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 010 In order to separate cellular membranes , we first performed differential centrifugation to obtain four membrane pellets containing different membrane markers ( Figures 4 and 5 ) . The PC level of each fraction was measured and normalized so that the lipidation activity for equal amounts of PC ( specific activity ) could be determined ( Figure 5A , B and C ) . The 25k and 100k fractions had significant lipidation activity ( Figure 5A , B ) . The total activity contained in each fraction was calculated by multiplying the specific activity by the PC level ( Figure 5D ) . The 25k pellet contained most ( >70% ) of the total activity , whereas the 100k pellet had little ( <10% ) total activity due to low levels of membrane . The 25k membrane was enriched in peroxisomes ( PMP70/ABCD3 ) , late endosomes ( LAMP2 ) and cis-Golgi ( GM130/GOLGA2 ) . This fraction also contained membranes from the ERGIC ( SEC22B and ERGIC53/LMAN1 ) , plasma membrane/early endosomes ( PM/Endo , TFR ) , ER ( RPN1 ) , ER exit sites ( ERES , active sites on the ER that generate COPII-coated vesicles , SEC12 ) , lysosomes ( Cathepsin D ) , and ATG9 vesicles . Low levels of a mitochondrial marker ( Prohibitin 1 ) and almost no trans-Golgi ( TGN38/TGOLN2 ) or nuclear ( Histone H4 ) compartments ( Figure 5A ) were detected . 10 . 7554/eLife . 00947 . 011Figure 5 . Separation of the total membrane by differential centrifugations . ( A–D ) A differential centrifugation experiment was performed as depicted in Figure 4 . The total PC of each fraction was measured and presented as a percentage of the total membrane ( C ) and adjusted to a concentration of 0 . 6 mg/ml . The T7-LC3 lipidation activity of each fraction was tested and immunoblot was performed to examine the generation of lipidated T7-LC3 as well as the distribution of the indicated membrane markers ( A ) . The level of each marker in the 25k pellet fraction was calculated as a percentage of the total membrane ( A ) . The specific activity ( the ability of each membrane fraction to induce LC3 lipidation with the equal amount of PC ) of each membrane fraction to trigger T7-LC3 lipidation was measured as a ratio of lipidated to unlipidated T7-LC3s ( B ) . The total activity recovered from each fraction was calculated by multiplying the specific activity by the corresponding PC level of each fraction and shown as a percentage of the total membrane ( D ) . Error bars represent standard deviations of at least three experiments . RPN1 , Ribophorin1; TFR , Transferrin receptor; Mem , membrane; Endo , endosome . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 011 To further fractionate the 25k membrane , we performed sucrose step gradient ultracentrifugation ( Figure 4 ) . This separated the 25k membrane to two distinct fractions , a light ( L ) fraction between the 1 . 1 M and 0 . 25 M layers of sucrose , and a pellet ( P ) fraction that sedimented to the bottom ( Figure 4 ) . The specific and total activity was determined as described above ( Figure 6 ) . Interestingly , the lipidation activity was almost exclusively retained in the L fraction , which was enriched in ERGIC , cis-Golgi , ATG9 vesicles and plasma membrane/early endosomes ( Figure 6A , B and D ) . In contrast , the P fraction , which mainly consisted of ER , ERES , mitochondria , lysosomes and peroxisomes , induced very little T7-LC3 lipidation ( Figure 6A , B and D ) . 10 . 7554/eLife . 00947 . 012Figure 6 . Separation of the 25k pellet fraction by sucrose gradient ultracentrifugation . ( A–D ) A sucrose step gradient ultracentrifugation to further separate the 25k pellet fraction was performed as depicted in Figure 4 . The total PCs of each fraction were measured and presented as a percentage of the 25k pellet membrane ( C ) and adjusted to a concentration of 0 . 6 mg/ml . The T7-LC3 lipidation activities of the L and P fraction were tested and immunoblot was performed as in Figure 5A . The level of each marker in the L fraction was calculated as a percentage of the total membrane ( A ) . The specific activity of each membrane fraction was measured as in Figure 5B . The total activity recovered from each fraction was calculated by multiplying the specific activity by the PC level of each fraction and shown as the percentage of 25k pellet membrane ( D ) . Error bars represent standard deviations of at least three experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 012 To further refine the membrane source of T7-LC3 lipidation activity , we centrifuged the L fraction on an OptiPrep gradient and collected ten fractions from the top ( Figure 4 ) . The lipidation activity was distributed in fractions two through four which co-distributed with SEC22B and ERGIC53 ( Figure 7 ) , two ERGIC markers ( Zhang et al . , 1999; Appenzeller-Herzog and Hauri , 2006 ) . Intriguingly , PE , the substrate for LC3 lipidation , was not enriched in the fractions that triggered T7-LC3 lipidation ( Figure 7B ) . The high activity of these membrane fractions was not caused by selective enrichment of the autophagic factors directly contributing to LC3 lipidation , as all of these factors are enriched in the cytosol , or by influencing the formation of the ATG5–12–16 complex essential for LC3 lipidation ( Fujita et al . , 2008 ) compared with other membrane fractions ( Figure 7—figure supplement 1 ) . These data indicate that factors other than those directly involved in catalyzing LC3 lipidation contribute to the high lipidation activity of these membranes . We further found that the lipidation reaction triggered by the ERGIC-enriched membranes was enhanced by cytosol from starved cells and was inhibited by wortmannin and FYVE peptide ( Figure 7—figure supplement 2 ) . These data suggest that the lipidation activity of the isolated membranes is controlled by the pathway ( s ) that regulate autophagy in vivo . 10 . 7554/eLife . 00947 . 013Figure 7 . Separation of the L fraction by OptiPrep gradient ultracentrifugation . ( A–B ) An OptiPrep gradient ultracentrifugation was used to resolve membranes in the L fraction , as depicted in Figure 4 . 10 fractions were collected . The total PCs of each fraction were measured and adjusted to a concentration of 0 . 6 mg/ml . The T7-LC3 lipidation activities of each fraction were tested and immunoblot was performed as in Figure 5A . The specific activity of each membrane fraction was measured similar to Figure 5 . The PE level of each normalized fraction was determined . A heat map showing the relative levels of the specific activity , PE and each of the indicated markers was generated ( B ) . In each group the fraction with the highest value was defined as 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 01310 . 7554/eLife . 00947 . 014Figure 7—figure supplement 1 . The ERGIC membrane promotes LC3 lipidation without altering cytosolic factors . ( A ) The major autophagy factors are cytosolic . Immunoblot of the fractions from Figure 7 and the cytosol ( Cyt ) used for the in vitro lipidation assay was performed with indicated antibodies . ( B ) The formation of ATG5–12–16 complex is not altered by ERGIC . Indicated membrane fractions obtained from the membrane fractionation were incubated with cytosol for the LC3 lipidation assay . Membranes were removed by centrifugation and the supernatant fractions were collected for size exclusion chromatography on a Superpose 6 column followed immunoblot analysis with the indicated antibodies . Most of the ATG5 is conjugated with ATG12 in wild-type cells ( data not shown ) . ERGIC , the combination of fractions 2–4 in the OptiPrep gradient fractionation; P , the pellet fraction from the sucrose gradient centrifugation . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 01410 . 7554/eLife . 00947 . 015Figure 7—figure supplement 2 . The lipidation activity of the ERGIC-enriched fractions are regulated by starvation and PI3K . ( A ) Lipidation of T7-LC3 from the ERGIC-enriched fractions is enhanced by starvation . An in vitro lipidation reaction similar to that in Figure 7 was performed with HEK293T cytosols from starved ( ST ) and non-starved ( NT ) cells followed by SDS-PAGE and immunoblot . ( B ) Wortmannin inhibits T7-LC3 lipidation triggered by the ERGIC-enriched fraction . Fractions 2–4 from Opti-prep gradient were collected and pooled . The in vitro lipidation reaction was performed with increasing concentrations of wortmannin ( Wtm ) followed by SDS-PAGE and immunoblot . ( C ) PI3P dependence of T7-LC3 lipidation triggered by the ERGIC-enriched fraction . The ERGIC-enriched membranes were pooled as shown in ( B ) . The in vitro lipidation reaction was performed with indicated concentrations of GST-FYVE and FYVE ( C/S ) followed by SDS-PAGE and immunoblot . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 015 The data from membrane fractionation indicate that an ERGIC-enriched membrane fraction induces LC3 lipidation . To determine if ERGIC is indeed the essential membrane for LC3 lipidation , we immunodepleted ERGIC from the L fraction with an anti-SEC22B antibody ( Figure 8A ) . After immunodepletion , both SEC22B and ERGIC53 membranes were reduced , whereas the plasma membrane/endosome membranes , indicated by transferrin receptor , were not affected . Significantly , the ability of the L fraction to induce T7-LC3 lipidation was reduced more than threefold ( Figure 8A ) , suggesting that ERGIC contributes to the high activity of LC3 lipidation in the L fraction . 10 . 7554/eLife . 00947 . 016Figure 8 . ERGIC directly triggers in vitro LC3 lipidation . ( A ) Immunodepletion of ERGIC membrane from L fraction reduces in vitro lipidation activity . The L fraction was prepared as shown in Figures 4 and 5 . An immunodepletion experiment with indicated combinations of anti-SEC22B antibody ( Ab ) and blocking peptide ( pep ) was performed . The membranes from the flow-through were collected and the in vitro lipidation reaction was performed . Equal amounts of membrane from each group were used for the lipidation reaction . T , total membrane from the L fraction . ( B ) Enrichment of lipidation activity on the SEC22B-positive membranes . Immunoisolation of SEC22B positive membranes from the L fraction of Atg5 KO MEFs was performed and the in vitro lipidation reaction was conducted with membranes bound to the beads as well as the total membrane ( T ) from the L fraction . The total membrane used was adjusted to the same amount of the membranes ( based on PC content ) specifically bound to the beads in the reaction . ( C ) Enrichment of lipidation activity on KDEL Receptor ( KDELR ) -positive membranes . Atg5 KO MEFs were transfected with a plasmid encoding the Flag-tagged KDELR protein . 48 hr after transfection , KDELR-positive membranes were immunoisolated with anti-Flag agarose and assayed for lipidation activity as in ( B ) . ( D–F ) Lipidation activity was not enriched in late endosome/lysosome , plasma membrane/endosome or ER membranes . Atg5 KO MEFs were transfected with plasmids encoding LAMP1-Flag ( D ) , Vangl2-Myc ( E ) or Flag-GFP-ER-TM ( F ) . 48 hr after transfection , the 25k pellet fractions ( for LAMP1-Flag and Flag-GFP-ER-TM ) or the L fraction were collected and immunoisolations were performed with anti-Flag agarose or anti-Myc agarose as described above and assayed for lipidation activity . Quantification of lipidation activity was shown as the ratio of LC3-II to LC3-I ( II/I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 01610 . 7554/eLife . 00947 . 017Figure 8—figure supplement 1 . Mitochondrial-associated endoplasmic reticulum membranes ( MAM ) are not active to trigger in vitro LC3 lipidation . ( A ) Indicated membrane fractions were prepared as described by Wieckowski et al . ( Wieckowski et al . , 2009 ) and in vitro lipidation was performed as shown in Figures 4–7 . T , total membrane; S , supernatant membrane separated from crude mitochondria preparation; CM , crude mitochondria fraction; Mito , mitochondria fraction; MAM , Mitochondrial-associated endoplasmic reticulum membranes , FACL4 , Long-chain acyl-CoA synthetase 4 . ( B ) Immunoblot analysis of the MAM fractions purified from non-treated ( NT ) and starved ( ST ) cells . Asterisk , non-specific band . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 017 To test the direct role of ERGIC on LC3 lipidation , we immunoisolated SEC22B or KDEL receptor ( KDELR , another ERGIC marker [Capitani and Sallese , 2009] ) -positive membranes ( Figure 8B , C ) for use in the in vitro lipidation assay . In both experiments , the immunoisolated membranes had substantially increased activity ( about twofold or more than fivefold in the SEC22B or KDELR immunoisolated membranes , respectively ) compared to the total input membrane ( Figure 8B , C ) . In contrast , immunoisolation of lysosomes , plasma membrane/endosomes or ER did not enrich the lipidation activity compared to the total input membrane ( Figure 8D , E and F ) . Therefore the data suggest that ERGIC is the most active membrane substrate for LC3 lipidation . A recent report suggested that starvation induces the recruitment of ATG14 to a zone of adhesion between the ER and mitochondria , termed mitochondrial-associated endoplasmic reticulum membranes ( MAM; Hamasaki et al . , 2013 ) . Hamasaki et al . suggested that this zone of adhesion might trigger the formation of the autophagosome ( Hamasaki et al . , 2013 ) . Such a specialized patch of the ER could be an immediate precursor of the autophagosome . Markers of this adhesion were separated from the ERGIC fraction and the adhesion membrane isolated by the protocol of Wieckowski et al . ( Wieckowski et al . , 2009 ) contained little lipidation activity ( Figure 8—figure supplement 1 ) , suggesting this membrane alone may not be the direct template for LC3 lipidation . Consistent with Hamasaki et al . , a fraction of ATG14 appeared on MAM in a starvation-stimulated manner ( Figure 8—figure supplement 1B ) . To further test the importance of ERGIC in LC3 lipidation , we used inhibitors to deplete ERGIC in cultured cells . H89 is a protein kinase A ( PKA ) inhibitor that blocks COPII-coated vesicle assembly by preventing SAR1 binding to ER membrane when used at a high concentration ( Chijiwa et al . , 1990; Aridor and Balch , 2000 ) . Brefeldin A ( BFA ) is a fungal metabolite that inhibits ARF1 activation ( Peyroche et al . , 1999 ) . Treatment of cells with a high concentration of H89 ( 100 µM ) led to the dispersal of ERGIC53 whereas GM130 , a cis-Golgi marker , remained in the perinuclear region , suggesting that the ERGIC is disrupted but not the cis-Golgi ( Figure 9—figure supplement 1 ) . A low concentration of H89 ( 10 µM ) , which is enough to inhibit PKA , did not affect the localization of either ERGIC53 or GM130 ( Figure 9—figure supplement 1 ) . BFA treatment collapsed the Golgi into puncta colocalizing with ERGIC53 ( Figure 9—figure supplement 1 ) . Treatment of cells with 100 µM H89 after BFA treatment dispersed both ERGIC53 and GM130 ( Figure 9—figure supplement 1 ) . To assess the membrane fractions for retention of lipidation activity , we treated cells with the indicated drugs and the total membrane from each sample was collected and incubated with cytosol from starved cells ( Figure 9 ) . Membrane from cells treated with a high but not a low concentration of H89 ( 100 µM vs 10 µM ) lost the ability to activate LC3 lipidation ( Figure 9A ) . Membranes from BFA-treated cells did not show diminished lipidation activity , nor did BFA mitigate or enhance the effect of H89 treatment on lipidation activity ( Figure 9A ) . Clofibrate is a peroxisome-proliferator activated receptor ( PPAR ) agonist that inhibits ER-to-Golgi transport and promotes retrograde transport of Golgi vesicles back to the ER through an unknown mechanism independent of PPAR activation ( de Figueiredo and Brown , 1999 ) . Clofibrate treatment led to the dispersal of ERGIC53 and GM130 ( Figure 9—figure supplement 1 ) . Like H89 , membranes from clofribate-treated cells failed to promote LC3 lipidation ( Figure 9B ) . As controls , kbNB142-70 , a PKD inhibitor ( Bravo-Altamirano et al . , 2011 ) , had no affect and Pitstop 2 , a clathrin inhibitor ( von Kleist et al . , 2011 ) , only moderately decreased in vitro LC3 lipidation ( Figure 9B ) . 10 . 7554/eLife . 00947 . 018Figure 9 . ERGIC is required for in vitro LC3 lipidation . ( A and B ) In vivo depletion of ERGIC abolishes the in vitro lipidation of LC3 . Atg5 KO MEFs were treated without or with 10 µg/ml Brefeldin A ( BFA ) for 30 min and then incubated with the indicated concentrations of H89 for 10 min ( A ) . Alternatively , cells were directly treated with the indicated drugs: 100 µM H89 , 500 µM clofibrate , 50 µM kbNB142-70 and 50 µM Pitstop2 ( B ) . Total membranes from the cells were collected , the lipidation reaction with T7-LC3 was performed and the products evaluated by SDS-PAGE and immunoblot . Ctr , control ( C ) Blocking ERGIC disruption preserved the in vitro lipidation of LC3 . Atg5 KO MEFs were treated with control , H89 or clofibrate ( Clofi ) in the absence or presence of nocodazole ( Noco ) . Lipidation reactions with the total membranes from the treated cells were performed . ( D–F ) The in vitro lipidation of LC3 recovers with restoration of ERGIC . Atg5 KO MEF cells were treated with BFA followed by 100 µM H89 . Cells were then washed with fresh medium to remove the drugs and , at indicated intervals , samples were collected for immunofluorescence ( D ) or total membrane collection for the in vitro lipidation reaction ( E ) . Quantification of the recovery of lipidation activity , ERGIC and Golgi are shown in ( F ) . Bar , 10 µm . Quantification of lipidation activity was shown as the ratio of LC3-II to LC3-I ( II/I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 01810 . 7554/eLife . 00947 . 019Figure 9—figure supplement 1 . Immunofluorescence showing the effect of indicated drugs on ERGIC and Golgi . Atg5 KO MEFs were treated with indicated drugs or drug combinations as depicted in Figure 9A , B . Cells were fixed for immunofluorescence with the indicated antibodies . Bar , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 019 Retrograde transport to the ER requires microtubules . Disruption of microtubules by nocodazole inhibited the retrograde transport of ERGIC53 towards ER induced by H89 or clofibrate , preserving its punctate localization ( Figure 9—figure supplement 1 ) . Correspondingly , the inhibitory effects of H89 and clofibrate were reversed in respect to the cell-free lipidation activity ( Figure 9C ) . To examine the reversibility of these effects , we treated cells with BFA and H89 for 20 min and then washed the cells into fresh medium for periods up to 40 min ( Puri and Linstedt , 2003 ) . ERGIC53 puncta began to reappear within 2 min and the normal localization was fully restored by 20 min . Cis-Golgi regeneration , indicated by the perinuclear accumulation of GM130 , was slower and was completed by about 30 min ( Figure 9D , F ) . Aliquots examined in the cell-free LC3 lipidation assay displayed a rapid return of activity correlating with the kinetics of ERGIC recovery ( Figure 9E , F ) . These results support our membrane fractionation results concerning the role of the ERGIC in lipidation of LC3 . To test the role of ERGIC in autophagosome formation , we starved cells in the presence or absence of ERGIC-depleting drugs H89 and clofibrate ( Figure 10A ) . We then monitored autophagosome biogenesis by immunofluorescence analysis of LC3 puncta formation ( Klionsky et al . , 2012 ) . Under normal conditions , LC3 is dispersed in the cell , indicating a low level of autophagy ( Figure 10A ) . Starvation-induced LC3 puncta formation was abolished by H89 ( 100 µM ) and clofibrate ( Figure 10A , B ) . Depressed formation of LC3 puncta was not due to enhanced autophagosome turnover because chloroquine , which blocks a late step in autophagy ( Klionsky et al . , 2012 ) , did not mitigate the effect of H89 ( 100 µM ) or clofibrate ( Figure 10A , B ) . Puncta formation of another phagophore marker , ATG16 ( Fujita et al . , 2008 ) , was also blocked by ERGIC depletion ( Figure 10—figure supplement 1 ) . 10 . 7554/eLife . 00947 . 020Figure 10 . ERGIC is required for starvation-induced LC3 puncta formation . ( A ) Drugs that disrupt ERGIC inhibit LC3 puncta formation . MEFs were transfected with plasmids encoding Myc-LC3 . After transfection ( 24 hr ) , the cells were either non-starved ( NT ) or starved ( ST ) in the absence or presence of the indicated drugs followed by immunofluorescence using anti-Myc antibody . Bar , 10 µm . ( B ) Quantification of the cells shown in ( A ) . Error bars represent standard deviations of three experiments . ( C ) Genetically disrupting ERGIC inhibits LC3 puncta formation . MEF cells were co-transfected with plasmids encoding Myc-LC3 and the indicated SAR1A variants . After transfection ( 24 hr ) , the cells were starved in the absence or presence of chloroquine followed by immunofluorescence using anti-Myc antibody . Bar , 10 µm . ( D ) Quantification of the cells shown in ( C ) . Error bars represent standard deviations of three experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 02010 . 7554/eLife . 00947 . 021Figure 10—figure supplement 1 . Drugs that disrupt ERGIC inhibit starvation-induced ATG16 puncta formation . ( A ) MEFs were transfected with plasmids encoding ATG16-Myc . After transfection ( 24 hr ) , the cells were either non-starved ( NT ) or starved ( ST ) in the absence or presence of the indicated drugs followed by immunofluorescence using anti-Myc antibody . Bar , 10 µm ( B ) Quantification of the cells shown in ( A ) . Error bars represent standard deviations of three experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 02110 . 7554/eLife . 00947 . 022Figure 10—figure supplement 2 . Effects of SAR1A variants on ERGIC . MEFs were transfected with plasmids encoding the indicated SAR1A-DsRed variants or control DsRed . After transfection ( 24 hr ) , immunofluorescence with indicated antibodies was performed . Bar , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 022 In addition to inhibiting vesicle traffic from the ER , H89 is a potent inhibitor of PKA . In order to determine whether the observed decrease in the number of LC3 puncta could be due to PKA inhibition and not loss of ERGIC , we tested a low concentration of H89 that inhibits PKA but has no effect on the ERGIC ( Figure 9—figure supplement 1 ) . In contrast to the effect of a high concentration of H89 ( 100 µM ) , cells treated with the low concentration ( 10 µM ) showed a moderate increase in LC3 puncta with or without chloroquine ( Figure 10A , B ) . A recent study also reported that PKA inhibition promotes autophagy ( Cherra et al . , 2010 ) , thus PKA inhibition appears not to be the basis of the effect of H89 on autophagy . As an additional test of the role of ERGIC in autophagosome formation , we introduced mutant forms of the SAR1A GTPase to inhibit the generation of COPII vesicles . A GTP-bound mutant of SAR1A ( H79G ) locks COPII membrane cargos on the ERES and a GDP-bound mutant , SAR1A T34N , completely blocks COPII coat formation ( Ward et al . , 2001 ) . Overexpression of either SAR1A H79G or T34N led to dispersed ERGIC53 localization ( Figure 10—figure supplement 2 ) and inhibited starvation-induced LC3 puncta formation in both control and chloroquine-treated cells ( Figure 10C , D ) . We conclude that the ERGIC is a precursor of or contributes to the formation of the autophagosome . ATG14 is the key mediator bridging upstream cytosolic signals and the autophagic membrane reorganization response . Upon starvation , ATG14 is recruited to a membrane along with the rest of the class III PI3K complex to generate PI3P ( Matsunaga et al . , 2009; Sun et al . , 2008; Zhong et al . , 2009; Matsunaga et al . , 2010 ) . These events can be visualized by the localization of ATG14 and DFCP1 to puncta in starved cells ( Axe et al . , 2008; Matsunaga et al . , 2010 ) . To test the role of ERGIC in this pathway , we treated starved cells with H89 . As shown in Figure 11 , 100 µM H89 but not 10 µM H89 prevented the formation of both ATG14 and DFCP1 puncta ( Figure 11A , B ) . Similar inhibition was also observed in starved cells expressing the two SAR1A mutants , H79G and T34N ( Figure 11C , D ) . 10 . 7554/eLife . 00947 . 023Figure 11 . ERGIC is required for the starvation-induced localization of ATG14 and DFCP1 to puncta . ( A ) H89 inhibits ATG14 and DFCP1 puncta formation . MEF cells were transfected with plasmids encoding EGFP-tagged ATG14 or DFCP1 . After transfection ( 24 hr ) , cells were starved in the absence or presence of the indicated concentrations of H89 followed by fixation and direct visualization of the EGFP signal . Bar , 10 µm . ( B ) Quantification of the cells shown in ( A ) . Error bars represent standard deviations of three experiments . ( C ) Expression of SAR1A mutants inhibits the formation of puncta that contain ATG14 and DFCP1 . MEF cells were co-transfected with plasmids encoding EGFP-tagged ATG14 or DFCP1 and indicated SAR1A-DsRed variants . After transfection ( 24 hr ) , cells were starved followed by fixation and direct visualization of EGFP signal . Bar , 10 µm . ( D ) Quantification of the cells shown in ( C ) . Error bars represent standard deviations of three experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 023 We developed a cell-free approach to measure the recruitment of ATG14 and DFCP1 to ERGIC membranes . Atg5 KO MEFs were treated with or without H89 ( 100 µM ) . Cells were lysed and membranes from a post-nuclear supernatant fraction were mixed with cytosol from starved HEK293 cells transfected with tagged forms of ATG14 and DFCP1 . The mixes were incubated in the presence of ATP and a regeneration system . Membranes were resolved on an OptiPrep buoyant density step gradient . We observed recruitment of tagged ATG14 and DFCP1 to a buoyant membrane fraction from untreated but much less from H89-treated cells ( Figure 12 ) . Recruitment was dependent on membranes , and stimulated by starvation ( Figure 12—figure supplement 1 ) . In starved cells , ATG14 acts upstream of PI3K activity whereas DFCP1 puncta formation requires PI3P generation ( Itakura and Mizushima , 2010 ) . Correspondingly , membranes from cells treated with PI3K inhibitors recruited ATG14 but not DFCP1 ( Figure 12—figure supplement 1B and C ) . In addition to the membrane recruitment result , immunofluorescence studies showed colocalization of ATG14 and DFCP1 with ERGIC53 after starvation ( Figure 12—figure supplement 2 ) . We conclude that the ERGIC membrane is an early site for the assembly of proteins responsible for the formation of the phagophore membrane . 10 . 7554/eLife . 00947 . 024Figure 12 . ERGIC is required for membrane recruitment of ATG14 and DFCP1 . ( A ) Disruption of ERGIC inhibits membrane recruitment of ATG14 and DFCP1 . Atg5 KO MEFs were either untreated or treated with H89 . Membranes were collected and incubated with cytosol of HEK293T cells expressing ATG14-HA and EGFP-DFCP1 . A buoyant density flotation assay was performed followed by immunoblot . ( B ) Quantification of the floated markers shown in ( A ) . The quantification of samples from the buoyant density gradient was calculated as the ratio of chemiluminescence in the first two fractions to the sum of all fractions . Error bars represent standard deviations of three experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 02410 . 7554/eLife . 00947 . 025Figure 12—figure supplement 1 . Establishment of the in vitro membrane recruitment assay . ( A ) Membrane dependence for the flotation of ATG14 and DFCP1 . Cytosol from HEK293T cells expressing ATG14-HA and EGFP-DFCP1 was incubated with or without membrane . A buoyant density gradient flotation assay was performed followed by SDS-PAGE and immunoblot . ( B ) Starvation and PI3K regulation of the membrane recruitment of ATG14 and DFCP1 . Atg5 KO MEF membranes were incubated with the cytosol from either non-starved or starved HEK293T cells expressing ATG14-HA and EGFP-DFCP1 in the presence of indicated drugs . A buoyant density flotation was performed and samples were evaluated by SDS-PAGE and immunoblot . ( C ) Quantification of the floated markers shown in ( B ) . Error bars represent standard deviations of three experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 02510 . 7554/eLife . 00947 . 026Figure 12—figure supplement 2 . Atg14L and DFCP1 puncta colocalize with ERGIC . ( A and B ) MEF cells were transfected with plasmids encoding ATG14-EGFP ( A ) or EGFP-DFCP1 ( B ) . 24 hr after transfection , the cells were starved for 20 min . Cells were then fixed and visualized by immunofluorescence with anti-ERGIC53 antibody . Insets show the magnified view of the boxed areas . Bar , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00947 . 026 In this study , we have identified the ERGIC as a key membrane determinant in the biogenesis of autophagosomes . We first developed a cell-free assay based on in vitro LC3 lipidation to measure autophagosome biogenesis ( Figures 1–3 ) . By combining this assay with membrane fractionation , we identified an ERGIC-enriched fraction as the most active membrane to trigger LC3 lipidation ( Figures 4–7 ) . Next we used organelle immuno-/inhibitor depletion and immunoisolation to demonstrate that the ERGIC is necessary and sufficient to support LC3 lipidation ( Figures 8 and 9 ) . Finally , we provided evidence that the ERGIC membrane acts by recruiting ATG14 to initiate PI3K activity , an early step essential for autophagosome biogenesis ( Figures 10–12 ) . Numerous morphological studies have indicated several possible sources of the autophagosomal membrane ( Burman and Ktistakis , 2010; Mari et al . , 2011; Weidberg et al . , 2011; Rubinsztein et al . , 2012 ) . Indeed , it is improbable that one organelle contributes all the membrane constituents that become part of a mature autophagosome . Nonetheless , it seems likely that one membrane responds to the signal that triggers autophagosome formation and the identity of that membrane has remained elusive . Our isolation of the ERGIC as the locus of LC3 lipidation is in line with morphological studies that describe an omegasome structure projecting directly from the ER membrane ( Axe et al . , 2008; Hayashi-Nishino et al . , 2009 ) . However , our results show clearly that the bulk ER membrane is not the site of lipidation , thus if the omegasome arises from the ER , it must become modified in some way to be active for LC3 lipidation . Starvation induces the membrane localization of soluble oligomeric proteins including ATG14 and the PI3K complex , followed by the recruitment of DFCP1 to generate the omegasome ( Axe et al . , 2008; Matsunaga et al . , 2009; Sun et al . , 2008; Zhong et al . , 2009; Matsunaga et al . , 2010 ) . This process occurs upstream of phagophore generation ( Itakura and Mizushima , 2010 ) . Our data show that in starved cells and in isolated membranes , the presence of ERGIC is required for the efficient membrane recruitment of ATG14 and DFCP1 ( Figures 11 and 12 ) . Thus the ERGIC may play a role in an early stage of phagophore formation by providing a platform to recruit the class III PI3K complex and provide precursor membranes for phagophore initiation , which may be further expanded in a special subdomain of ER . How and why the ERGIC is used to trigger phagophore formation remains unclear . Perhaps the tubular and curved structure of the ERGIC ( Appenzeller-Herzog and Hauri , 2006 ) in mammalian cells favors recruitment of the ATG14 complex and subsequently of other components . ATG14 has been reported to sense membrane curvature via an amphipathic alpha helix located in a C-terminal ‘BATS’ domain ( Fan et al . , 2011 ) . In yeast , it has been shown that highly curved membranes positive for ATG9 are delivered to the PAS ( Mari et al . , 2010; Nair et al . , 2011; Yamamoto et al . , 2012 ) . Subsequently , the curvature sensing protein Atg1 recruits Atg13 and the Atg17–31–29 protein complex to initiate the formation of the phagophore ( Ragusa et al . , 2012 ) . The suggested requirement for a tubular membrane together with the possible existence of an integral membrane protein ( s ) that triggers ATG14 recruitment are now open for biochemical analysis . The cell-free LC3 lipidation reaction responds to a starvation signal , likely originating in the cytosolic fraction . Fractionation of the cytosol should reveal the full range of biochemical requirements including regulatory components induced by starvation as well as the core proteins essential for LC3 lipidation . Furthermore , this approach could be exploited to evaluate the maturation of the phagophore through subsequent stages of morphological transformation including envelope closure and fusion with the lysosome . We obtained horseradish peroxidase-conjugated goat anti-mouse or anti-rabbit IgG from Jackson ImmunoResearch Laboratories ( West Grove , PA ) ; fluorescent secondary antibodies and Earle’s Balanced Salt Solution ( EBSS ) from Invitrogen ( Grand Island , NY ) ; PIP Strips from Echelon ( Salt Lake City , UT ) ; 3-methyladenine ( 3-MA ) , wortmannin , rapamycin , H89 and clofibrate from Sigma ( St . Louis , MO ) ; ATG4B from Boston Biochem ( Cambridge , MA ) ; SEC22B antibody blocking peptides ( sequence CG+HSEFDEQHGKKVPTVSRPYSFIEFDT ) from David King ( University of California , Berkeley ) ; Torin 1 and kbNB142-70 from Tocris ( Minneapolis , MN ) ; Pitstop 2 from Abcam ( Cambridge , MA ) ; reagents for PE measurement as described by Hokazono et al . ( Hokazono et al . , 2011 ) and other reagents from previously described sources ( Ge et al . , 2008 , 2011 ) . Amine oxidase was kindly provided by Eisaku Hokazono ( Kyushu University , Higashi-ku , Japan ) . Mouse anti-GM130 , transferrin receptor , PMP70 and rabbit anti-Prohibitin-1 , SEC22B and Ribophorin 1 antibodies were described before ( Ge et al . , 2008; Schindler and Schekman , 2009; Ge et al . , 2011 ) . We purchased mouse anti-Flag , rabbit anti-ERGIC53 , anti-LC3 , anti-LAMP2 , anti-ULK1 , anti-ATG14 and anti-BECN1 antibodies from Sigma ( St . Louis , MO ) ; mouse anti-T7 antibody from EMD ( Billerica , MA ) ; hamster anti-ATG9 , mouse anti-tubulin , rabbit anti-FACL4 and HRP-labeled anti-GST antibodies from Abcam ( Cambridge , MA ) ; rabbit anti-Cathepsin D from Epitomics ( Burlingame , CA ) ; rabbit anti-TGN38 from Novus Biologicals ( Littleton , CO ) ; mouse anti-ATG7 , anti-ATG3 , anti-ATG5 and anti-ATG16 from MBL ( Woburn , MA ) ; goat anti-SEC12 antibody from R&D Systems ( Minneapolis , MN ) ; mouse anti-Myc antibodies from Cell Signaling ( Boston , MA ) ; mouse anti-GFP antibody from Santa Cruz ( Dallas , Texas ) ; rabbit anti-Histone H4 antibody from the Robert Tjian lab ( University of California , Berkeley ) . The plasmids encoding ATG14-EGFP , ATG14-HA , Myc-LC3 and ATG16-Myc were kindly provided by Qing Zhong lab ( University of California , Berkeley ) . The EGFP-DFCP1 plasmid was kindly provided by Nicholas Ktistakis lab ( Babraham Institute , UK ) . And the LAMP1-RFP-Flag plasmid was from Addgene ( provided by the Sabatini lab , Whitehead Institute ) . The GST-FYVE , GST-FYVE ( C/S ) and T7-LC3 plasmids were constructed by subcloning the indicated inserts from the GFP-TM-FYVE , GFP-TM-FYVE ( C/S ) constructs ( Nicholas Ktistakis , Babraham Institute , UK ) and Myc-LC3 plasmids into pGEX4T1 and pET28a vectors . The encoded proteins were expressed in E . coli BL21 and affinity purified with glutathione or Ni Sepharose ( GE Healthcare Life Sciences , Piscataway , NJ ) . The Flag-GFP-ER-TM plasmid was generated by PCR insertion of a Flag tag into the GFP-ER-TM plasmid ( Nicholas Ktistakis , Babraham Institute , UK ) . The human Vangl2-myc plasmid was described before ( Guo et al . , 2013 ) . Inserts from the pGEX-Sar1As ( Kim et al . , 2005 ) were subcloned into the DsRed-Monomer-N1 vector to generate DsRed-tagged SAR1A plasmids . The cells were cultured to confluence and either untreated or starved in EBSS ( for lipidation using endogenous LC3 , MEF cells were starved for 30 min while HEK293T and COS-7 cells for 1 hr; for lipidation using T7-LC3 , HEK293T cells were starved for 1 . 5 hr ) . Then the cells were harvested by scraping and centrifuging at 600×g for 5 min , washed with PBS followed by another 600×g-spin for 5 min and homogenized by passing through a 22 G needle in a 1 . 5× cell pellet volume of B88 buffer ( 20 mM HEPES-KOH , pH 7 . 2 , 250 mM sorbitol , 150 mM potassium acetate and 5 mM magnesium acetate ) plus cocktail protease inhibitors ( Roche , Indianapolis , IN ) , phosphatase inhibitors ( Roche ) and 0 . 3 mM DTT . The cell homogenates were centrifuged at 160 , 000×g for 30 min , supernatant fractions were collected and the centrifugation was repeated three times to achieve a clarified fraction ( approximately 6–10 mg/ml of protein ) which was used in the lipidation reaction . E . coli BL21 cells with the indicated expression plasmids were cultured at 37°C overnight . The overnight culture was inoculated at 1:50 dilution to a one liter volume and shaken at 37°C to an OD600 of 0 . 6–0 . 8 . Protein expression was induced with 100 µM IPTG at 23°C for 5 hr and the cells were collected by centrifuging at 10 , 000×g for 10 min . The pellet was washed with 0 . 1 M PBS and suspended with 20 ml 0 . 2 M PBS ( pH 7 . 4 ) with 15 mM imidazole and 1x protease inhibitors ( Roche ) . Lysozyme was added to the cells at a concentration of 0 . 5 mg/ml and the digestion was performed on ice for 30 min after which Triton X-100 was added to a concentration of 0 . 5% . The cell suspension was sonicated with five to seven 15 s bursts until the solution was not viscous and the lysate was centrifuged at 100 , 000×g for 30 min . The supernatant was collected and incubated with 1 ml Ni Sepharose ( packed beads ) at 4°C for 2 hr . Then the beads were collected and washed with 70 vol of cold 0 . 2 M PBS with 25 mM imidazole and 0 . 2% Tween-20 followed by 10 vol of 0 . 2 M PBS with 25 mM imidazole . The bound proteins were eluted with 0 . 2 M PBS with 250 mM imidazole , buffer exchanged to 0 . 1 M PBS ( pH 7 . 4 ) and stored at −80°C . Thrombin digestion was performed in the presence of 1 U/ml of thrombin ( Roche ) at room temperature for 1 hr followed by adding 1 mg/ml AEBSF ( Santa Cruz ) to deactivate thrombin . For each reaction , cytosol ( 2 mg/ml final concentration ) , ATP regeneration system ( 40 mM creatine phosphate , 0 . 2 mg/ml creatine phosphokinase , and 1 mM ATP ) , GTP ( 0 . 15 mM ) ( Kim et al . , 2005 ) , 0 . 2 µg HisT7-LC3 ( 1-120 ) or T7-LC3 ( 1-120 ) generated by thrombin digestion and different membrane fractions ( 0 . 2 mg/ml PC content final concentration ) were incubated in a final volume of 30 µl . The reactions were performed at 30°C for the indicated times . Where indicated , compounds or proteins were added to the reactions . Cells ( ten 15-cm dishes ) were cultured to confluence , harvested and homogenized in a 2 . 7× cell pellet volume of buffer containing 20 mM HEPES-KOH , pH 7 . 2 , 400 mM sucrose and 1 mM EDTA by passing through a 22 G needle until ∼85% lysis analyzed by Trypan Blue staining . Homogenates were either centrifuged at 100 , 000×g for 45 min to collect total membranes or subjected to sequential differential centrifugation at 1 , 000×g ( 10 min ) , 3 , 000×g ( 10 min ) , 25 , 000×g ( 20 min ) and 100 , 000×g ( 30 min , TLA100 . 3 rotor , Beckman ) to collect the membranes sedimented at each speed . The PC content of each fraction was measured as described before ( Ge et al . , 2011 ) . Membrane fractions containing equal amounts of PC were used to test LC3 lipidation activity . The 25 , 000×g membrane pellet , which contained the highest activity , was suspended in 0 . 75 ml 1 . 25 M sucrose buffer and overlayed with 0 . 5 ml 1 . 1 M and 0 . 5 ml 0 . 25 M sucrose buffer ( Golgi isolation kit; Sigma ) . Centrifugation was performed at 120 , 000×g for 2 hr ( TLS 55 rotor , Beckman ) , after which two fractions , one at the interface between 0 . 25 M and 1 . 1 M sucrose ( L fraction ) and the pellet on the bottom ( P fraction ) , were separated . Activities of the two fractions were then tested as described above , and the L fraction was selected and suspended in 1 ml 19% OptiPrep for a step gradient containing 0 . 5 ml 22 . 5% , 1 ml 19% ( sample ) , 0 . 9 ml 16% , 0 . 9 ml 12% , 1 ml 8% , 0 . 5 ml 5% and 0 . 2 ml 0% OptiPrep each . Each density of OptiPrep was prepared by diluting 50% OptiPrep ( 20 mM Tricine-KOH , pH 7 . 4 , 42 mM sucrose and 1 mM EDTA ) with a buffer containing 20 mM Tricine-KOH , pH 7 . 4 , 250 mM sucrose and 1 mM EDTA . The OptiPrep gradient was centrifuged at 150 , 000×g for 3 hr ( SW 55 Ti rotor , Beckman ) and subsequently ten fractions , 0 . 5 ml each , were collected from the top . Fractions were diluted with B88 buffer and membranes were collected by centrifugation at 100 , 000×g for 1 hr . The activity of each fraction was determined and the distribution of the activity was compared with that of each membrane marker . Membranes containing an equal amount of PC from each fraction were also measured for PE content using an enzymatic assay ( Hokazono et al . , 2011 ) . Cells ( ten 15-cm dishes ) expressing indicated protein markers were cultured to confluence and harvested as indicated in the ‘Membrane fractionation’ section . Membranes from either the 25 , 000×g membrane pellet ( for Flag-GFP-ER-TM or LAMP1-RFP-Flag immunoisolation ) or the L fraction ( for SEC22B , KDELR-Flag or Vangl2-myc immunoisolation ) were collected , suspended in immunoisolation buffer containing 25 mM HEPES , pH 7 . 4 , 140 mM potassium chloride , 5 mM sodium chloride , 2 . 5 mM magnesium acetate , 50 mM sucrose and 2 mM EGTA ( Zoncu et al . , 2011 ) , and diluted to a PC content of 0 . 2 mg/ml . Anti-Flag ( 100 µl , packed volume ) or anti-Myc agarose ( Sigma ) was added to a 1 ml membrane suspension with or without 0 . 2 mg/ml blocking peptides ( 3xFlag peptide and Myc peptide; Sigma ) and mixed by rotation at 4°C overnight . For immunoisolation of endogenous SEC22B vesicles , 20 µl rabbit anti-SEC22B antibody was added to a 1 ml L fraction membrane suspension with or without 0 . 2 mg/ml SEC22B antibody blocking peptide and incubated for 3 hr at 4°C followed by addition of 100 µl ( packed volume ) Protein A Sepharose for overnight incubation at 4°C . Beads with the associated membranes were washed with 1 ml immunoisolation buffer three times and membranes bound to the beads were eluted by incubating with 0 . 5 mg/ml of the indicated competing peptides for 0 . 5 hr at room temperature . The eluted membranes were collected by ultracentrifugation . The sedimented activities were determined and compared to input membrane of equal PC content . For cytosol preparation , HEK293T cells were transfected with plasmids encoding the genes for the indicated proteins by X-tremeGene HP ( Roche ) . At 48 hr post-transfection , the cytosols were harvested as described above . For membrane preparation , Atg5 KO MEF cells were treated with indicated compounds and the cells were lysed . After a 1 , 000×g centrifugation , the total membranes from the supernatant were sedimented at 100 , 000×g for 1 hr . Similar reactions containing the cytosols , ATP regeneration system , GTP and the total membranes from different treatments were carried out in a final volume of 50 µl at 30°C for 1 hr . After the reaction , a membrane flotation experiment was performed . OptiPrep ( 200 µl of 50% ) diluted in B88 was added to the reaction mixture to make a 40% solution which was overlayed with 200 µl 35% OptiPrep and 50 µl B88 . The gradient was centrifuged at 150 , 000×g for 1 . 5 hr . Seven fractions , 80 µl each , were collected from the top . The bottom fractions no . 5 to 7 were combined and evaluated by SDS-PAGE and immunoblot to examine the distribution of indicated protein markers . Immunofluorescence was performed as previously described ( Ge et al . , 2008 , 2011 ) . Images were acquired with a Zeiss LSM 710 laser confocal scanning microscope . ERGIC recovery quantification was described previously ( Puri and Linstedt , 2003 ) . Golgi recovery was quantified by manually counting the percent of cells displaying a perinuclear location of GM130 . For each sample , more than 100 cells were counted . Immunoblot was performed as previously described ( Ge et al . , 2008 , 2011 ) and dot blot was carried out according to the PIP Strip user manual ( Echelon ) . Images were acquired and bands were quantified with Chemidoc MP Imaging System ( Bio-Rad ) . HEK293T cells were grown in a tissue culture facility . Atg5 KO and control MEFs ( Kuma et al . , 2004 ) , Atg3 KO , Atg7 KO and control MEFs ( Komatsu et al . , 2005; Sou et al . , 2008 ) , and Ulk1 KO and control MEFs ( Kundu et al . , 2008 ) were generously provided by Noboru Mizushima ( Tokyo Medical and Dental University , Japan ) , Masaaki Komatsu ( Tokyo Metropolitan Institute of Medical Science , Japan ) and Kundu Mondira ( St . Jude Children’s Research Hospital ) . The cells were grown in monolayer at 37°C in 5% CO2 and maintained in Dulbecco’s modified Eagle’s medium ( DMEM ) supplemented with 10% FBS . For starvation , the cells were incubated in EBSS for the indicated times in the absence or presence of the drugs indicated in the manuscript .	Cells continually adapt their behavior to accommodate changes in their environment . For example , when nutrients are abundant , cells can grow or proliferate; in times of scarcity , however , they must conserve resources for essential tasks . In particular , during periods of starvation , cells can cannibalize themselves in a process called autophagy , which literally means ‘self-eating’ . Structures called autophagosomes engulf bits of cytoplasm and carry the contents to the digestive compartment of the cell , the lysosome , to be broken down into their constituent parts . This can include the degradation of proteins into amino acids , which can then be recycled into other proteins needed by the cell . In cells , proteins are shipped to their destinations—which can be the plasma membrane or a specific organelle within the cell—via a delivery system known as the secretory pathway . This pathway begins in the endoplasmic reticulum ( ER ) , where many of these proteins are made . From the ER , the proteins move to a compartment called the Golgi apparatus , which then sends them to their destinations , or to the lysosome to be broken down . Between the ER and Golgi they pass through a structure called the ER–Golgi intermediate compartment ( ERGIC ) . Although the signaling pathways that initiate autophagy are known , less is understood about the actual formation of the autophagosomes . Now , Ge et al . have developed an in vitro system to study their formation , and gone on to identify a membrane that is both necessary and sufficient to create these structures . Previous studies have implicated a variety of membranes—including the plasma membrane and the membranes belonging to the ER , the Golgi apparatus , the lysosome and various other organelles—in the formation of autophagosomes . To identify which of these membranes might be involved , Ge et al . focused on a protein called LC3 that is a key marker for the formation of the autophagosome . This protein is recruited to the growing autophagosome by a lipid , so discovering which membranes can add a lipid to LC3 should shed light on the assembly process . By separating the full range of organelles in a cell lysate into fractions ( a process called fractionation ) , Ge et al . found that the ERGIC was the most active membrane to attach lipid to LC3 . Additionally , the lipid was only added when signaling pathways that stimulate autophagy—such as the PI3K pathway—were activated . Together , these results provide insight into the mechanism of autophagosome formation , and the structures in the cell that participate in this process .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2013	The ER–Golgi intermediate compartment is a key membrane source for the LC3 lipidation step of autophagosome biogenesis
Blood progenitors within the lymph gland , a larval organ that supports hematopoiesis in Drosophila melanogaster , are maintained by integrating signals emanating from niche-like cells and those from differentiating blood cells . We term the signal from differentiating cells the ‘equilibrium signal’ in order to distinguish it from the ‘niche signal’ . Earlier we showed that equilibrium signaling utilizes Pvr ( the Drosophila PDGF/VEGF receptor ) , STAT92E , and adenosine deaminase-related growth factor A ( ADGF-A ) ( Mondal et al . , 2011 ) . Little is known about how this signal initiates during hematopoietic development . To identify new genes involved in lymph gland blood progenitor maintenance , particularly those involved in equilibrium signaling , we performed a genetic screen that identified bip1 ( bric à brac interacting protein 1 ) and Nucleoporin 98 ( Nup98 ) as additional regulators of the equilibrium signal . We show that the products of these genes along with the Bip1-interacting protein RpS8 ( Ribosomal protein S8 ) are required for the proper expression of Pvr . Similar to vertebrates , blood cell differentiation in Drosophila is regulated in multiple hematopoietic environments , which include the head mesoderm of the embryo ( Tepass et al . , 1994; Lebestky et al . , 2000; Milchanowski et al . , 2004 ) , the specialized , tissue-associated microenvironments of the larval periphery ( e . g , body wall hematopoietic pockets ) ( Markus et al . , 2009; Makhijani et al . , 2011 ) , and the larval lymph gland , an organ dedicated to the development of blood cells that normally contribute to the pupal and adult stages ( Rizki , 1978; Shrestha and Gateff , 1982; Lanot et al . , 2001; Jung et al . , 2005 ) . Understanding how blood cell development is regulated in the lymph gland is the primary goal underlying the work presented here . Differentiating blood cells ( hemocytes ) of the lymph gland are derived from multipotent progenitors ( Jung et al . , 2005; Mandal et al . , 2007; Martinez-Agosto et al . , 2007 ) . These blood progenitors readily proliferate during the early growth phases of lymph gland development , which is followed by a period in which many of these cells slow their rate of division and are maintained without differentiation in a region termed the medullary zone ( MZ , Figure 1 ) ( Jung et al . , 2005; Mandal et al . , 2007 ) . During the same period , other progenitor cells begin to differentiate along the peripheral edge of the lymph gland to give rise to a separate cortical zone ( CZ ) ( Jung et al . , 2005 ) . How progenitor cell maintenance and differentiation are regulated during the course of lymph gland development has become a major area of exploration in recent years , and several different signaling pathways have been identified that maintain progenitor cells through the larval stages ( Lebestky et al . , 2003; Mandal et al . , 2007; Owusu-Ansah and Banerjee , 2009; Sinenko et al . , 2009; Mondal et al . , 2011; Mukherjee et al . , 2011; Tokusumi et al . , 2011; Dragojlovic-Munther and Martinez-Agosto , 2012; Pennetier et al . , 2012; Shim et al . , 2012; Sinenko et al . , 2012 ) . Wingless ( Wg; Wnt in vertebrates ) is expressed by blood progenitor cells in the lymph gland and has an important role in promoting their maintenance ( Sinenko et al . , 2009 ) , and reactive oxygen species ( ROS ) function in these cells to potentiate blood progenitor differentiation both in the context of normal development and during oxidative stress ( Owusu-Ansah and Banerjee , 2009 ) . Progenitor cell maintenance at late developmental stages is also dependent upon Hedgehog ( Hh ) signaling from a small population of cells called the posterior signaling center that functions as a hematopoietic niche ( PSC ) ( Lebestky et al . , 2003; Jung et al . , 2005 ) . 10 . 7554/eLife . 03626 . 003Figure 1 . Equilibrium signaling maintains hematopoietic progenitors in the developing lymph gland . The lymph gland primary lobe consists of three distinct cellular populations or zones . The medullary zone ( MZ ) contains blood progenitor cells while the nearby cortical zone ( CZ ) contains differentiating and mature blood cells . The posterior signaling center ( PSC ) functions as a supportive population ( a niche ) that expresses Hedgehog ( Hh ) and maintains the progenitor cells utilizing this ‘niche signal’ . The receptor tyrosine kinase ( RTK ) Pvr and the STAT ( STAT92E ) transcriptional activator are required in CZ cells for the proper expression and secretion of the extracellular enzyme ADGF-A , which keeps the extracellular adenosine levels relatively low by converting it to inosine . The Pvr ligand Pvf1 is made in PSC cells and is transported through the lymph gland to activate Pvr in CZ cells . Collectively , we refer to the system that generates ADGF-A from the differentiating cells as ‘equilibrium signaling’ , which is required independently of the niche-derived Hh signaling for the maintenance of progenitor blood cells in the MZ . Signaling events downstream of both ADGF-A and Hh ( dashed arrows ) cause the inhibition of Protein Kinase A ( PKA ) within progenitor blood cells , thereby promoting their maintenance . The individual components are color coded to match the schematic of the lymph gland . The equilibrium signal ADGF-A is blue , originating from the CZ; the niche signal Hh is magenta , originating in the PSC; PKA is gray , functioning in the MZ progenitor cells . Full details of this molecular pathway can be found in Mondal et al . , ( 2011 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03626 . 003 More recently , it has been discovered that the maintenance of lymph gland blood progenitors also requires a backward signal arising from the differentiating cells ( Mondal et al . , 2011 ) . This signal is controlled by a novel pathway that combines the function of the receptor tyrosine kinase Pvr and JAK-independent STAT ( STAT92E ) activation in differentiating cells , followed by the expression of ADGF-A ( Figure 1 ) , a secreted enzyme that converts adenosine to inosine ( Mondal et al . , 2011 ) . Extracellular adenosine is a well-established signal in mammalian systems in various contexts , particularly stress conditions ( Fredholm , 2007; Sheth et al . , 2014 ) , and an elevated adenosine level in Drosophila causes extensive blood cell proliferation ( Dolezal et al . , 2005; Mondal et al . , 2011 ) . It has been demonstrated that differentiating and mature cells express ( and are the primary source of ) ADGF-A , and that its enzymatic activity ( which converts adenosine to inosine ) is required for progenitor cell maintenance ( Mondal et al . , 2011 ) . As differentiation proceeds , ADGF-A expression ( activity ) increasingly promotes the maintenance of extant blood progenitors through the reduction of stimulatory adenosine . In this way , the differentiating cell population helps balance the progenitor/differentiating cell ratio and is the basis for our referring to ADGF-A as an ‘equilibrium signal’ . Loss of ADGF-A ( or STAT or Pvr ) from differentiating cells increases extracellular adenosine level and thereby increases Adenosine Receptor ( AdoR ) signaling and downstream Protein Kinase A ( PKA ) activity in progenitors , which causes these cells to proliferate ( Dolezal et al . , 2005; Mondal et al . , 2011 ) . PKA is a central regulator of progenitor maintenance because it integrates input from both the equilibrium signal ( ADGF-A ) and the niche signal ( Hh ) . PKA mediates the conversion of the transcriptional regulator Cubitus interruptus ( Ci , a homolog of vertebrate Gli ) from its full length form ( Ci155 ) , required for progenitor maintenance ( Mandal et al . , 2007 ) , to a cleaved form ( Ci75 ) that promotes proliferation . Signaling by Hh inhibits PKA and promotes Ci155 stabilization whereas adenosine/AdoR signaling activates PKA and promotes Ci75 conversion . Thus , both Hh ( the niche signal ) and ADGF-A ( the equilibrium signal which removes adenosine ) limit PKA activity and promote progenitor cell maintenance . Although niche and equilibrium signaling are both clearly important , details of their regulation and interaction are less clear . Thus , we performed a loss-of-function genetic screen to identify new genes involved in lymph gland blood progenitor maintenance , particularly those involved in equilibrium signaling . In this study , we report the results of this screen and the identification of three genes , bip1 , RpS8 , and Nup98 , as new components of the equilibrium signaling pathway . Unlike the adult eye or wing , analysis of internal larval tissues such as the lymph gland requires laborious dissection and processing . To circumvent this barrier to genetic screening , we generated a line of flies termed the Hand-Hemolectin Lineage Traced-gal4 line ( HHLT-gal4 UAS-2XEGFP , Figure 2A; see ‘Materials and methods’ for precise genotype ) in which the hematopoietic system is labeled by Gal4-dependent expression of EGFP , such that it can be visualized in live , whole animals ( Figure 2B–C ) . This line makes use of two gal4 drivers to target early lymph gland blood cells ( hemocytes; Hand-gal4 ) and circulating and sessile blood cells ( Hemolectin-gal4 or Hml-gal4 ) and incorporates a Gal4/FLP recombinase-dependent cell lineage tracing cassette to maintain Gal4 expression in the lymph gland after the Hand-gal4 driver itself is down-regulated during the first instar . The Hand-gal4 driver is expressed in the embryonic cardiogenic mesoderm from which the lymph gland is derived . Therefore , the dorsal vessel ( heart ) cardioblasts and the pericardial nephrocytes are also marked by the cell lineage tracing cassette ( Figure 2B ) . EGFP is not expressed in other larval tissues , except in the late third-instar salivary glands that are readily discernible from the hematopoietic system ( Figure 2B , E ) . 10 . 7554/eLife . 03626 . 004Figure 2 . The Hand-Hemolectin Lineage Tracing-gal4 line ( HHLT-gal4 UAS-2XEGFP ) and its use as an in vivo screening tool . ( A ) Schematic describing the key elements of the HHLT-gal4 driver line . ( B ) Image showing the hematopoietic system within a wandering stage third-instar HHLT > GFP larva ( dorsal view ) . Primary , secondary , and tertiary lobes of the lymph gland are readily discernible through overlying musculature , epidermal cells , and cuticle . Lymph gland lobes develop bilaterally , flanking the larval heart ( dorsal vessel , DV ) . Non-blood pericardial cells ( PC ) also express GFP due to early expression of Hand-gal4 . Circulating/sessile blood cells also express GFP due to Hml-gal4 and sessile groups are easily observable . GFP is also seen in ventrally located salivary glands ( SG , out of focus ) of larvae beyond the third-instar transition ( due to Hand-gal4 ) . ( C ) HHLT > GFP control larvae; ( D ) HHLT > GFP larvae overexpressing Ras85D ( LA 527 ) exhibit hyperproliferative lymph glands; ( E ) HHLT > GFP larvae overexpressing combgap ( LA 630 ) show little or no GFP expression in the lymph gland region . Arrows indicate GFP fluorescence from salivary glands ( SG ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03626 . 00410 . 7554/eLife . 03626 . 005Figure 2—figure supplement 1 . As a ‘proof-of-principle’ approach and to assess the effectiveness of HHLT-gal4 as a screening tool , HHLT-gal4 was crossed to lines harboring gain-of-function UAS transgenes known to cause excessive cellular proliferation , with the expectation that such transgenes would cause significant expansion of the hematopoietic tissues . ( A ) HHLT > GFP larvae ( outcrossed to w1118 ) showing baseline fluorescence . ( B–B′ ) HHLT > GFP larvae expressing UAS-human activated Raf ( h-RafACT ) show increased GFP fluorescence with the same exposure ( B ) as for control larvae ( A ) and exhibit enlarged lymph glands with reduced exposure time ( B′ ) . ( C ) HHLT > GFP larvae expressing UAS-activated Drosophila Alk ( DAlkACT ) also exhibit enlarged lymph glands . ( D and E ) Relative bleed cell densities from control animals ( D ) and animals expressing UAS-DAlkACT ( E ) which show an expansion . ( F ) HHLT > GFP larvae expressing UAS-diap1-RNAi exhibit reduced fluorescence in lymph gland and circulating cells . The obvious changes in the level of HHLT-gal4-mediated EGFP expression observed in these backgrounds shows that HHLT-gal4 is indeed a useful tool with which to assess the hematopoietic system in vivo , in the context of a genetic screen . DOI: http://dx . doi . org/10 . 7554/eLife . 03626 . 005 A screen was conducted in which HHLT-gal4 was used to independently misexpress 503 unique UAS-controlled Drosophila genes , with their effects on the hematopoietic system assessed in whole animals based upon EGFP expression . The particular collection of UAS-based gene misexpression lines used ( termed LA lines ) are mapped insertions ( against Flybase release 5 . 7 ) of the P{Mae-UAS . 6 . 11} element into endogenous gene loci that have been previously shown to cause developmental phenotypes upon misexpression ( Crisp and Merriam , 1997; Bellen et al . , 2004 ) . When crossed to HHLT-gal4 , 281 of these lines cause a scorable phenotype in either lymph glands or circulating blood cells of late third-instar larvae ( Supplementary file 1 ) . As an example , LA line 527 , which is predicted to misexpress Ras85D , causes a robust expansion of the lymph gland ( Figure 2D ) , consistent with the previously identified role of Ras85D in controlling hemocyte proliferation ( Asha et al . , 2003; Sinenko and Mathey-Prevot , 2004 ) . By contrast , LA line 630 misexpresses the gene combgap ( encoding a zinc finger transcription factor ) and causes a strong reduction in lymph gland size ( Figure 2E ) . We used these results from the misexpression screen as a means to select potentially relevant genes for the subsequent loss-of-function analyses by RNA interference ( using UAS-RNAi lines ) . We were able to obtain RNAi lines targeting 251 of the candidate genes identified by misexpression and found that 73 RNAi lines targeting 69 genes alter lymph gland size or morphology when crossed to HHLT-gal4 ( Supplementary file 2 ) . To characterize the RNAi phenotypes in more detail , the level of blood cell differentiation within the lymph gland was evaluated by immunostaining with anti-Peroxidasin ( Pxn ) antibodies ( Nelson et al . , 1994 ) . In wild-type lymph glands , expression of mature cell markers such as Pxn is restricted to the periphery of the primary lobe ( the cortical zone ) ( Jung et al . , 2005 ) . By contrast , when niche signaling or equilibrium signaling are compromised , progenitor cells are lost and differentiation markers , including Pxn , are expressed throughout the lymph gland primary lobes ( Mandal et al . , 2007; Mondal et al . , 2011 ) . Rescreening the 73 identified RNAi lines using HHLT-gal4 identified 20 genes ( 21 RNAi lines ) that , when knocked down , cause the expression of Pxn in cells throughout the lymph gland primary lobe ( Figure 3; Table 1 and Supplementary file 2 ) . Compared to controls , the progenitor population ( Pxn negative ) is either strongly reduced or absent in each RNAi background . This ‘expanded’ Pxn phenotype is interpreted as a loss-of-progenitor cell phenotype . 10 . 7554/eLife . 03626 . 006Figure 3 . Identification of RNAi lines that cause an expanded Peroxidasin phenotype when expressed throughout the lymph gland . Peroxidasin ( Pxn , red ) is normally restricted to cortical zone cells ( near the periphery ) ( A , control ) but is seen throughout the lymph gland in RNAi backgrounds ( B–V ) expressed by HHLT-gal4 . Line identifiers and gene targets are shown; additional details listed in Table 1 . Images represent a single middle confocal section taken from a Z-plane series through the entire primary lobe . DOI: http://dx . doi . org/10 . 7554/eLife . 03626 . 00610 . 7554/eLife . 03626 . 007Table 1 . RNAi lines and target genes causing an ‘expanded’ Peroxidasin expression phenotype with HHLT-gal4DOI: http://dx . doi . org/10 . 7554/eLife . 03626 . 007Line #UAS-RNAi IDRNAi targetGeneOff targetsLG size/qualityProtein function13859CG4214Syx50Small/missingGolgi SNARE26543CG7398Trn1Large/baggyhnRNP nuclear import39572CG5738lolal0SmallTranscription factor412574CG12052lola0Large/baggyTranscription factor512759CG6854CTPsyn0Small/baggyCTP synthase615886CG6376E2f1Small/normalTranscription factor717954CG10009Noa360Small/missingZinc finger nucleolar protein819485CG10009Noa360Small/missingZinc finger nucleolar protein922836CG10267Zif0Small/missingTranscription factor1024215CG8149CG81490BaggyDNA binding protein1126176CG3363CG33630SmallUnknown1226370CG4036CG40361LargeOxidoreductase1338472CG1129CG11290Small/missingPeptide transferase1440306CG31938Rrp400Small/normalRNA exosome1541009CG3836stwl0Small/normalTranscription factor1644606CG6778Aats-gly0Small/missingGlycyl-tRNA synthetase1749753CG33155CG331554Small/normalUnknown187574R-2CG7574bip10Small/baggyTranscription factor1910198R-1CG10198Nup98-960Small/missingNucleoporin2012030R-2CG12030Gale0SmallUDP-galactose 4'-epimerase2112765R-3CG12765fsd0Small/normalF-box proteinThis RNAi line targeting sequence overlaps with the putative mRpL53 gene in the same locus . Lines 1–17 from VDRC , lines 18–21 from NIG Japan . Using the pan-lymph gland HHLT-gal4 driver , we identified 21 RNAi lines that cause a loss of progenitor cells in the primary lobes at late stages of lymph gland development . In order to discern whether any of the associated candidate genes have a specific progenitor-maintenance function that is restricted to cells belonging to a single zone , we rescreened the 21 RNAi lines using cell-type-specific Gal4-expressing lines . Targeting RNAi to differentiating and mature cells using Hml-gal4 ( Sinenko and Mathey-Prevot , 2004 ) identified six genes ( CG6854 [CTPsyn] , CG7398 [Transportin] , CG7574 [bip1] , CG10009 [Noa36] , CG10198 [Nup98 , also known as Nup98-96] , and CG31938 [Rrp40] ) that cause an expansion of Pxn ( Figure 4A–G ) and Hml-gal4 , UAS EGFP ( Figure 4H–M ) expression . Since the function of these genes is needed in the CZ for the maintenance of the MZ progenitors , these six genes encode likely candidates for new components of the equilibrium signaling pathway . 10 . 7554/eLife . 03626 . 008Figure 4 . Identification of candidate genes that cause an expanded Peroxidasin expression phenotype within the lymph gland when knocked down by RNAi in differentiating and mature cells . RNAi from identified lines ( Figure 3/Table 1 ) was expressed in lymph glands using Hml-gal4 UAS-GFP ( Hml > GFP ) . In the control , Pxn ( A ) and GFP ( A′ ) are restricted to the cortical zone ( periphery ) . By contrast , knock down of six candidate genes causes extensive expression of Pxn ( B–G ) and Hml ( Hml > GFP ) ( B′–G′ ) throughout the lymph gland , indicating a loss of progenitors in these genetic backgrounds . The combined Pxn and Hml expression patterns for each genetic background are shown ( MERGE , A″–G″ ) . DNA ( blue ) is stained to mark nuclei . DOI: http://dx . doi . org/10 . 7554/eLife . 03626 . 00810 . 7554/eLife . 03626 . 009Figure 4—figure supplement 1 . RNAi lines causing an expanded Peroxidasin expression phenotype when expressed in progenitor cells using dome-gal4 . ( A ) Control ( dome-gal4 with no UAS-dsRNA ) with Pxn expression ( red ) limited to the cortical zone of the lymph gland primary lobe; DNA ( blue ) . ( B–L ) Individual candidate genes identified by RNAi knockdown directly in progenitor cells using dome-gal4 . RNAi for each gene causes the expansion of Pxn expression ( red ) throughout the primary lobes . DOI: http://dx . doi . org/10 . 7554/eLife . 03626 . 009 Screening with dome-gal4 ( Jung et al . , 2005 ) to target RNAi to the progenitor cells identified eleven genes ( Figure 4—figure supplement 1A–L ) , three of which ( Transportin , Noa36 , and Rrp40 ) are in common with those identified using Hml-gal4 . By contrast , use of Antennapedia-gal4 ( Antp-gal4 ) ( Mandal et al . , 2007 ) to target RNAi specifically to niche cells failed to identify any of the 21 lines as additional niche signaling components ( not shown ) . Lastly , seven of the 21 RNAi lines did not cause a phenotype when expressed with any of the zone-specific Gal4 driver lines used . Taken together , our screen identified three genes , CTPsyn , bip1 , and Nup98 , which cause a loss of lymph gland progenitor cells upon RNAi knock down in differentiating cells , but not in progenitor cells or in niche cells . As described below , it was ultimately possible to connect two of these genes , bip1 and Nup98 , to the equilibrium signaling pathway through the control of Pvr expression . The bip1 gene was originally identified through a yeast two-hybrid screen that showed that its encoded protein binds the BTB/POZ domain of the transcription factor Bric à brac 1 ( Bab1 ) ( Pointud et al . , 2001 ) , a protein that has several developmental roles including the formation of ovarian terminal filament cells that are required for germline stem cell maintenance ( Lin and Spradling , 1993; Sahut-Barnola et al . , 1995; Couderc et al . , 2002 ) . Analysis of the predicted Bip1 amino acid sequence ( InterPro ) ( Hunter et al . , 2012 ) identifies a THAP domain containing a C2CH-type zinc finger motif that is known to bind DNA ( Sabogal et al . , 2010 ) . As no known bip1 mutants exist , several different approaches were used to validate the bip1 RNAi results and elucidate the function of the bip1 gene in differentiating the blood cells . First , qRT-PCR confirmed that bip1 is expressed in the lymph gland and demonstrated that the bip1 RNAi line ( NIG 7574R-2 ) actually targets bip1 transcripts . Indeed , RNAi knock down of bip1 using Hml-gal4 ( Sinenko and Mathey-Prevot , 2004; Jung et al . , 2005 ) reduces bip1 mRNA levels in the lymph gland to approximately ten percent of that observed in controls ( Figure 5A ) . The bip1 RNAi blood phenotype is also suppressible by the simultaneous overexpression of bip1 ( UAS-bip1LA645; Figure 5B–B' ) , demonstrating the specific requirement for bip1 in maintaining progenitors . Driving bip1 RNAi with Pxn-gal4 , an alternative differentiating- and mature-cell driver to Hml-gal4 , also causes the loss of progenitor cells ( Figure 5C–C′ ) , thereby confirming that bip1 knock-down in differentiating cells is key to its associated phenotype . Additionally , the progenitor cell marker dome-MESO-lacZ ( Hombria et al . , 2005; Krzemien et al . , 2007 ) is strongly reduced relative to control lymph glands ( Figure 5D–D′ ) in the bip1 RNAi ( Hml-gal4 ) background . This result confirms that progenitor cells fail to be maintained in bip1 RNAi lymph glands , rather than ectopically upregulating Pxn and Hml-gal4 expression . 10 . 7554/eLife . 03626 . 010Figure 5 . Validation of the bip1 RNAi phenotype . ( A ) Quantitative RT-PCR demonstrates that bip1 is expressed in the lymph gland and that the RNAi line NIG 7574R-1 targeting bip1 indeed reduces bip1 transcript level when expressed using Hml-gal4 . Hml-gal4 expresses GFP throughout the primary lobes in bip1 RNAi lymph glands ( Hml > bip1-i , B ) , and this phenotype is suppressed by overexpression of bip1 ( B′ ) , restoring both the cortical and the medullary zones . ( C–C′ ) Expression of bip1 RNAi using Pxn-gal4 phenocopies obtained with Hml-gal4 , further supporting a cell-type-specific function of bip1 . Expression of the progenitor cell marker dome-MESO-lacZ ( D ) is strongly reduced in bip1 RNAi lymph glands ( D′ ) , demonstrating that the gain in differentiation markers is due to the loss of progenitor cells that normally express dome-MESO-lacZ . RNAi knock down of RpS8 , encoding a putative Bip1-interacting protein , causes the expansion of Pxn and Hml-gal4 expression throughout the lymph gland ( E–E′ ) , similar to that observed upon the loss of bip1 . DOI: http://dx . doi . org/10 . 7554/eLife . 03626 . 010 Several ribosomal components , including Ribosomal protein S8 ( RpS8 ) , have been shown to associate with chromatin at active transcription sites and to associate with nascent transcripts to form ribonucleoprotein complexes ( Brogna et al . , 2002 ) . Interestingly , RpS8 has also been identified in genomic-scale yeast two-hybrid analyses as a Bip1-interacting protein ( Giot et al . , 2003; Formstecher et al . , 2005; Stark et al . , 2006 ) , which suggests that Bip1 and RpS8 may function together in vivo to regulate gene expression . Consistent with this idea , RNAi knockdown of RpS8 also causes the expanded expression of both Pxn and Hml-gal4 UAS-GFP expression throughout the lymph gland primary lobes ( Figure 5E–E′ ) . This result reflects a specific function of RpS8 in these cells because knockdown directly in niche or progenitor cells ( using Antp-gal4 and dome-gal4 , respectively ) does not cause their loss to differentiation ( not shown ) . Thus , RpS8 RNAi effectively phenocopies bip1 RNAi , as both cause a loss of progenitor cells when knocked down in differentiating cells . Collectively , these data support a model in which Bip1 functions along with RpS8 in a protein complex within differentiating cells to maintain multipotent lymph gland progenitors at later stages of development , consistent with a potential function in the equilibrium signaling pathway . The progenitor maintenance function of Pvr signaling in differentiated cells requires the downstream function of the STAT transcriptional activator and the secreted enzyme ADGF-A ( Mondal et al . , 2011 ) , and , consistent with this relationship , overexpression of either activated STAT ( STATACT ) ( Ekas et al . , 2010 ) or ADGF-A in differentiating cells can suppress the Pvr loss-of-function phenotype ( Mondal et al . , 2011 ) . Likewise , we find that overexpression of STATACT or ADGF-A can suppress the bip1 RNAi phenotype ( Figure 6A–D′ ) . Furthermore , overexpression of Pvr also strongly suppresses the bip1 RNAi phenotype , returning lymph gland morphology and organization to essentially wild type ( Figure 6E–E′ ) . By contrast , overexpression of bip1 does not suppress the Pvr RNAi phenotype ( Hml-gal4 UAS-Pvr RNAi UAS-bip1LA645; not shown ) . Collectively , these results place bip1 function genetically upstream of Pvr and other equilibrium signaling components in lymph gland progenitor maintenance by differentiating cells . 10 . 7554/eLife . 03626 . 011Figure 6 . bip1 , RpS8 , and Nup98 control Pvr expression in the lymph gland . Expression of bip1 RNAi in differentiating cells ( Hml-gal4 or Hml> , B–B′ ) causes expansion of both Pxn and Hml-gal4 UAS-GFP throughout the lymph gland , as compared to controls ( A–A′ ) . Misexpression of either activated STAT ( STATACT , C–C′ ) , ADGF-A ( D–D′ ) , or Pvr ( E–E′ ) partially ( in the case of STAT activation or ADGF-A overexpression ) or fully ( in case of Pvr overexpression ) suppresses this bip1 phenotype , suggesting that bip1 functions upstream of these genes . Expression of Pvr in control third-instar lymph glands ( F–F′ ) and mid-second instar ( 40 hr post-hatching , G–G′ ) . Reduced expression of Pvr is already apparent in bip1 RNAi lymph glands by 40 hr ( H–H′ ) , and this loss is even stronger in homozygous animals expressing higher levels of RNAi ( I–I′ ) ; increased differentiation , based upon Hml-gal4 UAS-GFP expression , is also apparent ( I ) . Strong suppression of Pvr is also observed in homozygous bip1 RNAi lymph glands ( J–J′ ) . RNAi knockdown of RpS8 also causes differentiation and the loss of Pvr expression ( K–K′ ) . Likewise , RNAi knockdown of Nup98 also causes differentiation and the loss of Pvr expression ( L–L′ ) . ( M ) Control background ( Hml-gal4/+ ) showing normal expression of the differentiation marker Pxn in the cortical zone of the lymph gland . Progenitor cells in the MZ region are easily discerned by their lack of Pxn expression . By contrast , few progenitor cells ( Pxn-negative cells ) are observed in lymph glands when single-copy loss-of-function mutations of Pvr and Nup98 ( PvrC2195/+; Nup98Df ( 3R ) mbc-R1/+ ) are combined ( M′ ) , further indicating the close interaction between these genes . The middle-third ( confocal z-stack ) of the primary lobe is shown . Misexpression of bip1 in this background is sufficient to suppress these phenotypes and restore Pvr expression to the lymph gland ( N–N′ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03626 . 01110 . 7554/eLife . 03626 . 012Figure 6—figure supplement 1 . bip1 and RpS8 are required for normal Pvr transcript levels . Quantitative RT-PCR analysis demonstrating that RNAi knockdown of bip1 in lymph glands ( A ) and circulating cells ( B ) using Hml-gal4 causes a reduction in Pvr transcript levels to approximately 70% and 34% of the control ( Hml-gal4 only ) level , respectively . RpS8 RNAi using Hml-gal4 reduces Pvr to approximately 32% of the control level ( C , circulating cells ) . Note that Hml-gal4 is only expressed in a subset of cells in the lymph gland but in the vast majority of cells in circulation . DOI: http://dx . doi . org/10 . 7554/eLife . 03626 . 01210 . 7554/eLife . 03626 . 013Figure 6—figure supplement 2 . Pvr expression is regulated autonomously by bip1 , Nup98 , and RpS8 within the lymph gland . Mock FLP-out Gal4-expressing clones ( GFP-positive cells , green ) show no reduction in Pvr expression ( red , A–A′ ) , whereas similar clones expressing Pvr RNAi show a very strong reduction in Pvr protein expression ( B–B′ ) . FLP-out Gal4-expressing clones expressing either bip1 RNAi ( C–C′ ) or Nup98 RNAi ( D–D′ ) also reduce Pvr protein levels , with bip1 RNAi exhibiting somewhat stronger effects . FLP-out clones were made using Hand-gal4 UAS-FLP Ay-gal4 UAS-GFP at 25°C . Restricting the RNAi knockdown of bip1 ( F ) , Nup98 ( G ) , or RpS8 ( H ) to circulating cells with srpHemo-gal4 ( E ) shows no effect on lymph gland Pvr levels . DOI: http://dx . doi . org/10 . 7554/eLife . 03626 . 01310 . 7554/eLife . 03626 . 014Figure 6—figure supplement 3 . Loss of the nucleoporin Sec13 by RNAi neither causes a differentiation phenotype within the lymph gland nor the loss of Pvr expression . ( A ) Hml-gal4 control showing normal cortical zone expression ( UAS-GFP ) . ( A′ ) Pvr is expressed normally throughout the primary lobe in control lymph gland in ( A ) . ( B–B′ ) Loss of Sec13 by RNAi ( Hml-gal4 UAS-Sec13-i ) does not cause increased differentiation ( GFP expression ) or the loss of Pvr expression ( red ) . DNA , blue . DOI: http://dx . doi . org/10 . 7554/eLife . 03626 . 01410 . 7554/eLife . 03626 . 015Figure 6—figure supplement 4 . Loss of Pvr expression is not a common feature of highly differentiated lymph glands . The lymph gland marker P1 is normally restricted to the differentiated cells of the cortical zone . However , in collier1 ( col1 ) mutants that lack niche signaling , P1 expression ( A , green ) is observed throughout , indicating strong differentiation and a lack of progenitor cells . In this background , Pvr is expressed at normal levels ( A′ , compare with cortical zone levels in Figure 6F′ ) , indicating that loss of Pvr is not a general feature of highly differentiated lymph glands . DOI: http://dx . doi . org/10 . 7554/eLife . 03626 . 01510 . 7554/eLife . 03626 . 016Figure 6—figure supplement 5 . Overexpression of bip1 , Nup98 , and RpS8 , and RNAi knockdown of other nucleoporins does not affect Pvr levels in the lymph gland . Compared with control levels ( A , Hml-gal4 ) , overexpression of bip1 ( B , UAS-bip1LA645 ) , Nup98 ( C , UAS-Nup98 [Parrott et al . , 2011] ) , or RpS8 ( D , UAS-RpS8DP01446 [Staudt t al . , 2005] ) does not significantly affect Pvr levels ( red ) . Likewise , RNAi knockdown of Nup154 ( F ) , Nup214 ( G ) , or Nup358 ( H ) ( all with Hml-gal4 ) does not significantly alter Pvr level compared to controls ( E ) , further supporting the specific function of Nup98 in Pvr regulation . DOI: http://dx . doi . org/10 . 7554/eLife . 03626 . 016 The suppression of the bip1 RNAi phenotype by overexpression of Pvr suggested that bip1 may positively control Pvr expression during normal development . Indeed , a reduction in Pvr protein expression within the lymph gland is observed in the bip1 RNAi background by mid-second instar , soon after differentiation begins ( ∼40 hr post-hatching; Figure 6F–J′ ) . This reduction in Pvr expression is even stronger ( along with significantly increased differentiation , based upon Hml-gal4 expression ) at the same developmental time point in larvae having two copies of Hml-gal4 UAS-bip1 RNAi ( compare Figure 6I′ with Figure 6H′ ) , further supporting the model that bip1 RNAi causes the loss of Pvr expression . By the late third instar ( when the bip1 RNAi differentiation phenotype is most apparent ) , Pvr protein levels in the lymph gland remain strongly reduced ( Figure 6J–J′ ) . Knockdown of bip1 function in lymph glands by Hml-gal4-mediated RNAi reduces lymph gland Pvr transcript levels to approximately 70% of control levels ( assessed by qRT-PCR; Figure 6—figure supplement 1A ) , consistent with the observed loss of Pvr protein ( Figure 6H′–J′ ) . However , because not all lymph gland cells express Hml ( Hml-gal4 ) , the actual reduction of Pvr transcript levels in cells expressing bip1 RNAi is likely to be greater than the observed total reduction . In support of this idea , bip1 RNAi in circulating blood cells ( where greater than 90% express Hml-gal4 ) reduces Pvr transcript level to approximately 35% of the control level ( Figure 6—figure supplement 1B ) . Expression of bip1 RNAi in FLP-out Gal4-expressing cell clones made exclusively in the lymph gland strongly reduces Pvr levels compared to nearby cells not expressing RNAi as well as to mock clones ( Figure 6—figure supplement 2A–C ) , consistent with the autonomous regulation of Pvr by bip1 in the lymph gland . Furthermore , bip1 RNAi expression with srpHemo-gal4 ( Bruckner et al . , 2004 ) , which expresses in a large fraction of circulating cells but in few or no cells within the lymph gland , does not reduce lymph gland Pvr levels ( Figure 6—figure supplement 2E–H ) . Collectively , these data indicate that bip1 is required for proper Pvr protein expression , and therefore proper equilibrium signaling , within the developing lymph gland . As described above , RpS8 is a putative Bip1-interacting protein in vivo and RpS8 RNAi in differentiating lymph gland cells , like bip1 RNAi , causes the loss of progenitor cells ( Figure 5F–F′ ) . This effect is likely due to the loss of equilibrium signaling during development since RpS8 RNAi also reduces Pvr protein expression in the lymph gland ( Figure 6K–K′ ) . Knockdown of RpS8 by RNAi , as with knockdown of bip1 , also reduces Pvr transcript levels ( Figure 6—figure supplement 1C ) . Interestingly , a Drosophila RNAi screen using the blood-related S2 cell line previously identified both Pvr and RpS8 as regulators of cell size and division ( Sims et al . , 2009 ) . Although the relationship between Pvr and RpS8 was not explored , their results as well as ours are consistent with RpS8 having a regulatory role in Pvr expression in blood cells . In addition to bip1 , the screen described here identified Nup98 as a potential equilibrium signaling component because its knockdown in differentiating cells specifically causes a loss of progenitors cells ( Figure 3T and Figure 4F–F′′ ) . Although Nup98 is widely known as a general component of the nuclear pore complex , recent work has demonstrated that Nup98 and other nuclear pore components such as Sec13 and Nup88 , can regulate gene expression through the binding of target promoters ( Capelson et al . , 2010; Kalverda et al . , 2010; Liang et al . , 2013 ) . Moreover , chromatin immunoprecipitation experiments identified bip1 , RpS8 , and the equilibrium signaling genes Pvr and STAT ( STAT92E ) as in vivo Nup98 regulatory targets ( Capelson et al . , 2010 ) . Consistent with a function in regulation of equilibrium signaling genes , Nup98 knockdown specifically in differentiating cells of lymph glands causes a strong reduction in Pvr expression ( Figure 6L–L′ ) . By contrast , RNAi knockdown of the nucleoporin Sec13 in differentiating cells has no effect on the maintenance of progenitor cells or Pvr expression ( Figure 6—figure supplement 3 ) underscoring the specific role of Nup98 in Pvr expression control . Furthermore , the close genetic relationship between Nup98 and Pvr is illustrated by the fact that single-copy loss of these genes in combination causes extensive loss of progenitor cells to differentiation ( Figure 6M–M′ ) . Interestingly , overexpression of bip1 in Nup98 RNAi lymph glands ( Hml-gal4 UAS-Nup98 RNAi UAS-bip1LA645 ) is sufficient to restore Pvr protein expression and to suppress the loss of progenitors to differentiation ( based upon lymph gland morphology and Hml-gal4 expression; Figure 6N–N′ ) . As has been shown , knockdown of bip1 , Nup98 , or RpS8 in differentiating cells each causes a strong reduction in Pvr expression in the lymph gland . Our interpretation of this common phenotype is that each gene works in the equilibrium signaling pathway to control Pvr expression , although an alternative hypothesis is that the loss of Pvr expression is a common feature of highly differentiated lymph glands and is not specifically related to the function of these genes . To test this , Pvr expression was examined in collier ( col ) mutant lymph glands , which lack niche signaling and are strongly differentiated by late larval stages ( Crozatier et al . , 2004; Mandal et al . , 2007 ) , and was found to be normal ( Figure 6—figure supplement 4 , compare with Pvr expression in wild-type cortical zone differentiating cells in Figure 6F′ ) . Thus , Pvr requires bip1 , RpS8 , and Nup98 for proper developmental expression in the lymph gland . Several genetic screens , including overexpression and enhancer/suppressor screens of mutant or tumor phenotypes , have been conducted in the fly hematopoietic system ( Milchanowski et al . , 2004; Zettervall et al . , 2004; Stofanko et al . , 2008; Avet-Rochex et al . , 2010; Tan et al . , 2012; Tokusumi et al . , 2012 ) ; however , the screen described here represents the first loss-of-function screen targeting normal developmental mechanisms throughout the lymph gland . This was accomplished with the development and use of the pan-lymph gland expression tool HHLT-gal4 to drive UAS-mediated RNAi , which identified 20 different candidate genes that cause a loss of progenitor cells when knocked down within the lymph gland . From subsequent analyses using lymph gland zone-restricted Gal4 driver lines , we arrive at a model ( Figure 7 ) in which Bip1 , RpS8 , and Nup98 are required in differentiating blood cells upstream of Pvr to control its expression and function in the equilibrium signaling pathway that maintains blood progenitors within the lymph gland . Future analyses will be required to identify additional components of this important signaling pathway and to provide more information about how equilibrium signaling interacts with other pathways in the control of blood cell progenitor maintenance , cell fate specification , and proliferation . 10 . 7554/eLife . 03626 . 017Figure 7 . Schematic of the equilibrium signaling pathway demonstrating the proposed roles of Bip1 , RpS8 , and Nup98 in controlling Pvr . Bip1 , RpS8 , and Nup98 are independently required for the expression of Pvr ( direct arrows ) . Rescue of endogenous Pvr expression by misexpression of bip1 in the Nup98 RNAi background indicates that bip1 functions genetically downstream of Nup98 ( dashed arrow ) in the control of Pvr expression . Bip1 and RpS8 may work together in a complex ( dashed line ) to control Pvr expression in vivo . These components collectively comprise the known equilibrium signaling pathway working within the lymph gland to promote progenitor cell maintenance , along with the previously known Hh niche signaling mechanism . DOI: http://dx . doi . org/10 . 7554/eLife . 03626 . 017 The Pvr receptor , with its numerous developmental roles , is arguably one of the most important members of the Drosophila RTK family , yet most of what is known about Pvr stems from analyses of how it works in the context of intracellular signaling . Little is known about how Pvr gene or protein expression is regulated . Importantly , the work described here sheds new light upon this issue by demonstrating a role for bip1 , RpS8 , and Nup98 in the regulation of Pvr expression . Our data and that of others suggest that this regulation of Pvr is likely taking place at the gene level , although other mechanisms are also possible . Ribosomes are required for protein translation , however specific ribosomal components or subunits may selectively stabilize transcripts and/or mediate preferential translation ( Xue and Barna , 2012 ) , while nucleoporins control both nuclear entry of regulatory proteins and the exit of mRNAs to the cytoplasm , and specific subcomponents are known to exhibit differential functions in this regard ( Strambio-De-Castillia et al . , 2010 ) . Thus , RpS8 and Nup98 may selectively affect Pvr expression post-transcriptionally through transcript stabilization , transport , and translation . Although the specific mechanisms of molecular control of Pvr expression by bip1 , RpS8 , and Nup98 remain to be determined , their function is clearly critical in mediating proper equilibrium signaling and , therefore , proper blood progenitor maintenance within the lymph gland . The finding that bip1 regulates Pvr expression in the context of hematopoietic equilibrium signaling represents the first functional association for bip1 in Drosophila . The predicted Bip1 protein exhibits only one recognizable structural sequence , namely a THAP domain that contains a putative DNA-binding zinc finger motif . Our results suggest that Bip1 behaves as a positive regulator of Pvr transcription , but whether this occurs directly through Bip1 interaction with the Pvr locus will require further investigation . Understanding how progenitor cell maintenance and homeostasis is controlled over developmental time is crucial for understanding normal cellular and tissue dynamics , especially in the context of ageing or disease . The identification of Bip1 and Nup98 as regulators of hematopoietic progenitors in Drosophila may be indicative of important conserved functions of related proteins within the vertebrate blood lineages similar to what has been shown previously for GATA , FOG , and RUNX factors ( Waltzer et al . , 2010 ) . THAP-domain proteins are conserved across species and have been reported to have a variety of important functions in mammalian systems , including maintenance of murine embryonic stem cell pluripotency ( Cayrol et al . , 2007; Dejosez et al . , 2008 , 2010 ) . What role , if any , THAP-domain proteins have in vertebrate blood progenitor maintenance ( or hematopoiesis in general ) remains to be established . Likewise , Nup98 has not been implicated in any normal hematopoietic role despite being a well-studied protein in other contexts . With regard to the diseased state , mutations in the human THAP1 gene have been associated with dystonia ( Fuchs et al . , 2009; Paisan-Ruiz et al . , 2009; Kaiser et al . , 2010; Mazars et al . , 2010 ) , a neuromuscular disorder that causes repetitive , involuntary muscular contraction , and THAP1/Par4 protein complexes have been shown to promote apoptosis in leukemic blood cells in various experimental contexts in vitro ( Lu et al . , 2013; Zhang et al . , 2014 ) . Chromosomal translocations that generate Nup98 fusion proteins have been implicated in numerous human myelodysplastic syndromes and leukemias ( Nishiyama et al . , 1999; Ahuja et al . , 2001; Lin et al . , 2005; Nakamura , 2005; van Zutven et al . , 2006; Slape et al . , 2008; Kaltenbach et al . , 2010; Murayama et al . , 2013 ) , further underscoring the need to explore Nup98 function in the hematopoietic system . Therefore , the study of bip1 and Nup98 in Drosophila , a powerful molecular genetic system , will likely be of benefit to understand the function of related vertebrate genes in normal and disease contexts . Misexpression P{Mae-UAS . 6 . 11} inserts ( LA lines ) were obtained from John Merriam , UCLA ( Los Angeles , California ) . UAS-RNAi lines were obtained from the Vienna Drosophila RNAi Center ( VDRC , Vienna , Austria ) , the National Institute of Genetics ( NIG , Kyoto , Japan ) , and the Bloomington Drosophila Stock Center ( TRiP lines , BDSC , Bloomington , Indiana ) . The lines UAS-FLP . JD1 , UAS-2XEGFP , P{GAL4-Act5C ( FRT . CD2 ) . P}S , UAS-human RafACT , Df ( 3R ) mbc-R1 , UAS-RpS8PD01446 , and w1118 ( BDSC 5905 ) were from the BDSC . Pvrc02195 was from Exelixis ( available from BDSC , obtained from D Montell ) . HmlΔ-gal4 UAS-2XEGFP ( S Sinenko ) , Antp-gal4/TM6B Tb ( S Cohen ) , P{ubi-gal80 ts}10; Antp-gal4/TM6B Tb ( this lab ) , domeless-gal4 UAS-2XEYFP/FM7i ( this lab ) , UAS-DAlkACT ( R Palmer ) , dome-MESO-lacZ ( S Brown ) , Pxn-gal4 ( M Galko ) , UAS-STATACT ( E Bach ) , UAS-ADGF-A ( T Dolezal ) , collier1; P ( col5-cDNA ) /CyO-TM6B , Tb ( M Crozatier ) , srpHemo-gal4 ( K Brückner ) , and Hand-gal4 ( Z Han ) have been previously described . Second chromosome inserts of Hand-gal4 , HmlΔ-gal4 , UAS-FLP . JD1 , and UAS-2XEGFP were recombined onto a single chromosome and placed with P{GAL4-Act5C ( FRT . CD2 ) . P}S on Chromosome 3 . Because Gal4 reporter lines with specific , pan-lymph gland expression are unknown , we took advantage of a FLP-out lineage tracing approach that we have used previously to perpetually mark lymph gland cells ( Jung et al . , 2005; Evans et al . , 2009 ) . The Hand-gal4 reporter reflects the expression of the Hand gene , which is expressed in the cardiogenic mesoderm , from which the lymph gland is derived . Within the lymph gland , Hand-gal4 is expressed from the late embryo through the first larval instar but then is downregulated ( Han and Olson , 2005 ) . Using Hand-gal4 in conjunction with UAS-FLP and a FLP-out Gal4-expressing line ( P{GAL4-Act5C ( FRT . CD2 ) . P}S ) ( Pignoni and Zipursky , 1997 ) , lymph gland cells are perpetually with EGFP throughout all subsequent developmental stages . To express EGFP in circulating cells , we used Hemolectin-gal4 ( HmlΔ-gal4 ) ( Sinenko and Mathey-Prevot , 2004 ) , which is specific to mature blood cells both in circulation and in the lymph gland cortical zone ( Jung et al . , 2005 ) . HHLT-gal4 expression is easily detectable in lymph glands and circulating cells of whole animals throughout larval development . Due to the embryonic activity of Hand-gal4 , HHLT-gal4 also labels dorsal vessel cardioblasts and pericardial cells , although by late larval stages the expression of EGFP in the former is almost undetectable . HHLT-gal4 virgins were crossed to males from individual LA lines , RNAi lines , or w1118 as a control . All crosses were reared at 29°C to maximize Gal4 activity . Wandering third-instar larvae from control and experimental crosses were collected , washed with water , and placed in glass spot wells ( Fisher ) on ice to minimize movement . Animals were scored visually using a Zeiss Axioskop 2 compound fluorescence microscope . Non-screen images of HHLT > GFP larvae were collected with a Zeiss SteREO Lumar fluorescence microscope . Images were collected using either an AxioCam HRc or HRm camera with AxioVision software . Lymph glands were dissected and processed as previously described ( Jung et al . , 2005 ) . Briefly , lymph glands were dissecting from third-instar larvae in 1× PBS , fixed in 4% formaldehyde/1× PBS for 30 min , washed three times in 1×PBS with 0 . 4% Triton-X ( 1× PBST ) for 15 min each , blocked in 10% normal goat serum/1× PBST for 30 min , followed by incubation with primary antibodies in block . Primary antibodies were incubated with tissue overnight at 4°C and then washed three times in 1× PBST for 15 min each , reblocked for 15 min , followed by incubation with secondary antibodies for 3 hr at room temperature . Samples were washed three times in 1× PBST , with TO-PRO-3 iodide ( diluted 1:1000; Invitrogen , Carlsbad , California ) added to the last wash to stain nuclei . Samples were washed briefly with 1× PBS to remove excess TO-PRO-3 and detergent prior to mounting on glass slides in VectaShield ( Vector Laboratories , Burlingame , California ) . Mouse anti-Peroxidasin was a kind gift from John and Lisa Fessler ( UCLA ) and was used at 1:1500 dilution . Rat anti-Pvr was a kind gift from Benny Shilo and was used at 1:400 dilution . Secondary Cy3-labeled antibodies were obtained from Jackson ImmunoResearch Laboratories Inc . ( West Grove , Pennsylvania ) and used at 1:500 dilution . Lymph glands from 50 third-instar larvae were isolated by dissection . For fat body analysis , ten third-instar larvae were used . RNA was extracted from these tissues with the RNeasy mini kit ( Qiagen , Germantown , Maryland ) . Relative quantitative RT-PCR ( comparative CT ) was performed using Power SYBR Green RNA-to-CT 1-step kit ( Applied Biosystems , Carlsbad , California ) and a StepOne Real-Time PCR detection thermal cycler ( Applied Biosystems ) using primers specific for Pvr , bip1 , and rp49 . Primer sequences are: Pvr ( forward ) , 5′-TTCGGATTTCGATGGTGAAT-3′; Pvr ( reverse ) , 5′-CGGACACTAAGCTGGTCGAT-3′; bip1 ( forward ) , 5′-CGGAGTTTATGGACAGCACA-3′; bip1 ( reverse ) , 5′-CCTTAGCAGGAGGAGGAGGT-3′; rp49 ( forward ) , 5′-GCTAAGCTGTCGCACAAATG-3′; rp49 ( reverse ) , 5′-GTTCGATCCGTAACCGATGT-3′ .	Progenitor cells are cells that can either multiply to make new copies of themselves or mature into different specialized cell types—such as blood cells . In the fruit fly Drosophila , new blood cells are formed in several different locations , including in an organ called the lymph gland . In 2011 , researchers found that the fate of blood progenitor cells within the lymph gland is controlled by signals from two nearby sources—one from specialized , supportive ( ‘niche’ ) cells and the other from maturing blood cells . The signal from the maturing blood cells ensures that the relative amounts of progenitor and maturing blood cells are kept in the right balance . As a result , this signaling process has been called ‘equilibrium signaling’ . Questions remain as to how equilibrium signaling is regulated , and how it interacts with signals from the niche . To investigate this , Mondal et al . —including some of the researchers involved in the 2011 work—used various genetic techniques to create Drosophila larvae in which the tissues that become blood cells are made visible with fluorescent proteins . This meant that these tissues could be examined in live , whole animals by using a microscope . Mondal et al . then searched for the Drosophila genes involved in generating new blood cells in the lymph gland—particularly those involved in equilibrium signaling . This was done by switching on and off hundreds of genes , one by one , in the lymph gland , and any genes that caused changes to the generation of new blood cells were then investigated further . Following these investigations , Mondal et al . focused on three genes—and when each of these genes was switched off in maturing blood cells , the result was that fewer progenitor cells remained in the lymph gland . This effect was not seen when the genes were switched off in the progenitor or the niche cells , which suggested that the genes are likely to be components of the equilibrium signaling pathway . Switching off these genes in maturing blood cells also dramatically reduced the levels of a protein called Pvr , a key equilibrium signaling protein known from the 2011 study and an important player in blood cell development in several species . How the newly identified genes actually control Pvr protein levels to maintain proper equilibrium signaling in the lymph gland remains to be explored . However , this work provides a basis for investigating the role of related genes in blood cell development in vertebrate systems , namely humans .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2014	Pvr expression regulators in equilibrium signal control and maintenance of Drosophila blood progenitors
Human cytomegalovirus ( hCMV ) immediate early 1 ( IE1 ) protein associates with condensed chromatin of the host cell during mitosis . We have determined the structure of the chromatin-tethering domain ( CTD ) of IE1 bound to the nucleosome core particle , and discovered that the specific interaction between IE1-CTD and the H2A-H2B acidic patch impairs the compaction of higher-order chromatin structure . Our results suggest that IE1 loosens up the folding of host chromatin during hCMV infections . In eukaryotes , nuclear DNA is highly packaged into chromatin by histones . The nucleosome , the basic repeating unit of chromatin , typically assembles 146–147 bp of DNA wrapped around a histone octamer consisting of two copies each of H3 , H4 , H2A and H2B ( Luger et al . , 1997a ) . Nucleosomes connected by linker DNA form a 10-nm array resembling “beads-on-a-string” . The binding of linker histone further compacts the linear array into a more condensed 30-nm chromatin fiber . The interaction between the N-terminal tail of histone H4 and a specific surface region on the neighboring nucleosome termed the“acidic patch” plays crucial roles for the formation of 30-nm chromatin fiber ( Dorigo et al . , 2004; Luger et al . , 1997a; Song et al . , 2014 ) . The acidic patch is formed by a number of negatively charged residues of H2A and H2B , including Glu56 , Glu61 , Glu64 , Asp90 , Glu91 and Glu92 of H2A , and Glu102 and Glu110 of H2B ( Luger et al . , 1997a ) . Several nucleosome-binding proteins have been shown to specifically interact with the acidic patch . Therefore , it has been hypothesized that these proteins may play a role in regulating the higher-order chromatin structure by competing with the H4 N-terminal tails for binding to the acidic patch of the nucleosome ( Kalashnikova et al . , 2013; McGinty and Tan , 2015 ) . The 72-kDa immediately early 1 ( IE1 ) protein of human cytomegalovirus ( hCMV ) plays critical roles in the viral early gene expression and DNA replication ( Gawn and Greaves , 2002; Mocarski et al . , 1996 ) . In addition , IE1 has been long known to associate with condensed host chromatin during mitosis ( Ahn et al . , 1998; Dimitropoulou et al . , 2010; Huh et al . , 2008; Krauss et al . , 2009; Lafemina et al . , 1989; Nevels et al . , 2004; Reinhardt et al . , 2005; Shin et al . , 2012; Wilkinson et al . , 1998 ) . The chromatin-tethering domain ( CTD ) located at the very C-terminal end of IE1 ( a . a 476–491 ) is responsible for the association with the mitotic chromosome , specifically through the acidic patch of the nucleosome ( Mucke et al . , 2014; Reinhardt et al . , 2005 ) . However , the molecular mechanism underlying the association and the impact on the structure and function of host chromatin by IE1 remain to be determined . Here we provide an analysis of the structural basis for the interaction between the CTD of IE1 ( IE1-CTD ) and the nucleosome core particle ( NCP ) and explore the impact of their binding on the higher-order structure of chromatin . The complex of NCP with an IE1-CTD peptide ( a . a . 476–491 ) was obtained by soaking the peptide into preformed NCP crystals , and a 2 . 8 Å structure was solved by molecular replacement . The structure shows one molecule of IE1-CTD bound to the NCP ( Figure 1A; Figure 1—figure supplement 1 ) . The presence of only one IE1-CTD peptide in the complex structure is due to the availability of only one side of the NCP surface in the preformed crystal lattice , as in the case of NCP in complex with the latency-associated antigen ( LANA ) of Kaposi’s sarcoma-associated herpes virus ( KSHV ) ( Barbera et al . , 2006 ) . The IE1-CTD peptide adopts an extended , v-shaped conformation with a short α-helix at its C-terminus . In the complex structure , IE1-CTD is well positioned into the acidic patch of the nucleosome formed by H2A and H2B ( Figure 1A and B ) . IE1-CTD contacts histone H2B at two distinct sites , the C-terminal portion of α1 and the N-terminal half of αC , through van der Waals interaction and intermolecular hydrogen bonds via its mainchain groups . Specifically , the amide and carbonyl groups of Thr480 make hydrogen bonds with the mainchain carbonyl and the sidechain amide groups of Gln44 of H2B ( amino acid residue numbering following that in the reference of Luger et al . , 1997a ) ; the amide and carbonyl groups of Val484 bond the carboxylate group of Glu110 and the nitrogen atom of the imidazole ring of His106 , respectively ( Figure 1B ) . IE1-CTD contacts histone H2A via a number of sidechain contacts . His481 makes one hydrogen bond with Glu56 of H2A; Thr485 and Ser487 each makes one hydrogen bond with Glu64 of H2A; and Arg486 makes hydrogen bonds with Glu61 on α2 , as well as with Asp90 and Glu92 located on αC of histone H2A . In addition , Met483 of IE1-CTD is placed in a hydrophobic pocket consisting of Leu23 , and the aliphatic portion of Glu56 and Tyr57 . 10 . 7554/eLife . 11911 . 003Figure 1 . Structure of the IE1-CTD–NCP complex . ( A ) Location of the IE1-CTD binding site on the acidic patch of the nucleosomal surface . The histone octamer is shown as a surface representation colored according to electrostatic potential distribution ( positive , blue; neutral , white; negative , red ) . DNA is shown as a cartoon colored white , and IE1-CTD is shown as a stick model . ( B ) A detailed view of the interaction between IE1-CTD and NCP . Histones H2A and H2B are shown in a ribbon representation superimposed with selected residues ( in sticks ) involved in interaction with IE1-CTD . Dashed lines indicate hydrogen bonds . An enlarged view of the region surrounding His481 of IE1-CTD is shown in an inset at the bottom of the figure . ( C ) Superposition of IE1-CTD and LANA . Both peptides are shown as a ribbon representation superimposed with sidechains ( IE1-CTD , green; LANA , blue ) . ( D ) Structure-based alignment of IE1-CTD and LANA sequences . Residues colored in red are involved in similar interactions with the histones , and the two residues colored in magenta are engaged in IE1-CTD-specific interactions with histone H2A . DOI: http://dx . doi . org/10 . 7554/eLife . 11911 . 00310 . 7554/eLife . 11911 . 004Figure 1—figure supplement 1 . An omit electron density map of the bound IE1-CTD . A stereo view of the simulated annealing ( Fo-Fc ) omit map showing the presence of IE1-CTD . The map is contoured at 2 . 5 σ level . A stick model of IE1-CTD is superimposed . DOI: http://dx . doi . org/10 . 7554/eLife . 11911 . 004 The nucleosomal acidic patch is well known for hosting the binding of a number of proteins ( Armache et al . , 2011; Arnaudo et al . , 2013; Barbera et al . , 2006; Kato et al . , 2013; Makde et al . , 2010; McGinty et al . , 2014; Wang et al . , 2013; Yang et al . , 2013 ) . Most closely related to hCMV IE1 is LANA of KSHV ( Barbera et al . , 2006 ) . The N-terminal CTD of LANA forms a hairpin-like structure that pokes into the acidic patch of the nucleosome ( Figure 1C ) . The distinctly folded IE1-CTD and LANA-CTD share certain common features of nucleosome binding . Structural comparison reveals that three IE1-CTD residues , Met483 , Arg486 and Ser487 , occupy the same regions of the acidic patch and engage the same sets of histone residues as the corresponding LANA residues in binding the nucleosome ( Figure 1C , D , 2A and B ) . In particular , Arg486 interacts with the acidic patch in a manner commonly found in the structures of protein-nucleosome complexes known to date ( Figure 2A–F ) . A careful examination of the acidic patch reveals that it can be divided into three adjoining ligand-binding zones ( Figure 2C ) . Zone I is formed by Glu61 , Leu65 , Asp90 and Glu92 of histone H2A , and Glu102 and Leu103 of histone H2B; Zone II is formed by histone H2A residues Tyr57 , Ala60 , Glu61 and Glu64 , and the latter serves as a ridge separating zone I and II; and zone III is formed by Glu56 and Ala60 of H2A , and Val41 , Gln44 and Glu110 of histone H2B . The binding of an arginine in zone I is conserved among all protein-NCP complexes known thus far . Thr485 of IE1 and Leu8 of LANA are bound in zone II , which is unoccupied in other NCP complexes ( Figure 2 ) . Main differences accounting for the specific interaction between IE1-CTD and NCP appear to reside in His481 , which is bound in zone III and makes a hydrogen bond with Glu56 of H2A , and Thr485 , which is bound to zone II and forms a hydrogen bond with Glu64 of H2A . And finally , the binding of the N-terminal segment of IE1-CTD spanning residues 476–480 to α1 of H2B is unique to IE1 ( Figure 1B ) . 10 . 7554/eLife . 11911 . 005Figure 2 . Comparison of protein binding modes to the acidic patch of NCP . All NCP-binding peptides or protein segments , shown in a stick model superimposed onto a cartoon representation of the backbone , were superimposed onto the structure of NCP , shown in a surface representation colored according to electrostatic potential , of the IE1-CTD complex based on alignment of NCP structures . ( A ) The binding of IE1-CTD to NCP . ( B ) LANA ( PDB id: 1ZLA ) . ( C ) RCC1 segment ( PDB id: 3MVD ) . The green ovals indicate distinct binding zones of the acidic patch . The protruding ridge at the junction between zone I and zone II is also labeled . ( D ) Sir3 ( PDB1d: 4KUD ) . ( E ) CENP-C ( PDB id: 4X23 ) . ( F ) PRC1-RING1B ( PDB id: 4R8P ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11911 . 005 To reveal the determinants for the binding specificity of IE1-CTD , we carried out structure-guided mutagenesis of IE1-CTD and histones and analyzed their interactions using isothermal titration calorimetry ( ITC ) . Wild-type IE1-CTD bound to recombinant human NCP with a dissociation constant ( KD ) of 0 . 4 μM , while its mutant lacking the N-terminal segment interacting with α1 of H2B ( Δ476–480 ) reduced the nucleosome-binding affinity to a KD of 12 μM ( Figure 3A and B ) . To put the binding affinities in perspective , an approximately five-fold weaker binding of full-length IE1 than IE1-CTD alone to NCP was observed ( Figure 3—figure supplement 1A and B ) , possibly caused by self-inhibitory effects of other domains in the full-length protein . In comparison , IE1 lacking the CTD showed no detectable binding to NCP ( Figure 3—figure supplement 1C ) . With IE1-CTD , the most severe reduction of binding affinity was seen in the H481A mutant , which has a KD of 43 . 4 μM ( Figure 3C ) , while loss of one hydrogen bond by substituting Thr485 with a valine brought the KD to 11 . 3 μM ( Figure 3D ) . To gain further insights into the ~hundred-fold reduction of binding affinity introduced by the H481A mutation , we measured the effect of histone H2A mutation on Glu56 , which interacts with His481 via hydrogen bonding and Met483 by hydrophobic/van der Waals interaction . ITC measurement shows that the NCP reconstituted with the E56R mutant of H2A , a charge-swap mutant , lowered the binding to IE1-CTD to a level beyond detection ( Figure 3E ) . This change of histone H2A also brought about a conspicuous reduction of the binding of LANA to NCP from a KD of 0 . 25 to 23 . 4 μM ( Figure 3F and G ) . LANA and IE1 share the methionine-mediated interaction with Glu56 in zone III of the acidic patch ( Figure 2A and B ) , and the greater compromise of the binding of IE1-CTD to the H2A-E56R NCP reflects the importance of the His481-Glu56 hydrogen bond in the IE1-NCP complex , as compared to the van der Waals interaction between Thr14 of LANA and Glu56 of histone H2A ( Figure 1B ) . 10 . 7554/eLife . 11911 . 006Figure 3 . ITC measurements of peptide-NCP binding affinities . ( A–G ) Raw data and fitting curves of the integrated data for the indicated peptides and NCPs are shown together with the derived KD values and fitting errors . DOI: http://dx . doi . org/10 . 7554/eLife . 11911 . 00610 . 7554/eLife . 11911 . 007Figure 3—figure supplement 1 . Binding of full-length IE1 to NCP . Bindings of full-length IE1 ( A ) , IE1-CTD ( B ) , and IE1 lacking CTD ( C ) are measured by ITC at 150 mM NaCl concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 11911 . 007 The acidic patch of the nucleosome has been implicated in mediating higher-order chromatin folding via interaction with the N-terminal tail of histone H4 ( Luger et al . , 1997a; Schalch et al . , 2005 ) . Our previous cryo-EM structure of 30-nm chromatin fiber reveals that N-terminal tails of histone H4 are involved in inter-nucleosomal contacts between the tetranucleosomal structural units through the acidic patches of adjacent nucleosomes ( Song et al . , 2014 ) . Since IE1-CTD is bound at the acidic patch , it is conceivable that IE1 binding may interfere with proper folding of the 30-nm chromatin fiber . To determine the extent by which the folding of chromatin fiber is affected by IE1 binding , we incubated IE1-CTD with the in vitro reconstituted 30-nm chromatin fiber assembled with an array of 12 tandem nucleosomes carrying repeats of 177 bp 601 DNA in the presence of linker histone H1 , and analyzed the sample by analytical ultracentrifugation in sedimentation velocity ( AUC ) ( Song et al . , 2014 ) . The nucleosomal arrays used for reconstituting 30-nm chromatin fiber were highly saturated and homogeneous , as examined by micrococcal nuclease ( MNase ) digestion and electron microscopy ( Figure 4—figure supplement 1A and B ) . For comparison , the same batch of nucleosomal array ( without H1 ) and 30-nm chromatin fiber ( with H1 ) were used in AUC analysis . AUC experiments showed that , in the absence of IE1-CTD , the nucleosomal array sedimented with a median sedimentation coefficient Save ( sedimentation coefficient at 50% boundary fraction ) of 36 ± 1 S , and the 30-nm chromatin fiber sedimented at 51 . 5 ± 0 . 6 S ( Figure 3A ) . In the presence of IE1-CTD , the 10-nm nucleosomal array was unaffected while Save of the 30-nm chromatin fiber shifted from 51 . 5 to 48 S , indicating that the binding of IE1-CTD made the chromatin fiber more loosely folded ( Figure 4A ) . It should be emphasized that the IE1-CTD-containing chromatin fiber represents an altered chromatin state different from both the extended 10-nm nucleosomal array and the folded 30-nm chromatin fiber . This chromatin-alteration property of IE1-CTD is shared by the full-length IE1 and fully depends on the presence of CTD ( Figure 4B and C ) . Further AUC analyses with IE1-CTD mutants showed that they essentially retained the ability to decondense the 30-nm chromatin fiber , possibly due to their incomplete loss of NCP-binding abilities ( Figure 4D , E and F ) . By contrast , chromatin fibers reconstituted with the E56R mutant of histone H2A , previously shown to be unable to interact with IE1-CTD , displayed no alteration of chromatin folding by IE1-CTD ( Figure 4G ) . These observations indicate that the binding of IE1-CTD at the acidic patch of the nucleosome modulates the higher-order structure of chromatin . It should be pointed out that not all acidic patch-binding proteins affect chromatin folding , as LANA does not alter the folding of 30-nm chromatin fiber in our AUC analysis ( Figure 4H ) . 10 . 7554/eLife . 11911 . 008Figure 4 . Influence of IE1-CTD on higher-order chromatin structure . ( A ) AUC analyses showing that IE1-CTD has little effect on the folding of the 10-nm nucleosomal array . Green and black data points represent that of 10-nm nucleosomal arrays in the absence and presence of IE1-CTD , respectively . In contrast , sedimentation profile of the 30-nm chromatin fiber reconstituted in the presence of linker histone H1 ( blue squares ) was shifted with the addition of IE1-CTD ( red dots ) . ( B ) Full-length IE1 shares the property of IE1-CTD in selectively altering the folding of the 30-nm chromatin fiber . ( C ) A truncation variant of IE1 lacking CTD ( IE1ΔC ) does not alter chromatin structure . ( D–F ) Indicated IE1-CTD mutants retain the ability to impact the folding of the 30-nm chromatin fiber . ( G ) An E56R mutant of histone H2A renders IE1-CTD ineffective in altering the structure of the 30-nm chromatin fiber . ( H ) LANA-CTD ( red dots ) does not affect the folding of the 30-nm chromatin fiber . Instead , the 10-nm nucleosome array appears to be slightly affected with the addition of LANA-CTD peptide . DOI: http://dx . doi . org/10 . 7554/eLife . 11911 . 00810 . 7554/eLife . 11911 . 009Figure 4—figure supplement 1 . Assessment of the quality of reconstituted nucleosomal array . ( A ) Nucleosomal arrays corresponding to ~1 μg DNA were cleaved with indicated amount of micrococcal nuclease ( MNase ) . ( B ) EM analysis of the reconstituted nucleosomal array ( Bar: 100 nm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11911 . 009 Our structural analysis revealed that IE1-CTD binds the acidic batch of NCP in a distinct manner . A careful analysis of the landscape of the acidic patch reveals that it can be divided into distinct binding zones that host specific amino acid residues . The conserved binding mode of an arginine in zone I of NCP is important for the association of all NCP-binding partners known to date , while other zones of the acidic patch are involved in differential binding of individual partners . For example , zone III of the acidic patch hosts the binding of His481 of IE1-CTD and Thr14 of LANA , respectively . These two residues interact with Glu56 of histone H2A differently in the structure: while the τ nitrogen of the imidazole ring of His481 makes a hydrogen bond with the carboxylate group of Glu56 of histone H2A at a distance of approximately 2 . 9 Å , Thr14 of LANA contacts Glu56 of H2A approximately 5 Å away via van der Waals interaction . This difference in peptide-NCP interactions perhaps account for our observations that an E56R mutant affected the binding of IE1-CTD more severely than that with LANA . For the purpose of dissecting the functions of individual NCP binding partners , one would ideally like to be able to isolate histone mutants that affect the binding of one protein but not others . The most promising such sites appear to lie in zone II and III , as they are less frequently occupied among known NCP binding proteins ( Figure 2 ) . However , the task is more challenging to distinguish IE1 and LANA bindings , as they both interact with all binding zones of the acidic patches . Nevertheless , the identification of the E56R variant of histone H2A as an IE1-noninteracting mutant has already served as a useful tool for assessing IE1’s ability to modulate the higher-order structure of chromatin . Interestingly , this chromatin-modulating activity is not shared by LANA . It was shown previously that the LANA peptide promotes the compaction of nucleosomal arrays , judged by an assay in which the folding of nucleosomal arrays was induced by magnesium ions ( Chodaparambil et al . , 2007 ) . In qualitative agreement with the previous observation , we saw that the addition of LANA made the nucleosomal arrays sedimented slower . Nevertheless , unlike IE1-CTD , LANA has no effect on the preformed H1-containing 30-nm chromatin fiber . There are two possible reasons for the different behavior of LANA , one possibility is that LANA is bound to the chromatin fiber but did not cause any changes to the folding , and the other possibility is that LANA failed to bind 30-nm chromatin fiber . In either case , IE1 and LANA differ in their ability to affect the folding of 30-nm chromatin fiber . The contrasting properties of IE1 and LANA suggest that distinct modes of protein binding to the acidic patch of the nucleosome could exert differential influences to chromatin folding . The discovery of the chromatin modulating activity of IE1 should facilitate further mechanistic understanding of its functions in viral pathogenesis , particularly its impact on the chromatin structure of the host genome , such as transcriptional regulation . To express histones , cDNA fragments encoding Xenopus laevis histones H3 . 3C and H4 , H2A type-1 and H2B 1 . 1 , human H3 . 1 and H4 , human H2A type 1-B/E and H2B type 1-J were cloned into pCDFDuet-1 vectors ( Novagen ) to generate four bicistronic plasmids for co-expression of Xenopus and human H3-H4 and H2A-H2B pairs , respectively , in the BL21 ( DE3 ) -RIL strain of E . coli at 37°C . Bacterial cells overexpressing H3-H4 and H2A-H2B were mixed and lysed together . Xenopus and human histone octamers were then purified as described ( Kingston et al . , 2011 ) . Xenopus NCP was assembled with a 146-bp palindromic DNA fragment derived from human α-satellite DNA according to a described procedure ( Luger et al . , 1997b ) . Human NCP was assembled with the 147-bp Widom 601 DNA sequence and the human histone octamer following the same procedure . Human NCP carrying the E56R mutation of histone H2A was generated by mutagenesis using the TaKaRa MutanBEST kit ( TaKaRa , China ) . NCP reconstituted with Xenopus histones was crystallized by sitting-drop vapor diffusion at 16°C in a condition containing 50 mM sodium cacodylate , pH 6 . 2 , 100 mM magnesium acetate , and 11% 2-methyl-2 , 4-pentanediol . The co-crystal structure was obtained from a NCP crystal soaked with a chemically synthesized IE1-CTD peptide ( a . a . 476–491 , SciLight Biotechnology ) at 2 mg/ml for 24 hr in a buffer containing 50 mM sodium cacodylate , pH 6 . 4 , 100 mM magnesium acetate , 24% 2-methyl-2 , 4-pentanediol and 5% trehalose . X-ray diffraction data were collected at 100K at Beamline BL18U of Shanghai Synchrotron Radiation Facility ( SSRF ) using a Pilatus 6 M detector at a wavelength of 1 . 0308 Å , and the data was processed using the HKL2000 package ( Otwinowski and Minor , 1997 ) . The structure was solved by molecular replacement with PHASER ( McCoy et al . , 2007 ) using the Xenopus NCP structure ( PDB ID: 1AOI ) as the search model . The electron density for the IE1-CTD peptide was clear after refinement with REFMAC ( Murshudov et al . , 1997 ) allowing unambiguous building of the IE1-CTD model using COOT ( Emsley and Cowtan , 2004 ) . The model was then refined with PHENIX ( Adams et al . , 2010 COOT ) . The Rwork and Rfree of the final model were 19 . 5% and 24 . 4% , respectively . Detailed statistics for crystallographic analyses are shown in Table 1 . 10 . 7554/eLife . 11911 . 010Table 1 . Statistics of crystallographic analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 11911 . 010Data collection statisticswavelength ( Å ) 1 . 0308space groupP212121unit cell ( Å ) a = 106 . 70 , b = 109 . 47 , c = 181 . 98resolution ( Å ) 30 . 00–2 . 80 ( 2 . 90–2 . 80 ) Rmerge0 . 133 ( 0 . 611 ) I/σI12 . 5 ( 3 . 3 ) Completeness ( % ) 99 . 9 ( 100 . 0 ) Total/Unique reflections346679/52496Refinement statisticsRwork/Rfree0 . 195/0 . 244rmsd bonds ( Å ) 0 . 008rmsd angles ( º ) 0 . 935No . of Atoms Protein6116 DNA5982 Peptide104 Ion4 Water230B factor ( Å2 ) Protein35 . 9 DNA87 . 8 Peptide56 . 9 Ion47 . 5 Water36 . 0Ramachandran plot favored750 ( 98 . 7% ) allowed8 ( 1 . 1% ) outlier2 ( 0 . 3% ) Plasmids for expressing the full-length hCMV ( Towne ) IE1 and its truncation variant lacking CTD ( IE1ΔCTD , a . a . 1–475 ) were obtained from Dr . Michael Nevels ( Mucke et al . , 2014 ) . Both fragments were expressed as a GST-fusion protein at 16°C in the BL21 ( DE3 ) -RIL strain of E . coli . They were purified with glutathione-Sepharose resins , followed by cleavage of the GST-tag and further purification with a HiTrap Q HP column ( GE Healthcare ) . ITC experiments with IE1-CTD and LANA peptides were performed at 20°C , with the peptide solutions titrated into human NCP solutions in a buffer containing 10 mM Tris-HCl , pH 7 . 5 , and 50 mM NaCl . An NCP concentration of 0 . 02 mM was used in all experiments , except for the titration of wild-type IE1-CTD into wild-type NCP , in which case 0 . 018 mM NCP was used . The peptide concentrations used were , wild-type IE1-CTD at 0 . 87 mM , T485V at 0 . 99 mM , and the rest of IE1-CTD mutants and LANA all at 0 . 59 mM . Detailed procedures follow a protocol published previously ( Yang et al . , 2013 ) . For the set of ITC experiments involving full-length IE1 and IE1ΔCTD , a buffer containing 10 mM Hepes , pH 7 . 4 , and 150 mM NaCl was used to minimize background heat generation . Background heat measured through titrating samples from the syringe into the buffer without NCP was subtracted from the integrated data . For comparison , the binding of IE1-CTD to NCP under the same condition was also measured . In the set of experiments , an NCP concentration of 0 . 015 mM was used , and the concentrations of IE1 , IE1ΔCTD and IE1-CTD used were 0 . 51 , 0 . 52 and 0 . 59 mM , respectively . Recombinant human core histones were prepared as described above . Linker histone H1 . 4 and DNA templates of 12 tandem 177 bp repeats of the 601 sequence were cloned and purified , and reconstituted into chromatin as previously described ( Chen et al . , 2013; Dyer et al . , 2003; Li et al . , 2010; Song et al . , 2014 ) . The reconstituted chromatin samples were subject to AUC analysis in a buffer containing 10 mM HEPES , pH 8 . 0 and 0 . 1 mM EDTA . All AUC experiments were performed with nucleosomal array and chromatin fiber concentrations at 0 . 25 μM , and peptides/proteins to NCP at 5:1 molar ratio , whenever applicable , on a Beckman Coulter ProteomeLab XL-I , and the data were analyzed using enhanced van Holde-Weischet analysis and the Ultrascan II 9 . 9 revision 1504 as previously described ( Chen et al . , 2013 ) . MNase digestion of nucleosomal arrays follows a procedure described previously ( Li et al . , 2010 ) . In brief , chromatin sample containing the equivalent of 1 μg DNA were incubated with micrococcal nuclease ( Sigma ) in a 50 μl reaction ( 10 mM HEPES , pH 7 . 5 , 25 mM KCl , 0 . 2 mM EDTA , 10% Glycerol and 2 mM CaCl2 ) at 37°C for 4 min . The digestion was terminated by addition of 50 μl stop buffer ( 200 mM NaCl , 2% SDS , 10 mM EDTA ) . Each sample were treated with 0 . 1 mg/ml Proteinase K and reacted at 55°C for 45 min . MNase digested DNA was purified and analyzed on a 1 . 3% agarose gel . EM analysis by metal shadowing with tungsten was performed as previously described ( Chen et al . , 2013 ) . The samples were examined using a FEI Tecnai G2 Spirit 120 kV transmission electron microscope . The coordinates and diffraction data have been deposited in PDB under the accession code 5E5A .	Most of the DNA in a cell is tightly wrapped around groups of proteins called histones , which gives the impression of beads on a string . These bead-like structures are called nucleosomes , and interactions between histones in different nucleosomes can link one nucleosome to another , to package the DNA into a very condensed form . Viruses sometimes interact with this condensed DNA; for example , a virus called human cytomegalovirus is known to attach to condensed DNA when cells are preparing to divide . But the consequences of these interactions are not always clear . Now , Fang , Chen et al . have worked out the three-dimensional structure of a protein from the cytomegalovirus while it is attached to a nucleosome . This structure revealed that the viral protein connects to same part of the histones that otherwise helps pull the nucleosomes together . Further experiments then compared how the cytomegalovirus protein attaches to nucleosomes with the interaction between nucleosomes and a similar protein from a different virus . Both viral proteins were seen to interact with the same part of the histone protein , but in different ways . Next , Fang , Chen et al . showed that the DNA is more loosely packed when the cytomegalovirus protein is attached to the nucleosomes . This was not the case for the similar protein from the other virus . The experiments show that small differences in the ways viral proteins interact with condensed DNA can change their effects on DNA packaging . Additionally , these findings may help scientists to better understand how the binding of the cytomegalovirus protein to the nucleosomes might affect this virus’s ability to infect or cause illness in humans .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "short", "report", "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2016	Human cytomegalovirus IE1 protein alters the higher-order chromatin structure by targeting the acidic patch of the nucleosome
microRNA-1 ( miR-1 ) is an evolutionarily conserved , striated muscle-enriched miRNA . Most mammalian genomes contain two copies of miR-1 , and in mice , deletion of a single locus , miR-1-2 , causes incompletely penetrant lethality and subtle cardiac defects . Here , we report that deletion of miR-1-1 resulted in a phenotype similar to that of the miR-1-2 mutant . Compound miR-1 knockout mice died uniformly before weaning due to severe cardiac dysfunction . miR-1-null cardiomyocytes had abnormal sarcomere organization and decreased phosphorylation of the regulatory myosin light chain-2 ( MLC2 ) , a critical cytoskeletal regulator . The smooth muscle-restricted inhibitor of MLC2 phosphorylation , Telokin , was ectopically expressed in the myocardium , along with other smooth muscle genes . miR-1 repressed Telokin expression through direct targeting and by repressing its transcriptional regulator , Myocardin . Our results reveal that miR-1 is required for postnatal cardiac function and reinforces the striated muscle phenotype by regulating both transcriptional and effector nodes of the smooth muscle gene expression network . Cardiac gene expression is cooperatively regulated by an intertwined network of transcription factors and microRNAs ( miRNAs ) ( Srivastava , 2006; Cordes and Srivastava , 2009 ) . Perturbations in the activity or expression of factors within this network result in cardiac structural and functional defects in animal models and in humans . Among these , serum response factor ( SRF ) and Myocardin ( Myocd ) cooperate to directly regulate the myogenic gene program in both cardiac and smooth muscle ( Treisman , 1987; Wang et al . , 2001; Chen et al . , 2002; Sepulveda et al . , 2002 reviewed in Wang and Olson , 2004 ) . These factors transcriptionally regulate numerous miRNAs that , in turn , regulate transcription factors to reinforce specific cellular decisions and behavior . ( Kwon et al . , 2005; Zhao et al . , 2005; Niu et al . , 2008; Cordes et al . , 2009 ) . miRNAs are small , ∼21 nucleotide ( nt ) , single-stranded RNAs that negatively regulate the stability and translation of mRNA transcripts . miRNAs target sequences within the 3′ UTRs of mRNA transcripts that are highly complementary to the miRNA seed sequence ( nt 2–8 ) and have imperfect complementarity outside of the seed region ( Valencia-Sanchez et al . , 2006 ) . Due to the degenerate nature of miRNA:mRNA interactions , a single miRNA may have hundreds of mRNA targets ( Bartel , 2009 ) . Often miRNAs target multiple genes in a common pathway , thereby amplifying the effect of an individual miRNA on a given biological process ( Fish et al . , 2008; Cordes et al . , 2009 ) . microRNA-1 ( miR-1 ) is a highly conserved miRNA and its expression is enriched specifically in cardiac and skeletal muscle . In mice , it is expressed in the heart and somites of the developing embryo during myogenic differentiation , beginning around embryonic day ( E ) 8 . 5 ( Zhao et al . , 2005; Liu et al . , 2007 ) . The cardiac expression of miR-1 increases during development , with a dramatic rise in the postnatal period . RNA sequencing has revealed that miR-1 is the most abundant miRNA in the adult mouse heart , representing up to 40% of all miRNA transcripts ( Rao et al . , 2009 ) . miR-1 is transcribed as part of a bicistronic cluster with another striated muscle-enriched miRNA , miR-133a . In the genomes of most mammals , a duplication event has occurred resulting in two copies of the miR-1/133a locus , with miR-1-2 and miR-133a-1 on chromosome 18 and miR-1-1 and miR-133a-2 on chromosome 2 of the murine genome ( Figure 1—figure supplement 1 ) . Both precursors are transcriptionally regulated by several key myogenic transcription factors , including Myogenin , MYOD , SRF , MYOCD ( Kwon et al . , 2005; Zhao et al . , 2005; Rao , 2006 ) and MEF2 ( Liu et al . , 2007 ) . When processed , both the miR-1-2 and miR-1-1 precursors give rise to identical mature miR-1 species , suggesting evolutionary pressure on both alleles ( Figure 1—figure supplement 2 ) . An additional miRNA cluster encoding miR-133b and miR-206 is expressed uniquely in skeletal muscle , with the mature sequence of miR-206 sharing a common seed with miR-1 , but varying by 4 nts outside of the seed region . Deletion of miR-1-2 in mice ( Zhao et al . , 2007 ) , which reduces the total expression of cardiac miR-1 by roughly 50% , results in a spectrum of cardiac defects on a pure 129 background , including incompletely penetrant lethality , cardiomyocyte proliferative defects , and electrophysiological abnormalities . In flies , loss of the single miR-1 gene results in abnormal myogenic differentiation and cell polarity defects in cardiac progenitors ( Kwon et al . , 2005; Sokol , 2005; King et al . , 2011 ) ; however , the consequences of complete loss of miR-1 in mammals are unknown . In this study , we report that targeted deletion of the miR-1-1 locus results in a phenotype similar to that described for miR-1-2 null mice , and that the complete loss of miR-1 is uniformly lethal before weaning due to cardiac dysfunction . We show that the loss of miR-1 results in perinatal heart failure with myocardial sarcomeric defects , hypophosphorylation of Myosin Light Chain 2 , and ectopic expression of Telokin , a smooth muscle-restricted inhibitor of Myosin Light Chain 2 phosphorylation . Furthermore , we found the SRF co-factor , MYOCD , which is critical for transcriptional activation of both the cardiac and smooth muscle gene programs in vivo ( Li et al . , 2003; Hoofnagle et al . , 2011; Wang et al . , 2001; Chen et al . , 2002 ) , is directly targeted by miR-1 . The smooth muscle isoform of Myocd was preferentially upregulated in the absence of miR-1 and likely contributed to ectopic activation of the smooth muscle gene program in the heart . Our findings reveal that miR-1 is embedded in an SRF-dependent cardiac gene program that promotes sarcomerogenesis and myogenic differentiation , while simultaneously repressing the smooth muscle program . We used homologous recombination to delete one allele of the miR-1-1 precursor in embryonic stem ( ES ) cells , with a floxed neomycin cassette used for positive selection . ( Figure 1—figure supplement 3 and ‘Materials and methods’ ) . Injection of targeted ES cells into blastocysts resulted in high-percentage chimeras that transmitted the targeted allele through the germline . Intercrosses of miR-1-1 heterozygous mice revealed that approximately half of all miR-1-1 homozygous-null mice died before weaning when bred onto a pure 129 strain , similar to miR-1-2 null mice ( Figure 1A , upper ) . This lethality was strain dependent , as miR-1-1 null animals on a mixed background ( 129/BL6 ) survived at normal Mendelian ratios until weaning . ( Figure 1A , lower ) . By quantitative RT-PCR ( qPCR ) , we found that total cardiac miR-1 levels were decreased in miR-1-1 knockout animals by about 40% at postnatal day ( P ) 2 ( Figure 1B ) . 10 . 7554/eLife . 01323 . 003Figure 1 . Viability and cardiac function of miR-1-1−/− mice . ( A ) Genotypes of offspring generated from miR-1-1+/− intercrosses on either a pure 129 background ( upper ) or a mixed BL6/129 strain ( lower ) . Numbers of expected and observed genotype ratios are given for weaning-age ( 3-week-old ) pups . ( B ) qPCR of mature miR-1 and miR-133a in post-natal day 2 hearts . N = 6 per genotype . ( C ) Analyses of cardiac function by echocardiography of adult animals of indicated genotypes on a pure 129 background . N = 5 per genotype . LVEDD , left ventricular end-diastolic dimension; LVESD , left ventricular end-systolic dimension . ( D ) Adult wildtype ( upper ) and miR-1-1−/− ( lower ) hearts on a pure 129 background ( I ) . RA , right atrium; LA , left atrium; RV , right ventricle; LV , left ventricle . Hematoxylin and eosin images taken at 1 . 25X magnification ( II ) ; and 40X magnification; scale bar indicates 25 μm ( III ) . Masson trichrome stain of miR-1-1 knockout myocardium , images taken at 40X; scale bar 50 μm ( IV ) ( E ) Analyses of cardiac conduction by electrocardiogram ( EKG ) of adult animals of indicated genotypes on a pure 129 background . p<0 . 05; p<0 . 01; ns , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 00310 . 7554/eLife . 01323 . 004Figure 1—figure supplement 1 . Schematic of the miR-1/133a genomic loci . The mature miR-1 sequence ( depicted in blue ) is processed from two distinct miR-1 precursors encoded in the murine genome , miR-1-1 on chromosome 2 and miR-1-2 on chromosome 18 . Both miR-1 copies are co-transcribed with miR-133a ( mature depicted in green ) with the genomic distances between the two miRNAs indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 00410 . 7554/eLife . 01323 . 005Figure 1—figure supplement 2 . Identical sequences of mature miR-1-1 and miR-1-2 . The miR-1 seed sequences ( nt 2–8 ) is indicated in red . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 00510 . 7554/eLife . 01323 . 006Figure 1—figure supplement 3 . Left , targeting scheme for deletion of the miR-1-1 locus . Location of genotyping primers indicated . Right , PCR genotyping results from miR-1-1 wild-type , heterozygous , or homozygous knockout animals . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 00610 . 7554/eLife . 01323 . 007Figure 1—figure supplement 4 . Averaged electrocardiogram tracings tracing from lead I of an adult wild-type or miR-1-1 knockout animal on a pure 129 background . Green lines represent multiple overlaid EKG tracings with average indicated by black line . PR and QRS intervals indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 00710 . 7554/eLife . 01323 . 008Figure 1—figure supplement 5 . Electrocardiogram tracings of an adult wild-type or miR-1-1 knockout animal on a pure 129 background . The miR-1-1−/− tracing ( right ) indicates the presence of an arrhythmia not observed in the wild-type control ( left ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 00810 . 7554/eLife . 01323 . 009Figure 1—figure supplement 6 . qPCR for the miR-1 target , Irx5 , in adult miR-1-1−/− and wild-type hearts . N = 5 per genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 00910 . 7554/eLife . 01323 . 010Figure 1—figure supplement 7 . qPCR for mature miR-133a in adult wild-type and miR-1-2 knockout hearts . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 010 To evaluate cardiac function , we performed echocardiography on adult miR-1-1 null or wild-type littermates . We found a reduction in fractional shortening , as well as an increase in left ventricular end-diastolic and end-systolic dimension in the miR-1-1 knockout animals , indicating ventricular dilation ( Figure 1C ) . Mild ventricular dilation was confirmed histologically ( Figure 1D I and II ) . Additionally , we observed areas of fibrosis in the ventricular myocardium of the miR-1-1 knockouts ( Figure 1D III and IV ) . Like miR-1-2 knockouts , miR-1-1 knockout animals exhibited subtle conduction abnormalities , including prolonged ventricular depolarization and repolarization , indicated by a broader QRS complex and longer QT interval than controls . Additionally , miR-1-1 knockout mice showed broader P waves without alteration of the PR interval ( Figure 1E , Figure 1—figure supplement 4 ) . Intermittent atrial arrhythmias were observed in two of five knockout animals analyzed , but were not observed in wild-type animals ( Figure 1—figure supplement 5 ) . Thus , mice lacking miR-1-1 were grossly similar to those lacking miR-1-2 , in that they exhibited partial lethality as well as subtle conduction abnormalities . Cardiac conduction defects in the miR-1-2 knockout mice were at least partially ascribed to dysregulation of the miR-1 target , Irx5 . Similarly , we found that Irx5 was upregulated in the miR-1-1 knockout mice ( Figure 1—figure supplement 6 ) . miR-1-1 knockout mice also had a partial decrease in mature miR-133a levels ( Figure 1B ) . Previously , using semi-quantitative RT-PCR , the miR-133a precursor levels were reported to be unchanged in the miR-1-2 knockout animals . In this study , the analysis of the mature species by qPCR revealed a slight decrease in miR-133a in weaning-age miR-1-2 null hearts , although not statistically significant ( Figure 1—figure supplement 7 ) . Importantly , miR-133a expression was maintained at a level described to be inconsequential in previous reports ( Liu et al . , 2008 ) . To investigate the consequences of complete loss of miR-1 , we intercrossed miR-1-1 and miR-1-2 mutant mice ( Zhao et al . , 2007 ) to generate double-heterozygous mice in a 129/BL6 mixed background . At weaning , no lethality was observed in the double-heterozygous mice ( Figure 2—figure supplement 1 ) similar to the miR-1-1 knockout on a mixed background . Using gene expression microarray analyses , we found 201 genes that were dysregulated in the single knockouts or double-heterozygous mice ( Figure 2—figure supplement 2 ) . Of those , 24 genes were coordinately dysregulated in animals of all three genotypes . The majority of genes ( 195/201 ) were similarly altered between the double heterozygotes and at least one of the single knockouts . There were , however , some differences in gene expression between these groups , which may suggest minor functional differences of the miR-1 loci . Double-heterozygous mice were subsequently intercrossed to generate knockout animals with only a single intact allele ( miR-1-1−/−: miR-1-2+/− and miR-1-1+/−: miR-1-2−/− ) . Most of these mice were viable and fertile on the mixed background , though mice lacking both the copies of miR-1-2 were under-represented at weaning ( Figure 2—figure supplement 3 ) , indicating some difference in the compensatory ability of the two loci . Single-allele mice ( ¾ alleles knocked-out ) were intercrossed to generate mice completely lacking miR-1 . miR-1 compound-null mice on a mixed background were born at slightly less than Mendelian ratios and were of normal birth weight ( Figure 2A , Figure 2—figure supplements 4 and 5 ) . Roughly a quarter of the double-knockout animals died very soon after birth . In a subset of these animals , we observed ventricular septal defects ( VSDs ) and misalignment of the aorta over the ventricular septum ( overriding aorta ) , likely accounting for their lethality ( Figure 2—figure supplement 6 ) . Surviving miR-1 double-knockouts failed to thrive post-natally , with no double-knockout animals surviving beyond P10 ( Figure 2B , Figure 2—figure supplement 5 ) . Examination of surviving miR-1 double-knockouts revealed expansion of the superior portion of the right ventricle ( conus ) and enlargement of the atria , when compared to wild-type mice beginning at P0 ( Figure 2C , asterisk ) . By P4 , and more extensively by P10 , the dilation of all cardiac chambers was observed , with a particularly notable enlargement of the right atria . Echocardiography revealed severely impaired fractional shortening by P2 ( Figure 2D and E ) with poor systolic function . Frequent ventricular thrombi were observed by P4 , consistent with impaired cardiac function in these animals ( Figure 2C ) . Functional analysis of the conduction system by electrocardiogram ( EKG ) revealed a spectrum of abnormalities , including a prolonged QRS complex , and prolonged PR and QT intervals ( Figure 2F , Figure 2—figure supplement 7 ) . The presence of frequent sinus pauses was observed in all knockout animals analyzed ( Figure 2—figure supplement 8 ) . These morphological , functional , and electrophysiological data indicate that the postnatal lethality in miR-1 knockout animals is due to cardiac dysfunction . 10 . 7554/eLife . 01323 . 011Figure 2 . Compound miR-1 knockout mice exhibit lethality due to a spectrum of cardiac defects . ( A ) Genotypes of offspring generated from miR-1-1−/+:miR-1-2−/− X miR-1-1−/−: and miR-1-2+/− intercrosses on a mixed BL6/129 background . Numbers of expected and observed genotype ratios are given for post-natal day 0 ( P0 ) and weaning-age ( 3-week-old ) pups . ( B ) Kaplan-Meier survival curve of miR-1 double-knockout animals and double-heterozygous littermates . ( C ) Abnormal cardiac morphology in postnatal miR-1 double-knockout mice includes an elongated outflow tract , as evidenced by bulging of the conus at P0 ( asterisk ) . By P4 , chamber dilation and thinning of the myocardium was apparent and ventricular clots were commonly observed . RA , right atrium; LA , left atrium; RV , right ventricle; LV , left ventricle . Middle panel images taken at 1 . 25X magnification; lower panel images taken at 40X magnification . ( D ) Echocardiography showed abnormal cardiac function in P2 miR-1 null compared to wild-type animals . N = 7 for wild-type mice and N = 4 for miR-1 null mice . LVEDD , left ventricular end-diastolic dimension; LVESD , left ventricular end-systolic dimension . ( E ) Representative M-mode image by echocardiography indicating diastolic and systolic dimensions . ( F ) Electrocardiographic analysis revealed conduction abnormalities in miR-1 double-knockout animals by P2 , including a decreased heart rate and elongated QRS relative to wild-type controls . p<0 . 05; p<0 . 01; p<0 . 001;**p<0 . 0001; ns , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 01110 . 7554/eLife . 01323 . 012Figure 2—figure supplement 1 . Genotypes of offspring generated from miR-1-1+/−: miR-1-2+/+ X miR-1-1+/+: and miR-1-2+/− intercrosses on a mixed BL6/129 background . Numbers of expected and observed genotypes are given for weaning-age ( 3-week-old ) pups . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 01210 . 7554/eLife . 01323 . 013Figure 2—figure supplement 2 . Cluster analysis of relative gene expression changes in miR-1-1−/− and miR-1-2−/− single knockout and miR-1-1+/−:miR-1-2+/− double-heterozygous hearts . Data presented as Log2 of the fold change compared to averaged wild-type controls . Genes that were coordinately regulated among all three genotypes are indicated . N = 3 per genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 01310 . 7554/eLife . 01323 . 014Figure 2—figure supplement 3 . Genotypes of offspring generated from miR-1-1−/−: miR-1-2+/− X miR-1-1+/−: miR-1-2−/− intercrosses on a mixed BL6/129 background at weaning . Numbers of expected and observed genotypes are given for weaning-age ( 3-week-old ) pups . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 01410 . 7554/eLife . 01323 . 015Figure 2—figure supplement 4 . Genotypes of offspring generated from miR-1-1−/−: miR-1-2+/− X miR-1-1+/−; and miR-1-2−/− intercrosses on a mixed BL6/129 background at birth . Numbers of expected and observed genotypes are given for postnatal day 0 pups . Animals discovered dead at birth are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 01510 . 7554/eLife . 01323 . 016Figure 2—figure supplement 5 . miR-1 double-knockouts fail to thrive . ( A ) Images of miR-1 knockout animals and double-heterozygous littermates at indicated ages . ( B ) Mass of P0 double-knockout animals ( n = 11 ) or double-heterozygous littermates ( n = 19 ) . ( C ) Post-natal weight gain as a percentage of birth weight of miR-1 double-knockouts or double-heterozygous littermates . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 01610 . 7554/eLife . 01323 . 017Figure 2—figure supplement 6 . Abnormal cardiac morphology in miR-1 double-knockouts found dead at birth compared to wild-type control . A subset of miR-1 null mice found dead at birth had ventricular septal defects ( arrows ) . All animals displayed dilated atria . RA , right atrium; LA , left atrium; RV , right ventricle; LV , left ventricle . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 01710 . 7554/eLife . 01323 . 018Figure 2—figure supplement 7 . Averaged electrocardiogram tracings from lead I of P2 miR-1 null ( right ) or wild-type ( left ) mice . Green lines represent multiple overlaid EKG tracings with average indicated by black line . PR , QRS and QT intervals indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 01810 . 7554/eLife . 01323 . 019Figure 2—figure supplement 8 . Electrocardiogram tracings of a postnatal wild-type or miR-1 null animal on a mixed background . The miR-1 null tracing ( right ) indicates the presence of multiple arrhythmias not observed in the wild-type control ( left ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 01910 . 7554/eLife . 01323 . 020Figure 2—figure supplement 9 . Left , qPCR of mature miR-1 or miR-133a in E12 . 5 wild-type or miR-1 null hearts ( n = 3 per genotype ) . Right , mature miRNA expression in P0 wild-type or miR-1 null hearts ( n = 5 per genotype ) . ND = not detected . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 02010 . 7554/eLife . 01323 . 021Figure 2—figure supplement 10 . qPCR to detect the miR-1-2/133a-1 or miR-1-1/133a-2 promoter sequences , or an intergenic genomic sequence , following chromatin immunoprecipitation ( ChIP ) of RNA polymerase II ( RNA Pol II ) in P2 wild-type or miR-1 null hearts . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 021 qPCR confirmed that miR-1 was not detectable in hearts of miR-1 compound-null mice . Notably , miR-133a expression was decreased in miR-1 double-knockout animals with the dysregulation of miR-133a becoming more pronounced with age as the locus normally becomes more actively transcribed ( Figure 2—figure supplement 9 ) . As miR-1 lies transcriptionally upstream of miR-133a , we cannot exclude the possibility that the decrease in miR-133a expression is in part due to impaired transcriptional read-through downstream of the miR-1 targeting event . However , chromatin immunoprecipitation revealed reduced RNA polymerase II ( Pol II ) occupancy at the miR-1-1/miR-133a-2 , and miR-1-2/miR-133a-1 promoters in P2 miR-1 double-knockout mouse hearts , suggesting that transcriptional initiation of these loci is reduced secondary to loss of miR-1 expression ( Figure 2—figure supplement 10 ) . This suggests a feedback mechanism whereby miR-1 maintains appropriate expression of the miR-1/133a loci . Given the relatively normal cardiac morphogenesis in surviving miR-1 knockouts , we hypothesized that their impaired cardiac contractility was due to a primary myocardial defect . In agreement , transmission electron microscopy of ventricular tissue revealed the areas of extensive sarcomeric disruption at P0 , before the onset of ventricular dilation ( Figure 3A ) . Mitochondrial morphology was also abnormal , including overall decreased mitochondrial size ( Figure 3—figure supplements 1 and 2 ) and decreased complexity of mitochondrial cristae ( Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 01323 . 022Figure 3 . Sarcomere disruption in miR-1 null cardiomyocytes . ( A ) Transmission electron microscopy ( TEM ) of P0 wild-type or miR-1 double-knockout myocardium . Representative areas of sarcomeric disarray are indicated ( ) . Distance between Z-lines is indicated with lines . Arrows indicate disrupted Z-line structures . Scale bar , 1 μm . ( B ) Immunofluorescence of sarcomeric structures in isolated P0 cardiomyocytes with Phalloidin ( F-actin cytoskeleton [orange] ) and sarcomeric alpha-actinin ( green ) DAPI ( blue ) indicates nuclei . Images captured at 40X magnification . ( C ) Percentage of cardiomyocytes of individual sarcomeric classes observed . N = 3 for miR-1 null mice; N = 2 for miR-1-1−/−:miR-1-2+/− mice; and n = 6 for miR-1 double-heterozygous mice , with a minimum of 50 cells classified per animal . ( D ) Representative analysis showing percentage of P5 miR-1 null cardiomyocytes of individual sarcomeric classes observed 24 hr after transfection of a miR-1 mimic or control RNA mimic . Roughly 50 cardiomyocytes per condition were evaluated . p<0 . 05; p<0 . 01; p<0 . 001;*p<0 . 0001; ns , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 02210 . 7554/eLife . 01323 . 023Figure 3—figure supplement 1 . Transmission electron microscopy ( TEM ) reveals mitochondrial morphology defects in miR-1 null hearts . TEM of P0 wild-type or two miR-1 double-knockout animals . Arrowheads indicate mitochondria with low cristae density . Scale bars equal 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 02310 . 7554/eLife . 01323 . 024Figure 3—figure supplement 2 . Quantification of mitochondrial area from TEM images reveals a reduction in mitochondrial area in both miR-1 double-knockouts compared to controls , although the degree to which mitochondrial area is reduced varies . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 02410 . 7554/eLife . 01323 . 025Figure 3—figure supplement 3 . Sarcomeric organization classification scheme . Description of sarcomeric organization scheme used in Figure 3C and D with corresponding representative images . Sarcomeric structure highlighted with immunostaining for the Z-line protein α-actinin ( green ) . Nuclei stained with DAPI ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 025 Sarcomeric morphology was further analyzed in isolated P0 neonatal cardiomyocytes by immunostaining for the Z-line protein α-Actinin , as well as by Phalloidin staining to visualize the filamentous actin cytoskeleton . Individual cardiomyocytes were classified on a scale from I–V based on their sarcomeric organization , with class-I cells showing highly ordered sarcomeres , and class-V cells having no sarcomeric organization ( Figure 3—figure supplement 3 ) . Sarcomeres in miR-1 null cardiomyocytes showed significant disruption when compared to those from miR-1 double-heterozygous littermates . Only 10% of double-knockout cardiomyocytes had highly organized sarcomeres , compared to 45% of miR-1 double-heterozygous cells . The degree of sarcomeric organization correlated with miR-1 dosage as cardiomyocytes lacking three out of four copies of miR-1 showed an intermediate degree of organization ( Figure 3C ) . To determine if reintroduction of miR-1 was sufficient to improve sarcomere organization in cardiomyocytes from post-natal hearts , we transiently transfected a miR-1 RNA mimic into cultured cardiomyocytes isolated from P5 animals . Indeed , adding miR-1 partially rescued this phenotype and enhanced sarcomeric organization in miR-1 double-knockout cardiomyocytes , compared to those treated with a control RNA mimic ( Figure 3D ) . We performed RNA sequencing of late embryonic stage ( E18 ) miR-1 null and wild-type hearts to identify genes that were dysregulated in the absence of miR-1 . We selected this time point in order to reveal primary changes in gene expression due to the loss of miR-1 and not those that may arise secondary to heart failure . Given that many direct miRNA targets are upregulated at the protein , but not transcript level , we expected that sequencing analysis of this stage would identify pathways that are dysregulated in the miR-1 knockout , some of which may involve direct miR-1 targets . We utilized the GREAT interface ( McLean et al . , 2010 ) to evaluate the enrichment of miRNA targets within the set of genes that were upregulated in the miR-1 null hearts , compared to wild-type controls ( Figure 4—figure supplement 1 ) . We found that genes containing miR-1/206 seed sequence complementarity were most significantly enriched in this data set . Genes targeted by miR-495 , miR-518a-2 , miR-501 and miR-409 were also enriched , though to a lesser degree . Notably , mRNAs with seed sequence complementarity to miR-133a were not enriched , suggesting that the reduction in miR-133a levels did not reach the threshold for significant dysregulation of genes in the miR-1 knockout . We next performed a gene ontology analysis using the GO-Elite interface ( Zambon et al . , 2012 ) to determine at a functional level which dysregulated genes may be phenotypically relevant . We found that genes participating in metabolic and mitochondria-related pathways were downregulated in miR-1 knockout , compared to wild-type hearts , consistent with the morphological abnormalities visualized by electron microscopy ( Figure 4A ) . Interestingly , many upregulated genes fell into the ‘regulation of actin cytoskeletal’ pathway ( Figure 4B ) . As the dysregulation of actins and other cytoskeletal genes contribute to cardiomyopathies , in a complementary analysis , we investigated if any of the upregulated cytoskeletal genes were direct miR-1 targets . miRNA targets remain challenging to predict computationally; therefore , we utilized three different target prediction algorithms—Targetscan ( www . targetscan . org ) , PITA ( http://genie . weizmann . ac . il/pubs/mir07/mir07_dyn_data . html ) , and Pictar ( http://pictar . mdc-berlin . de/cgi-bin/PicTar_vertebrate . cgi ) —to identify consensus predicted targets ( Figure 4C , Figure 4—source data 1 ) . Of the genes that were expressed at an equal or higher level in the knockout compared to wild-type hearts , 89 genes were predicted miR-1 targets by all three algorithms . Of these , 13 genes were significantly upregulated at the mRNA level in the miR-1 null hearts . ( Figure 4C , Figure 4—source data 1 ) . 10 . 7554/eLife . 01323 . 026Figure 4 . Pathway analysis of genes dysregulated in E18 miR-1 null hearts . ( A ) Pathway analysis of genes downregulated in E18 miR-1 null hearts identified via Go-elite and GenMAPP . Pathways related to metabolism are enriched . ( B ) Pathway analysis of genes upregulated in the miR-1 null hearts identified via Go-elite and GenMAPP . Values represent normalized mean centered log2 of FPKM for each genotype . ( C ) Venn diagram depicting overlap of miR-1 targets as predicted by three prediction algorithms: Targetscan , PITA and Pictar . The genes analyzed were expressed at a relative quantity of ≥1 in miR-1 null vs wild-type hearts based on their FPKM ( upper ) . Consensus targets predicted by all three algorithms , showing significant upregulation in the miR-1 null vs wild-type hearts are presented as descending log2 fold change in miR-1 null over wild-type hearts ( lower ) . Mylk ( MLCK ) in red was identified as a regulator of the actin cytoskeleton ( B ) and a significantly upregulated , predicted miR-1 target ( C ) . dKO , double-knockout; WT , wild-type; FPKM , fragments per kilobase per million . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 02610 . 7554/eLife . 01323 . 027Figure 4—source data 1 . miR-1 targets as predicted by three algorithms . Putative miR-1 target genes as predicted by Targetscan , PITA and Pictar . Genes used for this analysis were expressed at a relative quantity of ≥1 based on FPKM in miR-1 null vs wild-type hearts . Genes indicated in bold were significantly upregulated ( FDR<0 . 1 ) in miR-1 null vs wild-type hearts . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 02710 . 7554/eLife . 01323 . 028Figure 4—figure supplement 1 . MicroRNA target enrichment analysis of genes that were expressed at a relative quantity of ≥1 in the miR-1 null vs wild-type hearts . The miR-1/206 seed but not the miR-133a seed was enriched in this gene set . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 028 In the miR-1 null animals , myosin light chain kinase ( MLCK ) was a putative miR-1 target of particular interest , as it was highly upregulated and had a known role in regulating the cytoskeleton ( Figure 4B , C ) . MLCK , which is encoded within the Mylk locus , regulates the cytoskeleton by phosphorylating the regulatory Myosin Light Chain 2 ( MLC2 ) . MLC2 is associated with Myosin Heavy Chain ( MHC ) and is situated adjacent to the actin interaction domain of the globular myosin head . In smooth muscle , MLC2 phosphorylation is sufficient to induce contraction , and in the striated muscle of the heart , phosphorylation increases the rate and magnitude of contractile force ( Davis et al . , 2001 , reviewed in Kamm and Stull ( 2001 ) ) . The 31 exons of the Mylk locus give rise to four distinct but overlapping gene products ( Figure 5A ) . Two 220-kD MLCK isoforms , which vary only by alternate inclusion of exon 1 , are expressed in non-muscle cell types , a 130-kD broadly expressed isoform and a 17-kD isoform , called Telokin , which lacks a functional kinase domain but shares an identical carboxy-terminal domain , and is specifically expressed in smooth , but not cardiac , muscle . The transcription of each isoform is regulated by individual promoters , which direct their spatially and temporally restricted expression ( Herring et al . , 2006 ) . When the 3′ UTR region of Mlck/Telokin containing the predicted miR-1 binding site was cloned downstream of a luciferase reporter , luciferase activity was repressed in the presence of a miR-1 mimic . The repression was alleviated when the target site was deleted , validating this 3’ UTR of Mlck/Telokin as a direct miR-1 target ( Figure 5B ) . 10 . 7554/eLife . 01323 . 029Figure 5 . Dysregulation of Telokin and Myosin light chain phosphorylation in miR-1 null hearts . ( A ) Diagram of the gene products encoded within the Mylk locus . Independent promoters preceding exons are indicated by arrows . Exon-spanning qPCR primers used are indicated in red ( Mlck/Telokin ) or blue ( Mlck ) . ( B ) Luciferase activity of a reporter construct containing ∼200 bp of the Mlck/Telokin 3′-UTR surrounding the predicted miR-1 binding site with or without the site deleted . The constructs were co-transfected into H9C2 myoblasts with a miR-1 mimic or a control mimic . Sequence of the putative miR-1 target site as predicted by Targetscan and site conservation between human ( Hs ) and mouse ( Mm ) is indicated . ( C ) Western blot of heart lysates ( top ) and qPCR of RNA ( bottom ) from P0 wild-type or miR-1 null mice . ( N = 5 per group ) . ( D ) Model of Telokin function in smooth muscle to promote the activity of myosin light chain phosphatase and inhibit the activity of the myosin light chain kinase . ( E ) Western blot of total myosin light chain 2 ( MLC2 ) and phosphorylated myosin light chain ( p-MLC2 ) in P0 wild-type or miR-1 null hearts; GAPDH serves as loading control . ( F ) qPCR of the Telokin promoter sequence or an intergenic genomic sequence following chromatin immunoprecipitation ( ChIP ) of RNA polymerase II in P2 wild-type or miR-1 null hearts . For the Telokin promoter , two non-overlapping probe sets were used , indicated as Telokin promoter A and B . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 02910 . 7554/eLife . 01323 . 030Figure 5—figure supplement 1 . Putative miR-1 targets dysregulated in miR-1 null hearts were not affected in miR-133a double-knockout hearts . qPCR of indicated transcripts in adult hearts from miR-133a double-knockout or double-heterozygous animals . N = 2 per genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 030 To determine which isoforms were upregulated in miR-1 double-knockout hearts , we examined the protein expression profiles of the various isoforms in wild-type and miR-1 double-knockout hearts by western blot ( Figure 5C ) . The non-muscle 210-kD isoform was undetectable , and the expression of the 130-kD MLCK isoform was not significantly altered in miR-1 double-knockout hearts . Telokin protein , as previously reported ( Herring et al . , 2001 ) , was not detectable in wild-type hearts , but was highly expressed in hearts of miR-1 double-knockouts . qPCR confirmed the normally smooth muscle-restricted Telokin was aberrantly expressed in miR-1 null myocardium ( Figure 5C ) . Notably , while smooth muscle gene expression is reported to be dysregulated in miR-133a-1:miR-133a-2 double-knockout hearts ( Liu et al . , 2008 ) , we found that Telokin expression was not upregulated in those animals ( Figure 5—figure supplement 1 ) , suggesting that the misexpression was specifically due to the loss of miR-1 . Despite lacking a catalytic kinase domain , Telokin plays an important role in the regulation of MLC phosphorylation in smooth muscle by inhibiting MLCK and promoting activity of the MLC phosphatase ( Choudhury et al . , 2004; Khromov et al . , 2012 ) ( Figure 5D ) . Consistent with aberrant Telokin expression in cardiomyocytes , MLC phosphorylation was dramatically decreased in miR-1 double-knockout hearts , likely contributing to the observed cardiac dysfunction ( Figure 5E ) . While direct repression by miR-1 may help to inhibit cardiac translation and stability of Telokin transcripts , it was unclear how loss of miR-1 resulted in the preferential upregulation of Telokin and not the full-length Mlck , which is thought to share a common 3′ UTR . To determine if the Telokin promoter was aberrantly active in miR-1 double-knockout hearts , we performed RNA Pol II chromatin immunoprecipitation and assayed for occupancy at the Telokin promoter . In wild-type hearts , Pol II occupancy at the Telokin promoter was equivalent to that of an untranscribed intergenic region , in agreement with Telokin’s published smooth muscle-restricted expression pattern ( Herring et al . , 2001 ) ( Figure 5F ) . In contrast , Pol II actively bound the Telokin promoter in miR-1 double-knockout hearts ( Figure 5F ) . These data indicate that miR-1 normally acts to repress Telokin expression in the heart by both directly targeting the Telokin 3′ UTR and by negatively regulating Telokin transcription . To gain mechanistic insight into how miR-1 may be negatively regulating Telokin transcription and to identify , at a more general level , transcriptional networks perturbed in the absence of miR-1 , we again utilized the GREAT interface ( McLean et al . , 2010 ) to identify known transcription factor motifs within the Msig database that were enriched within regulatory elements of the genes dysregulated in the double-knockouts . Interestingly , a disproportionate number of the genes upregulated in miR-1 double-knockout hearts contained CArG boxes ( CC/ATn/GG ) , a motif bound by SRF ( Treisman , 1986 ) ( Figure 6A , Figure 6—figure supplement 1 ) . SRF is a critical regulator of the cardiac and smooth muscle transcriptome and regulates sarcomere formation in cardiomyocytes ( Li et al . , 1997 , 2003; Wang et al . , 2001; Niu et al . , 2007; Hoofnagle et al . , 2011 ) . SRF is also a highly conserved direct upstream regulator of the miR-1/133a transcript ( Kwon et al . , 2005; Zhao et al . , 2005 ) and is itself a target of miR-133a , being upregulated in the hearts of miR-133a-1:miR-133a-2 double-knockout animals ( Chen et al . , 2005; Liu et al . , 2008 ) . However , SRF was not dysregulated at the RNA or protein level in miR-1 double-knockout hearts ( Figure 6B ) . This finding suggests that a simple increase in SRF expression is not responsible for the upregulation of SRF target genes and , furthermore , that the level of miR-133a expressed in miR-1 double-knockouts is sufficient to maintain normal SRF expression levels . 10 . 7554/eLife . 01323 . 031Figure 6 . Upregulation of SRF targets in miR-1 null hearts . ( A ) Promoter motif enrichment in genes upregulated in miR-1 double-knockout hearts . Motif sequences and transcriptional regulators are indicated ( left ) with the log10 of the hypergeometric p value graphed . Multiple CArG box sequences , the motif bound by SRF , were identified ( red ) . ( B ) Western blot ( top ) , and qPCR ( bottom ) for Srf expression in post-natal miR-1 null or wild-type hearts . ( C ) Myocardin dependent and independent SRF target gene expression in postnatal miR-1 null or wild-type hearts by qPCR ( left ) . Western blots of selected Myocardin-dependent SRF target genes ( right ) . Tagln , Transgelin/Sm22; Csrp1 , 2 , Cysteine And Glycine-Rich Protein 1 , 2; Cnn1 , 2 , Calponin1 , 2; Acta2 , smooth muscle alpha actin; miR-145 , microRNA-145; Fos , FBJ Murine Osteosarcoma Viral Oncogene Homolog; Egr1 , Early growth response 1 . p<0 . 05; p<0 . 01; p<0 . 001;**p<0 . 0001; ns , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 03110 . 7554/eLife . 01323 . 032Figure 6—figure supplement 1 . Dysregulation of SRF targets in miR-1 double-knockout hearts . Relative expression of SRF regulated genes identified via GREAT analysis . Smooth muscle genes are highlighted in red . Values indicated represent the normalized gene expression of the log2 FPKM . FPKM , fragments per kilobase per million . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 03210 . 7554/eLife . 01323 . 033Figure 6—figure supplement 2 . ( A ) qPCR of SRF co-factors in miR-1 wild-type or double-knockout hearts . N = 5 . ( B ) Western blot for protein expression of SRF co-factors . ns , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 033 One mechanism by which the specificity of SRF targets is conferred is through the expression and activity of tissue-specific cofactors . Myocardin ( MYOCD ) , which is expressed predominantly in cardiac and smooth muscle ( Wang et al . , 2001 ) , is one such cofactor . MYOCD is necessary and sufficient for smooth muscle differentiation in vitro and in vivo ( Li et al . , 2003; Chen et al . , 2002; Wang et al . , 2003 ) and is necessary for the differentiation and maintenance of ventricular cardiomyocytes ( Wang et al . , 2001; Hoofnagle et al . , 2011 ) . Among the SRF target genes that were dysregulated in the miR-1 double-knockouts , many , including Telokin , were known targets of the MYOCD/SRF complex ( Figure 6C , Figure 6—figure supplement 1 ) , while MYOCD independent SRF target genes , such as Fos and Egr1 , were not upregulated ( Figure 6C ) . Notably , the expression of numerous other known SRF co-factors , including NKX2-5 , MRTF-A , BRG1 , GATA4 and GATA6 were unchanged ( Figure 6—figure supplement 2 ) . These findings suggest that MYOCD/SRF dependent gene expression is specifically upregulated in the miR-1 knockout hearts . Using qPCR to compare the expression of Myocd in miR-1 double-knockout hearts and controls , we found that the expression of this critical transcription factor was increased roughly twofold upon deletion of miR-1 ( Figure 7A ) . Using Targetscan , we identified a potential miR-1 binding site in the Myocd 3′-UTR and demonstrated , using a luciferase assay , that miR-1 directly targets the Myocd 3′ UTR ( Figure 7B ) . Previous studies have shown that the Telokin promoter is more responsive to MYOCD than the full-length 130-kD Mlck promoter ( Herring et al . , 2006 ) . Thus , as we observed , an increase in MYOCD expression would be predicted to preferentially upregulate the transcription of the Telokin isoform , while only modestly increasing transcription of the full-length transcript . 10 . 7554/eLife . 01323 . 034Figure 7 . miR-1 regulates the SRF co-factor , Myocardin . ( A ) qPCR for Myocardin ( Myocd ) expression . Probe sets are specific for inclusion of exon 2a for smooth muscle Myocd ( smMyocd ) , exclusion of exon 2a for cardiac Myocd ( cMyocd ) , or for downstream exons common to both transcripts ( Total Myocd ) . P0 hearts were analyzed , N = 5 per genotype . ( B ) Luciferase activity of a reporter construct containing the miR-1 putative target site derived from the Myocd 3′ UTR or a deleted site . The constructs were co-transfected into H9C2 myoblasts with either a miR-1 mimic or a control mimic . The sequence of the putative miR-1 target site as predicted by Targetscan and site conservation between human ( Hs ) and mouse ( Mm ) is indicated . ( C ) The expression of luciferase driven by the Telokin+370 promoter when transfected into Cos cells , alone or with either full-length smMyocd or cMyocd . p<0 . 05; p<0 . 01; p<0 . 001;***p<0 . 0001; ns , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 034 There are two distinct Myocd isoforms , the full-length cardiac isoform ( cMYOCD ) and the truncated smooth muscle isoform ( smMYOCD ) ( Creemers et al . , 2006 ) . smMyocd is produced by alternate splicing and inclusion of exon2a , which encodes a translational stop and necessitates the use of an alternate downstream start site . This splicing event results in a truncated protein , lacking the N-terminal–MEF2 interaction domain of the full-length cMYOCD . Although the functional significance of these two isoforms is not well understood , the promoters of some smooth muscle-restricted genes , such as Tagln ( Sm22 ) , are more sensitive to smMYOCD than cMYOCD , while some cardiac promoters , such as Myh6 ( α-Mhc ) , are more responsive to cMYOCD ( Imamura et al . , 2010 ) . Normally , smMyocd expression is relatively low in the neonatal heart; however , we discovered using qPCR that in miR-1 double-knockout hearts the expression of smMyocd was fivefold higher than control hearts , while cMyocd was upregulated roughly 1 . 8-fold ( Figure 7A ) . More specific upregulation of smMyocd was not evident with statistical confidence in our RNA-seq data likely due to insufficient sequencing depth to detect the small 44 bp exon 2a inclusion within the relatively low overall abundance of smMyocd transcript . While cMyocd still represents the predominant isoform in miR-1 double-knockout hearts ( data not shown ) , subtle changes in transcription factor expression can result in dramatic changes in gene expression . Using a luciferase reporter cloned downstream of the mouse Telokin promoter , we found that smMYOCD was twice as effective as cMYOCD at activating the Telokin promoter ( Figure 7C ) . More broadly , we found that many of the MYOCD-dependent genes upregulated in the miR-1 knockouts were smooth muscle genes ( Figure 6—figure supplement 1 ) . Smooth muscle gene upregulation was also reported in the myocardium of miR-133a knockout mice , due in part to the upregulation of SRF . Neither total Myocd nor smMyocd , however , was upregulated in the myocardium of miR-133a double-knockouts , indicating that the miR-133a levels do not affect Myocd expression ( Figure 5—figure supplement 1 ) . These data suggest a model whereby miR-1 and miR-133a cooperate to repress smooth muscle gene transcription in the heart by repressing smMyocd and Srf , respectively , thereby reinforcing the striated muscle phenotype ( Figure 8B ) . 10 . 7554/eLife . 01323 . 035Figure 8 . miR-1 regulation of Myosin light chain phosphorylation and smooth muscle gene expression . ( A ) Dual regulatory model by which miR-1 acts to normally repress Telokin expression in the heart by directly targeting its 3′ UTR and its upstream transcriptional regulator , smooth muscle myocarding ( smMYOCD ) . MLCK , myosin light chain kinase; MLCP , myosin light chain phosphatase; MLC , myosin light chain; MHC , myosin heavy chain; P , phosphorylation . ( B ) Regulatory model by which the miR-1/133a cluster cooperates to suppress smooth muscle gene expression in the heart . DOI: http://dx . doi . org/10 . 7554/eLife . 01323 . 035 In ex vivo cardiomyocytes , restoring miR-1 expression alone was sufficient to partially rescue the sarcomeric defects of miR-1 double-knockout hearts , demonstrating that the loss of miR-1 plays a causative role in this phenotype , but not ruling out a contribution from decreased miR-133a expression . In this and previous studies ( Mishima et al . , 2009 ) , genes related to the actin cytoskeleton are highly responsive to alterations in miR-1 expression . MLC phosphorylation is critical for sarcomere assembly and regulating the speed and force of contraction in the heart ( Aoki et al . , 2000; Davis et al . , 2001 ) . In mammals , there are multiple MLC2 isoforms , but deletion of the single cardiac zebrafish isoform results in a lack of thick filament assembly ( Rottbauer et al . , 2006 ) . At a molecular level , phosphorylation of MLC2 results in a conformational shift in the associated MHC , bringing it into closer proximity to the thin filament and increasing the probability of cross-bridge formation . Not surprisingly , in humans , MLC2 mutations are associated with myopathies ( Poetter et al . , 1996; Davis et al . , 2001 ) . Thus , decreased phosphorylation of MLC in the absence of miR-1 likely plays a role in the sarcomeric disruption and cardiac dysfunction in miR-1 double-knockouts . We found that one mechanism by which miR-1 normally acts to maintain p-MLC in the heart is through the repression of the usually smooth muscle-restricted protein Telokin . Telokin expression in smooth muscle is thought to maintain cells in a less contractile state , and it is interesting that the heart employs miR-1 to ensure the absence of Telokin through both transcriptional and post-transcriptional mechanisms ( Figure 8A ) . Although it is not possible to distinguish the contribution of transcriptional regulation of Telokin vs direct miR-1 targeting of the Telokin 3′ UTR , the dramatic degree of Telokin protein upregulation compared to the more moderate Telokin mRNA increase is consistent with some degree of translational control by miR-1 ( Figure 4C ) . How Telokin misexpression affects cardiac function has not been directly evaluated; however , ectopic expression in the myocardium has been reported in a doxorubicin induced rat model of cardiomyopathy , although the precise mechanism by which Telokin contributes to this pathology is unknown ( Dudnakova et al . , 2003 ) . While the focus of this study was to explore the function of miR-1 in the heart , miR-1 likely plays a more subtle role in the skeletal muscle . It is worth noting that electron microscopy of P0 miR-1-null skeletal muscle showed the presence of ordered sarcomeric structures ( data not shown ) and neonatal mice were mobile . While we cannot rule out a more subtle skeletal muscle phenotype due to the early lethality , it is possible that the skeletal muscle specific miR-1 family member , miR-206 , can compensate for the loss of miR-1 in this tissue . A clear understanding of the role that the miR-1 family plays in the context of skeletal muscle will require the generation a skeletal muscle-specific compound deletion of miR-1-1 , miR-1-2 and miR-206 . The marked downregulation of miR-133a in the miR-1 double-knockouts is a critical consideration with respect to the interpretation of some of the phenotypes that we observed . While targeted disruption of miR-1 might adversely affect transcriptional read-through of the miR-1-1/miR-133a-2 locus , secondary downregulation of miR-133a expression from regulatory feedback mechanisms also appears to occur ( Figure 2—figure supplement 10 ) . miR-1 and miR-133a are dramatically upregulated during the differentiation of cardiac and skeletal muscle ( Chen et al . , 2005; Ivey et al . , 2008 ) . While the current study shows definitively that miR-1 is not required for initial cardiomyocyte differentiation in vivo , the loss of miR-1 expression may impair further cardiomyocyte differentiation and/or maturation . This would predictably result in decreased developmental upregulation of the miR-1/133 loci . Consistent with this model , the loosely organized sarcomeric morphology and the abnormal mitochondrial morphology we observed by TEM are similar to those of more immature cardiomyocytes ( Hom et al . , 2011 ) . The upregulation of members of the cardiac fetal gene expression program , such as Nppa ( Figure 6—figure supplement 1 ) , and even the upregulation of smooth muscle genes , which are normally expressed in embryonic cardiomyocytes , supports this paradigm . As such , the expression of miR-133a , a key developmentally regulated miRNA , may be secondarily downregulated in miR-1 double-knockout mice as a result of a subtle cardiac differentiation or maturation defect , although the exact mechanism is unknown . The power of miRNAs to affect biological processes is amplified by the ability of a single miRNA to regulate multiple nodes within a genetic network . Srf is a direct target of miR-133a , and its expression , as well as that of many smooth muscle target genes , is upregulated in the hearts of miR-133a-knockout animals ( Chen et al . , 2005; Liu et al . , 2008 ) . We found that SRF mRNA and protein expression in miR-1 double-knockout hearts was unaffected , despite the partial downregulation of miR-133a . Nevertheless , some of the smooth muscle genes upregulated in miR-133a double-knockout hearts were also upregulated in the miR-1 double-knockout animals ( Acta2 , Cald1 , Csrp2 , Cnn1 , Tagln ) , while others , including Telokin , cMyocd and smMyocd , were uniquely upregulated in the miR-1 double-knockouts . Our data suggest that miR-1 and miR-133a , transcribed together from bicistronic miRNA clusters , cooperate to repress SRF-dependent smooth muscle gene expression in the heart by independently regulating SRF or its co-factor , MYOCD , respectively ( Figure 8B ) . We also found that the Telokin promoter is preferentially responsive to the smMYOCD isoform , upregulated in the absence of miR-1 . This is consistent with the observation that , while both MYOCD isoforms share an identical SRF-interacting domain , they are not functionally equivalent with respect to mediating the transcription of SRF target genes ( Creemers et al . , 2006; Imamura et al . , 2010 ) . While yet to be carefully evaluated , cMyocd and smMyocd are thought to share a common 3′ UTR; thus , it is unlikely that direct targeting by miR-1 alone is responsible for the observed isoform switch . The splice factor ( s ) that regulate the inclusion or exclusion of the 2a exon in the smooth muscle or heart , respectively , are unknown . Therefore , a potential mechanism by which miR-1 may regulate Myocd splicing is through the direct repression of a splice factor that normally mediates alternate inclusion of the 2a exon in smooth muscle . Further studies will be required to elucidate the mechanism by which miR-1 regulates this critical transcription factor and ultimately smooth muscle gene expression . The miR-1-1 targeting vector was generated by flanking 5′ and 3′ genomic DNA around a floxed neomycin resistance gene ( Neo ) driven by a pGK promoter . Targeting of the miR-1-1 locus was accomplished through homologous recombination in E14 ES cells and resulted in replacement of the genomic sequence containing the pre-miR-1-1 ( ∼280 bp ) by the Neo gene . 672 colonies were screened , and three targeted clones were identified . miR-1-1 heterozygous embryonic stem cells were injected into D3 . 5 BL6 blastocysts . For the establishment of the miR-1-1-targeted allele on a 129 background , chimeric males were mated to wild-type 129 ( Jackson Lab ) females . To generate compound miR-1 knockouts , miR-1-1-targeted animals were crossed to a previously described miR-1-2 targeted mouse maintained on a mixed ( 129/BL6 ) background . Animals double heterozygous for both miR-1 alleles ( miR-1-1+/−:miR-1-2+/− ) were subsequently intercrossed to generate animals lacking three out of four alleles of miR-1 ( miR-1-1−/−:miR-1-2+/− or miR-1-1+/−:miR-1-2−/− ) . These animals were subsequently intercrossed to generate compound miR-1 knockout animals ( miR-1-1−/−:miR-1-2−/− ) . Genotyping was performed by PCR [Supplementary file 1] with primers that specifically recognized the wild-type or targeted miR-1-1 or miR-1-2 alleles . The animals were sacrificed by decapitation for neonates or by CO2 , followed by cervical dislocation for adult animals . All animal care and experimental protocols were reviewed and approved by the Institutional Animal Care and Use Committee of the University of California , San Francisco ( UCSF ) . For electron microscopy , tissue was fixed in 2% glutaraldehyde , 1% paraformaldehyde in 0 . 1 M sodium cacodylate buffer pH 7 . 4 , post fixed in 2% osmium tetroxide in the same buffer , en block stained in 2% aqueous uranyl acetate , dehydrated in acetone , infiltrated , and embedded in LX-112 resin ( Ladd Research Industries , Williston , VT ) . Toluidine blue stained semi-thin sections were made to locate the areas of interest . The samples were ultrathin sectioned on a Reichert Ultracut S ultramicrotome and counter stained with 0 . 8% lead citrate . Grids were examined on a JEOL JEM-1230 transmission electron microscope ( JEOL USA , Inc . , Pleasanton , CA ) and photographed with the Gatan Ultrascan 1000 digital camera ( Gatan Inc . , Pleasanton , CA ) . Adult mouse echocardiography was performed under anesthesia as described ( Qian et al . , 2011 ) . Electrocardiograms were performed as described ( Zhao et al . , 2007 ) . Neonatal studies were performed as described above without anesthesia . Tissue was collected at indicated times and fixed using 10% formalin overnight at 4°C and stored subsequently in 70% ethanol . Paraffin embedding and staining was performed using standard histological techniques . Sarcomeric staining was performed using anti-sarcomeric alpha-actinin ( 1:400; Sigma , St . Louis , MO ) and rhodamine conjugated Phalloidin ( Clontech , Mountain View , CA ) . Hearts were isolated from P0 animals and rinsed several times in 1X HBSS with Pen-Strep . The great vessels and atrium were removed and discarded , and the ventricles were minced manually with scissors and further disassociated enzymatically with collagenase digestion . 1 mg/ml collagenase II was added to the minced ventricles and briefly incubated at 37°C for 3–5 min with agitation . The digests were allowed to settle without agitation briefly ( 3 min ) before the supernatant of this initial digestion ( mostly blood cells ) was discarded and another digestion was performed for 15–20 min . The supernatant of this subsequent digestion ( cardiomyocytes ) was added to 3X volume of FBS , and centrifuged for 5 min at 300 × g in a tabletop centrifuge . The resulting cell pellet was resuspended in DMEM/F12 medium with 10% FBS and Pen-Strep and passed through a 70-μM filter before plating on 1% gelatin with fibronectin . Cardiomyocytes for sarcomeric analysis were fixed and stained 24 hr post-plating . The cardiomyocytes used for sarcomeric rescue studies were transfected with miR-1 mimic or control ( Ambion/Life Technologies , Carlsbad , CA ) roughly 12 hr post-plating with Lipofectamine 2000 ( Life Technologies ) and maintained in serum-containing medium . The cells were fixed for analysis 24 hr post-transfection . Roughly 200 bp surrounding the predicted miR-1 target sites in the Mlck/Telokin or Myocd 3′ UTR were amplified directly from cDNA generated from miR-1 double-knockout hearts with the primers listed ( Supplementary file 1 ) and subcloned into the PGL3 ( Promega , Madison , WI ) firefly luciferase vector 3′ of the reporter gene with XbaI . Luciferase constructs were transfected along with a Renilla normalization vector into the H9C2 rat myoblast cell line with Lipofectamine 2000 ( Life Technologies ) , according to manufacturer’s instructions . Briefly , 12-well plates were transfected at roughly 60% confluency and analyzed 20 hr later . Each well received 3 μl of Lipofectamine 2000 , 800 ng of PGL3-Target , and 200 ng of Renilla vector . Experimental wells received 10 pmols of miR-1 mimic ( Ambion/Life Technologies ) , while control wells received 10 pmols of a non-targeting control mimic ( Ambion/Life Technologies ) . Promoter activity assays were performed with the Telokin+370 promoter ( a gift from Dr Paul Herring , Indiana University ) . cMyocd and smMyocd constructs were as described in Wang and Olson ( 2004 ) Cordes et al . , ( 2009 ) . Promoter constructs ( 500 ng ) , Myocd ( 250 ng ) or control LacZ expression plasmid and 50 ng of Renilla were transfected into Cos cells with Lipofectamine 2000 at 50% confluency . Luciferase intensity was analyzed 20 hr post-transfection . Firefly and Renilla luciferase activities were quantified in lysates with the Dual Luciferase Reporter Assay kit ( Promega ) on a Victor 1420 Multilabel Counter ( PerkinElmer , Madison , WI ) . Firefly luciferase values were normalized to Renilla to control for transfection efficiency . RNA was isolated with TRIzol reagent according to the manufacturer’s protocol . Quantitative real-time PCR for microRNAs was performed with the TaqMan miRNA assay kit ( Applied Biosystems/Life ) , and TaqMan probes for miR-1 ( Applied Biosystems/LIfe ) , miR-133a ( Applied Biosystems/Life ) were used according to the manufacturer’s protocols . cDNA for mRNA quantification was generated using Superscript with oligodT and random hexamers ( Invitrogen ) . Detection of splice variants for Myocd was performed using oligo dT generated cDNA ( Invitrogen/Life ) . qPCR probes for the Telokin promoter , and Myocd splice variants were designed using Primerquest software and synthesized by Integrated DNA Technology ( IDT , San Jose , CA ) . Intergenic region primers were designed based on sequence published previously . The probe sequences and part numbers are listed in Supplementary file 1 . Expression values were normalized to the expression of U6 ( Applied Biosystems/Life ) for miRNA analysis or Gapdh ( Applied Biosystems/Life ) for mRNA quantification , and fold change was determined using the ΔΔCT method with SDS RQ Manager software ( Applied Biosystems/Life ) . Whole hearts from E18 wild-type and miR-1 double-knockout animals were isolated and total RNA was extracted with TRIzol ( Invitrogen/Life ) , following the manufacturer’s suggested protocol . Genomic DNA was removed using a gDNA eliminator column ( Qiagen , Hilden , Germany ) . RNA from three hearts of each genotype was pooled , and library preparation and sequencing were performed by the Beijing Genomics Institute ( BGI ) . In brief , polyA transcripts were enriched and paired-end reads were sequenced on a High-seq 2000 ( Illumina , San Diego , CA ) . Reads were mapped to the mm 9 genome , Ensembl v 59 annotation , with TopHat2 ( Kim et al . , 2013 ) . Rank expectation ( Thomas et al . , 2011 ) was used to identify genes that were differentially expressed between the two backgrounds , using a false discovery rate threshold of 0 . 1 . Neonatal hearts ( P2 ) from wildtype , miR-1-1 null and miR-1-2+/−:miR-1-1+/− double heterozygote animals were isolated and total RNA was extracted with TRIzol ( Invitrogen/Life ) , following the manufacturer’s suggested protocol . Genomic DNA was removed using a gDNA Eliminator column ( Qiagen ) . The samples were hybridized to Affymetrix Mouse Genechip ST 2 . 0 arrays . All arrays were RMA normalized and differentially expressed genes were identified using Limma . Gene changes were compared to an existing data set ( Zhao et al . , 2007 ) for miR-1-2 null animals . 822 genes were statistically changed in one of the three genotypes when compared to the wild-type control ( p value of 0 . 0025 and a fold change of greater than 1 . 5 or less than 0 . 5 ) . Of these differentially expressed genes , 201 genes could be evaluated across all genotypes due to the difference in array probe sets . Hearts were isolated from animals of the indicated genotypes between P0 and P3 and rinsed with 1X PBS . The tissue was resuspended in RIPA buffer and disassociated in a Bullet Blender ( Next Advance ) . After a clarification and sonication step , the lysates were loaded on to a 4–20% SDS-PAGE ( Biorad , Hercules , CA ) gel and blotted using standard protocols . Primary antibodies against GAPDH ( 1:1000; Abcam ) , SRF ( 1:200; Santa Cruz Bio . , Dallas , TX ) , CSRP1 ( 1:500; Abcam ) , TAGLN/Sm22 ( 1:500; Abcam ) , p-MLC ser18/thr19 ( 1:1000; Cell Signaling ) , MLCK/Telokin ( 1:1000; Abcam , Cambridge , England ) , ACTA2 ( 1:2000; Sigma ) . Visualization and quantification of blots was done on a Licor Odyssey system with fluorescently conjugated secondary antibodies ( Licor , Lincoln , NE ) , according to manufacturer’s instructions . Pol II ChIP was performed as in Lee et al . ( 2006 ) with minor modifications . qPCR primer sequences described in Supplementary file 1 . Data were analyzed with Prism and/or Excel and an unpaired t test was used to determine statistical significance . Predicted genotype ratios were calculated by chi-square analysis . Relative quantity for qPCR analysis was determined using ΔΔCT method . Values for experimental animals normalized to the average of controls .	MicroRNAs are tiny RNAs that do not encode proteins . Instead , they regulate the expression of genes by preventing protein-encoding messenger RNAs from being translated into protein . MicroRNAs are expressed throughout the body , including the heart , where the most abundant microRNA is called miR-1 . This is encoded by two nearly identical genes: miR-1-1 and miR-1-2 . Mice that lack the miR-1-2 gene have various heart abnormalities , but generally survive because they still produce some miR-1 from their remaining miR-1-1 gene . Now , Heidersbach et al . have generated the first mice that specifically lack both miR-1 genes , and shown that these animals die before weaning . When viewed under the electron microscope , heart muscle from miR-1 double knockout mice lacks the characteristic ‘striped’ , or striated , appearance of normal heart muscle . Additionally , miR-1 double knockout hearts have some gene expression characteristics more similar to the smooth muscle found in the gut and in the walls of blood vessels . Smooth muscle differs from striated muscle in that it lacks sarcomeres: these are bands of fibrous proteins , such as myosin , that are essential for muscle contraction . In normal mice , an enzyme called MLCK contributes to the formation and function of sarcomeres by adding phosphate groups to myosin molecules . By contrast , in smooth muscle an enzyme called Telokin promotes phosphate group removal , and thus affects the function of sarcomeres . Heidersbach et al . showed that miR-1 interacts directly with Telokin mRNA to prevent its expression in the heart , and simultaneously represses a protein called Myocardin , which directly activates transcription of Telokin . However , when miR-1 is absent , as in the miR-1 double knockout mice , Telokin is expressed in heart muscle , along with many other genes characteristic of smooth muscle . As well as improving our understanding of the development and functioning of the heart , these findings should shed new light on the role of microRNAs in maintaining the patterns of gene expression that characterize unique cell fates .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "developmental", "biology" ]	2013	microRNA-1 regulates sarcomere formation and suppresses smooth muscle gene expression in the mammalian heart
Goal-directed behavior involves distributed neuronal circuits in the mammalian brain , including diverse regions of neocortex . However , the cellular basis of long-range cortico-cortical signaling during goal-directed behavior is poorly understood . Here , we recorded membrane potential of excitatory layer 2/3 pyramidal neurons in primary somatosensory barrel cortex ( S1 ) projecting to either primary motor cortex ( M1 ) or secondary somatosensory cortex ( S2 ) during a whisker detection task , in which thirsty mice learn to lick for water reward in response to a whisker deflection . Whisker stimulation in ‘Good performer’ mice , but not ‘Naive’ mice , evoked long-lasting biphasic depolarization correlated with task performance in S2-projecting ( S2-p ) neurons , but not M1-projecting ( M1-p ) neurons . Furthermore , S2-p neurons , but not M1-p neurons , became excited during spontaneous unrewarded licking in ‘Good performer’ mice , but not in ‘Naive’ mice . Thus , a learning-induced , projection-specific signal from S1 to S2 may contribute to goal-directed sensorimotor transformation of whisker sensation into licking motor output . Primary sensory cortex processes incoming sensory information flexibly in an experience , context and task-dependent manner ( Gilbert and Li , 2013; Harris and Mrsic-Flogel , 2013 ) . Functionally-tuned sensory information is signaled from primary sensory cortex to distinct cortical areas ( Movshon and Newsome , 1996; Sato and Svoboda , 2000; Chen et al . , 2013; Glickfeld et al . , 2013; Yamashita et al . , 2013 ) , but the cellular mechanisms underlying specific cortico-cortical signals during goal-directed behavior are poorly understood . Neuronal activity in primary somatosensory barrel cortex ( S1 ) is known to participate in the execution of a simple whisker-dependent detection task , in which thirsty mice learn to lick a spout in order to obtain a water reward ( Sachidhanandam et al . , 2013 ) . In well-trained mice , putative excitatory neurons in layer 2/3 of S1 , on average , have a long-lasting biphasic depolarization after whisker deflection in hit trials , whereas in miss trials the late depolarization is smaller in amplitude ( Sachidhanandam et al . , 2013 ) . However , there is considerable variability across different recordings ( Sachidhanandam et al . , 2013 ) , which could in part relate to distinct types of excitatory projection neurons . Layer 2/3 of S1 barrel cortex has major anatomical ipsilateral cortico-cortical connections to primary whisker motor cortex ( M1 ) and secondary somatosensory cortex ( S2 ) ( Aronoff et al . , 2010 ) . M1-projecting ( M1-p ) and S2-projecting ( S2-p ) neurons in layer 2/3 of S1 are likely to be distinct cell-types exhibiting differential patterns of gene expression ( Sorensen et al . , 2015 ) , distinct intrinsic electrophysiological properties in vivo ( Yamashita et al . , 2013 ) , and carrying functionally different signals ( Sato and Svoboda , 2010; Chen et al . , 2013; 2015; Yamashita et al . , 2013 ) . Retrograde labeling suggests that M1-p and S2-p neurons in S1 are largely non-overlapping types of excitatory neurons ( Sato and Svoboda , 2010; Chen et al . , 2013; Yamashita et al . , 2013 ) . Here , we investigate the cellular basis of selective signaling of sensorimotor information in distinct cortico-cortical pathways during the whisker detection task through membrane potential ( Vm ) recordings of M1-p and S2-p neurons , finding that task learning induces a licking-related depolarization specifically in S2-p neurons . Thirsty mice were trained to lick for water reward in response to a 1 ms deflection of the right C2 whisker ( Sachidhanandam et al . , 2013; Sippy et al . , 2015 ) , and whole-cell Vm recordings were targeted through two-photon microscopy to fluorescently-labelled M1-p and S2-p neurons in layer 2/3 of the C2 barrel column in S1 of the left hemisphere ( Yamashita et al . , 2013 ) ( Figure 1A , B ) . We used two types of mice for recordings: ( 1 ) ‘Good performer’ mice that exhibited a high discriminability between test trials and catch trials ( for details see Materials and Methods ) during recordings ( 59 recordings in 27 mice; hit rate , 0 . 77 ± 0 . 03; false alarm rate , 0 . 17 ± 0 . 01; d’ = 2 . 12 ± 0 . 09; d’ > 1 . 1 for each recording; Figure 1C ) learned through training sessions ( typically 7–13 daily sessions prior to the recording day , but some mice learned more quickly ) , responding with a reaction time of 317 ± 17 ms ( time from whisker deflection to tongue contact with the water spout ) ( Figure 1D ) ; and ( 2 ) ‘Naive’ mice that were used for recordings on the first day of being exposed to the task and showed no apparent discrimination ( 36 recordings in 16 mice; hit rate , 0 . 31 ± 0 . 03; false alarm rate , 0 . 28 ± 0 . 03; d’ = 0 . 03 ± 0 . 09; d’ < 0 . 9 , for each recording; Figure 1C ) , with a mean reaction time of 369 ± 23 ms which was significantly slower than ‘Good performer’ mice ( p=0 . 0014; Figure 1E ) . 10 . 7554/eLife . 15798 . 003Figure 1 . Target-specific Vm dynamics in S1 projection neurons during task performance . ( A ) Top , the experimental setup . Bottom , a representative two-photon image of a CTB-labeled M1-p neuron ( green ) with a recording pipette ( red ) . ( B ) Detection task trial structure ( FA: false alarm , CR: correct rejection ) . ( C ) Behavioral performance during whole-cell recordings ( HR: hit rates , FAR: false alarm rates , GP: ‘Good performer’ , N: ‘Naive’ ) . ( D , E ) Left , grand average changes in Vm ( thick line: mean , thin lines: ± sem ) and action potential ( AP ) firing rate for hit trials recorded from S2-p neurons ( red ) and M1-p neurons ( blue ) in ‘Good performer’ ( D ) and ‘Naive’ ( E ) mice ( Arrow: 1 ms stimulation of the C2 whisker ) . A box plot indicates reaction time ( 1st lick ) . Right , box plots for postsynaptic potential ( PSP ) amplitude , secondary late Vm depolarization quantified at 0 . 05–0 . 25 s , Vm depolarization during the lick period ( at 0 . 25–1 . 0 s ) , and evoked AP rates at early ( 0–0 . 05 s ) , late ( 0 . 05–0 . 25 s ) and lick ( 0 . 25–1 . 0 s ) periods . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 00310 . 7554/eLife . 15798 . 004Figure 1—source data 1 . Data values and statistics underlying Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 00410 . 7554/eLife . 15798 . 005Figure 1—source data 2 . Data values and statistics underlying Figure 1—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 00510 . 7554/eLife . 15798 . 006Figure 1—source data 3 . Data values and statistics underlying Figure 1—figure supplement 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 00610 . 7554/eLife . 15798 . 007Figure 1—figure supplement 1 . Hit Vm traces from S1 projection neurons in ‘Good performer’ mice . ( A ) Left and middle , example Vm traces obtained from two individual S2-p neurons on two representative hit trials . Right , the averaged subthreshold Vm trace is superimposed with individual traces . APs are truncated . PSTHs for corresponding AP rates and the distribution of the first lick timings are also shown . ( B ) Same as A , but for M1-p neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 00710 . 7554/eLife . 15798 . 008Figure 1—figure supplement 2 . Hit Vm traces from S1 projection neurons in ‘Naive’ mice . ( A ) Left and middle , example Vm traces obtained from two individual S2-p neurons on two representative hit trials . Right , the averaged subthreshold Vm trace is superimposed with individual traces . APs are truncated . PSTHs for corresponding AP rates and the distribution of the first lick timings are also shown . ( B ) Same as A , but for M1-p neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 00810 . 7554/eLife . 15798 . 009Figure 1—figure supplement 3 . Average hit Vm traces and PSTHs . ( A ) Average hit Vm traces obtained from individual cells ( thin lines ) from S2-p ( left ) and M1-p ( right ) neurons of ‘Good performer’ mice . The grand average Vm traces ( thick lines ) are superimposed . The dotted line indicates the stimulus onset . ( B ) Left , grand average of subthreshold responses from S2-p ( red ) and M1-p ( blue ) neurons are shown at high temporal resolution ( superimposed with the baseline Vm subtracted ) . An arrow indicates the onset of whisker stimulation . Middle , grand average PSTHs at high temporal resolution with the baseline AP rates subtracted . Time 0 is the onset of whisker stimulation . Right , quantification of evoked AP rate at initial 20 ms after whisker deflection . ( C and D ) Same as A and B , but for ‘Naive’ mice . ( E ) Quantification of △Vm and △AP rate at 0 . 05 – 0 . 35 s and 0 . 35 – 1 . 0 s on hit trials in ‘Naive’ mice . ( F ) PSP amplitude , Vm depolarization at the late ( 0 . 05–0 . 25 s ) and lick ( 0 . 25 – 1 . 0 s ) periods and evoked AP rates at early ( 0 – 0 . 05 s ) , late ( 0 . 05 – 0 . 25 s ) and lick ( 0 . 25 – 1 . 0 s ) periods were analyzed on a mouse-by-mouse basis for ‘Good performer’ mice ( n = 9 mice for M1-p neurons; n = 17 mice for S2-p neurons ) . ( G ) Same as F , but for ‘Naive’ mice ( n = 6 mice for M1-p neurons; n = 9 mice for S2-p neurons ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 00910 . 7554/eLife . 15798 . 010Figure 1—figure supplement 4 . Target-specific changes of hit responses with task learning . ( A and B ) Left , grand average changes in Vm ( thick line: mean , thin lines: ± sem ) and AP rate for hit trials recorded from S2-p neurons ( A ) and M1-p neurons ( B ) in ‘Good performer’ ( colored ) and ‘Naive’ ( black ) mice ( Arrow: 1 ms stimulation of the C2 whisker ) . Right , box plots for PSP amplitude , Vm depolarization at the late ( 0 . 05–0 . 25 s ) and lick ( 0 . 25 – 1 . 0 s ) periods and evoked AP rates at early ( 0 – 0 . 05 s ) , late ( 0 . 05 – 0 . 25 s ) and lick ( 0 . 25 – 1 . 0 s ) periods . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 010 In trained ‘Good performer’ mice , whisker stimulation evoked a biphasic Vm depolarization in hit trials for both M1-p and S2-p neurons ( Figure 1D , Figure 1—figure supplement 1 , 3 ) . The early sensory response was not different comparing M1-p and S2-p neurons ( p=0 . 40 ) , but S2-p neurons had significantly larger Vm depolarization ( △Vm ) during the late phase ( △Vm at 0 . 05 – 0 . 25 s after whisker deflection: S2-p=4 . 00 ± 0 . 59 mV , n = 31; M1-p=1 . 68 ± 0 . 44 mV , n = 22; p=0 . 0035 ) and during the subsequent lick period ( △Vm at 0 . 25 – 1 . 0 s after whisker deflection: S2-p=3 . 10 ± 0 . 50 mV , n = 31; M1-p=0 . 73 ± 0 . 30 mV , n = 22; p=0 . 00038 ) ( Figure 1D , Figure 1—figure supplement 1 , 3 ) . The evoked action potential ( AP ) rate of S2-p neurons compared to M1-p neurons was also significantly higher during the lick period ( p=0 . 015 ) , but not during early ( p=0 . 57 ) or late ( p=0 . 98 ) response periods ( Figure 1D , Figure 1—figure supplement 1 , 3 ) . In randomly licking ‘Naive’ mice ( Figure 1C ) , M1-p neurons , compared to S2-p neurons , exhibited larger postsynaptic potentials ( PSPs ) in response to whisker stimulation in hit trials ( PSP amplitude: S2-p=8 . 41 ± 0 . 69 mV , n = 14; M1-p=12 . 20 ± 1 . 25 mV , n = 12; p=0 . 043 ) and larger depolarizations during the licking phase ( △Vm at 0 . 25 – 1 . 0 s after whisker deflection: S2-p=0 . 45 ± 0 . 62 mV , n = 14; M1-p=2 . 19 ± 0 . 57 mV , n = 12; p=0 . 043 ) ( Figure 1E , Figure 1—figure supplement 2 , 3 ) . The AP rates in ‘Naive’ mice during the lick period were also significantly larger in M1-p neurons compared to S2-p neurons ( p=0 . 046 ) ( Figure 1E , Figure 1—figure supplement 2 , 3 ) . Therefore , analyzed for hit trials , S2-p neurons were more strongly excited during licking compared to M1-p neurons in ‘Good performer’ mice , but , interestingly , the opposite was true for ‘Naive’ mice in which M1-p neurons were more excited during licking compared to S2-p neurons . Notably , the secondary long-lasting depolarization in S2-p neurons after whisker stimulation was seen only in ‘Good performer’ mice , not in ‘Naive’ mice ( Figure 1D , E , Figure 1—figure supplement 4 ) , while the small sustained depolarization of M1-p neurons in ‘Naive’ mice was attenuated in ‘Good performer’ mice ( Figure 1D , E , Figure 1—figure supplement 4 ) . We next examined whether the Vm dynamics of S1 projection neurons correlated with task execution on a trial-by-trial basis . In S2-p neurons of ‘Good performer’ mice , the amplitude of PSPs and the late △Vm were slightly larger in hit compared to miss trials ( PSPs increased by 20% , p=0 . 026; late △Vm increased by 39% , p=0 . 029 ) ( Figure 2A , Figure 2—figure supplement 1 and 2 ) . Furthermore , the △Vmin S2-p neurons during the licking period was substantially larger ( 270% increase ) in hit compared to miss trials ( △Vm at 0 . 25 – 1 . 0 s: hit 2 . 63 ± 0 . 55 mV , miss 0 . 71 ± 0 . 47 mV , n = 19 , p=0 . 0014 ) ( Figure 2A , Figure 2—figure supplement 1 and 2 ) . Thus , the Vm dynamics of S2-p neurons after whisker deflection were correlated with task execution in trained mice . However , hit and miss trials were not significantly different in M1-p neurons of ‘Good performer’ mice in early ( p=0 . 23 ) , late ( p=0 . 43 ) or licking ( p>0 . 99 ) phases ( Figure 2B , Figure 2—figure supplement 1 and 2 ) . 10 . 7554/eLife . 15798 . 011Figure 2 . Target-specific Vm correlation with task execution . ( A ) Left , grand average Vm traces ( thick line: mean , thin lines: ± sem ) of S2-p neurons during hit ( red ) and miss ( black ) trials for ‘Good performer’ mice . Right , data for each cell ( thin lines ) and box plots for PSP amplitude and Vm depolarization at the late ( 0 . 05–0 . 25 s ) and lick periods ( 0 . 25–1 . 0 s ) on hit ( H ) and miss ( M ) trials . ( B ) Same as A , but for M1-p neurons . ( C ) Same as A , but for S2-p neurons in ‘Naive’ mice . ( D ) Same as C , but for M1-p neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 01110 . 7554/eLife . 15798 . 012Figure 2—source data 1 . Data values and statistics underlying Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 01210 . 7554/eLife . 15798 . 013Figure 2—source data 2 . Data values and statistics underlying Figure 2—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 01310 . 7554/eLife . 15798 . 014Figure 2—figure supplement 1 . Representative hit/miss Vm traces . ( A ) Example average subthreshold Vm traces on hit ( colored ) and miss ( black ) trials from four individual S2-p neurons in trained ‘Good performer’ mice . Arrows: the onset of whisker stimulus . Insets: PSPs of corresponding traces at high temporal resolution . ( B ) Same as A , but for M1-p neurons . ( C ) Same as A , but for S2-p neurons in ‘Naive’ mice . ( D ) Same as C , but for M1-p neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 01410 . 7554/eLife . 15798 . 015Figure 2—figure supplement 2 . Mouse-by-mouse analysis of hit/miss responses . ( A ) Data from S2-p neurons in ‘Good performer’ mice averaged for each mouse ( thin lines , n = 14 mice ) and box plots for PSP amplitude and Vm depolarization at the late ( 0 . 05–0 . 25 s ) and lick periods ( 0 . 25–1 . 0 s ) on hit ( H ) and miss ( M ) trials . ( B ) Same as A , but for M1-p neurons ( n = 7 mice ) . ( C ) Same as A , but for S2-p neurons in ‘Naive’ mice ( n = 9 mice ) . ( D ) Same as C , but for M1-p neurons ( n = 6 mice ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 015 In contrast , for ‘Naive’ mice , S2-p neurons did not distinguish hit and miss trials in early ( p=0 . 24 ) , late ( p=0 . 67 ) or licking ( p=0 . 71 ) response phases ( Figure 2C , Figure 2—figure supplement 1 and 2 ) . M1-p neurons in ‘Naive’ mice also had similar hit and miss responses during early ( p=0 . 42 ) and late ( p=0 . 13 ) periods . However , M1-p neurons in ‘Naive’ mice had significantly larger △Vm during the lick period in hit trials compared to misses ( △Vm at 0 . 25 – 1 . 0 s: hit 2 . 19 ± 0 . 57 mV , miss 0 . 30 ± 0 . 38 mV , n = 12 , p=0 . 034 ) ( Figure 2D , Figure 2—figure supplement 1 ) . Thus , S2-p neurons , but not M1-p neurons , in ‘Good Performer’ mice had larger depolarizing responses in hit trials compared to misses , whereas in ‘Naive’ mice M1-p neurons , but not S2-p neurons , had a larger depolarization during licking in hit trials compared to misses . Some S2-p neurons depolarized strikingly during spontaneous unrewarded licking ( Figure 3—figure supplement 1 ) . We therefore examined licking-related Vm dynamics and found that S2-p neurons of ‘Good performer’ mice depolarized during spontaneous unrewarded licking , peaking at around the time when the tongue first contacted the water spout , ( △Vm at ± 0 . 1 s around tongue-spout contact: 3 . 48 ± 0 . 62 mV , n = 20 ) . Licking-related depolarization was significantly ( p=0 . 0045 ) smaller in M1-p neurons of ‘Good performer’ mice ( △Vm: 0 . 83 ± 0 . 44 mV , n = 10 ) ( Figure 3A , Figure 3—figure supplement 1 , 3 ) . S2-p neurons of ‘Good performer’ mice also increased firing rate significantly during licking compared to M1-p neurons ( p=0 . 027 ) . Licking-related Vm and AP modulation was weak in ‘Naive’ mice , and it was not significantly different comparing S2-p and M1-p neurons ( △Vm , p=0 . 060; △AP , p=0 . 30 ) ( Figure 3B , Figure 3—figure supplement 2 , 3 ) . 10 . 7554/eLife . 15798 . 016Figure 3 . Target-specific Vm depolarization during spontaneous unrewarded licking . ( A , B ) Left , grand average change in Vm ( thick line: mean , thin lines: ± sem ) and AP rate aligned at the onset of detected spontaneous licking ( dotted line ) in M1-p and S2-p neurons of ‘Good performer’ ( A ) and ‘Naive’ ( B ) mice . Right , quantifications at ± 0 . 1 s around the detected lick onset . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 01610 . 7554/eLife . 15798 . 017Figure 3—source data 1 . Data values and statistics underlying Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 01710 . 7554/eLife . 15798 . 018Figure 3—source data 2 . Data values and statistics underlying Figure 3—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 01810 . 7554/eLife . 15798 . 019Figure 3—figure supplement 1 . Licking-related Vm dynamics of S2-p and M1-p neurons in ‘Good performer’ mice . ( A ) Left , an example Vm trace ( below ) from an S2-p neuron during spontaneous unrewarded licking and the corresponding lick-sensor signal ( above ) . Right , individual lick-sensor signals ( above ) and Vm traces ( below , thin lines; APs are truncated ) aligned and superimposed at the onset of detected licking ( the time when the tongue first contacted the water spout; dotted line ) , obtained from the cell shown on the left . The average subthreshold Vm trace ( thick line ) is superimposed . ( B ) Another example from an S2-p neuron different from the cell shown in A . ( C ) Same as A , but for an M1-p neuron . ( D ) Another example from an M1-p neuron different from the cell shown in C . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 01910 . 7554/eLife . 15798 . 020Figure 3—figure supplement 2 . Licking-related Vm dynamics of S2-p and M1-p neurons in ‘Naive’ mice . ( A ) Left , an example Vm trace ( below ) from an S2-p neuron during spontaneous unrewarded licking and the corresponding lick-sensor signal ( above ) . Right , individual lick-sensor signals ( above ) and Vm traces ( below , thin lines; APs are truncated ) superimposed and aligned at the onset of detected licking ( the time when the tongue first contacted the water spout; dotted line ) , obtained from the cell shown on the left . The average subthreshold Vm trace ( thick line ) is superimposed . ( B ) Another example from an S2-p neuron different from the cell shown in A . ( C ) Same as A , but for an M1-p neuron . ( D ) Another example from an M1-p neuron different from the cell shown in C . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 02010 . 7554/eLife . 15798 . 021Figure 3—figure supplement 3 . Further analysis of Vm dynamics during spontaneous unrewarded licking . ( A , B ) Left , grand average change in Vm ( thick line: mean , thin lines: ± sem ) and AP rate aligned at the onset of detected spontaneous licking ( dotted line ) in S2-p ( A ) and M1-p ( B ) neurons of ‘Good performer’ and ‘Naive’ mice . Right , quantifications at ± 0 . 1 s around the detected lick onset . ( C ) The lick-related change of Vm and AP rate was analyzed on a mouse-by-mouse basis for ‘Good performer’ mice ( n = 7 mice for M1-p neurons; n = 15 mice for S2-p neurons ) . ( D ) Same as C , but for ‘Naive’ mice ( n = 7 mice for M1-p neurons; n = 9 mice for S2-p neurons ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 021 S2-p neurons , but not M1-p neurons , in ‘Good performer’ mice are therefore excited during spontaneous licking , whereas in ‘Naive’ mice there was little spontaneous licking-related activity in S2-p or M1-p neurons . The licking-related depolarization in S2-p neurons was significantly larger in ‘Good performer’ mice compared to that in ‘Naive’ mice ( Figure 3—figure supplement 3 ) , suggesting emergence of projection-specific excitation related to licking after task learning . Our projection-specific Vm measurements in mice with different levels of task proficiency suggest that cortico-cortical signals originating from S1 are bi-directionally modulated by task learning in a pathway-specific manner ( Figure 4 ) . In ‘Naive’ mice , whisker stimulation evoked the strongest signals in M1-p neurons during hit trials , whereas in ‘Good performer’ mice S2-p neurons showed the strongest excitation during hit trials . The largest differences in activity during task performance between S2-p and M1-p neurons were observed during the lick period , and task learning was accompanied by enhanced excitation during spontaneous licking specifically in S2-p neurons . 10 . 7554/eLife . 15798 . 022Figure 4 . Schematic summary of target-specific output signals from S1 during execution of the whisker detection task , together with speculative hypothesis relating to possible cortico-cortical signalling pathways from S1 to tongue/jaw area of motor cortex . Left , In ‘Good performer’ mice , S2-p neurons , not M1-p neurons , in S1 develop depolarization correlated with task performance . The activities of S2-p neurons could be routed toward the tongue/jaw area of M1/M2 ( tjM1/M2 ) . Right , In ‘Naive’ mice , M1-p neurons , not S2-p neurons , exhibit depolarization correlated with task execution . wM1: whisker M1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15798 . 022 The lack of task-correlated activity in M1-p neurons in ‘Good performer’ mice is consistent with results from a previous study of a closely-related whisker detection task in which inactivation of whisker M1 did not reduce hit rates in trained animals , but rather increased false-alarm rates ( Zagha et al . , 2015 ) . Thus signals from S1 to whisker M1 may not be essential for task execution . Optogenetic inactivation of S1 during the late phase impairs task performance ( Sachidhanandam et al . , 2013 ) , suggesting a causal role for late excitation . S2-p neurons exhibited a learning-induced depolarization at the late and lick phases of hit trials and during spontaneous unrewarded licking . The grand-averaged , late depolarization in S2-p neurons on hit trials peaked at 261 ms after whisker stimulation ( Figure 1D ) , which was earlier than the mean reaction time ( 317 ms ) . The licking-related depolarization in S2-p neurons started shortly ( 260 ± 46 ms , n = 18 ) before tongue-spout contact during spontaneous licking ( Figure 3A ) , which is consistent with the larger depolarization at early and late phases of their responses in hit compared to miss trials ( Figure 2A ) . Interestingly , S2 has been suggested to be reciprocally connected to a tongue/jaw-related M1/M2 area ( also termed anterior lateral motor cortex , ALM ) a neocortical region known to be involved in goal-directed licking ( Oh et al . , 2014; Guo et al . , 2014; Li et al . , 2015 ) . We therefore speculate that the licking-related signals in S2-p neurons in S1 might contribute to exciting neurons in tongue/jaw-related M1/M2 via S2 through reciprocally connected networks of excitatory long-range projection neurons , thus contributing to driving licking motor output ( Figure 4 ) . Consistent with such a hypothesis involving reciprocal excitation between S1 and S2 , axons from S2 innervating S1 were found to exhibit strong task-related hit vs miss modulation in a closely-related whisker detection task ( Yang et al . , 2016 ) . It is important to note that there are many possible sensory/motor signals that might contribute to the lick-related depolarization of S2-p neurons in trained mice: movement of jaw and tongue must begin before tongue-spout contact , and we did not quantify orofacial movements during task performance . Furthermore , rodents are known to have reward-expecting orofacial movements such as sniffing and whisking ( Deschênes et al . , 2012 ) . However , transection of the facial motor nerve that controls whisker movements has no impact on task performance or the late phase Vm ( Sachidhanandam et al . , 2013 ) , suggesting that the late phase Vm might be generated by internal brain circuits rather than sensory reafference coming from associated whisker movements . In this study we primarily compare neurons projecting to different targets in mice with the same level of task proficiency ( i . e . S2-p vs M1-p in ‘Good performer’ mice , or S2-p vs M1-p in ‘Naive’ mice ) , and the differences found comparing these projection neurons can therefore not reflect differences in sensorimotor behavior . In future experiments , it will be important to examine causal roles of S2-p neurons , as well as to investigate the synaptic mechanisms driving the target-specific Vm dynamics in M1-p and S2-p neurons associated with task learning . Implantation of a metal head-restraint post on male C57BL6J mice ( 6-week-old or older ) , identification of the locations of the S1-C2 barrel column and whisker-S2 of the left hemisphere by intrinsic optical signal imaging , and the injection of CTB conjugated with Alexa-Fluor 488 or 594 ( 0 . 5% in PBS , weight/volume , Invitrogen ) into left whisker M1 ( 1 mm anterior , 1 mm lateral from Bregma ) and left S2 were performed as previously described ( Yamashita et al . , 2013 ) . The injection volume of CTB was 50 – 100 nl for M1 and 25 – 50 nl for S2 at the depths of 300 and 800 µm , giving a total volume of 100 – 200 nl for M1 and 50 –100 nl for S2 . Animals were kept with a light/dark cycle ( 12 hr/12 hr ) in cages of four mice or less . Experiments were typically performed during the dark period . At least one day after CTB injection , mice started to be water-restricted . The mice were adapted to head restraint on the recording setup through initial training to freely lick the water spout for receiving water reward ( 3 – 5 sessions , one session per day ) . Mice were then taught to associate whisker deflection with water availability through daily training sessions , essentially as described previously ( Sachidhanandam et al . , 2013; Sippy et al . , 2015 ) . For whisker stimulus , we used a brief ( 1 ms ) magnetic pulse to elicit a vertical deflection of the right C2 whisker transmitted by a small metal particle glued on the whisker . The reward time window was 1 s after the whisker stimulus throughout training . Trials with whisker stimulation ( test trials ) or those without whisker stimulation ( catch trials ) were started without preceding cues at random inter-trial intervals ranging from 2 – 10 s . Catch trials were randomly interleaved with test trials , with 40 – 50% probability of all trials . If the mouse licked in the 2 s ( or 3 s in some experiments ) preceding the time when the trial was supposed to occur , then the trial was aborted . Catch trials were present from the first day of training . After each training session , 1 . 0 – 1 . 5 g of wet food pellet was given to the mouse in order to keep its body weight more than 80% of the initial value . Behavioral control and behavioral data collection were carried out with custom-written computer routines using a National Instruments board interfaced through LabView . Whole-cell patch-clamp recordings ( 95 recordings in total ) were targeted to cell bodies of CTB-labeled neurons in the center of the C2 barrel column ( as identified with intrinsic optical signal imaging ) of adult C57BL6J mice ( 8-week-old or older ) under visual control using a custom-built two-photon microscope , as previously described ( Yamashita et al . , 2013 ) . Recordings were made at the subpial depth of 120 – 270 µm , and the recording depths for M1-p and S2-p neurons were similar . The recording pipettes had resistances of 5 – 7 MΩ and were filled with a solution containing ( in mM ) : 135 potassium gluconate , 4 KCl , 10 HEPES , 10 sodium phosphocreatine , 4 MgATP , 0 . 3 Na3GTP ( adjusted to pH 7 . 3 with KOH ) . For targeting CTB-labeled neurons , Alexa 488 or 594 ( 1 – 20 μM ) was added to the pipette solution , depending on the color of the targeted cells . In most experiments , we targeted either M1-p or S2-p neurons . In one mouse , we injected CTB-Alexa 488 in M1 and CTB-Alexa 594 in S2 and targeted both M1-p and S2-p neurons . The Vm was measured using a MultiClamp 700B amplifier with a 10 kHz low pass Bessel filter , and digitized at 20 kHz by a National Instruments board . Vm was not corrected for liquid junction potential . Short ( 1 min ) sweeps of the Vm and the behavioral signals from the lick sensor together with TTL signals to control the water valve and the electromagnetic coil were recorded using Ephus in Matlab ( Suter et al . , 2010 ) . We used two types of mice for recordings: ( 1 ) ‘Good performer’ mice that exhibited a high discriminability between test trials and catch trials during recordings ( 59 recordings in 27 mice; hit rate , 0 . 77 ± 0 . 03; false alarm rate , 0 . 17 ± 0 . 01; d’ = 2 . 12 ± 0 . 09 ) learned through training sessions ( typically 7–13 daily sessions prior to the recording day , but some mice learned more quickly ) ; and ( 2 ) ‘Naive’ mice that were used for recordings on the first day of being exposed to the task and showed no apparent discrimination ( 36 recordings in 16 mice; hit rate , 0 . 31 ± 0 . 03; false alarm rate , 0 . 28 ± 0 . 03; d’ = 0 . 03 ± 0 . 09; d’ < 0 . 9 , for each recording ) . For calculating d’ when hit rate or false alarm rate was measured as 1 . 0 or 0 . 0 , each value was corrected by subtracting or adding 1/ ( 2N ) , where N is the trial number . Each recording typically lasted ~20 min or less , and we made multiple whole-cell recordings from one animal in most of the experiments . For each recording we routinely monitored the level of task performance by calculating d’ and discarded data with d’ < 1 . 1 in ‘Good performer’ mice or those with d’ > 0 . 9 in ‘Naive’ mice . The d’ values for recordings of M1-p and S2-p neurons was not significantly different ( p=0 . 73; for ‘Good performer’ mice; p=0 . 50 for ‘Naive’ mice ) . Subthreshold postsynaptic potentials ( PSPs ) were analyzed after removing APs by median-filtering ( Crochet and Petersen , 2006 ) . For analysis of Vm changes evoked by task-relevant whisker deflection , baseline Vm was defined as the mean Vm at 0 – 5 ms before the stimulus onset . The amplitude of PSPs was defined as the difference between the baseline Vm and the peak Vm of averaged traces . The △Vm at the late and lick periods was estimated as the difference between the baseline Vm and the mean Vm of the averaged traces at 0 . 05 – 0 . 25 s ( late ) or 0 . 25 – 1 . 0 s ( lick ) after whisker stimulus . APs evoked by whisker stimulation were estimated by subtracting spontaneous AP rate from the AP rate measured in the early ( 0 – 0 . 05 s ) , late ( 0 . 05 – 0 . 25 s ) or lick ( 0 . 25 –1 . 0 s ) periods after the stimulation for each cell . Baseline AP rates were computed as the mean of no-lick periods ( 2 s before test/catch trials ) totaling over 16 s . Peristimulus time histograms ( PSTHs ) were computed by counting AP number in each 50 ms ( or 10 ms ) bin for each cell and averaging the number across cells recorded . Grand average PSTHs are shown in Hz after subtracting baseline AP rates . On average , 31 ± 2 hit trials ( n = 53 cells ) and 18 ± 2 miss trials ( n = 29 cells ) per recording were analyzed for ‘Good performer’ mice , and 14 ± 1 hit trials ( n = 26 cells ) and 31 ± 3 miss trials ( n = 26 cells ) per recording were analyzed for ‘Naive’ mice . In some recordings the well-trained mouse showed few misses and in such cases we only analyzed hit responses . Lick bouts that occurred at least 3 s after whisker stimulation , and at least 1 s after the cessation of previous lick bouts were selected for analysis of Vm modulation induced by spontaneous unrewarded licking . On average , 64 ± 6 lick bouts ( n = 30 cells ) of ‘Good performer’ mice and 31 ± 3 lick bouts ( n = 26 cells ) of ‘Naive’ mice were analyzed for each recording . The individual Vm traces aligned at the onset of detected lick bouts ( lick onset ) were median-filtered to remove APs . Baseline Vm was defined as the mean Vm at 1 . 0 – 0 . 6 s before the lick onset , and the magnitude of Vm modulation was estimated by the difference between the baseline Vm and the mean Vm at ± 0 . 1 s around the lick onset . APs evoked during lick events were calculated by subtracting baseline AP rate ( averaged at 0 . 6 – 1 . 0 s before the detected lick onset ) from the AP rate measured within ± 0 . 1 s from the lick onset . PSTHs around lick events are shown in Hz after subtracting the baseline AP rate . The onset of the licking-related Vm depolarization was computed as the time point where Vm increased over 3 x SD of the baseline Vm for the 18 out of 20 S2-p cells with pre-lick depolarization . All values ( except for box plots ) are presented as mean ± sem . Box plots indicate median and 1st/3rd quartile , with Tukey’s whiskers showing maximal and minimal data points within 1 . 5 times interquartile range away from 1st/3rd quartile . Statistical testing using two-tailed Wilcoxon rank-sum test for unpaired data ( for example , M1-p vs S2-p for Figure 1D , E and Figure 3A , B; ‘Good performer’ vs ‘Naive’ for Figure 1C ) and two-tailed Wilcoxon signed rank test for paired data ( for example , hit vs miss trials for Figure 2; hit vs false-alarm rates for Figure 1C ) was performed in IgorPro ( WaveMetrics ) without excluding any data points . Testing for the normality of data distribution was performed in IgorPro and we found that at least one of the samples in every two-sample comparison was not normally distributed . Non-parametric tests were therefore used for all figures . We analyzed data on a cell-by-cell basis unless otherwise noted . Neither randomization nor blinding was done for data collection or analysis .	Many animals can learn quickly to associate specific behaviors with rewards , such as food . Often , the animal’s senses of smell , taste , and touch trigger the behavior , and this allows an animal to respond favorably to changes in its environments . However , it is not clear exactly what happens in the animal’s brain to reinforce a behavior that ends in a reward , or how its senses help trigger the rewarding behavior . Yamashita and Petersen have now studied what happens in the brains of mice that were taught to complete a task to get a reward . In the training , thirsty mice learned that they would receive a reward of water if they licked a water spout after they were briefly touched on one of their whiskers . Then , Yamashita and Petersen measured electrical changes in the brain cells of the trained mice and compared those with measurements from the brain cells of untrained mice . The measurements specifically focused on the brain cells that receive sensory information from the whiskers . These cells are in a region of the brain called the primary sensory cortex , which is known to help mice carry out the task . This brain area in turn sends signals to many downstream areas of the brain . Yamashita and Petersen found that learning the task appeared to enhance the signaling of some cells in this area of the mouse brain . However , this was only the case for the cells that send signals to a region of the brain that further processes the sensory information ( the so-called secondary sensory cortex ) . Other cells that are intermingled in this region but send signals to the part of the brain that controls movement ( the motor cortex ) were not affected in this way . Together the data suggest that routing signals from the primary sensory cortex to specific downstream areas might allow animals to learn tasks that depend on responding to sensory cues . More studies are now needed to understand exactly how these signals are generated and whether they contribute to triggering the licking behavior in the mice .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "short", "report", "neuroscience" ]	2016	Target-specific membrane potential dynamics of neocortical projection neurons during goal-directed behavior
Recent genome analyses have identified recurrent mutations in the cohesin complex in a wide range of human cancers . Here we demonstrate that the most frequently mutated subunit of the cohesin complex , STAG2 , displays a strong synthetic lethal interaction with its paralog STAG1 . Mechanistically , STAG1 loss abrogates sister chromatid cohesion in STAG2 mutated but not in wild-type cells leading to mitotic catastrophe , defective cell division and apoptosis . STAG1 inactivation inhibits the proliferation of STAG2 mutated but not wild-type bladder cancer and Ewing sarcoma cell lines . Restoration of STAG2 expression in a mutated bladder cancer model alleviates the dependency on STAG1 . Thus , STAG1 and STAG2 support sister chromatid cohesion to redundantly ensure cell survival . STAG1 represents a vulnerability of cancer cells carrying mutations in the major emerging tumor suppressor STAG2 across different cancer contexts . Exploiting synthetic lethal interactions to target recurrent cohesin mutations in cancer , e . g . by inhibiting STAG1 , holds the promise for the development of selective therapeutics . Cohesin is a highly conserved ring-shaped protein complex that is thought to topologically embrace chromatid fibers ( Peters and Nishiyama , 2012 ) , which is essential for sister chromatid cohesion and chromosome segregation in eukaryotes . In addition , cohesin participates in DNA repair , genome organization and gene expression ( Losada , 2014 ) . The cohesin subunits SMC1 , SMC3 and RAD21 ( also called SCC1 ) comprise the core ring of the complex . A fourth universally conserved subunit , a HEAT repeat protein of the Scc3/STAG family , peripherally associates with the core cohesin ring by binding to RAD21 ( Tóth et al . , 1999 ) , and is required for the dynamic association of cohesin with chromatin ( Hu et al . , 2011; Murayama and Uhlmann , 2014 ) . Human somatic cells express two paralogs of this protein , called STAG1 and STAG2 ( Losada et al . , 2000; Sumara et al . , 2000 ) . Recent cancer genome studies identified recurrent mutations in cohesin subunits and regulators in approximately 7 . 3% of all human cancers ( Lawrence et al . , 2014; Leiserson et al . , 2015; Solomon et al . , 2011 ) . STAG2 , the most frequently mutated cohesin subunit , emerges as one of only 12 genes that are significantly mutated in four or more major human malignancies ( Lawrence et al . , 2014 ) . STAG2 mutations have been reported in ~6% of acute myeloid leukemias and myelodysplastic syndromes ( Kon et al . , 2013; Thota et al . , 2014; Walter et al . , 2012 ) , 15–22% of Ewing’s sarcomas ( Brohl et al . , 2014; Crompton et al . , 2014; Tirode et al . , 2014 ) , and in up to 26% of bladder cancers of various stages and grades ( Balbás-Martínez et al . , 2013; Guo et al . , 2013; Solomon et al . , 2013; Taylor et al . , 2014 ) . The deleterious nature of most STAG2 mutations strongly suggests that the gene represents a new tumor suppressor ( Hill et al . , 2016 ) . STAG2 mutations were initially thought to promote tumorigenesis due to defects in sister chromatid cohesin leading to genome instability ( Barber et al . , 2008; Solomon et al . , 2011 ) . However , the vast majority of cohesin-mutated cancers are euploid ( Balbás-Martínez et al . , 2013; Kon et al . , 2013 ) , indicating that cohesin mutations may promote tumorigenesis through altering different cohesin functions such as genome organization and transcriptional regulation ( Galeev et al . , 2016; Mazumdar et al . , 2015; Mullenders et al . , 2015; Viny et al . , 2015 ) . Regardless of the mechanisms driving cohesin mutant tumors , the recent success of poly ( ADP-ribose ) polymerase inhibitors in the treatment of BRCA-mutated ovarian and prostate cancer demonstrates that exploiting tumor suppressor loss by applying the concept of synthetic lethality in defined patient populations can impact clinical cancer care ( Castro et al . , 2016; Kim et al . , 2015; Mirza et al . , 2016; Oza et al . , 2015 ) . The estimated half a million individuals with STAG2-mutant malignancies would greatly profit from exploring specific dependencies of these cancers . We hypothesized that STAG2 loss could alter the properties and function of the cohesin complex leading to unique vulnerabilities of STAG2 mutated cells . To identify factors whose inactivation would be synthetic lethal with loss of STAG2 function , we first used CRISPR/Cas9 to inactivate STAG2 in near-diploid , chromosomally stable HCT 116 colon carcinoma cells ( Figure 1A ) . Two clones , STAG2- 505c1 and 502c4 , harboring deleterious mutations in STAG2 and lacking detectable STAG2 protein expression were selected for analyses ( Figure 1—figure supplement 1 and Supplementary file 1 ) . The isogenic parental and STAG2- HCT 116 cells were transfected with short-interfering RNA ( siRNA ) duplexes targeting 25 known cohesin subunits and regulators . After normalization to the non-target control siRNA ( NTC ) , the effects of siRNA duplexes targeting individual genes were compared in parental and STAG2- cells . Depletion of the known essential cohesin regulator SGOL1 had a detrimental impact on viability of both parental and STAG2- cells . Remarkably , depletion of STAG1 strongly decreased cell viability in STAG2- cells , while being tolerated by the isogenic parental cells ( Figure 1B ) . The pronounced selective effect of STAG1 depletion on STAG2- cells was confirmed in individual transfection experiments and colony formation assays ( Figure 1C , D , E ) . Expression of an siRNA-resistant STAG1 transgene alleviated the anti-proliferative effect of STAG1 but not of SGOL1 siRNA duplexes in STAG2- HCT 116 cells demonstrating the specificity of the siRNA treatment ( Figure 1—figure supplement 2 ) . Double depletion of STAG1 and STAG2 by siRNA in parental cells confirmed their synthetic lethal interaction ( Figure 1—figure supplement 3 ) . Co-depletion of p53 and STAG1 indicated that the dependency of STAG2- cells on STAG1 was independent of p53 ( Figure 1—figure supplement 4 ) . In contrast to the loss of essential cohesin subunits or regulators , depletion of STAG1 had no effect on cell viability in non-transformed telomerase-immortalized human retinal pigment epithelial cells ( hTERT RPE-1 ) ( Figure 1—figure supplement 5 ) . This result is supported by a large-scale genetic loss-of-function study that found that neither STAG1 nor STAG2 is essential for the proliferation of hTERT-RPE1 cells ( Hart et al . , 2015 ) . To corroborate our genetic interaction findings using an independent strategy , we introduced Cas9 into parental and STAG2- HCT 116 cells as well as KBM-7 leukemia cells for competition assays ( Figure 1F and Figure 1—figure supplement 1 ) . Transduction of lentiviruses co-expressing mCherry and single guide RNAs ( sgRNAs ) targeting essential cohesin subunit genes , such as RAD21 and SMC3 , resulted in the rapid loss of the infected and mCherry-positive cells from the population irrespective of STAG2 genotype ( Figure 1F ) . In striking contrast , transduction with sgRNAs targeting STAG1 caused the depletion of STAG2- HCT 116 and KBM-7 cells but not of their parental STAG2 proficient counterparts ( Figure 1F ) . Collectively , these experiments identify STAG1 as a vulnerability of STAG2 mutated cells in engineered solid cancer and leukemia models . STAG1 inactivation has little if any impact on the viability and proliferation of STAG2 wild-type cancer cells and non-transformed cells , but is essential for survival in the absence of STAG2 . 10 . 7554/eLife . 26980 . 003Figure 1 . Identification of STAG1 as a genetic vulnerability of STAG2 mutated cells . ( A ) Engineering of an isogenic HCT 116 cell model by CRISPR/Cas9-mediated inactivation of STAG2 . The position of the sgRNAs used to create deleterious insertion and deletion mutations in the STAG2 coding sequence is indicated . ( B ) Parental HCT 116 cells and two STAG2 mutant clones ( STAG2- ) were subjected to an siRNA screen . Pools of 4 siRNA duplexes targeting 25 known cohesin subunits and regulators were transfected into the three cell lines . Cell viability was measured 7 days after transfection . Following normalization to non-target control ( NTC ) values , the average cell viability of siRNA pools in parental HCT 116 cells and two STAG2- clones was plotted against each other ( n = 2 or more independent experiments with 2 biological repeats each ) . ( C ) HCT 116 parental cells and STAG2- clones were transfected with the indicated siRNAs . Protein lysates were prepared 72 hr after transfection and analyzed by immunoblotting . ( D ) Cell viability was assessed 8 days after siRNA transfection using a metabolic assay ( n = 4 biological repeats , error bars denote standard deviation ) and ( E ) using crystal violet staining . ( F ) Cas9-GFP expressing isogenic parental and STAG2- HCT 116 and KBM-7 cells were transduced with a lentivirus encoding mCherry and sgRNAs targeting the indicated genes . The percentage of mCherry-positive cells was determined over time by flow cytometry and normalized to the fraction of mCherry-positive cells at the first measurement and sequentially to a control sgRNA ( n = 1 experimental replicate ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26980 . 00310 . 7554/eLife . 26980 . 004Figure 1—figure supplement 1 . Characterization of CRISPR/Cas9-generated STAG2 mutated clones used in this study . ( A ) Four different sgRNAs co-expressed with Cas9 from an all-in-one plasmid or a lentiviral vector were used to generate deleterious frameshift insertions or deletion mutations ( indels ) in STAG2 coding exons . sgRNAs sg502 and sg16 target exon 8 of STAG2 , while sgRNAs sg505 and sg19 target exon 9 of the gene . Cognate sgRNA target sequences are underlined . Associated protospacer adjacent motifs ( PAMs ) are marked by boxes . The predicted Cas9 cleavage sites are marked by triangles . ( B ) CRISPR/Cas9-generated STAG2 indels in HCT 116 and KBM-7 cell clones as detected by Sanger sequencing . STAG2 is located on the X chromosome . Therefore , only one allele needed to be inactivated in HCT 116 ( male ) and KBM-7 ( near-haploid ) cells . ( C ) Protein lysates prepared from the indicated parental and mutant clones were analyzed for STAG2 protein expression by immunoblotting using antibodies directed against different STAG2 epitopes . DOI: http://dx . doi . org/10 . 7554/eLife . 26980 . 00410 . 7554/eLife . 26980 . 005Figure 1—figure supplement 2 . Rescue of the synthetic lethal interaction between STAG1 and STAG2 by expression of an siRNA-resistant FLAG-STAG1 transgene . HCT 116 parental cells , a STAG2 wild-type clone ( 502wt ) , and two STAG2- clones ( 505c1 and 502c4 ) were transduced with a lentivirus encoding no transgene ( empty vector ) or an siRNA-resistant and 3xFLAG-tagged STAG1 transgene ( FLAG-STAG1 ) . Stably selected cell pools were used for the analysis . ( A ) Immunofluorescence analysis of FLAG-STAG1 transgene expression and nuclear localization in HCT 116 STAG2- 505c1 cells . Scale bar , 20 μm . ( B ) Protein extracts prepared from HCT 116 STAG2- 505c1 cells expressing no transgene ( empty vector ) or a FLAG-STAG1 transgene were subjected to anti-FLAG immunoprecipitation . The input , unbound and precipitated fractions were analyzed by immunoblotting . Co-precipitation of cohesin subunits was only detected in FLAG-STAG1 expressing cells indicating specific incorporation of the transgenic protein into the cohesin complex . ( C ) Protein extracts prepared from the indicated cell lines that were transduced with an empty vector or a FLAG-STAG1 transgene were analyzed by immunoblotting ( left panel ) and transfected with NTC , STAG1 and SGOL1 siRNA duplexes ( right panel ) . Cell viability was measured 7 days after transfection and is plotted normalized to the viability of NTC siRNA transfected cells ( n ≥ 5 biological repeats , error bars denote standard deviation ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26980 . 00510 . 7554/eLife . 26980 . 006Figure 1—figure supplement 3 . Double depletion experiment confirms STAG1-STAG2 synthetic lethality . ( A ) HCT 116 parental cells were co-transfected with NTC siRNA and siRNA duplexes targeting one of the following genes: NTC , PLK1 , RAD21 , CDCA5 , SGOL1 , STAG1 or STAG2 . HCT 116 parental cells were also co-transfected with siRNA duplexes targeting STAG1 and STAG2 . Cell viability was determined 8 days after transfection and is plotted normalized to cell viability of NTC siRNA-transfected cells ( n = 3 biological repeats , error bars denote standard deviation ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26980 . 00610 . 7554/eLife . 26980 . 007Figure 1—figure supplement 4 . Double depletion experiments indicate that the STAG1-STAG2 genetic interaction is independent of p53 . Parental HCT 116 cells and STAG2- 505c1 cells were transfected with NTC ( - ) or TP53 ( + ) siRNA duplexes . Protein extracts were prepared 4 days after transfection and analyzed by immunoblotting ( left panel ) . Parental HCT 116 cells and STAG2- 505c1 cells were co-transfected with the indicated siRNA duplexes ( right panel ) . Cell viability was determined 8 days after transfection and is plotted normalized to the cell viability of NTC+NTC or NTC+TP53 siRNA-co-transfected cells ( n = 3 biological repeats , error bars denote standard deviation ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26980 . 00710 . 7554/eLife . 26980 . 008Figure 1—figure supplement 5 . Depletion of STAG1 does not reduce viability in hTERT RPE-1 cells . Human telomerase-immortalized retinal pigment epithelial cells ( hTERT RPE-1 ) were transfected with the indicated siRNA duplexes . Protein extracts were prepared 4 days after transfection and analyzed by immunoblotting ( left panel ) . Cell viability was determined 5 days after transfection and is plotted normalized to the cell viability of NTC siRNA-transfected cells ( n = 2 independent experiments with 3 biological replicates each , error bars denote standard deviation ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26980 . 008 To elucidate the mechanistic basis for this synthetic lethal interaction , we hypothesized that the combined loss of STAG1 and STAG2 , in contrast to loss of either component alone , could severely impair cell division . Chromosome alignment and segregation during mitosis rely on sister chromatid cohesion , the central function of the cohesin complex ( Peters and Nishiyama , 2012 ) . Depletion of STAG1 resulted in an increase in the mitotic index and a prolongation of the duration of mitosis in STAG2- but not wild-type cells ( Figure 2A and Figure 2—figure supplement 1 ) . Immunofluorescence microscopy revealed a failure to align chromosomes at the metaphase plate upon STAG1 loss in STAG2- cells ( Figure 2B ) . In mitotic chromosome spread analysis STAG2 inactivation caused a partial loss of centromeric cohesion in HCT 116 cells as previously reported ( Canudas and Smith , 2009; Kim et al . , 2016; Remeseiro et al . , 2012; Solomon et al . , 2011 ) ( Figure 2C ) . Depletion of the essential centromeric cohesin protection factor SGOL1 resulted in a complete loss of sister chromatid cohesion in most chromosome spreads irrespective of STAG2 genotype . In striking contrast , STAG1 depletion selectively abrogated sister chromatid cohesion in STAG2- but not parental cells ( Figure 2C , single chromatids ) . The severe mitotic defects observed upon loss of STAG1 in STAG2- cells were accompanied by the emergence of aberrantly sized and shaped interphase nuclei ( Figure 2—figure supplement 2 ) and by a progressive increase in apoptosis ( Figure 2D ) . These results provide a mechanistic basis for the synthetic lethal interaction between STAG1 and STAG2 . STAG1 inactivation abrogates sister chromatid cohesion exclusively in STAG2- cells resulting in catastrophic mitotic failure , abnormal cell division and apoptosis . To hold sister chromatids together , cohesin can tolerate the loss of either STAG1 or STAG2 alone but not the loss of both . 10 . 7554/eLife . 26980 . 009Figure 2 . Loss of STAG1 function causes severe mitotic defects , abrogates sister chromatid cohesion and triggers apoptosis in STAG2- but not parental HCT 116 cells . ( A ) Parental and STAG2- 505c1 HCT 116 were transfected with NTC and STAG1 siRNA duplexes . Immunofluorescence analysis was performed 72 hr after transfection to determine the mitotic index by scoring the fraction of histone H3 phosphoSer10-positive ( H3pS10+ ) cells ( n ≥ 1323 cells , error bars denote standard deviation of three independent experiments ) , and ( B ) to investigate mitotic spindle geometry and chromosome alignment . Cropped and magnified examples of chromosome alignment are shown in ( B ) . Scale bars , 20 μm . ( C ) Giemsa-stained mitotic chromosome spreads were prepared 48 hr after transfection of parental and STAG2- HCT 116 cells with the indicated siRNA duplexes . The status of sister chromatid cohesion of individual metaphase spreads was categorized into normal , partial loss of cohesion or single chromatid phenotypes ( n = 100 spreads , error bars denote standard deviation of two independently analyzed slides ) . Scale bar , 10 μm . ( D ) Caspase activity was tracked over time using a live-cell caspase 3/7 substrate cleavage assay in parental and STAG2- 505c1 HCT 116 cells after transfection with the indicated siRNAs or after treatment with 0 . 3 μM doxorubicin at t = 0 hr ( n = 2 independent experiments with 4 biological repeats for NTC and STAG1 siRNA each and 1 biological repeat for doxorubicin each ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26980 . 00910 . 7554/eLife . 26980 . 010Figure 2—figure supplement 1 . The depletion of STAG1 prolongs mitosis in STAG2 mutated but not parental HCT 116 cells . Parental and STAG2- 505c1 HCT 116 were transfected with NTC and STAG1 siRNA duplexes . Cells were tracked from 0 . 5 to 72 hr ( imaging interval 30 min ) after transfection by bright-field live-cell imaging . The duration of mitosis ( time elapsed from mitotic cell rounding to anaphase onset ) and the fate of individual cells were determined ( n ≥ 43 cells ) . The duration of mitosis of individual cells is plotted . Blue lines denote mean , red dots denote cells that died in mitosis . Significance levels were quantified using unpaired t test . DOI: http://dx . doi . org/10 . 7554/eLife . 26980 . 01010 . 7554/eLife . 26980 . 011Figure 2—figure supplement 2 . Aberrant nuclear morphology in STAG2-mutated but not parental HCT 116 cells upon depletion of STAG1 . Parental and STAG2- 505c1 HCT 116 were transfected with NTC and STAG1 siRNA duplexes , fixed and stained with Hoechst 72 hr after transfection . Nuclei were scored for aberrant size ( >20 μm ) or morphology ( multi-nucleated cells and cells with polylobed nuclei; n ≥ 612 nuclei , error bars denote standard deviation between two independent experiments ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26980 . 011 We next expanded our analysis to patient-derived STAG2 mutations and STAG2-mutant cancer cell lines in order to investigate the disease relevance of the observed synthetic lethality ( Figure 3 ) . STAG1 depletion by siRNA abrogated both cell viability and sister chromatid cohesion in HCT 116 cell clones , in which three patient-derived deleterious mutations had been engineered into the STAG2 locus ( Kim et al . , 2016 ) , but not in parental HCT 116 cells ( Figure 3—figure supplement 1 ) . Among solid human cancers , STAG2 mutational inactivation is most prevalent in urothelial bladder cancer and Ewing sarcoma . Therefore , we assembled a panel of 16 bladder cancer cell lines: 11 STAG2-positive , three with deleterious STAG2 mutations ( UM-UC-3 , UM-UC-6 and VM-CUB-3 ) , one in which STAG2 was inactivated by CRISPR/Cas9 ( UM-UC-5 STAG2- 505c6 ) ( Figure 3—figure supplement 2 ) , and two with no detectable STAG2 expression ( LGWO1 and MGH-U3 ) ( Supplementary file 2 ) ( Balbás-Martínez et al . , 2013; Solomon et al . , 2013 ) . The STAG2 protein expression status in the panel of bladder cancer cell lines was confirmed using immunoblotting ( Figure 3A ) . siRNA experiments revealed that STAG2 status represented a predictive marker for the sensitivity to STAG1 depletion across the bladder cancer cell line panel . Whereas all cell lines were highly sensitive to depletion of the key mitotic kinase PLK1 , STAG1 siRNA reduced cell viability in STAG2-negative bladder cancer cells but had little or no effect on STAG2-positive bladder cancer cell lines ( Figure 3B ) . STAG1 depletion prevented colony formation and abolished sister chromatid cohesion selectively in STAG2 mutated UM-UC-3 ( F983fs ) but not in STAG2 wild-type UM-UC-5 bladder cancer cells ( Figure 3C , D ) . In contrast , SGOL1 depletion abrogated cell growth and cohesion in both cell lines . Consistent with the results obtained in bladder cancer cells , STAG2 mutation status was also linked to STAG1 dependency in a panel of four Ewing sarcoma cell lines ( Figure 3E , F and Supplementary file 2 ) ( Solomon et al . , 2011; Tirode et al . , 2014 ) . Lentiviral transduction of a FLAG-STAG2 transgene into STAG2 mutated UM-UC-3 bladder cancer cells resulted in the restoration of STAG2 expression , nuclear localization of the transgenic protein and its incorporation into the cohesin complex ( Figure 3—figure supplement 3 ) . Crucially , restoration of STAG2 expression alleviated the STAG1 dependency of UM-UC-3 cells providing a causal link between STAG2 loss and STAG1 dependency ( Figure 3G ) . These results demonstrate that the synthetic lethal interaction between STAG1 and STAG2 that we discovered in isogenic cell pairs is recapitulated in disease-relevant bladder cancer and Ewing sarcoma cell models . 10 . 7554/eLife . 26980 . 012Figure 3 . The synthetic lethal interaction between STAG1 and STAG2 is manifested in disease-relevant bladder cancer and Ewing sarcoma cell lines . ( A ) The indicated bladder cancer cell lines were analyzed for STAG2 expression by immunoblotting . ( B ) The indicated bladder cancer cell lines were transfected with NTC , STAG1 and PLK1 siRNA duplexes . Viability was determined 7 or 10 days after transfection and normalized to the viability of NTC siRNA transfected cells ( n = 2 independent experiments with 5 biological repeats each , error bars denote standard deviation ) . ( C ) STAG2 wild-type UM-UC-5 and STAG2 mutated UM-UC-3 cells were transfected with NTC , STAG1 and SGOL1 siRNA duplexes . Colony formation was analyzed 7 days after transfection by crystal violet staining . ( D ) 72 hr after siRNA transfection into UM-UC-5 and UM-UC-3 cells , Giemsa-stained chromosome spreads were prepared and analyzed for sister chromatid cohesion phenotypes ( n = 100 spreads , error bars denote standard deviation of two independently analyzed slides ) . ( E ) The indicated Ewing sarcoma cell lines were analyzed for STAG2 protein expression by immunoblotting . ( F ) The indicated Ewing sarcoma cell lines were transfected with NTC , STAG1 and SGOL1 siRNA duplexes . Viability was measured 6 days after transfection and normalized to the viability of NTC siRNA transfected cells ( n = 3 independent experiments with 3 biological replicates each , error bars denote standard deviation ) . ( G ) STAG2 mutated UM-UC-3 cells were transduced with a lentivirus encoding a FLAG-STAG2 transgene . Stably selected cell pools were subsequently transfected with NTC , STAG1 or SGOL1 siRNA duplexes . Viability was measured 7 days after transfection and normalized to the viability of NTC siRNA transfected cells ( n = 4 biological replicates , error bars denote standard deviation ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26980 . 01210 . 7554/eLife . 26980 . 013Figure 3—figure supplement 1 . Patient-derived STAG2 mutations cause STAG1 dependency in engineered isogenic HCT 116 cells . HCT 116 cell lines engineered to harbor the indicated deleterious patient-derived STAG2 mutations were transfected with NTC , STAG1 and SGOL1 siRNA duplexes . ( A ) Protein extracts were prepared 48 hr after transfection and analyzed by immunoblotting . ( B ) Cell viability was measured 7 days after siRNA transfection and plotted normalized to the viability of NTC-transfected cells ( n = 4 independent experiments with 5 biological repeats each , error bars denote standard deviation ) . ( C ) Sister chromatid cohesion phenotypes were analyzed in Giemsa-stained mitotic chromosome spreads that were prepared 48 hr after siRNA transfection ( n = 100 chromosome spreads , error bars denote standard deviation between two independently analyzed slides ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26980 . 01310 . 7554/eLife . 26980 . 014Figure 3—figure supplement 2 . Characterization of CRISPR/Cas9-generated STAG2 knockout in UM-UC-5 bladder cancer cell line . ( A ) sgRNA 505 co-expressed with Cas9 from an all-in-one plasmid was used to generate deleterious frameshift insertions or deletion mutations ( indels ) in STAG2 coding exons . The cognate sgRNA target sequence is underlined . Associated protospacer adjacent motifs ( PAMs ) are marked by boxes . The predicted Cas9 cleavage site is marked by a triangle . ( B ) CRISPR/Cas9-generated STAG2 indels in the UM-UC-5 cell clone STAG2- 505c6 as detected by Sanger sequencing . STAG2 is located on the X chromosome . Therefore , two alleles had to be inactivated in UM-UC-5 ( female ) cells . ( C ) Protein lysates prepared from the indicated parental and mutant UM-UC-5 clones were analyzed for STAG2 protein expression by immunoblotting . DOI: http://dx . doi . org/10 . 7554/eLife . 26980 . 01410 . 7554/eLife . 26980 . 015Figure 3—figure supplement 3 . Restoration of STAG2 protein expression in UM-UC-3 bladder cancer cells . STAG2-deficient UM-UC-3 cells were transduced with lentiviral particles encoding no transgene ( empty vector ) or a 3xFLAG-tagged STAG2 transgene ( FLAG-STAG2 ) . Stable selected cell pools were used for further analysis . ( A ) Expression and nuclear localization of the transgenic FLAG-STAG2 protein was assayed by immunofluorescence microscopy . ( B ) Non-transduced UM-UC-3 cells , UM-UC-3 cells transduced with an empty vector and UM-UC-3 cells transduced with a FLAG-STAG2 transgene were transfected with NTC or STAG1 siRNA duplexes . Protein lysates prepared 72 hr after siRNA transfection were analyzed by immunoblotting . ( C ) Protein extracts prepared from UM-UC-3 cells expressing no transgene ( empty vector ) or a FLAG-STAG2 transgene were subjected to anti-FLAG immunoprecipitation . The input , unbound and precipitated fractions were analyzed by immunoblotting . Co-precipitation of cohesin subunits was only detected in FLAG-STAG2 expressing cells indicating successful incorporation of the transgenic STAG2 protein into the cohesin complex . DOI: http://dx . doi . org/10 . 7554/eLife . 26980 . 015 Here we identify STAG1 as a strong genetic vulnerability of cells lacking the major emerging tumor suppressor STAG2 ( Figure 4 ) . The paralog dependency between STAG1 and STAG2 has recently also been reported in an independent study ( Benedetti et al . , 2017 ) . We show that the synthetic lethal interaction between STAG1 and STAG2 is observed in isogenic HCT 116 and KBM-7 cells as well as in bladder cancer and Ewing sarcoma cell lines . Thus , the genetic interaction between STAG paralogs is conserved in three major human malignancies: carcinoma , leukemia and sarcoma . Importantly , the finding that cancer cells harboring deleterious STAG2 mutations remain exquisitely dependent on STAG1 demonstrates that this genetic vulnerability is maintained throughout the process of carcinogenesis and not bypassed by adaptive processes , such as the transcriptional activation of the germline-specific paralog STAG3 ( Pezzi et al . , 2000; Prieto et al . , 2001 ) . Furthermore , expression analysis did not reveal upregulation of STAG1 protein or mRNA as a major compensatory mechanism for the loss of STAG2 function in cancer cell lines or patient tumors ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 26980 . 016Figure 4 . Model for the synthetic lethal interaction between STAG1 and STAG2 . In wild-type cells , both cohesin-STAG1 and cohesin-STAG2 complexes redundantly contribute to sister chromatid cohesion and successful cell division . Loss of STAG1 is tolerated in these cells as cohesin-STAG2 complexes alone suffice to support sister chromatid cohesion for cell division . In cancer cells in which STAG2 is mutationally or transcriptionally inactivated , sister chromatid cohesion is now entirely dependent on cohesin-STAG1 complexes . Inactivation of STAG1 in STAG2 mutated cells therefore results in a loss of sister chromatid cohesion followed by mitotic failure and cell death . DOI: http://dx . doi . org/10 . 7554/eLife . 26980 . 01610 . 7554/eLife . 26980 . 017Figure 4—figure supplement 1 . Analysis of STAG1 protein or mRNA expression in cancer cell lines and patient tumors . ( A ) , Immunoblot analysis of protein extracts prepared from the indicated cancer cell lines . ( B ) , STAG1 mRNA expression levels ( transcripts per million , TPM ) in tumor samples obtained from The Cancer Genome Atlas ( TCGA ) pan-cancer datasets ( left graph ) and bladder cancer datasets ( right graph ) . Genetic alteration data were obtained from cBioPortal . The results shown here are in whole or part based upon data generated by the TCGA Research Network: http://cancergenome . nih . gov/ . Samples containing alterations resulting in STAG2 amino acid changes ( missense mutations , nonsense point mutations or small insertions/deletions ) were catagorized as STAG2 mutated ( STAG2 mut ) . The boxplot bodies indicate the first quartile ( 25% ) , median , and third quartile ( 75% ) . The whiskers extend no further than 1 . 5 times the distance between the first and the third quartiles to the lowest and largest extremes from the boxplot bodies . All samples exceeding the whisker limits are shown individually as points . Significance tests were performed with the wilcox . test function of the ggsignif v0 . 2 . 0 R v3 . 3 . 2 package . Significance levels were labeled as follows: n . s . >0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 26980 . 01710 . 7554/eLife . 26980 . 018Figure 4—figure supplement 2 . Expression levels of STAG1 and STAG2 mRNA in normal human tissues . Expression data were obtained from The Genotype-Tissue Expression ( GTEx ) project ( www . gtexportal . org ) . STAG1 and STAG2 mRNA expression levels ( transcripts per million , TPM ) in the indicated normal human tissues are plotted . The boxplot bodies indicate the first quartile ( 25% ) , median , and third quartile ( 75% ) . The whiskers extend no further than 1 . 5 times the distance between the first and the third quartiles ( inter-quartile range , IQR ) to the lower and larger extremes from the boxplot bodies . All samples exceeding the whisker limits are shown individually as points . The dotted line indicates a TPM value of 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 26980 . 018 Our experiments strongly suggest that the loss of sister chromatid cohesion followed by aberrant cell division and cell death is the mechanistic basis underlying the synthetic lethality between STAG1 and STAG2 ( Figure 4 ) . Both paralogs associate with the cohesin complex in a mutually exclusive manner ( Losada et al . , 2000; Sumara et al . , 2000 ) . Although STAG1 and STAG2 may confer distinct functionalities to the cohesin complex ( Canudas and Smith , 2009; Remeseiro et al . , 2012 ) , STAG1 and STAG2 containing complexes act redundantly to ensure sufficient sister chromatid cohesion to support cell division in human somatic cells . While the loss of one paralog is compatible with cell viability and proliferation , the loss of both paralogs abrogates cohesin’s ability to hold sister chromatids together , which results in lethality . These mechanistic findings are consistent with previous studies that employed double siRNA depletion experiments to indicate functional redundancy between STAG1 and STAG2 in maintaining sister chromatid cohesion ( Canudas and Smith , 2009; Hara et al . , 2014; Remeseiro et al . , 2012 ) . Since STAG1 inactivation has little or no effect on the proliferation of STAG2 proficient cells , selective targeting of STAG1 could offer a large therapeutic window . The fact that STAG2 mRNA is expressed in all normal human tissues analyzed ( Figure 4—figure supplement 2 ) supports the hypothesis that selective inhibition of STAG1 function would indeed spare most non-cancerous tissues . Potential approaches for therapeutic targeting of STAG1 include the inhibition of the interaction between STAG1 and the cohesin ring subunit RAD21 ( Hara et al . , 2014 ) and the selective degradation of STAG1 using proteolysis-targeting chimera ( PROTAC ) technology ( Deshaies , 2015 ) . The high degree of homology between the STAG1 and its paralog STAG2 will be a key challenge to overcome . The mechanisms by which mutations in STAG2 and other cohesin subunits drive tumorigenesis in solid and hematological tissues are not yet firmly established . Our work highlights the fact that such knowledge is not a prerequisite for the identification of selective vulnerabilities . Both deleterious STAG2 mutations and the loss of STAG2 expression are strong predictive biomarkers for STAG1 dependence in cell models and could be utilized for patient stratification in the future . Our work demonstrates that unique genetic dependencies of cohesin mutated cancer cells exist . Such vulnerabilities hold the promise for the development of selective treatments for patients suffering from STAG2 mutated cancer . The following antibodies were used: C-term . pAb goat anti-STAG2 ( Bethyl , Montgomery , TX , US; A300-158A and A300-159A ) , N-term . mAb rabbit anti-STAG2 ( Cell Signaling , Danvers , MA , US; 5882 ) , full length pAb rabbit anti-STAG2 ( Cell Signaling , Danvers , MA , US; 4239 ) , mouse anti-GAPDH ( Abcam , UK; ab8245 ) , rabbit anti-STAG1 ( GeneTex , Irvine , CA , US; GTX129912 ) , mouse anti-FLAG ( Sigma-Aldrich , St . Louis , MO , US; F3165 and F1804 ) , mouse anti-p53 ( Calbiochem , San Diego , CA , US; OP43 ) , rabbit anti-H3pS10 ( Merck Millipore , Billerica , MA , US; 06570 ) , FITC Conjugated mouse anti-Tubulin ( Sigma-Aldrich , St . Louis , MO , US; F2168 ) , rabbit anti-SGOL1 ( Peters laboratory ID A975M ) , SMC3 ( Peters laboratory ID 845 ) , mouse anti-RAD21 ( Merck Millipore , Billerica , MA , US; 05–908 ) , rabbit anti-SMC1 ( Bethyl , Montgomery , TX , US; A300-055A ) , and secondary rabbit ( P0448 ) , mouse ( P0161 ) and goat ( P0160 ) anti-IgG-HRP ( all Dako , Denmark ) . sgRNA sequences used in this study are listed in Supplementary file 1 . The following lentiviral vectors were used to introduce mutations in STAG2 in HCT 116 and UM-UC-5 cells: CRISPR STAG2 Hs0000077505_U6gRNA-Cas9-2A-GFP and CRISPR STAG2 Hs0000077502_U6gRNA-Cas9-2A-GFP ( Sigma-Aldrich , St . Louis , MO , US ) . CRISPR sgRNA STAG2_16 and sgRNA STAG2_19 were co-expressed from U6gRNA 16-U6gRNA 19-EF1αs-Thy1 . 1-P2A-neo to introduce mutations in STAG2 in KBM-7 cells . HCT 116 stably expressing Cas9-GFP were obtained by lentiviral transduction with pLentiCRISPR-EF1αs-Cas9-P2A-GFP-PGK-puro . KBM-7 infected with dox-inducible Cas9 were obtained by sequential retroviral transduction with pWPXLd-EF1A-rtTA3-IRES-EcoRec-PGK-Puro and pSIN-TRE3G-Cas9-P2A-GFP PGK-Blast . pLVX-3xFLAG-STAG1r-IRES-Puro and pLVX-3xFLAG-STAG2r-IRES-Puro lentiviral vectors for siRNA-resistant transgene expression were generated by gene synthesis ( GenScript , China ) based on the STAG1 cDNA sequence NCBI NM_005862 . 2 and STAG2 cDNA sequence NCBI NM_001042749 . 2 followed by cloning into the parental pLVX-IRES-Puro vector ( Clontech , Mountain View , CA , US ) . Silent nucleotide changes were introduced into the STAG1 and STAG2 coding sequences within the siRNA target sequences to render the transgenes siRNA-resistant . For competition assays , U6-sgRNA-EF1αs-mCherry-P2A-neo lentiviral vectors were used . HCT 116 cells were cultured in McCoy’s 5A w/glutamax medium supplemented with 10% fetal calf serum ( FCS ) , KBM-7 cells were cultured in IMDM medium supplemented with 10% FCS , sodium butyrate , L-glutamine and antibiotics ( all Invitrogen , Waltham , MA , US ) . hTERT RPE-1 cells were cultured in DMEM:F12 ( ATCC: 30–2006 ) +10% FCS+0 , 01 mg/ml hygromycin B . Bladder cancer cell lines UM-UC-5 , UM-UC-6 , UM-UC-18 , LGWO1 , and MGHU-3 were cultured in DMEM +10% FCS w/NEAA , Glutamax and NaPyruvat , 5637 , 639 V , 647 V , J82 , JMSU-1 , KU-19–19 , RT4 T24 , UM-UC-3 and VM-CUB-3 were cultured according to ATCC instructions ( ATCC , Manassas , VA , USA ) . All Ewing sarcoma cell lines were cultured in RPMI +10% FCS . Lentiviral particles were produced using the Lenti-X Single Shot system ( Clontech ) . Following lentiviral infection , stably transduced pools were generated using puromycin selection ( HCT 116: 2 µg/ml , UM-UC-3: 2 µg/ml , UM-UC-5: 3 µg/ml ) . Sources , STAG2 status and authentication information ( STR fingerprinting at Eurofins Genomics , Germany ) of cell lines used in this study are provided in Supplementary file 2 . All cell lines were tested negatively for mycoplasma contamination and have been authenticated by STR fingerprinting . Bladder cancer cell lines J82 , RT4 and VM-CUB-3 and Ewing sarcoma cell line SK-N-MC are contained within the ICLAC list of commonly misidentified cell lines but have been STR verified for this study . For knockdown experiments , cells were transfected with ON-TARGETplus SMARTpool siRNA duplexes ( Dharmacon , Lafayette , CO , US ) and the Lipofectamine RNAiMAX reagent according to the manufacturer’s instructions ( Invitrogen , Waltham , MA , US ) . HCT 116 chromosome spreads , apoptosis assay , immunoblotting , immunofluorescence and live cell imaging experiments were performed using a final siRNA concentration of 20 nM . Cell viability and crystal violet staining assays were performed using 10 nM siRNA . hTERT RPE-1 cells and bladder cancer cells were transfected with 10 nM siRNA for cell viability assays , crystal violet staining and chromosome spreads . The UM-UC-3 FLAG-STAG2 cell line was transfected with 20 nM siRNA for immunoblotting . Ewing sarcoma cell lines were co-transfected with 50 nM ON-TARGETplus SMARTpool siRNAs ( Dharmacon , Lafayette , CO , US ) plus pRetro-Super ( Brummelkamp et al . , 2002 ) using Lipofectamin Plus reagent ( Invitrogen , Waltham , MA , US ) . The next day cells were subjected to puromycin ( 1 µg/ml ) selection for 72 hr ( Ban et al . , 2014 ) , and cultured for two additional days . Viability was determined using CellTiter-Glo ( Promega , Madison , WI , US ) , and by staining with crystal violet ( Sigma-Aldrich , St . Louis , MO , US; HT901 ) . For sgRNA competition assays , Cas9-GFP was expressed constitutively ( HCT 116 ) or was induced by doxycycline addition ( KBM-7 ) . mCherry and sgRNAs were introduced by lentiviral transduction . The fraction of mCherry-positive cells was determined at the indicated time points using a Guava easycyte flow cytometer ( Merck Millipore , Germany ) and normalized to the first measurement and sequentially to control sgRNAs ( non-targeting for HCT 116 and STAG2_19 for KBM-7 ) . Apoptosis was analyzed using the IncuCyte Caspase-3/7 Apoptosis Assay ( Essen BioScience , Ann Arbour , MI , US ) . Cell pellets were resuspended in extraction buffer ( 50 mM Tris Cl pH 8 . 0 , 150 mM NaCl , 1% Nonidet P-40 supplemented with Complete protease inhibitor mix ( Roche , Switzerland ) and Phosphatase Inhibitor cocktails ( Sigma-Aldrich , St . Louis , MO , US; P5726 and P0044 ) ) ; for Figure 1—figure supplement 1 ( KBM-7 ) and Figure 3—figure supplement 1 , pellets were resuspended in ( 25 mM Tris-HCl pH 7 . 5 , 100 mM NaCl , 5 mM MgCl2 , 0 . 2% NP-40 , 10% glycerol , 1 mM NaF , Complete protease inhibitor mix ( Roche , Switzerland ) , Benzonase ( VWR , Radnor , PA , US ) ) and lysed on ice . For Figure 1—figure supplement 2B and Figure 3—figure supplement 3C , lysates were spun down for 10 min , followed by FLAG-immununoprecipitation using anti-FLAG M2-Agarose Affinity Gel ( Sigma-Aldrich , St . Louis , MO , US ) for two hours and washing with lysis buffer . Input lysates and immunoprecipitates were resuspended in SDS sample buffer and heated to 95°C . For immunofluorescence , cells were fixed with 4% paraformaldehyde for 15 min , permeabilized with 0 . 2% Triton X-100 in PBS for 10 min and blocked with 3% BSA in PBS containing 0 . 01% Triton X-100 . Cells were incubated with primary and secondary antibody ( Alexa 594 , Molecular Probes , Eugene , OR , US ) , DNA was counterstained with Hoechst 33342 ( Molecular Probes , Eugene , OR , US; H3570 ) and tubulin was sequentially stained with an FITC-conjugated mouse anti-tubulin antibody ( Sigma-Aldrich , St . Louis , MO , US; F2168 ) ) . Coverslips and chambers were mounted with ProLong Gold ( Molecular Probes , Eugene , OR , US ) . Images were taken with an Axio Plan2/AxioCam microscope and processed with MrC5/Axiovision software ( Zeiss , Germany ) . An IncuCyte ( EssenBioScience , Ann Arbor , MI , US ) imaging system was used to record live cells , and duration of mitosis was determined by measuring the time from mitotic cell rounding until anaphase onset . Data analysis was performed with Microsoft Excel 2013 and GraphPad Prism 7 ( GraphPad Sofware , La Jolla , CA , US ) . Significance levels were quantified using unpaired t test . For chromosome spread analysis , nocodazole was added to the medium for 60 min at 100 ng/ml . Cells were harvested and hypotonically swollen in 40% medium/60%tap water for 5 min at room temperature . Cells were fixed with freshly made Carnoy's solution ( 75% methanol , 25% acetic acid ) , and the fixative was changed three times . For spreading , cells in Carnoy's solution were dropped onto glass slides and dried . Slides were stained with 5% Giemsa ( Merck , Germany ) for 4 min , washed briefly in tap water and air dried . For chromosome spread analysis two independent slides were scored blindly for each condition . STAG2 mRNA expression in tumors ( TCGA Research Network: http://cancergenome . nih . gov/ ) and STAG1 and STAG2 expression in normal human tissues ( The Genotype-Tissue Expression ( GTEx ) project , www . gtexportal . org ) ( GTEx Consortium , 2013 ) were analyzed based on the data available from UCSC Xena version 2016-04-12 . Expression levels correspond to transcripts per million ( TPM ) . The downloaded values were log2 ( TPM +0 . 001 ) values which were transformed to TPM values before generating the figures . STAG2 mutation data were downloaded from cBioPortal ( Cerami et al . , 2012; Gao et al . , 2013 ) git repository in Sep . 2016 . Samples were labeled as STAG2 mutated if they carried an alteration resulting in an amino acid change in STAG2 ( i . e . missense or nonsense point mutation or small insertion or deletion ) . Gene deletions or amplifications were not considered .	A big challenge for cancer research is to find drugs and treatments that kill cancer cells without harming the other cells of a patient . Cancer cells contain genetic mutations that cause them to grow and divide more rapidly than healthy cells . About half a million cancer patients worldwide have tumors that feature mutations to the gene that produces a protein called STAG2 , a component of a large protein ring called cohesin . These mutations are particularly common in bladder cancers and Ewing sarcoma , a childhood bone cancer . The cohesin ring holds together duplicated chromosomes during cell division , establishing the iconic X-shape of chromosomes in dividing cells . It is not clear exactly how mutations that affect STAG2 make cancer more likely to develop . However , it is possible that these cancer-specific mutations make cancer cells vulnerable in ways that healthy cells are not . Using a genetic screening approach , van der Lelij , Lieb et al . searched for genes whose inactivation would harm only those cells that have mutant STAG2 proteins . This search found that one such gene encodes a protein called STAG1 , a close relative of STAG2 . Reducing the amount of STAG1 protein in cells with mutant forms for STAG2 caused these cells to start dying , whereas healthy cells were unaffected . Van der Lelij , Lieb et al . then conducted biochemical and cell biological experiments on bladder cancer and Ewing sarcoma cells to show that the cells need at least one of STAG1 or STAG2 to hold replicated chromosomes together . Without either protein , the X-shape of the chromosomes was lost and the cells died when they tried to divide . Thus , human cells can survive without STAG1 or STAG2 but not without both , a concept known as synthetic lethality . More research is now needed to identify how the STAG1 protein could be prevented from working . This knowledge could ultimately be used to develop drugs that would kill off only those cancer cells that have mutations that affect STAG2 .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "short", "report", "cell", "biology", "cancer", "biology" ]	2017	Synthetic lethality between the cohesin subunits STAG1 and STAG2 in diverse cancer contexts
Transcription start-site ( TSS ) selection and alternative promoter ( AP ) usage contribute to gene expression complexity but little is known about their impact on translation . Here we performed TSS mapping of the translatome following energy stress . Assessing the contribution of cap-proximal TSS nucleotides , we found dramatic effect on translation only upon stress . As eIF4E levels were reduced , we determined its binding to capped-RNAs with different initiating nucleotides and found the lowest affinity to 5'cytidine in correlation with the translational stress-response . In addition , the number of differentially translated APs was elevated following stress . These include novel glucose starvation-induced downstream transcripts for the translation regulators eIF4A and Pabp , which are also translationally-induced despite general translational inhibition . The resultant eIF4A protein is N-terminally truncated and acts as eIF4A inhibitor . The induced Pabp isoform has shorter 5'UTR removing an auto-inhibitory element . Our findings uncovered several levels of coordination of transcription and translation responses to energy stress . Transcription start site ( TSS ) selection and alternative promoter ( AP ) usage increase transcriptome diversity and its regulation . For example the level of transcription initiation can vary between different transcription start sites ( TSS ) under different growth conditions , in response to a specific signal or in different cell types and tissues . In addition , mRNA isoforms with different 5’ leaders can vary in their translation efficiency or their half-lives . Likewise , AP usage can lead to the generation of protein isoforms that differ in their N-termini and as a result have different or even opposite biological functions . Recent large-scale promoter analysis in hundreds of human and mouse primary cell types shed light on the prevalence of AP usage in mammals ( Forrest et al . , 2014 ) . Several studies have examined translation and stability of transcript isoforms of the same gene ( Arribere and Gilbert , 2013; Floor and Doudna , 2016; Wang et al . , 2016 ) but little is known about the contribution of AP usage to the translational response to stress . The process of protein synthesis is highly energy consuming and tightly regulated by the availability of nutrients , oxygen and growth factors . Downregulation of the translation machinery is a major mechanism that allows cells to preserve energy and cope with environmental deficiencies . Under these conditions translation of many mRNAs is inhibited but that of others is unchanged or even enhanced in order to survive the stress . The translation inhibition response is mediated by several mechanisms , in particular by the impairment of key initiation factors eIF2 and eIF4E ( Sonenberg and Hinnebusch , 2009 ) . Stresses such as growth factor , energy and amino acid deficiencies affect the formation of the cap binding complex eIF4F , comprising of the initiation factors eIF4E , eIF4G and eIF4A . Under these conditions , eIF4E-Binding Proteins ( 4EBPs ) which bind eIF4E with high affinity and interfere with its binding to eIF4G , is activated by dephosphorylation resulting in inhibition of cap-dependent translation . 4EBP is controlled by the mammalian target of rapamycin ( mTOR ) , a protein kinase that phosphorylates and diminishes its ability to bind eIF4E . Under circumstances of limited nutrients or other stresses mTOR activity is inhibited , 4EBP activity is enhanced and cap-dependent translation is suppressed ( Sonenberg and Hinnebusch , 2009 ) . The effect of the suppression of the general translation initiation factors under stress appears to vary from gene to gene , and is dictated by specific regulatory elements present in the mRNAs . A well-characterized mRNA feature associated with strong translational inhibition is the TOP element ( 5’ Terminal Oligo Pyrimidine ) , an uninterrupted stretch of 4–15 pyrimidines , starting with cytidine at the most 5’end of the mRNA ( for a review see Meyuhas and Kahan [2015] ) . A large fraction of TOP mRNAs code for proteins that are associated with translation and they are strongly translationally repressed following various physiological stresses that inhibit mTOR signaling by a mechanism that is not fully understood . It is thought that specific factors that bind the 5’ polypyrimidine track mediate the positive and negative translation regulation of the TOP mRNAs . While several features of cellular mRNAs involved in specific translational response to metabolic stress were characterized , the impact of specific TSS usage on the cellular response to stress is poorly investigated . In the present study we aimed to obtain a global view of the effects of TSS selection on translation following metabolic energy stress . To this end , we combined polysomal profiling with quantitative assessment of the 5’ ends of mRNAs . By comparing transcript isoforms that differ in their 5’ end we identified hundreds of genes with APs that are differentially translated , in particular following energy stress , suggesting that a major determinant of the differential response is associated with transcription-induced APs . We also determined the contribution of TSS nucleotides to translation . Strikingly , while the cap-proximal nucleotides have no significant effect on translation under optimal growth conditions , they display dramatic effects on the translational response to stress . We demonstrate that eIF4E levels drop following the stress and that the binding affinity of eIF4E towards capped mRNA with different first nucleotide varies significantly , with a relatively lower affinity to 5’ polypyrimidine as in TOP mRNAs . We next characterized two genes encoding translation regulatory factors with differentially translated APs . The first is eIF4A , the helicase subunit of the cap complex eIF4F , in which we identified a novel glucose-starvation-induced intronic promoter . The induced isoform , which has distinct 5’UTR and initiating AUG , is efficiently translated in energy deficient cells in spite of the global translation inhibition . The resultant protein is N-terminally truncated and acts as an eIF4A inhibitor , most likely to facilitate the stress response . The second is poly-A binding protein ( Pabp , Pabpc1 ) in which repression of the major TOP-containing isoform following stress is coupled with the induction of a downstream TSS that generates an mRNA isoform with a much shorter 5’UTR that is highly translated . Interestingly , the induced isoform lacks the well-characterized Pabp auto-inhibitory element ( Sachs et al . , 1986; Bag and Wu , 1996; Wu and Bag , 1998; Hornstein et al . , 1999 ) . Our findings expand the understanding of the regulatory mechanisms that coordinate the cellular response to metabolic energy stress both in transcription and translation . To obtain a global and quantitate view of the impact of transcription start site ( TSS ) selection on translation efficiency following cellular stress we combined polysomal profiling with quantitative assessment of the 5’ ends of mRNAs ( Figure 1A ) ( note that ribosomal footprinting is not suitable for our purpose since it primarily records coding sequences ) . Mouse Embryonic Fibroblasts ( MEFs ) were subjected to glucose deprivation , a stress that causes inhibition of global translation at the initiation and elongation levels mediated by signaling pathways that are AMPK-dependent ( Bolster et al . , 2002; Dubbelhuis and Meijer , 2002; Krause et al . , 2002; Reiter et al . , 2005; Shenton et al . , 2006 ) and independent ( Inoki et al . , 2003; Kalender et al . , 2010; Sinvani et al . , 2015 ) . Cells were then lysed and subjected to sucrose gradient sedimentation . As expected , the ribosome profile in response to glucose starvation ( GS ) was dramatically changed as the relative amount of 80S monoribosome was increased while polysomal fractions were decreased ( Figure 1B ) , indicating global inhibition of translation . The fractions from the gradient were merged to create three major pools: Polysome-free , from the top of the gradient to a single ribosome; Light , two to five ribosomes per mRNA; and Heavy , six or more ribosomes per mRNA . RNA was extracted from the pooled fractions and equivalent RNA volume was taken from each fraction for transcription start site library preparation . At this stage , two RNA spikes , GFP and Luciferase ( transcribed and capped in vitro ) were added into each fraction pool , to serve as controls for sample handling . For TSSs sequencing , we used the CapSeq method as previously described ( Gu et al . , 2012 ) . In this method an adaptor is added only to the cap site of the mRNA and is used for deep sequencing . Thus the sequence information reports on the 5’ end usage of transcripts within the polysomal profile . Samples from two independent biological replicates were subjected to Illumina sequencing . The 5’ ends of mapped reads predominantly corresponded to annotated and CAGE-defined transcription start sites , showing that our methodology provides quantitative single-base resolution view of the TSS landscape ( Figure 1C ) . 10 . 7554/eLife . 21907 . 003Figure 1 . Experimental design and general analysis of the impact of TSS selection on translation efficiency . ( A ) A schematic flowchart of the biological experiment and sample preparation for the CapSeq analysis . ( B ) Polysomal profiling of MEF cells subjected to glucose starvation ( GS ) for 8 hr ( dashed red ) or untreated ( blue ) . ( C ) Metagene analysis of CapSeq reads relative to the annotated TSS of Refseq and the summit of FANTOM5 TSSs at low and high resolutions . Only TSS regions ( −200 . . 200 ) with at least ten bases covered by reads were considered , and the coverage in each region was normalized to mean zero and standard deviation of one . The normalized coverage was then summed across all regions . ( D ) The relative distribution of genes with the indicated number of promoters per gene . ( E ) The relative global sum of reads for all promoters ( 9286 promoters with >500 reads ) in the polysome-free , light and heavy polysomal fractions in basal ( cont ) and GS conditions . The presented data are the mean of two independent replicates . ( F ) The fold change of the global mRNA levels between the basal ( cont ) and GS conditions . The presented data are the mean of two independent replicates . ( G ) The number and percentage of promoters translationally affected by GS . Promoters that had ribosome occupancy ( RO ) change of two-fold or more in both repeats were considered affected . ( H ) A scheme demonstrating the differential translation and transcription of transcripts with alternative TSSs/promoters . The TSSs are shown as arrows and the size of the arrow denotes the strength of the TSS relative to other TSSs of the same gene . The number of ribosomes occupying each mRNA represents the extent of its translation . ( I ) Promoters from the same gene were paired and their ROs were compared . Pairs of promoters that had an RO difference of two-fold or more in control or GS conditions in both repeats , independently , were considered as differentially translated promoters . The numbers of genes with at least one pair of differentially translated isoforms in control and GS conditions in both repeats are presented in a Venn diagram . ( J ) Boxplot presentation of the distributions of the 5′UTR lengths of differentially translated isoforms in each promoter pair ( as presented in I ) in control and GS conditions . The bottom and the top whiskers represent 5% and 95% of the distribution , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 21907 . 00310 . 7554/eLife . 21907 . 004Figure 1—figure supplement 1 . The response 5’TOP promoters to GS . ( A ) The cumulative CapSeq reads of 414 identified promoters with TSS TOP element in the polysome-free , light and heavy fractions in basal ( cont ) and GS conditions . The presented data are the mean of two independent CapSeq replicates . ( B ) The change of the mRNA levels of TOP promoters in basal ( cont ) and GS conditions . The presented data are the mean of two independent CapSeq replicates . ( C ) The impact of the GS-induced isoforms on 5’UTR length and ORF start . DOI: http://dx . doi . org/10 . 7554/eLife . 21907 . 004 When the results of two independent experiments were combined and normalized using internal controls , 9286 promoters that corresponded to 6861 genes had at least 500 reads per promoter and were considered in further analysis ( Supplementary file 1 ) . Under these thresholds , three quarters of the genes had a single expressed promoter in these cells and the rest had two or more ( Figure 1D ) . To validate that the TSS reads are in accordance with the expected biological outcome we first analyzed the global translational and transcriptional response to GS . We summed all the reads from each fraction for all the promoters and found that the overall translation was reduced upon GS as evident from the increase and decrease of the polysome-free ( Free ) and polysomal fractions ( Heavy ) , respectively ( Figure 1E ) , consistent with the polysomal profile ( Figure 1B ) . The overall mRNA levels were reduced by ~15% after the treatment ( Figure 1F ) , which most likely reflect the coupling between translation and mRNA stability ( Radhakrishnan and Green , 2016 ) . Next , we quantified the translational response of the TOP mRNAs ( CYYYY sequence is in the promoter summit ) , since these genes are known to be particularly sensitive to stresses . As expected , TOP mRNAs were preferentially inhibited upon GS on the translational level ( Figure 1—figure supplement 1A ) while the changes of the mRNA levels were comparable with those of other mRNAs ( Figure 1—figure supplement 1B ) . These data assured us in the ability of our approach to quantitatively characterize the translational properties of individual TSSs . We next analyzed the translational response of individual promoters following the GS stress . For this we determined the ribosome occupancy ( RO , the ratio between the reads in the Heavy+Light polysomal fractions to the polysome-free fraction ) of transcripts emanating from each promoter and calculated the effect of GS on the RO ( referred to as the RO effect ) for each experiment . A promoter was considered as translationally affected if its RO was affected by at least two-fold in each of the two biological repeats ( Supplementary file 1 ) . With these settings , we found that following glucose starvation , 22 . 5% of promoters showed a reduction of two-fold or more in translation whereas only 6 . 6% of promoters were induced by at least two-fold ( Figure 1G ) . Our approach enabled us to compare the translation efficiency of transcript isoforms emanating from APs of the same gene as schematically shown in Figure 1H . In genes having multiple promoters , we separately considered each pair of promoters with sufficient reads . In basal conditions 495 such pairs from 327 genes differed in their RO by two-fold or more ( Figure 1I ) . Of these , the mRNAs of 85 . 5% of promoter-pairs likely give rise to the same annotated ORF and the rest ( 14 . 5% ) are predicted to result in proteins with alternative N-termini . When the same analysis was performed following GS , 672 pairs of promoters from 465 genes displayed RO difference of two-fold or more , resulting in 42% increase in differential translation of APs ( Supplementary file 1 ) . Interestingly 70% of these differentially translated pairs do not overlap with those in basal conditions ( Figure 1I ) . These findings suggest that the increase in differential translation of transcript isoforms is part of the cellular stress response . Among APs corresponding to the same ORF , those resulting in shorter 5’UTRs had higher RO under basal conditions ( Figure 1J ) . On the other hand , no significant difference in 5’UTR length was seen between the more and the less efficiently translated pairs following stress ( Figure 1J ) . It therefore appears that the translation regulatory features following stress differ from those operating in basal conditions . We also compared the mRNA levels of the transcript isoforms following GS stress . We calculated the change in overall mRNA levels of each promoter following GS , and compared the effects on the APs of each gene ( Supplementary file 1 ) . 140 genes had at least one AP pair that showed ≥2 fold change in mRNA levels in response to stress in both experiments . Of the promoters preferentially induced following stress , 53% encoded a transcript with the same predicted ORF but shorter 5’UTR while the others were split between transcripts with the same ORF and longer 5’UTR and transcripts with different predicted ORF starts ( Figure 1—figure supplement 1C ) . The CapSeq data can provide insights into the impact of the exact nucleotide context of the TSS on translation efficiency . To determine the effect of the initiating nucleotides on translation efficiency we first analyzed all the non-redundant TSS positions in our data . We found that in MEFs , under basal and GS conditions , ~34% of TSSs initiated with adenosine ( A ) , ~23% with cytidine ( C ) , ~30% with guanosine ( G ) and ~13% with thymidine ( T ) ( Figure 2—figure supplement 1A ) , similar to previous reports ( Carninci et al . , 2006; Yamashita et al . , 2006; Forrest et al . , 2014 ) . Under basal conditions no significant difference was found in the RO of transcripts that vary in their first nucleotide ( Figure 2A ) . In contrast , the identity of the first base was correlated with significant differences in the response to GS ( Figure 2B ) . The RO of transcripts initiating with pyrimidines ( C and U ) was significantly reduced while those initiating with purines ( A and G ) appeared refractory to the stress . We performed the same analysis for initiating trinucleotides and found that under basal conditions , the RO of most trinucleotides is similar ( Figure 2—figure supplement 1B ) . Notable exceptions are UGA , UGC , UGG and UGU that display higher RO and the pyrimidine-rich trinucleotides CCT , CTC , CTT , that correspond to somewhat lower RO ( Figure 2—figure supplement 1B ) . The latter trinucleotides are most likely part of the TOP ( CYYYY ) element that was shown to confer slightly reduced translation even under optimal growth conditions ( Patursky-Polischuk et al . , 2009 ) . Interestingly following GS ( Figure 2C ) the majority the trinucleotides with initiating pyrimidines showed reduced translation , while those with purines were largely unaffected ( Figure 2C ) . Among the inhibitory trinucleotides were those that are part of the TOP element such as CCT and CTT , but also others that deviate from the TOP such as CAA , CAT , CTG , TAC etc . 10 . 7554/eLife . 21907 . 005Figure 2 . The impact of the TSS nucleotides on mRNA translation . ( A ) The RO distributions of transcripts that initiate with the indicated nucleotide in basal conditions . ( B ) The distribution of the RO effect ( log transformed ) of transcripts that initiate with the indicated nucleotide . The horizontal line indicates the overall median value . ( C ) The effect of GS on the RO for each initiating trinucleotide . The horizontal line indicates the overall median value . All the data presented in this figure are the mean of the two independent replicates . The bottom and the top whiskers represent 5% and 95% of the distribution , respectively . The number of TSSs starting with the indicated trinucleotide is indicated near each of the trinucleotide sequence . DOI: http://dx . doi . org/10 . 7554/eLife . 21907 . 00510 . 7554/eLife . 21907 . 006Figure 2—figure supplement 1 . The frequency of the initiating nucleotides and their impact on basal translation . ( A ) The percentage of each initiating nucleotide position as non-redundant TSSs in the CapSeq analysis in control and GS as indicated . ( B ) Boxplot presentation of the RO of each initiating trinucleotide in basal conditions . The horizontal line indicates the overall median value . The bottom and the top whiskers represent 5–95% of the population , respectively . The presented data are the mean of the two CapSeq experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 21907 . 006 Representative examples of TSSs that are separated by just a few bases within the same promoter and display differential translational reaction to GS are shown in Figure 3 . The overall translation and mRNA levels of Atp5a1 ( ATP Synthase , mitochondrial ) are downregulated upon GS ( Figure 3A , upper right panel ) . This promoter has several strong TSSs , three of which are adjacent to each other and designated 1 , 2 and 3 . Translation of TSS#1 that begins with CAT was much more strongly inhibited than that of the nearby TSSs that begin with ATT and TTT , regardless of the effect on transcription . 10 . 7554/eLife . 21907 . 007Figure 3 . Examples of the effect of the first nucleotide on translation efficiency . ( A , B ) Upper left panel – The chromosomal location , genomic structure and CapSeq data of the indicated genes in each fraction in control and GS conditions . The scale of the normalized and row reads ( in parentheses ) is shown . The nucleotide sequence of the major alternative TSSs within the promoter are marked by arrows and numbered . The upper right panel shows the relative levels of all reads for the indicated promoter and the lower panels for the designated specific TSSs in basal ( cont ) and GS conditions . The presented data are the mean of the two independent replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 21907 . 007 Another example is the p1 promoter of the Sars ( Seryl-tRNA synthetase ) gene that has multiple TSSs ( Figure 3B , upper panel shows the major two ) . The overall translation of Sars was downregulated after GS ( Figure 3B , upper right panel ) . The two major TSSs , separated only by 5nt , had a clearly differential response to GS . While in basal conditions TSS#1 and TSS#2 showed the same pattern of ribosome occupancy , following GS the translation of TSS#2 that initiates with CTC , was downregulated to a greater extent than TSS#1 that initiates with ACA . The overall translational response to GS of Sars p1 promoter was intermediate between the responses of each TSS . These examples clearly show that in addition to differential promoter regulation , the initiating sequence of an mRNA is also important for the translation efficiency of the transcript . To gain insight into the underlying basis for the effect of cap-proximal nucleotides on translation , we analyzed eIF4E , the cap binding subunit of eIF4F . eIF4E has a single promoter that drives high level of transcription and translation under basal conditions ( Figure 4A and B ) . Following GS eIF4E expression is substantially downregulated at the mRNA and translation levels ( Figure 4A and B ) , which leads to reduction in eIF4E protein levels ( Figure 4C ) . 10 . 7554/eLife . 21907 . 008Figure 4 . The effect of cap-proximal nucleotides on eIF4E binding affinity and activity . ( A ) The relative level of eIF4E reads ( mean of two independent replicates ) in the indicated fractions in basal ( cont ) and GS conditions . ( B ) The relative mRNA levels ( mean of two independent replicates ) of eIF4E in basal ( cont ) and GS conditions . ( C ) Representative immunoblot of total lysate using eIF4E and Tubulin antibodies following GS of MEFs for the indicated times . ( D ) Upper panel – the change in the intrinsic fluorescence of eIF4E ( 300 nM ) in response to increasing concentrations of capped RNA ligands ( 1 . 25 nM–5 μM ) or cap analog ( 1 . 25 nM–10 μM ) . The graphs represent the mean of three independent experiments with two different protein preparations . The bottom panel shows the calculated dissociation constant values ( Kd ) of eIF4E binding affinity to the indicated RNA ligands . ( E ) The effect of eIF4E knockdown on GFP expression driven by mRNA with C or A as the initiating nucleotides . HEK293T cells were transfected with either eIF4E siRNA or a non-targeting siRNA ( 10 nM ) . 48 hr later cells were transfected again with GFP reporter genes driven either by RPL18 ( C ) or CMV ( A ) promoter . Cells were harvested 24 hr after the second transfection and analyzed by western blot with GFP , eIF4E and Tubulin antibodies as indicated . ( F ) HEK293T cells were transfected RPL18 and CMV driven GFP reporter gene . Six hours later , increasing amounts of 4EGI-1 were added to the media as indicated . Cells were harvested 24 hr after transfection and subjected to western blot using GFP and Tubulin antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 21907 . 00810 . 7554/eLife . 21907 . 009Figure 4—figure supplement 1 . Complementary data for the analysis of eIF4E binding affinity . ( A ) The purified eIF4E protein used for the binding studies was subjected to 10% SDS-PAGE and Coomassie blue staining . ( B ) The p-values of the differences in binding affinities between the indicated capped RNAs that differ by their first nucleotide . DOI: http://dx . doi . org/10 . 7554/eLife . 21907 . 009 A possible link between eIF4E levels and the effect of cap-proximal nucleotides under stress may involve differential binding affinity . We therefore analyzed eIF4E binding to RNAs with different first nucleotides . Most previous studies of eIF4E binding affinity used cap analogs such as m7GTP or m7GpppG and reported Kd values ranging from low μM to few nM ( Ueda et al . , 1991; Carberry et al . , 1992; Minich et al . , 1994; Sha et al . , 1995; Hagedorn et al . , 1997; Miyoshi et al . , 1999; Wieczorek et al . , 1999; Hsu et al . , 2000; von Der Haar et al . , 2000; Niedzwiecka et al . , 2002; Scheper et al . , 2002; Ghosh et al . , 2008; Thillier et al . , 2012 ) . We chemically synthesized three capped RNA oligos as previously described ( Lavergne et al . , 2008; Thillier et al . , 2012 ) , 10 nt long each that differ in their first nucleotide . Human eIF4E was expressed as His-tag fusion in E . coli and purified from the soluble fraction by nickel agarose beads and subsequent gel filtration ( Figure 4—figure supplement 1A ) . Using the purified eIF4E we performed intrinsic fluorescence intensity measurements in the presence of increasing concentrations of either m7GpppG cap analog or the capped RNA oligos whose first nucleotide is C , A or G . The titration data of the RNA oligos and the cap analog are shown in Figure 4D . The measured Kd for the m7GpppG was 561 nM ( Figure 4D ) . Interestingly , under the same conditions , the Kd of all capped RNA oligos was substantially lower , indicating that the RNA moiety increases the binding affinity ( Figure 4D ) . Furthermore , significant differences in the Kd values between the RNA oligos were observed ( Figure 4D , p-values are shown in Figure 4—figure supplement 1B ) . The oligo with polypyrimidine CCU as initiating nucleotides display lower affinity compared to oligos with ACU and GCU as first nucleotides . These Kd values of the mRNAs are correlated with the data of the relationship of the first nucleotides and the RO response to GS ( Figure 2C ) . These findings suggest that eIF4E binding affinity contributes to the level of translation , in particular when its intracellular concentrations become limiting as in GS ( Figure 4C ) . To gain further support to this idea we downregulated the levels of eIF4E using siRNA and introduced into these cells GFP reporter genes driven either by Rpl18 or CMV promoters that direct transcription start site at a C ( this study ) or an A ( Elfakess and Dikstein , 2008 ) nucleotide , respectively . Both , the C and the A initiated transcripts were downregulated but the effect is substantially more pronounced for the C-initiated transcript ( Figure 4E ) . We also analyzed these GFP reporter genes in cells treated with low concentrations of 4EGI-1 , a drug that inhibits eIF4E by disrupting its interaction with eIF4G1 ( Moerke et al . , 2007 ) . Here again the C-initiating transcript was much more vulnerable to this drug compared to the A-initiated one ( Figure 4F ) , consistent with the notion that a C as an initiating nucleotide is highly responsive to fluctuations in eIF4E availability . Next we looked at several genes with more than one promoter displaying differential translation and transcription . The global inhibition of translation in response to stress is mediated by the 5’TOP element that is present in almost all ribosome subunits and many of the translation initiation factors . An exception is eIF4A , the helicase subunit of eIF4F . As seen in the mapped CapSeq reads ( Figure 5A ) , four promoters ( p1 , p2 , p3 and p5 ) account for most of eIF4A mRNAs . p1 and p2 are positioned upstream of the annotated eIF4A protein start codon and in basal conditions these are the major promoters producing highly transcribed and efficiently translated mRNAs ( Figure 5B and C ) . On the other hand , p3 and p5 promoters are intronic and positioned downstream of the annotated protein start codon so that transcription from these promoters generates isoforms with different 5’UTRs and an alternative truncated N-terminus . Upon GS p1 and p2 translation was slightly reduced ( Figure 5B ) , whereas the intronic p3 and p5 promoters were significantly upregulated transcriptionally and translationally ( Figure 5B and C ) . While p3 and p5 constitute a small fraction of the overall eIF4A under basal conditions , they account for more than half of eIF4A transcripts upon GS . 10 . 7554/eLife . 21907 . 010Figure 5 . Characterization of the GS-induced AP of eIF4A . ( A ) The chromosomal location , genomic structure and CapSeq data of eIF4A in each fraction in basal ( cont ) and GS conditions ( uniquely mapped reads ) . The scale of the normalized and row reads ( in parentheses ) is shown . The positions of the FANTOM5 promoters are also indicated . ( B ) The relative levels of the reads from the indicated promoters ( mean of two independent replicates ) in the three polysomal fractions in control and GS conditions . ( C ) The number of CapSeq reads of the indicated promoters in control and GS conditions ( mean of two independent replicates ) . ( D ) Analysis of the p1 and p3 promoters of eIF4A by 5’RACE . Upper panel shows schematic presentation of the relevant eIF4A genomic region , the positions of the analyzed promoters and the 5’RACE reverse primers ( shown by arrowheads ) . The lower panel is the analysis of the 5’RACE PCR products by 6% PAGE . ( E ) Analysis of p1 and p3 transcript isoforms levels by 5’RACE in the indicated fractions of the polysome profile of control and glucose starved cells . ( F ) Representative immunoblot of total lysate of MEFs with eIF4A ( C-terminal epitope ) and Tubulin antibodies in control and GS . ( G ) The 5’UTR of the eIF4A from the p3 TSS to Met121 was cloned downstream of the CMV promoter and upstream of the GFP reporter gene as shown in the scheme . This construct was transfected into MEFs , which were then subjected to GS . Expression of GFP was monitored by western blot with GFP antibody . ( H ) Representative immunoblot of HEK293T cells that were co-transfected with GFP reporter having hairpin structure within the 5’UTR together with increasing amounts of WT or ΔN-eIF4A as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 21907 . 010 To investigate further the induction of the intronic promoters and their function we first confirmed with 5’RACE that p3 is indeed strongly induced by GS ( Figure 5D ) and is differentially translated compared to p1 as evident from their distribution in sucrose gradient fractions ( Figure 5E ) . The putative 5’UTR of the intronic isoform contains multiple AUGs possibly generating short upstream ORFs . To gain insight into the translation initiation site of the induced protein isoform we analyzed eIF4A protein using C-terminus specific antibody . We observed an energy-stress induced polypeptide of 33 kDa that is ~13 kDa shorter than the main protein ( Figure 5F ) . This truncated protein can be synthesized either from Met 121 or Met 127 of the main ORF . We used a GFP reporter that is preceded by the entire novel eIF4A 5’UTR starting from p3 up to Met121 . This construct drove the expression of a single polypeptide corresponding to translation initiation at Met121 , which was also induced following GS ( Figure 5G ) . To examine the function of this truncated protein we constructed an eIF4A expression plasmid lacking the first 120 amino acids ( eIF4AΔN ) . This truncation is expected to impair the helicase activity of eIF4A but to retain its interaction with eIF4G1 ( Dominguez et al . , 2001 ) . Cells were transfected with either full-length ( WT ) or N-terminally truncated eIF4A together with a GFP reporter gene preceded by a 5’UTR bearing a cap-proximal secondary structure ( Figure 5H ) . While the WT eIF4A has no effect on GFP expression , eIF4AΔN inhibited GFP protein levels in a dose dependent manner , indicating that eIF4AΔN acts as an inhibitor of eIF4A . Thus AP usage in eIF4A diminishes its activity following GS , reminiscent of the down regulation of other initiation factors following stress . Poly ( A ) binding protein ( Pabp ) , a central translation initiation regulatory factor , has several annotated promoters . The major promoter under basal conditions is p1 , which generates a TOP mRNA ( Figure 6A ) . This mRNA isoform is expressed 40-folds higher than the next most transcribed isoform , driven by promoter p2 ( Figure 6C ) . After GS , p1 mRNA levels and translation were both strongly reduced ( Figure 6B and C ) , while p2 transcription was elevated by eight-fold ( Figure 6C ) and the vast majority of p2 transcripts were heavily translated ( Figure 6B , right ) . In Pabp , both p1 and p2 are positioned upstream of the annotated protein start codon , hence the differential regulation of the APs upon GS changes solely the 5’UTR length ( from 467 nt to just 63 nt ) . Interestingly , it was previously reported by several groups that the 5’UTR of the Pabp major isoform consists of conserved A-rich sequences ( ARSs ) which serve as binding sites of Pabp , resulting in translational repression ( Bag and Wu , 1996; Hornstein et al . , 1999; Kini et al . , 2016 ) . These ARSs are missing in the shorter GS-induced mRNA isoform . Using 5’RACE we validated the induction of p2 mRNA isoform upon GS ( Figure 6D ) and the differential translational response of p1 and p2 to GS ( Figure 6E ) . Analysis of Pabp protein levels revealed that despite the strong translational inhibition of the primary isoform its protein levels remained stable , suggesting that the highly translated GS-induced transcript isoform compensates for the diminished translation of transcripts produced from p1 ( Figure 6F ) . Pabp plays an important role in translation by promoting RNA circularization mediated by interaction with the cap complex eIF4F ( Wells et al . , 1998; Kaye et al . , 2009; Yanagiya et al . , 2009; Park et al . , 2011 ) . However , tight control of Pabp levels is critical as excess of Pabp is inhibitory for translation in vivo and in vitro ( Yanagiya et al . , 2010 ) . As eIF4F activity and the overall mRNA levels ( dictating the overall abundance of poly-A tails ) are downregulated in response to GS , the stoichiometry of Pabp relative to eIF4F and to the mRNA poly-A tails is elevated upon the stress and thus contributes to translation inhibition . 10 . 7554/eLife . 21907 . 011Figure 6 . Characterization of Pabp APs . ( A ) The chromosomal location , genomic structure and CapSeq data of Pabp in each fraction in basal ( cont ) and GS conditions . The scale of the normalized and row reads ( in parentheses ) is shown . The positions of the FANTOM5 promoters are also shown . ( B ) The relative levels of the indicated promoter reads of the polysomal fractions in control and GS conditions ( mean of two independent replicates ) . ( C ) The total CapSeq reads of the indicated promoters in control and GS conditions ( mean of two independent replicates ) . ( D ) 5’RACE analysis of p1 and p2 promoters of Pabp . Upper panel shows a schematic presentation of Pabp region containing the p1 and p2 promoters , their positions , 5’RACE reverse primers ( shown as arrowheads ) and adenine-rich autoregulatory sequences ( ARSs ) . The lower panel is the analysis of the 5’RACE PCR products by 6% PAGE . ( E ) Analysis of p1 and p2 transcript isoforms levels by 5’RACE in the indicated fractions of the polysome profile of control and glucose starved cells . ( F ) Representative immunoblot of total lysate of MEFs with anti-Pabp and anti-Tubulin in control and GS conditions . ( G ) Characterization of Pabp p2 upstream region as a promoter . A scheme of p2 regulatory sequences of the indicated lengths ( starting from the AUG ) that were cloned upstream to a promoter-less Renilla luciferase ( RL ) reporter gene is shown on the left . These constructs were transfected into MEFs together Firefly luciferase reporter gene that served as an internal control . Renilla and Firefly luciferase activities were measured 24 hr after transfection . The results represent average ± SE of 4 transfection experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 21907 . 01110 . 7554/eLife . 21907 . 012Figure 6—figure supplement 1 . Examples of differential transcription and translation response to GS of APs . ( A , B ) Upper left panel - The chromosomal location , genomic structure , the positions of the FANTOM5 promoters and CapSeq data of the indicated genes in each fraction in basal ( cont ) and GS conditions . The scale of the normalized and row reads ( in parentheses ) is shown . Upper right panel - The total number of CapSeq reads of the indicated promoters in cont and GS conditions . Lower panels - The relative CapSeq reads of the indicated APs/TSSs for the polysome-free , light and heavy polysomal fractions in basal ( cont ) and GS conditions . The presented data are the mean of the two independent replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 21907 . 01210 . 7554/eLife . 21907 . 013Figure 6—figure supplement 2 . Additional examples of differential transcriptional and translational response to GS of APs . ( A–E ) Upper panel – A schematic representation of the APs of the indicated genes and their locations . The CapSeq reads of the indicated APs of the polysome-free , light and heavy polysomal fractions in basal ( cont ) and GS conditions ( mean of two independent replicates ) . Right panel - The CapSeq reads of the indicated promoters in control and GS conditions . Each example is accompanied by a brief summary of the expected outcome of the APs . DOI: http://dx . doi . org/10 . 7554/eLife . 21907 . 013 We examined the potential of p2 upstream region , which also serves as the 5’UTR of the major Pabp transcript ( p1 ) , to act as a promoter . The upstream sequences of p2 were placed upstream to a promoter-less Renilla luciferase ( RL ) gene and transfected into MEFs ( Figure 6G ) . The results revealed that progressive addition of upstream sequences up to 282 nt relative to the initiating ATG , substantially enhanced RL activity . Further extension of p2 upstream sequences up to 456 nt resulted in diminished RL activity . The results confirm that the p2 promoter overlaps the 5’UTR of the major Pabp transcript and uncovered positive and negative regulatory elements . Arhgef2 ( Rho/Rac guanine nucleotide exchange factor 2 ) is another interesting example of a gene with APs that are differentially regulated by GS ( Figure 6—figure supplement 1A , showing the five strongest promoters of Arhgef2 ) . The p1 and p3 promoters give rise to the longest Arhgef2 isoforms and their mRNA levels are slightly inhibited following GS . p2 directs the synthesis of the shortest Arhgef2 isoform with distinct 5’UTR and N-terminus and its translation and mRNA levels are moderately upregulated . p4 and p10 also generates an alternative N-terminus isoform with different 5’UTR as well . Following GS , their mRNA levels were induced by an average of 15-folds and their translation was also induced . Thus the GS stress shifted Arhgef2 protein synthesis from one N-terminal isoform to another . An intriguing example of alternative TSSs selection within the same promoter following stress is shown in Figure 6—figure supplement 1B . Hcfc2 ( Host cell factor C2 ) has one major expressed promoter ( p1 ) in MEFs . Upon GS , the p1 promoter mRNA contribution increased by 1 . 7-folds alongside a mild increase in ribosome occupancy ( Figure 6—figure supplement 1B , bottom panel ) . Interestingly , one of the weak TSSs within the Hcfc2 p1 promoter rose dramatically following GS ( TSS #2 ) . This TSS is positioned only three nucleotides upstream to the only annotated Hcfc2 protein’s start codon , creating 5’UTR of only 3nt . Although Hcfc2 has a TISU ( Translation Initiator of Short 5'UTR ) element surrounding its initiating AUG ( Elfakess and Dikstein , 2008; Elfakess et al . , 2011 ) , a 5’UTR of 3nt is too short even for TISU mediated translation initiation . Hence , the translation of the GS-induced TSS #2 is expected to start at a downstream in-frame AUG , creating N-terminally truncated isoform ( 34 aa shorter ) . Several additional interesting examples of AP-mediated response affecting both transcription and translation are described in Figure 6—figure supplement 2 . In the present study , we provide a global view and mechanistic insights of the impact of TSS selection on translation following metabolic energy stress . Our findings uncover the critical importance of the exact TSS/cap-proximal nucleotides in the translational response to energy stress . In basal condition , the initiating nucleotides have no significant effect while the 5’UTR length appears to be an important determinant that governs the differential translation efficiency of transcript isoforms , with longer 5’UTRs diminishing translation efficiency . The exact opposite is seen during energy stress , as the effect of 5’UTR length was insignificant while the initiating nucleotides appear to be responsible , at least in part , for the difference in the RO effect . This phenomenon was clearly evident from cases in which adjacent nucleotides from the same promoter display dramatic differences in the translational response to the stress . In these instances the 5’UTR length and sequence are almost identical . The differential sensitivity to the stress is particularly apparent between the purine and pyrimidine nucleotides . Although the first nucleotide contributed greatly to this differential effect , the following nucleotides are also important . The nucleotides that confer inhibition of translation following the stress start in most cases with cytidine , as does the TOP element . Intriguingly , the following nucleotides in GS-repressed trinucleotides do not always follow the TOP consensus of uninterrupted stretch of pyrimidines . Starting purines seem to confer greater resistance to the inhibition of translation . While the reporter gene assays clearly show the importance of the first nucleotide and isoforms generated from AP for differential translation ( Figures 4E , F and 5G ) , it is possible that other features in endogenous transcripts may contribute to translation . Furthermore eIF4G1 and eIF4A , the partners of eIF4E in eIF4F , may also contribute to the differential affinity via their RNA binding domains . Thus , our findings expand the repertoire of regulatory sequences that mediate the effect of the stress on translation . We provide a potential mechanistic link between eIF4E and the effect of TSS nucleotides on translation following GS . First , eIF4E translation and protein levels are downregulated . This is in addition to the inhibition of its activity by 4EBP upon this stress ( Bolster et al . , 2002; Dubbelhuis and Meijer , 2002; Krause et al . , 2002; Reiter et al . , 2005 ) . The translation inhibition of eIF4E may be a backup mechanism for 4EBP-mediated inhibition . Second , mRNAs that differ in their cap-proximal nucleotides display differential affinity towards eIF4E in correlation with the observed sensitivity to the stress . Specifically , the TOP-like CCU that is highly repressed following GS exhibits much lower affinity compared to RNAs that initiate with an A or G which we found to be more resistant in their translation to the stress . Several previous studies that addressed the regulatory mechanism of the TOP element uncovered several trans-acting factors , positive and negative , that mediate the translational control of these mRNAs ( Meyuhas and Kahan , 2015 ) . Our findings add an additional layer of regulation of these mRNAs by demonstrating that under energy stress conditions , their inhibition/resistance is , at least in part , the outcome of the intrinsic properties of the translation machinery . It is possible that the signal-modulated trans-acting factors of the TOP element act to modify the eIF4E-mRNA affinity . For instance LARP1 interacts with both the TOP element and PABP to stimulate TOP mRNA translation ( Tcherkezian et al . , 2014 ) . As an RNA binding protein with preference to pyrimidine-rich sequences this factor can greatly increase the recruitment of eIF4E to these mRNAs via PABP , which also interacts with eIF4G1 . The involvement of the binding affinity of TOP mRNAs to eIF4E was previously studied by Shama et al using in vitro translation assays of TOP and non-TOP mRNAs in the presence of the cap analog m7GpppG ( Shama et al . , 1995 ) . As the two types of mRNA were inhibited to a similar extent by the cap analog it was concluded that the TOP and non-TOP mRNAs have similar eIF4E binding affinity . The discrepancy between this study and our results can be explained by the use of different assays . In the present study the binding affinity was measured directly using known amounts of RNA and eIF4E protein . Shama et al . inferred the relative affinity from an indirect assay in which the actual amount of eIF4E and the mRNAs are not known and the concentration of the cap analog was very high , above the saturating range . Another study does suggest a role of eIF4E in mTOR-regulated TOP mRNA translation on the basis of short-term pharmacological inhibition of mTOR in 4EBP-deficient cells ( Thoreen et al . , 2012 ) . However , this conclusion was challenged since other stresses ( oxygen and amino acid deficiencies ) that also diminish mTOR activity resulted in TOP mRNAs repression in these cells ( Miloslavski et al . , 2014 ) . It is possible that the reduction in eIF4E availability by means other than 4EBP may contribute to this regulation . We also uncovered hundreds of genes in which their APs were differentially translated . Remarkably , this phenomenon was particularly apparent following GS as the number of genes with 5’end isoforms displaying differential RO was substantially elevated . The detailed analyses of several intriguing examples of transcript isoforms with differential translation reveal the potential of translational control via APs . Specifically , we found that differential transcription and translation of two central translation initiation factors , eIF4A and Pabp , contribute to the global inhibition of translation following energy stress . Both have GS-induced isoforms that result in up-regulation of inhibitory eIF4A and Pabp proteins . The GS-induced downstream intronic promoter of eIF4A drives the expression of an isoform with long 5’UTR that consists of 10 uAUGs , yet this isoform is efficiently translated under GS . One possibility is that the GS-induced translational activity involves an internal ribosome entry site ( IRES ) that can bypass the inhibitory effect of uAUGs . Alternatively , translation can be activated by uORF-mediated mechanisms in which phosphorylation of the subunit of eIF2α favors translation of genes containing multiple uAUGs ( Young and Wek , 2016 ) . In summary , by measuring isoform-specific translation in basal and energy stress conditions we uncovered previously unknown regulatory mechanisms that broaden our understanding of how stress alters cellular translatome and transcriptome and how these processes are coordinated . We anticipate that further analysis of the data and focusing on specific examples will provide additional novel insights of transcription-translation links . Untreated and 8 hr glucose starved MEFs were incubated with 100 μg/ml Cycloheximide ( Sigma ) for 5 min and then washed twice with cold buffer containing 20 mM Tris pH 8 , 140 mM KCl , 5 mM MgCl2 and 100 μg/ml Cycloheximide . The cells were collected and lyzed with 500 μl of same buffer that also contains 0 . 5% Triton , 0 . 5% DOC , 1 . 5 mM DTT , 150 units RNAse inhibitor ( Eurx ) and 5 μl of protease inhibitor ( Sigma ) . The lyzed samples were centrifuged at 12 , 000g at 4°C for 5 min . The cleared lysates were loaded onto 10–50% sucrose gradient and centrifuged at 41 , 000 RPM in a SW41 rotor for 90 min at 4°C . Gradients were fractionated and the optical density at 254 nm was continuously recorded using ISCO absorbance detector UA-6 . The collected samples were then merged to create three fractions: Polysome-free ( F ) , Light ( L ) and Heavy ( H ) . The use of three pools is sensitive to large changes in translation but relatively insensitive to smaller changes . RNA was isolated for each fraction using Trizol and Direc-Zol RNA mini-prep kits ( Zymo Research ) . RNA spikes ( 0 . 25 ng GFP and 1pg Luciferase ) that were transcribed ( P1300 , Promega ) and capped ( S1407 , NEB ) in-vitro using T7 RNA Polymerase were added into each fraction . Equivalent RNA volume was taken from each fraction for the library preparation for TSSs sequencing using the CapSeq method ( with modifications ) as previously described ( Gu et al . , 2012 ) . Specifically , the RNA samples were treated with Terminator 5´-Phosphate-Dependent Exonuclease ( TER51020 , Epicenter ) . Then , reaction volume was enlarged and CIP ( M0290 , NEB ) was added to dephosphorylate non-mRNA RNAs ( tRNA and 5S rRNA ) . DNaseI was added in this step . In order to remove mRNA 5’Cap , the samples were treated with TAP ( T19050 , Epicenter ) and then RNA linker was ligated at the 5’end of the formerly capped mRNA by T4 RNA ligase ( M0204 , NEB ) . cDNA was prepared from each RNA sample by using hexamer random primers linked to illumina 3’ Rd2 seq primer , index ( unique barcode for each sample ) , 4nt-long UMI and P7 illumina adaptor . To increase the cDNA quantity , a linear PCR amplification ( second strand synthesis ) was performed with forward primer containing the ligated linker sequence and carries P5 illumina adaptor . Final library amplification PCR step was performed after calibrating the numbers of PCR cycles . Size selection with magnetic beads ( Ampure XP , according to the manufacturer guidelines ) was performed after cDNA synthesis for removal of fragments < ~150 nt and also after linear amplification and final PCR steps , in which sizes of >200 bp and <500 bp were selected . The deep sequencing was performed with HiSeq High-Throughput Sequencing System . Reads were mapped to the mouse genome ( mm9 assembly ) using STAR aligner and to the GFP and luciferase sequences using Bowtie2 . The BAM file was altered to contain just the first base of the alignment using a custom Java script ( Supplementary file 2 ) and coverage tracks were prepared using genomeCoverageBed . The tracks were then normalized by the numbers of reads mapping to the GFP/LUC ( spikes ) and the polysomal normalization factors . The numbers of reads mapping to each TSS defined by the FANTOM5 project ( phases 1 + 2 , permissive set ) was counted using htseq-count . Log2-transformed ratios were computed after adding a pseudo-count of 0 . 5 to the each normalized read number . Promoters were assigned to Ensembl 75 transcript models ( ENSTXXX ) . A promoter was considered the ‘annotated’ promoter of a transcript if it was within 100 nt of its 5' end and ‘Exonic’ if it overlapped exons of the 5'UTR of the transcript . Ribosome occupancy was defined as the ratio of Light+Heavy fractions and the polysome-free fraction . RO effect=ROGSROcont; mRNA levels change after GS=GS reads ( free+Light+Heavy ) Cont reads ( free+Light+Heavy ) . Calculation of the differences of the RO between promoter pairs is shown in Figure 1I and for the mRNA levels ( Figure 1—figure supplement 1C ) is as follows: mRNA differential effect = mRNA levels change after GS ( p1 ) mRNA levels change after GS ( p2 ) . TOP element promoters were defined as promoter summits initiated with CYYYY sequence . The data can be accessed in Gene Expression Omnibus ( GEO ) accession no . GSE93981 or tracked using UCSC genome browser at: ftp://ftp-igor . weizmann . ac . il/pub/hubANA . txt MEFs from a WT mouse ( from Benois Viollet , INSERM , Paris ) and HEK293T ( from ATCC ) were maintained in DMEM supplemented with 10% fetal calf serum , 100 units/ml penicillin , 100 mg/ml streptomycin and 1 mM Sodium pyruvate ( for MEFs only ) . Cells were tested negative for mycoplasma by PCR . Cells were harvested after the indicated period of glucose starvation [glucose free media ( 11966025 , Gibco ) with 10% dialyzed FBS] or 24 hr after transfection and lysates were subjected to SDS-polyacrylamide gel electrophoresis followed by western blot . The levels of proteins were determined using the following antibodies: anti-GFP ( ab290 , Abcam ) , anti-eIF4A ( NBP2-24632 , Novus; AAS65480C , Antibody verify ) , anti-eIF4E ( ab33766 , Abcam ) , anti-HA ( ab9110 , Abcam ) , anti-Pabp ( sc-32318 , Santa Cruz ) and anti-Tubulin ( Sigma ) . The pEGFP-N1 ( Clontech ) , pEGFP-N1 with 5’UTR bearing secondary structure and the HA-eIF4A plasmids were previously described ( Elfakess et al . , 2011 ) . To construct the N-terminal truncated eIF4A we used the T-PCR method ( Erijman et al . , 2011 ) with HA-eIF4A as a template . Sequences of eIF4A starting from p3 TSS to Met121 were amplified from MEFs genomic DNA ( intronic part ) and cDNA ( exonic part ) and cloned downstream to the CMV promoter of in the pEGFP-N1 using two steps RF-cloning . The p2 upstream sequences of Pabp were amplified from MEFs cDNA and used to replace the CMV promoter in CMV-RL construct using T-PCR . The GFP reporter under the control of the Rpl18 promoter was described ( Sinvani et al . , 2015 ) . All the constructs were verified by sequencing . For knocking-down eIF4E , HEK293T cells were transfected with Dharmacon siGENOME SMART pool siRNA ( M-002000–00 , Thermo Scientific ) using DharmaFECT1 transfection reagent . The Dharmacon ON-TARGETplus Nontargeting siRNA #3 was used as a negative control . 48 hr after the initial transfection , cells were transfected with the indicated GFP reporter plasmids . Cells were harvested 24 hr after the second transfection . eIF4e was purified from E . coli BL21 ( DE3 ) bacteria transformed with eIF4E-pET-30a construct ( kindly provided by Franck Martin , Université de Strasbourg , France ) . Bacteria were grown with Kanamycin at 37°C up to OD ( 600 ) = 0 . 8 . Then , IPTG ( 0 . 5 mM ) was added and bacteria were harvested after overnight incubation at 18°C . The samples were lysed using sonication in TPA buffer ( 20 mM Tris-HCl , pH7 . 5 , 10% Glycerol , 0 . 1 mM EDTA , 1 mM DTT ( Sigma ) , 100 mM KCl ) added with 2 . 5 mM PMSF , 10 mM β-mercaptoethanol and protease inhibitor cocktail ( Sigma ) . All the buffers were prepared using Diethylpyrocarbonate ( DEPC ) -treated double distilled water . The lysate was subjected to centrifugation and subsequent ultracentrifugation in order to remove membrane associated RNAses . The soluble fraction was purified on gravity column using Ni-NTA His-Bind Resin ( 70666 , Novagen ) , eluted with the TPA buffer containing 250 mM imidazole and subsequently dialyzed . Fractions containing eIF4E protein underwent gel filtration chromatography ( Superdex 75 , GE Healthcare ) . Capped RNA oligos were chemically synthesized as previously described ( Lavergne et al . , 2008; Thillier et al . , 2012 ) . For the label-free fluorescence measurements , purified eIF4E ( 300 nM ) in TPA buffer with 0 . 1% Pluronic F-127 ( NanoTemper ) and 50 μg/ml yeast tRNA , was mixed with increasing concentrations of cap analog ( 1 . 25 nM to 10 μM; S1407 , NEB ) or capped RNA oligos ( 1 . 25 nM to 5 μM ) . The reaction was centrifuged at 12 , 000 rpm for 5 min and then premium-coated capillaries ( MO-Z005 , NanoTemper ) were loaded with the samples , and intrinsic fluorescence measurements were performed in a Monolith NT . LabelFree instrument ( NanoTemper ) . Data analysis was performed with the Monolith NT . Analysis software ( NanoTemper ) . Total RNA extracted from control and 8 hr glucose starved cells was used as a template to create cDNA with superscript II ( Invitrogen ) according to the manufacturer’s instructions using gene-specific primers . The cDNA was purified using a PCR purification kit ( Qiagen , Germany ) , followed by addition of polyG to the 5’-end using TdT enzyme ( Promega ) for 1 hr at 37°C . The reaction was terminated by heat inactivation for 15 min at 65°C , and the products were purified with the PCR purification kit ( Qiagen ) . The modified cDNA was used as a template for PCR with Phusion polymerase ( NEB ) using nested reverse primers and forward PolyC primer . PCR products were run on 6% polyacrylamide gel .	The production of new proteins is a complex process that occurs in two steps known as transcription and translation . During transcription , the cell copies a section of DNA to make molecules of messenger ribonucleic acid ( or mRNA for short ) in the nucleus of the cell . The mRNA then leaves the nucleus and enters another cell compartment called the cytoplasm , where it serves as a template to make proteins during translation . A mRNA molecule contains a sequence of building blocks known as nucleotides . There are four different types of nucleotides in mRNA and the order they appear in the sequence determines how the protein is built . Both transcription and translation consume a lot of energy so they are highly regulated and sensitive to environmental changes . However , since transcription and translation happen in different cell compartments , it is not known if and how they are coordinated under stress . Tamarkin-Ben-Harush et al . studied transcription and translation in mouse cells that were starved of glucose . The experiments show that the identity of the very first nucleotide in the mRNA – which is dictated during transcription – has a dramatic influence on the translation of the mRNA , especially when the cells are starved of glucose . This first nucleotide affects the ability of a protein called eIF4E , which recruits the machinery needed for translation , to bind to the mRNA . The experiments also show that there is a dramatic increase in the number of distinct mRNAs that are transcribed from the same section of DNA but translated in a different way during glucose starvation . The findings of Tamarkin-Ben-Harush et al . show that transcription and translation are highly coordinated when cells are starved of glucose , allowing the cells to cope with the stress . The next step is to further analyze the data to find out more about how transcription and translation are linked .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "biochemistry", "and", "chemical", "biology" ]	2017	Cap-proximal nucleotides via differential eIF4E binding and alternative promoter usage mediate translational response to energy stress
The cerebellum , a crucial center for motor coordination , is composed of a cortex and several nuclei . The main mode of interaction between these two parts is considered to be formed by the inhibitory control of the nuclei by cortical Purkinje neurons . We now amend this view by showing that inhibitory GABA-glycinergic neurons of the cerebellar nuclei ( CN ) project profusely into the cerebellar cortex , where they make synaptic contacts on a GABAergic subpopulation of cerebellar Golgi cells . These spontaneously firing Golgi cells are inhibited by optogenetic activation of the inhibitory nucleo-cortical fibers both in vitro and in vivo . Our data suggest that the CN may contribute to the functional recruitment of the cerebellar cortex by decreasing Golgi cell inhibition onto granule cells . The cerebellum plays a key role in the fine temporal control of posture and movements as well as in cognitive processes ( Ito , 1993; Leiner et al . , 1993 ) . Current cerebellar theories ( Apps and Garwicz , 2005; Jacobson et al . , 2008; Dean and Porrill , 2011 ) mainly discuss cerebellar computation from the point of view of its cortical circuitry , where both pre-cerebellar mossy fibers ( MFs ) and inferior olive ( IO ) -originating climbing fibers ( CFs ) modulate Purkinje neuron ( PN ) spiking . Sensory-motor signal processing in the main cerebellar output structure , the cerebellar nuclei ( CN ) , has received less attention . Modulation of spike frequency and timing in the CN projection neurons is considered to be mostly determined by the massive inhibitory cortico-nuclear projection of PNs ( Chan-Palay , 1977; De Zeeuw and Berrebi , 1995; Gauck and Jaeger , 2000; Telgkamp and Raman , 2002; Pedroarena and Schwarz , 2003; Telgkamp et al . , 2004; Person and Raman , 2012; Najac and Raman , 2015 ) . However , certain aspects of cerebellar function persist even when the cerebellar cortex is selectively inactivated or damaged ( Thompson and Steinmetz , 2009; Aoki et al . , 2014; Clopath et al . , 2014; Longley and Yeo , 2014 ) . Thus , a better understanding of the information processing in the CN as well as its influence on cerebellar cortical computation is needed . Currently , the cerebellar cortex and the CN are known to interact through two circuits . The best known is the nucleo-olivary ( NO ) circuit ( Apps and Garwicz , 2005; Apps and Hawkes , 2009; Chaumont et al . , 2013 ) , where the small GABAergic CN cells , subject to PN inhibition ( Najac and Raman , 2015 ) , project to the contralateral IO ( Fredette and Mugnaini , 1991 ) . This pathway regulates olivary activity ( Chen et al . , 2010; Bazzigaluppi et al . , 2012; Chaumont et al . , 2013; Lefler et al . , 2014 ) and thereby complex spike activity in the PNs and cerebellar cortical plasticity ( Hansel and Linden , 2000; Coesmans et al . , 2004; Bengtsson and Hesslow , 2006; Medina and Lisberger , 2008 ) . A less-known nucleo-cortical circuit is formed by the glutamatergic neurons of the CN which , in addition to projecting to various premotor and associative regions of the brain ( Tsukahara and Bando , 1970; Asanuma et al . , 1980; Angaut et al . , 1985; Sultan et al . , 2012; Ruigrok and Teune , 2014 ) , send axonal collaterals to the cerebellar granule cell layer ( GrCL; Houck and Person , 2015 ) . These collateral fibers form MF-like terminals contacting granule cell ( GrC ) and Golgi cell dendrites ( see also Tolbert et al . , 1976 , 1977 , 1978; Hámori et al . , 1980; Payne , 1983 ) . The functional significance of this excitatory nucleo-cortical ( eNC ) pathway , loosely following the modular arrangement of the cerebellum ( Dietrichs and Walberg , 1979; Gould , 1979; Haines and Pearson , 1979; Tolbert and Bantli , 1979; Buisseret-Delmas , 1988; Provini et al . , 1998; Ruigrok , 2010 ; reviewed by Houck and Person , 2013 ) , is likely related to efference copying of motor commands to the cerebellar cortex ( Sommer and Wurtz , 2008; Houck and Person , 2015 ) . In addition to the pathways linking the CN with the cerebellar cortex mentioned above , evidence has occasionally emerged for an inhibitory nucleo-cortical ( iNC ) pathway . GABAergic neurons have been shown to be labeled in the CN by retrograde tracing from the cerebellar cortex ( Batini et al . , 1989 ) , and nucleo-cortical terminals with non-glutamatergic ultrastructural features have been found to contact putative Golgi cell dendrites ( Tolbert et al . , 1980 ) . More recently , it was demonstrated that GlyT2-expressing CN neurons extend axons toward the cerebellar cortex ( Uusisaari and Knöpfel , 2010 ) , suggesting that the iNC pathway might be identifiable by its glycinergic phenotype . While the iNC projection is likely to have significant impact on cerebellar computation , its postsynaptic targets and its functional organization remain unknown . To establish the existence and prevalence of an inhibitory connection between the CN and the cerebellar cortex , we employed specific viral targeting of GABAergic and glycinergic neurons in the CN of GAD-cre and GlyT2-cre transgenic mouse lines , respectively ( Taniguchi et al . , 2011; Husson et al . , 2014 ) . We found that the GABA-glycinergic CN neurons form an extensive plexus of iNC axons , which contact Golgi cells in the cerebellar granular and molecular layers . Specific optogenetic activation of the iNC axons inhibited spikes in a distinct subpopulation of Golgi cells , characterized by their spontaneous firing , high neurogranin immunoreactivity , and negligible GlyT2 expression . As the functional significance of the iNC pathway is likely to be amplified by the high divergence of Golgi cells , which target thousands of GrCs ( Hámori and Somogyi , 1983; Jakab and Hamori , 1988; Andersen et al . , 1992; Korbo et al . , 1993 ) , as well as the remarkable mediolateral extent of the iNC axons , the CN might play a key role in the regulation of the information flow through the GrCL . In order to identify the iNC projection neurons , we specifically labeled the GABAergic and glycinergic CN neurons by injecting floxed adeno-associated virus ( AAV ) in the CN of GAD-cre and GlyT2-cre transgenic mouse lines , respectively . As shown in Figure 1A1 , B1 , these procedures resulted in the expression of the fluorophores ( mCherry in GAD-cre and YFP in GlyT2-cre mice ) in a subset of CN neurons . In the GAD-cre mice , the labeled neurons displayed a wide range of sizes and shapes , including both globular and multipolar morphologies ( Figure 1A2 , arrow and arrowhead , respectively ) . In contrast , in GlyT2-cre mice , the labeled neurons were predominantly large ( Figure 1B2 , arrowhead ) and multipolar , often with a thick principal dendrite ( Figure 1B2 , arrows ) . To examine the morphological difference between CN cells labeled in GAD-cre and GlyT2-cre mice , we measured and compared their soma sizes . The size distribution in GAD-cre CN was best fitted with a two-component Gaussian model ( Figure 1D , red bars and line; Gaussian peaks at 11 . 9 µm and 16 . 2 µm; R-square 0 . 97 , n = 650 cells in 6 animals ) , suggesting it is composed of two separate populations . In contrast , the optimal fit to the size distribution of GlyT2-cre neurons was obtained with a single-component Gaussian model ( Figure 1D , yellow bars and line; peak at 16 . 6 µm , R-square = 0 . 83 , n = 118 cells in 4 animals; KS-test GAD vs GlyT2 , p < 0 . 0001 ) . The peak of this GlyT2-fit matched well with the right-most peak in the GAD-cre distribution ( GAD-cre , second peak confidence interval , 13 . 6–18 . 8 µm; GlyT2-cre confidence interval , 16 . 0–17 . 1 µm ) . 10 . 7554/eLife . 06262 . 003Figure 1 . Targeted viral labeling of the GABAergic and glycinergic neurons in the CN reveals dense and wide-spread network of nucleo-cortical axons in the cerebellar cortex . ( A1–B1 ) Confocal images of coronal cerebellar sections in mice , where floxed virus was injected into the cerebellar nuclei of GAD-cre and GlyT2-cre mice . ( A1 ) Flocculus , coronal view , 40× confocal scan tiles . ( B1 ) Posterior vermis , horizontal view . ( A2–B2 ) Higher magnification confocal images of the cerebellar nuclei show transfected GABAergic and glycinergic neurons , respectively . Note the lower density of labeled neurons in GlyT2-cre brain . In the GAD-cre mice , both small , globular and larger , multipolar neurons ( arrows and arrowheads in A2 , respectively ) were seen . In the GlyT2-cre mice , only large , multipolar neurons were observed , characteristic to the glycinergic CN neurons ( arrowheads in B2 ) . In the GAD-cre mice , the transfected neurons included the GABAergic NO neurons , as evidenced by the fluorescent axons in the contralateral IO ( A3 ) . ( C ) Injection of hSyn-GFP-virus into the IO retrogradely labeled small , round NO neurons in the contralateral CN . ( D ) Comparison of soma sizes among the three labeled populations . The histograms are fitted with single ( GlyT2+ and NO ) or double ( GAD+ ) Gaussians ( thick lines ) , showing that the GlyT2+ neurons distribution matches the second peak in GAD + fit and that the GlyT2+ and NO neurons form distinct populations that contribute to the GAD + population . ( E ) Confocal composite images showing virally transfected CN neurons in GAD-cre ( E1 ) and GlyT2-cre ( E2 ) mice , and their axons ( arrows ) extending across the WM surrounding the CN into the GrCL of the cerebellar cortex . ( F1 ) Confocal composite image of a caudal coronal section of GlyT2-cre cerebellum where the lateral CN was virally transfected ( location of the CN is drawn schematically on top of the image ) . The wide distribution of the NC axons in medio-lateral direction , including parts of the contralateral cerebellum , is shown in yellow color . ( F2 ) Confocal composite image of a horizontal section at the level of the CN in GlyT2-cre cerebellum with transfection of the glycinergic neurons in the medial CN . The axons of the labeled neurons can be seen extending through wide areas of the vermal cortex . The inset shows a single inhibitory nucleo-cortical ( iNC ) axon forming axonal swellings across several hundreds of μm in the GrCL . Abbreviations: GrCL , granule cell layer; CN , cerebellar nuclei; mCN , medial cerebellar nuclei; IO , inferior olive; NO , nucleo-olivary; WM , white matter . Scale bars: A1 , B1: 50 μm; A2 , B2 , C , 10 μm; E , 100 μm , F1 , 2: 400 μm; F2 , inset: 100 μm . See also Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06262 . 00310 . 7554/eLife . 06262 . 004Figure 1—figure supplement 1 . iNC neurons receive functional GABAergic Purkinje Neuron inputs . ( A ) In the GlyT2-eGFP mouse , GFP-positive neurons ( in green ) are contacted by VIAAT-positive varicosities ( in red ) . Points of contacts are indicated by arrows . ( Z-projection thickness: 4 . 6 µm , scale bar: 10 µm ) . ( B ) Optogenetical stimulation of Purkinje neurons ( PNs ) axonal varicosities was performed by whole-field LED ( 470 nm ) illumination of the cerebellar nuclei of L7-ChR2-YFP mice bred with GlyT2-eGFP mice . In these double-positive mice , PNs expressed specifically the channelrhodopsin ( ChR2 ) ( Chaumont et al . , 2013 ) while iNC neurons can be easily targeted for patch-clamp recording using epifluorescence for GlyT2-eGFP . ( C ) One millisecond illumination ( indicated by blue box ) elicited large inhibitory responses in the iNC neurons ( mean amplitude of 416 . 5 ± 332 . 1 pA , n = 10 cells ) , in presence of glutamate receptors blockers ( APV 50 µM , NBQX 10 µM ) , which were blocked by 1 µM gabazine ( 98 . 1 ± 1 . 4 % block , n = 9 cells ) . The kinetics of the PN-originating inhibitory post-synaptic currents ( IPSCs ) in glycinergic CN cells ( decay time constant 3 . 28 ± 0 . 72 ms , n = 10 cells ) were similar to those of IPSCs at Purkinje cell synapses on glutamatergic projection neurons ( Telgkamp and Raman , 2002; Person and Raman , 2012; Husson et al , 2014; Kawaguchi and Sakaba , 2015 ) but much faster than at Purkinje cell synapses on NO cells ( Najac and Raman , 2015 ) , suggesting differential control of the GABAergic NO cells and the larger , GABA-glycinergic CN cells by PNs . DOI: http://dx . doi . org/10 . 7554/eLife . 06262 . 004 The difference between the GAD-cre and GlyT2-cre populations , corresponding to the left-most peak in the GAD-cre distribution ( Figure 1D ) , likely corresponds to the NO cells that are also transfected in the GAD-cre model , as evidenced by the presence of fluorescent axons in the IO ( Figure 1A3; see Lefler et al . , 2014 ) . To confirm this , we retrogradely labeled the NO cells via viral injections in the IO ( Figure 1C ) . The size distribution of the NO neurons ( mean: 12 . 8 ± 2 . 4 µm; n = 193 cells in 4 animals; see also Najac and Raman , 2015 ) was significantly different from the GlyT2 cells ( NO vs GlyT2 KS-test , p < 0 . 0001 , Figure 1D ) . Furthermore , the NO size distribution was well fitted with a single Gaussian with a peak closely resembling the left-most peak of the GAD-cre distribution ( Figure 1D , green bars and line; peak at 12 . 3 µm , confidence interval 12 . 0–12 . 7 µm; R-square 0 . 93; n = 193 cells in 4 animals ) . Thus , we conclude that the mixed GABA-glycinergic neurons form a separate population from the purely GABAergic NO neurons that are not transfected in adult GlyT2-cre animals ( Husson et al . , 2014 ) . These glycinergic neurons , like all other CN neurons , receive functional inputs from PN axons ( Figure 1—figure supplement 1 ) , as previously suggested by immunohistochemical and optogenetic studies ( De Zeeuw and Berrebi , 1995; Teune et al . , 1998 ) . In contrast to the NO axons , which leave the CN towards the brainstem , we found that axons of the large multipolar GAD and GlyT2-positive neurons projected across the white matter surrounding the CN and into the cerebellar cortex ( as shown in Figure 1E1 , 2 for the GAD-cre and GlyT2-cre cerebella , respectively ) . In the vermis , the projections regularly crossed the midline and extended into the contralateral cortex , but otherwise the projection was predominantly ipsilateral . The divergence of nucleo-cortical axons in the cortex varied depending on the extent and localization of viral transfection , coarsely following the known cerebellar modules ( Apps and Garwicz , 2005; Pijpers et al . , 2005; Apps and Hawkes , 2009 ) . Lateral CN injections labeled axons in the lateral and intermediate hemispheres and the flocculi ( Figure 1F1 ) , whereas medial CN injections yielded labeled axons predominantly in the vermal cerebellum ( Figure 1F2 ) . Surprisingly , individual nucleo-cortical axons could be seen to travel long distances in the medio-lateral direction ( up to several millimeters; see inset in Figure 1F2 ) forming boutons within the GrCL . The nucleo-cortical axons formed dense meshes in the GrCL ( Figure 2A1 , B1 ) . As seen in high-magnification images ( Figure 2A2 , B2 ) , the axons formed large swellings that were also seen in the molecular layer ( ML; Figure 2A3 , B3 ) . The axons in the two cre lines were remarkably similar in their appearance , even though the swellings labeled with mCherry in the GAD-cre line appeared nearly identical to the varicosities labeled by YFP in the GlyT2-cre line ( Figure 2A2 , B2 , C1; cross-sectional areas in GAD-cre , red bars , 2 . 1 ± 0 . 9 µm2 , n = 400 varicosities; in GlyT2-cre , yellow bars , 1 . 87 ± 0 . 9 µm2 , n = 415 varicosities; KS-test , p = 0 . 013 ) . Also , no large differences were evident among boutons found in the GrCL or ML ( Figure 2C2; KS-test , p = 0 . 023 ) . These anatomical similarities imply that the axons labeled in the two transfection models represent the projections of a specific population of CN neurons with a mixed GABA-glycinergic phenotype ( Husson et al . , 2014 ) . Indeed , immunostaining revealed that virtually all the nucleo-cortical fibers in the GAD-cre transfected mice were immunoreactive for GlyT2 ( Figure 2D1–2; 94 . 6 ± 6 . 2% , n = 2 animals , n = 9 stacks , n = 422 varicosities; Figure 2D ) , while those in GlyT2-cre cerebella were reactive for GAD65-67 ( 93 . 9 ± 5 . 0% , n = 3 animals , n = 7 stacks , n = 565 varicosities; Figure 2E1–2 ) . These results unequivocally demonstrate the dual neurotransmitter phenotype of the nucleo-cortical projection . Notably , neither rosette-like terminals nor evidence of contacts within cerebellar glomeruli was found . This indicates that they differ both in shape and location from the excitatory MFs and the glutamatergic nucleo-cortical fibers described earlier in the literature , both forming rosette-like terminals within the glomeruli ( Tolbert et al . , 1978; Hámori et al . , 1980; Batini et al . , 1992; Houck and Person , 2015 ) . 10 . 7554/eLife . 06262 . 005Figure 2 . iNC axons are found in cerebellar granule cell and molecular layers and contain GAD65-67 and GlyT2 . ( A–B ) Confocal composite images of sections through the flocculus in GAD-cre ( A1 ) and posterior vermis in GlyT2-cre ( B1 ) mice , showing dense iNC axons in the GrCl as well as sparse axons in the ML ( arrows ) . 40× composite tiles . Large axonal swellings from both GABAergic ( A2 ) and glycinergic ( B2 ) axons are found in the GrCL . Both GABAergic ( A3 ) and glycinergic ( B3 ) iNC axons occasionally rise into the lower ML . ( C ) Comparison of iNC axonal bouton sizes between the GABAergic and glycinergic axons ( C1 ) and between the boutons in the GrCL and ML ( C2 ) shows nearly identical distributions . ( D ) Merged confocal image ( Z-projection thickness: 12 . 2 μm ) showing iNC axons in GAD-cre mice injected with AAV-flox-EYFP ( green ) are co-stained for GlyT2 ( red ) ( D1 ) . Higher magnification of axonal swellings ( arrows ) co-stained for EYFP and GlyT2 ( D2 ) . ( E ) iNC boutons ( green ) transfected with AAV-flox-EYFP in GlyT2-cre mice are stained for GAD65-67 ( red , E1 , Z-projection thickness: 8 . 2 µm ) . ( E2 ) Higher magnification of iNC axon ( arrows ) co-stained for EYFP and GAD65-77 ( Z-projection thickness: 2 . 4 µm ) . Abbreviations: GrCL , granule cell layer; ML , molecular layer; PNL , Purkinje neuron layer; WM , white matter; n . s . , non-significant . Scale bars: A1 and B1: 100 μm; A2 and B2: 5 μm; A3 and B3: 50 μm . D1: 20 μm . D2–4: 5 μm . E1a-e: 10 μm; E2a-e: 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06262 . 005 Having demonstrated the existence of a GABA-glycinergic projection from the CN to the cerebellar cortex , generated by a distinct cell type of the CN , we proceeded to identify the targets of this iNC pathway . Golgi cells , which are the only ubiquitous cerebellar neurons that express glycine receptors in the cerebellar cortex ( Dieudonné , 1995 ) , as well as the only neurons with dendrites both in the granular and molecular layers , constitute the most likely targets for iNC axons . To investigate this possibility , we introduced a non-specific GFP-expressing virus to the cerebellar cortex of GAD-cre mice transfected as above in the CN , to be able to visualize neurons in the GrCL . This procedure labeled Golgi cells and indeed we found axonal swellings of iNC fibers apposed along the proximal dendrites and cell bodies of Golgi cells ( Figure 3A , arrows ) . 10 . 7554/eLife . 06262 . 006Figure 3 . Optogenetic stimulation of iNC axons in cerebellar slices inhibits Golgi cells’ spiking . ( A ) Confocal image ( left ) and reconstruction ( right ) of GrCL in GAD-cre mouse injected with AAV2-flox-ChR2-mCherry to the CN and AAV-GFP to the cerebellar cortex . iNC fibers ( red ) branch in the GrCL and form axonal swellings ( arrows ) on GoCs ( green ) . ( B ) Left: schematic drawing of the in vitro experimental arrangement . GoCs were recorded in GlyT2-cre or GAD-cre animals where iNC axons express both ChR2 and a fluorescent marker ( YFP in GlyT2-cre , mCherry in GAD-cre ) . Right: GoC patched and filled with Neurobiotin ( green ) in GlyT2-Cre mice surrounded by transfected axons ( red ) ( Z-projection thickness: 36 . 4 μm; Sagittal view ) . ( C ) Optogenetic stimulation of the iNC fibers . ( C1 ) An example of averaged IPSCs ( n = 30 ) recorded in Golgi cell induced by 5-ms illumination ( indicated by blue line ) , blocked by successive bath application of 300 nM strychnine ( str; orange ) , and 2 µM gabazine ( gbz , blue ) . ( C2 ) Summary plot of the percentage of inhibitory current blocked by strychnine and gabazine ( n = 9; p = 0 . 0039 ) . ( D ) Example voltage traces from a recorded Golgi cell with a 50-ms single light pulse in GAD-Cre mice . The averaged IPSP response ( ± STD ) of all 6 traces is magnified in the inset . ( E ) iNC activation delays spike generation in GoCs . ( E1 ) The traces recorded without ( top; ctrl; black ) and with ( bottom; stim; blue ) light stimulation are aligned either on the first spike in the sweep ( top ) or on the spike preceding the stimulus ( bottom; red , dashed line ) to emphasize the increased ISI in response to iNC stimulation . Average inhibition delay ( ± STD ) for the example cell is marked above traces in Box-and-whiskers symbols . ( E2 ) Comparison of the average ISI without and with light pulse ( ± STD ) normalized to the average ISI . ( p = 0 . 0001; n = 15 ) . ( F ) Example voltage traces from a recorded GoC during train of light pulses stimulation of iNC axons showing no spikes occurring during the illumination ( upper panel ) . Peri-stimulus time histograms ( PSTH ) of the GoCs ( n = 16 , 100-ms bin ) shows a decrease in the number of spikes after train pulse stimulation . Baseline average marked in dashed red line and STD values in red area ( all cells normalized to baseline frequency; lower panel ) . Scale bars: A: 20 µm , B: 50 µm . ‘Asterisks’ indicate statistical significance . Abbreviations: WM , white matter; GrCL , granule cell layer; GoC , Golgi cell; ML , molecular layer; ISI , inter-spike interval . See also Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06262 . 00610 . 7554/eLife . 06262 . 007Figure 3—figure supplement 1 . iNC activation modulate spike times in a fraction of non-responsive s-Golgi cells . ( A ) An example Golgi cell current-clamp recording showing spike-timing response to train of light pulses ( upper panel , 6 traces ) . Spike-timing analysis shows a significant decrease in the voltage variability between the different recorded traces ( bottom panel; red dashed line indicate z-score = 3 , p = 0 . 05 ) . ( B ) Population spike-timing analysis showing a decrease in the voltage variability in GoCs after optogenetic stimulation of iNC fibers ( n = 7; p = 0 . 05 ) . Black dots above traces mark spike times . DOI: http://dx . doi . org/10 . 7554/eLife . 06262 . 007 To confirm the presence of functional inhibitory synaptic connections between the CN and the cerebellar cortex , we selectively activated channelrhodopsin 2 ( ChR2 ) expressed in the iNC axons in acute slices ( as shown schematically in Figure 3B , left panel ) . First , we performed voltage-clamp whole-cell recordings from Golgi cells surrounded by transfected iNC fibers in GlyT2-cre mice ( Figure 3B , right panel ) . With the use of small collimated beams of light ( see ‘Materials and methods’ ) , we stimulated locations near Golgi cell dendrites with a single , short ( 5 ms ) pulses . Inhibitory post-synaptic currents ( IPSCs ) were evoked in 9 out of 38 recorded Golgi cells ( 23 . 7%; Figure 3C1 ) with a mean amplitude of 40 ± 28 pA . Given the ionic composition of our experimental solutions , the estimated reversal potential of −74 mV with the permeabilities of bicarbonate and chloride taken into account and the holding potential of −50 mV , the chord synaptic conductance was 1 . 7 ± 1 . 2 nS and the slope conductance was 1 . 9 ± 1 . 3 nS according to the Goldman–Hodgkin–Katz ( GHK ) equation . We calculated that the equivalent peak conductance of the iNC synapse measured in symmetrical chloride conditions would have been of 8 . 5 ± 5 . 9 nS . The light-evoked IPSCs had a 10–90% rise time of 2 . 5 ± 1 . 3 ms and a bi-exponential decay ( τ1 = 8 . 2 ± 1 . 9 ms , 52 . 3 ± 18 . 3%; τ2 = 34 . 8 ± 9 . 2 ms , n = 9 cells ) . Application of strychnine at a concentration selective for glycine receptors ( 300 nM ) decreased the amplitude of the IPSC by 24 ± 25% ( p = 0 . 039 , n = 9 cells ) without affecting the time course of the IPSCs ( rise time: 2 . 4 ± 0 . 9 ms; decay: τ1 = 9 . 2 ± 3 . 0 ms , 52 . 5 ± 23 . 5% , τ2 = 40 . 6 ± 21 . 2 ms; p = 0 . 91 , p = 0 . 65 and p = 1 . 00 respectively , n = 9 cells ) . These results confirm the presence of a glycinergic component at the iNC-Golgi cell synapses albeit with large variability in its magnitude ( range: 0–63%; Figure 3C2 , left ) . Subsequent application of a GABAA-receptor antagonist ( gabazine , 2 µM ) , almost completely blocked the response , decreasing the amplitude of the IPSC by 96 . 5 ± 2 . 9% ( p = 0 . 0078 , n = 8 cells; Figure 3C1 , 2 ) . These results confirm the mixed GABAergic-glycinergic nature of the iNC axons . To characterize the functional effect of the iNC-originating inhibitory currents on Golgi cells’ firing , we recorded Golgi cells in the current-clamp mode in acute slices obtained from GAD-cre mice . Whole-field light stimulation of the iNC axons had a significant effect on the spiking in 24 out of 86 recorded Golgi cells ( 27%; Figure 3D–G ) . A single 5-ms light pulse elicited clear inhibitory responses in most of cases , involving a hyperpolarizing post-synaptic potential ( n = 12 cells; IPSP amplitude 2 . 1 ± 1 . 5 mV; Figure 3D ) and/or prolongation of the inter-spike interval ( ISI ) during which the stimulation occurred ( n = 15 cells; 60 ± 30% increase; on average , 223 ± 123 ms ISI increased to 355 ± 286 ms; cells; p = 0 . 0001 , paired t-test; Figure 3E1 , E2 ) . This inhibition of spiking was more pronounced when iNC fibers were activated with a train of 4–5 light pulses at 50 Hz ( pulse duration 10 ms ) , eliciting a longer spike delay ( 71 ± 44% increase; n = 16 cells , p < 0 . 05; Figure 3F ) . In some of the recorded Golgi cells , trains of light pulses elicited a time locking of intrinsic spikes ( n = 7 , Figure 3—figure supplement 1 ) without clear inhibitory effect , suggesting a network effect mediated through the gap junctions among Golgi cells ( Dugué et al . , 2009; Vervaeke et al . , 2010 ) . Taken together , these results demonstrate that the iNC pathway inhibits spiking in Golgi cells . The rate of success in finding responsive Golgi cells was rather low ( 24% and 27% of all recorded Golgi cells in GlyT2-cre and GAD-cre mice , respectively ) suggesting that iNC axons might preferentially or exclusively inhibit a certain subpopulation of Golgi cells . In the following , we will present both electrophysiological and immunochemical evidence supporting this possibility . While analyzing our current-clamp recordings in GAD-cre mice , we noted considerable variability in Golgi cells' properties , with their spontaneous spiking showing the most striking difference ( Figure 4A ) . While 64 out of 85 recorded Golgi cells fired spontaneously at low rates ( mean frequency: 9 . 0 ± 6 . 5 Hz ) , the other 21 Golgi cells were quiescent and had a resting membrane potential negative to the spiking threshold ( resting potential −55 ± 2 . 4 mV , spike threshold −43 ± 1 mV , n = 13 cells ) . When evaluating iNC effects in Golgi cells , it became obvious that only the spontaneously active Golgi cells ( ‘s-Golgi cells’ ) were responsive to iNC stimulation . Specifically , 24 out of 64 ( 37 . 5% ) s-Golgi cells were inhibited by iNC activation ( Figure 4A–B; blue ) , whereas none of the 21 not-spontaneously spiking Golgi cells ( ‘ns-Golgi cells’; green ) were affected by the stimulation . Repeating these stimulations during depolarizing current injections that drove the ns-Golgi cells to continuous spiking also failed to reveal iNC effect ( Figure 4A , top right ) . These findings suggested that iNC fibers inhibit preferentially s-Golgi cells . It should be noted that our virus injections were unlikely to result in transfection of the entire iNC population , hence the observed fraction of inhibited s-Golgi cells is bound to be underestimated . 10 . 7554/eLife . 06262 . 008Figure 4 . Golgi cell subtypes differ in their sensitivity to iNC input and have different intrinsic properties . ( A ) Left: a spontaneously active GoC ( s-GoC ) is inhibited by optogenetic iNC activation ( light-blue bars above traces ) . Six superimposed traces with no holding current . ( A ) Right: a not-spontaneously spiking GoC ( ns-GoCs ) shows no response to iNC activation . Bottom trace: without depolarizing current injection; top trace: with +8 pA current injection to evoke spiking . Black dots above traces in both panels mark spike times . ( B ) Percentage and numbers of s-GoCs ( blue ) and ns-GoCs ( green ) that are inhibited by iNC axons ( left ) and those that are not ( right ) . ( C ) Example traces of three GoCs’ responses to positive current steps . Cell 1 , blue: s-GoC ( Cm = 173 . 5 pF ) ; Cell 2 , dark green: a large ns-GoC ( Cm = 125 . 3 pF ) ; Cell 3 , light green: small ns-GoC ( Cm = 73 . 0 pF ) . ( D ) AP waveforms differ between s-GoCs and ns-GoCs . Left: superimposed , grand average AP shapes ( ± STD ) obtained from s-GoCs ( n = 31 cells , blue ) and ns-GoCs ( n = 13 cells , green ) during steady-state firing . Right: APs are peak-normalized ( ± SEM ) . s-GoCs show faster spike repolarization as well as faster after-hyperpolarization ( arrowhead ) . ( E ) Comparison of AP parameters shows that s-GoCs ( n = 31 cells ) spikes are faster than those in ns-GoCs ( n = 13 cells ) . ( F ) Current-to-firing frequency ( IF ) relationship of s-GoCs ( blue , n = 23 ) and ns-GoCs ( green , n = 11 ) is not significantly different . The solid and dashed lines show fitted single polynomials and the confidence intervals , respectively . The current injection values are normalized to the estimated Cm of the cells . ( G ) Cm and input resistance of s-GoCs ( blue; n = 23 ) and ns-GoCs ( green; n = 11 ) . ( H ) Comparison of instantaneous firing frequency accommodation during a depolarizing step . Left: box-plot chart showing the development of frequency accommodation in s-GoCs ( blue ) and ns-GoCs ( green ) . For visual clarity , the s-GoC bars are slightly shifted to the right in respect to the ns-GoC . Right: box plots of the steady-state accommodation among s-GoCs ( blue ) and ns-GoCs ( green ) show that s-GoCs have a smaller range of accommodation than ns-GoCs ( t-test , p = 0 . 008 ) . ‘Asterisks’ denote statistical significance . Abbreviations: s-GoC , spontaneously spiking Golgi cell; ns-GoC , not-spontaneously spiking Golgi cell , AP , action potential; AHP , after-hyperpolarization; acc val , accommodation value . DOI: http://dx . doi . org/10 . 7554/eLife . 06262 . 008 The s-Golgi cells differ from ns-Golgi cells also in action potential ( AP ) shape ( Figure 4D ) recorded during steady-state firing . The observation that the AP waveform ( composed of AP and after-hyperpolarization [AHP] ) was shorter in s-Golgi cells compared to ns-Golgi cells ( Figure 4E; see Table 1 ) led us to seek for other distinguishing electrophysiological features . Compared to the s-Golgi cells , the ns-Golgi cells showed significantly larger variability in all of the AP shape measurements , although no significant differences were found in their average values ( see Table 1 ) . Also , no differences were found in population averages of the input–output relationship of the two Golgi cell groups , as evidenced by nearly identical current-to-firing frequency ( I-F ) curves ( Figure 4F; Table 1 ) . ns- and s-Golgi cells did not differ as a population in their estimated capacitance ( Cm ) , nor in their input resistance , but analysis of their variance showed clear differences between the groups ( Figure 4G , compare the significance values obtained with Wilcoxon and F-tests in Table 1 ) . The population variability became most visible when comparing the steady-state frequency accommodation ( Figure 4H , left , compare the significance values obtained with Wilcoxon and F-tests in Table 1 ) : the s-Golgi cells accommodated very uniformly to roughly half of the initial firing frequency , while ns-Golgi cells showed either no adaptation ( evidenced by steady-state accommodation values around 90% of control ) or adapted even more than the s-Golgi cells ( to 35% of control; compare the widths of blue and green bars in Figure 4H ) . The large variability of ns-Golgi cells in frequency accommodation , AHP time , AP half width , Cm , input resistance , and AP shape suggest that the ns-Golgi cells form a heterogeneous group of cells consisting of several functionally distinct subpopulations . 10 . 7554/eLife . 06262 . 009Table 1 . Summary of s-Golgi and ns-Golgi cells spiking parametersDOI: http://dx . doi . org/10 . 7554/eLife . 06262 . 009s-Golgins-GolgiN ( s ) N ( ns ) p-value ( Wilcoxon ) p-value ( F-test ) AP half-width ( ms ) 0 . 8 ± 0 . 21 . 2 ± 0 . 433110 . 010 . 001AP threshold ( mV ) 38 . 9 ± 6 . 04−36 . 4 ± 9 . 433110 . 80 . 1AP amplitude ( mV ) 51 . 2 ± 943 . 5 ± 10 . 733110 . 060 . 4AP peak voltage ( mV ) 22 . 4 ± 8 . 716 . 8 ± 13 . 0333110 . 10 . 17AHP min voltage ( mV ) −51 . 9 ± 4 . 5−51 . 5 ± 8 . 833110 . 80 . 01AHP time ( ms ) 2 . 1 ± 0 . 84 . 0 ± 0 . 533110 . 0090 . 03AHP amplitude ( mV ) 24 . 3 ± 4 . 223 . 1 ± 8 . 433110 . 90 . 003I-F slope ( r2 ) 0 . 50 . 42311–0 . 3 ( cov-analysis ) Freq . Acc . ( % ) 53 ± 8%59 ± 29%23110 . 31 . 1 × 10−6Cm ( pF ) 127 . 5 ± 48 . 3118 . 9 ± 78 . 323110 . 0520 . 0018Rm ( MΩ ) 185 . 5 ± 43 . 2161 ± 75 . 523110 . 0548 . 3 × 10−5 The results described above suggest that iNC fibers specifically target a subpopulation of spontaneously active Golgi cells with uniform electrophysiological properties . It is well established that Golgi cells are neurochemically heterogeneous ( Ottersen et al . , 1988; Simat et al . , 2007; Pietrajtis and Dieudonné , 2013 ) . While the majority of Golgi cells express the glycine transporter GlyT2 , with most of them being of mixed GABA-glycinergic phenotype , about 15–20 % of Golgi cells are purely GABAergic ( Ottersen et al . , 1988; Simat et al . , 2007 ) . GlyT2-expressing Golgi cells have previously been reported to be intrinsically silent ( Dugué et al . , 2009 ) . We , thus , hypothesized that the iNC-inhibited Golgi cells , all of which fire spontaneously , may correspond to the purely GABAergic Golgi cells . As all of the Golgi cells that express the calcium-binding protein , neurogranin are also GABAergic ( Simat et al . , 2007 ) , we used neurogranin in conjunction with GlyT2-eGFP expression to differentiate pure GABAergic Golgi cells from the mixed GABA-glycinergic and pure glycinergic Golgi cell populations ( Simat et al . , 2007; Pietrajtis and Dieudonné , 2013 ) . We designed a strategy to identify the subtypes of Golgi cells targeted by iNC axons . GlyT2-eGFP transgenic mice ( Zeilhofer et al . , 2005 ) , in which both mixed GABA-glycinergic and pure glycinergic Golgi cells are labeled with eGFP , were mated with GlyT2-cre animals . The CN of the offspring carrying both transgenes were injected with a floxed AAV expressing the red fluorescent protein tdTomato . Cerebellar cortical sections from these mice were then stained for neurogranin to differentiate between the Golgi cells subtypes . iNC axons were easily identified by their co-expression of eGFP and tdTomato and were found to preferentially contact cell bodies and dendrites of Golgi cells that were intensely stained for neurogranin ( Figure 5A–B ) . This selective targeting of neurogranin-positive cells by iNC fibers extended to the ML ( indicated by arrowheads in Figure 5C ) , as iNC fibers were seen to climb along the apical dendritic shafts of neurogranin-positive Golgi cells to the ML . Similarly , in GAD-cre animals , iNC fibers transfected with YFP and co-stained for GAD65-67 were found to impinge on the dendrites and cell bodies of Golgi cells strongly expressing neurogranin ( Figure 5D , arrowheads ) . 10 . 7554/eLife . 06262 . 010Figure 5 . iNC fibers contact preferentially a neurochemically distinct subtype of Golgi cells expressing neurogranin . ( A–C ) iNC fibers were transfected with AAV-flox-tdTomato in GlyT2-Cre X GlyT2-eGFP mice . iNC fibers were identified in the cerebellar cortex by their co-labeling for both GFP ( green ) and tdTomato ( red ) . ( A–B ) In the GrCL , iNC fibers contact somata and proximal dendrites of neurogranin-expressing ( blue ) GoCs , either devoid of GFP staining or exhibiting a faint GFP staining at their somata ( indicated by ‘asterisks’ ) . ( C ) In the ML , GFP-positive / tdTomato-positive iNC fibers ( arrowheads ) were also seen apposed to GoC apical dendrites stained for neurogranin ( blue ) and virtually devoid of GFP staining ( green ) . ( D ) iNC fibers , transfected with AAV-flox-YFP virus ( green ) in GAD-cre mice , contacted neurogranin-positive ( blue ) GoCs . iNC varicosities ( arrowheads ) are co-stained for GAD65-67 ( red ) . ( E ) Plot of the mean GlyT2-eGFP intensities over mean neurogranin intensities allows statistical discrimination ( k-means 2D ) between two GoCs populations . A first population ( blue ) was distinguished from the second population ( green ) by its none-to-low levels of GlyT2-eGFP staining , as seen with the bimodal distribution of mean GFP intensities ( F ) , while the mean neurogranin intensities were less discriminative ( G ) . According to the color-coded number of iNC inputs received by each GoC ( H ) , most of the ‘iNC-contacted’ GoCs were found in the neurogranin-positive/GlyT2-eGFP negative GoC population ( blue ) ( I ) . ( J ) Schematic drawing of the percentages obtained for each GoC subtypes . Z-projection thickness: A: 34 µm; B: 18 . 4 µm; C: 16 . 3 µm; D: 30 µm; D close up: 2 . 4 µm . Scale bar: A: 20 µm , B–D: 10 µm , D close up: 2 µm . Abbreviation: GoC , Golgi cell . DOI: http://dx . doi . org/10 . 7554/eLife . 06262 . 010 In most cases , the innervated Golgi cells were devoid of eGFP staining , suggesting that they are non-glycinergic Golgi cells . However , a few of the iNC-contacted Golgi cells exhibited a low level of eGFP staining in their somata ( examples are indicated by asterisks in Figure 5A , B ) . To distinguish between the eGFP positive and negative Golgi cell subpopulations in a more objective manner , we quantified the normalized mean GlyT2-eGFP and neurogranin staining intensities at the somata of Golgi cells ( n = 317 cells , n = 13 stacks n = 4 animals; see ‘Materials and methods’ ) . The 2D distribution of the two staining intensities was separated into two populations based on k-means statistical clustering ( Figure 5E; green population: n = 238 Golgi cells ( 75% ) ; blue population: n = 79 Golgi cells ( 25% ) ; see ‘Materials and methods’ ) . Golgi cells in the two groups differed mainly in their eGFP staining , as illustrated by the bimodal distribution of the mean eGFP intensities ( Figure 5F ) . The normalized neurogranin intensities were approximately 50% higher in the low-eGFP population ( Wilcoxon test p < 0 . 00001 ) , even though distributions overlapped extensively ( Figure 5G ) . We , thus , consider that the blue population of Figure 5 constituting of Golgi cells expressing neurogranin and none or low levels of GlyT2-eGFP corresponds to the population of Golgi cells releasing principally or exclusively GABA ( Aubrey et al . , 2007 ) . For the sake of brevity , we will refer to these cells as ‘GABAergic Golgi cells’ in the following . We counted the iNC appositions found on the somata and large proximal dendrites of each Golgi cell ( see ‘Materials and methods’; color-coded in Figure 5H ) as a measure of connection strength . Most of the Golgi cells that were contacted by at least one iNC bouton were GABAergic Golgi cells ( 80% , n = 32 out of the 40 ‘iNC-contacted’ cells; Figure 5H , I ) . In contrast to the glycinergic Golgi cells that were only rarely apposed to iNC boutons ( 3% , n = 8 out of 238 all Golgi cells; on average 1 . 88 ± 1 . 12 appositions per cell; max = 4 appositions; 15 appositions found overall; Figure 5I , blue dots ) , 41% of the GABAergic Golgi cells were contacted by on average 7 . 65 ± 4 . 69 appositions ( n = 32 out of 79 Golgi cells , max = 20 appositions; 245 appositions found overall; Figure 5I , blue dots ) . These results demonstrate that iNC fibers contact almost exclusively the GABAergic Golgi cell population , as summarized in the schematic drawing of Figure 5J ( 97% of iNC terminals contact the GABAergic Golgi cell population ) . To the best of our knowledge , this is the first evidence for differential connectivity among subpopulations of Golgi cells . The impact of a neuronal pathway depends on properties of transmission at its synapses as well as the firing pattern of its neurons . In the case of the iNC pathway , repetitive stimulation of the axons evoked stronger inhibition of Golgi cells ( Figure 3G ) . To investigate the physiological relevance of such burst activation of iNC axons , we examined the responses of iNC neurons to optogenetic stimulation . First , using acute slices from GlyT2-cre mice transfected as above , we performed extracellular recordings from iNC neurons , identified by their YFP fluorescence . iNC neurons were silent ( n = 11 cells ) in contrast with the other cell types in the CN ( Uusisaari and Knöpfel , 2010 , 2012 ) . Optogenetic excitation of iNC neurons by short light pulses ( 1 ms ) evoked high-frequency bursts of spikes ( Figure 6A1 ) . Increasing the illumination power resulted in an increased number of spikes and in mean burst frequency , which saturated around 450 Hz for a light power of 1–2 mW/mm² ( Figure 6A2; n = 11 cells ) . 10 . 7554/eLife . 06262 . 011Figure 6 . iNC neurons exhibited a burst firing phenotype and their optogenetic stimulation has inhibitory effects on Golgi cells firing in vivo . ( A1 ) Extracellular recordings of iNC neurons in GlyT2-Cre mice transfected with ChR2 virus during increasing intensity of stimulation ( 1-ms duration pulse; blue bars ) . ( A2 ) iNC neurons exhibit a burst firing phenotype with increase of mean number of spikes per burst ( left ) and mean burst frequency ( right ) when increasing illumination intensity . ( B1 ) Whole-cell current-clamp recording of GAD-Cre iNC neurons transfected with ChR2 during 10- , 50- , or 200-ms long light pulse ( blue bars; light intensity: 1 . 3 mW/mm2 ) has burst firing phenotype . ( B2 ) GAD-Cre transfected iNC neurons show increase of their mean number of spikes per burst ( left ) and mean burst duration ( right ) with increasing illumination duration . ( C1 ) Schematic drawing of the experimental system for in vivo recordings: extracellular recordings of GoCs during 25-ms pulse illumination of the CN in anesthetized GlyT2-Cre mice injected with ChR2 in the CN . ( C2 ) Raster plots of two GoCs recorded at the same time and of one PN recorded in the same area , with their corresponding PSTH . Light pulse start at 0 ms . ( D1 ) All superimposed smoothed PSTHs of responsive GoCs ( 18 out of 86 recorded GoCs ) , normalized to their mean firing rate ( FR ) , with the population average trace ( red ) . Two individual traces are highlighted ( orange and blue ) , illustrating the high variability of the inhibition period parameters . Smoothed PSTHs are obtained by convolving 1-ms time bin PSTHs with a Gaussian kernel with 3 ms standard deviation . ( D2 ) Characterizing parameters of the responses . ( D3 ) Light stimulation of iNC neurons decreased the firing rate . ( D4 ) Comparison of responsive and non-responsive GoCs firing rates . Abbreviations: GoC , Golgi cell . DOI: http://dx . doi . org/10 . 7554/eLife . 06262 . 011 To further characterize intrinsic bursting , we performed whole-cell current-clamp recordings from iNC neurons ( identified by their fluorescence and lack of spontaneous activity ) in slices from transfected GAD-cre mice ( Figure 6B1 , n = 3 cells ) . At saturating illumination intensity ( 1 . 3 mW/mm2 ) , stereotypical high-frequency bursts of spikes were evoked with short light durations ( 10 ms ) , resembling the extracellular recordings in the GlyT2-cre slices . These bursts were riding on a depolarized plateau , which outlasted the illumination period , and were often followed by a prolonged depolarized after-potential and low-frequency firing . Increasing the duration of the light pulse extended the burst duration without affecting the intra-burst frequency ( Figure 6B2 ) . These results indicate that high-frequency bursting of APs could constitute the main firing mode of iNC neurons in response to excitatory synaptic inputs . To investigate the physiological significance of the iNC pathway in an intact cerebellum , we implanted an optical fiber in the CN of virally transfected GlyT2-cre mice to optically activate the iNC neurons , while recording Golgi cell activity ( Figure 6C1 ) . Based on our in vitro calibration ( Figure 6A , B ) , single 25-ms long light pulses are expected to evoke short bursts of firing in the iNC neurons . This illumination protocol suppressed spiking in 18 out of 86 recorded Golgi cells ( 21% , Figure 6C2 , left ) . The rest of the Golgi cells ( 79% , Figure 6C2 , middle ) as well as PNs ( n = 50 cells , Figure 6C2 , right ) did not show any significant modulation of the spiking frequency following illumination . The time course of the inhibition in the responsive Golgi cells was variable ( duration: 23 . 4 ± 11 . 7 ms; onset latency: 14 . 5 ± 7 . 2 ms; peak latency: 25 . 4 ± 14 . 1 ms; n = 18 , Figure 6D2 ) as exemplified with colored traces from individual cells in Figure 6D1 . The variability of the inhibitory effect can be explained by the variability in iNC spike-burst duration that depends on the distance from the optic fiber and thereby stimulation light intensity ( Figure 6A ) . Regardless of this variability , Golgi cells’ firing was robustly suppressed ( frequency decreased to 1 . 58 ± 1 . 46 Hz from a baseline of 10 . 9 ± 3 . 9 Hz , n = 18 cells , Figure 6D3 ) . Interestingly , the average firing rate ( FR ) of responsive Golgi cells was significantly higher than the average FR of non-responsive Golgi cells ( 10 . 5 ± 3 . 5 Hz , n = 18 cells vs 8 . 2 ± 4 . 2 Hz , n = 68 cells , respectively; Wilcoxon test: p = 0 . 036; Figure 6D4 ) . While we cannot make a direct link between the lower FR of non-responsive Golgi cells in vivo and the quiescence of ns-Golgi cells in vitro , these results are supporting the notion that the iNC pathway is targeting a distinct group of Golgi cells . Overall , our results provide the first functional evidence for an iNC pathway suppressing GABAergic Golgi cell spiking . This pathway likely modulates the inhibitory control of GrCs and thereby gating of sensori-motor inputs into the cerebellar cortex . Anatomical demonstrations of nucleo-cortical pathways have appeared in literature already decades ago ( Tolbert et al . , 1976; Gould and Graybiel , 1976; Dietrichs and Walberg , 1979; Hámori et al . , 1980; Buisseret-Delmas , 1988; Batini et al . , 1992; reviewed in Haines and Manto , 2009; Houck and Person , 2013 ) . These classical studies , often ignorant of the afferents’ neurotransmitter type , described a range of nucleo-cortical axonal morphologies including rosette-like and simple terminals ( Hámori et al . , 1980; Tolbert et al . , 1980 ) . It was only later established that both glutamatergic ( Tolbert et al . , 1980; Payne , 1983; Batini et al . , 1992; Houck and Person , 2015 ) and GABAergic ( Hámori and Takács , 1988; Batini et al . , 1989 , 1992; Houck and Person , 2015 ) CN neurons project to the cortex . Here , using targeted viral transfection and labeling , we demonstrate that the iNC axons originate from a population of mixed GABA-glycinergic CN neurons . The iNC axon terminals were simple in their morphology , and rosette-like structures were never observed . Thus , the GABAergic rosette-like terminals found in GrCL glomeruli described in earlier works ( Chan-Palay et al . , 1979; Hámori and Takács , 1988 ) must arise from extracerebellar sources . The morphology and spread of the iNC axons as well as the axonal bouton size was also different from both the Golgi and Lugaro axons ( Dieudonné , 1998; Dumoulin et al . , 2001 ) . Our study discards the suggestion that iNC axons would emerge as collaterals of GABAergic NO neurons ( Figure 1; Tolbert et al . , 1978; Haines , 1988 ) . The neurons transfected in the GlyT2-cre animals do not include NO cells , as evidenced by the lack of labeling in the IO ( Husson et al . , 2014; see also De Zeeuw et al . , 1994 ) and the clear difference in cell body size between GlyT2-cre and NO neurons ( Figure 1B–D ) . While viral transfection protocols used in the GAD-cre mice also transfect NO cells ( Lefler et al . , 2014; Figure 1A3 ) , all the fibers found in the cortex were GlyT2 immunopositive , demonstrating that only those GABAergic CN cells that also express GlyT2 project to the cortex . Also , as the purely glycinergic neurons of the medial CN nucleus projecting to the vestibular nuclei ( Bagnall et al . , 2009 ) are not found in the main targets of viral transfections in the present study ( interpositus and lateral CN ) , they are unlikely to be the source of the iNC axons . The nucleo-cortical axons in both GlyT2-cre and GAD-cre models were very similar in shape and function and co-stained for GAD and GlyT2 , respectively ( Figure 2 ) . The small differences observed are likely to originate from variability in fixation procedures and wavelength dependence of optical resolution . Therefore , iNC fibers undoubtedly represent the axons of a single mixed CN neuron type . The density of nucleo-cortical fibers in the GAD-cre model was somewhat higher than in the GlyT2-cre model ( compare panels 2A1 and 2B1 ) , most likely due to the mosaic expression of cre in only 50% of mixed neurons in the GlyT2-cre mice ( Husson et al . , 2014 ) as well as the stronger expression levels obtained with the AAV9 serotype virus used in the GAD-cre model . The iNC neurons described in the present work show clear morphological ( Figure 1B2 ) and electrophysiological ( Figure 6 ) resemblances to the CN glycinergic neurons described in two recent studies as spontaneously inactive , mixed GABA-glycinergic neurons ( compare the present results with Figure 1 in Uusisaari and Knöpfel , 2010 and Figure 7Ab in Husson et al . , 2014 ) . Thus , we conclude that the iNC neurons , the ‘Gly-I’ neurons , and the mixed GABA-glycinergic neurons are the same cells . While these neurons also have local axon collaterals within in the CN ( Husson et al . , 2014 ) , their projection to the cortex is their most distinguishing feature . Thus , we propose that they should be referred to as ‘iNC’ neurons . Golgi cells have previously been shown to receive inhibitory synapses from both Lugaro and other Golgi cells in the cerebellar cortex . We demonstrate here that single iNC projection axons form numerous terminal swellings on the somata and dendrites of Golgi cells ( Figure 2 ) , somewhat reminiscent of the climbing fiber articulation on Purkinje cells . Specific optogenetic stimulation of the iNC axons evokes IPSCs mediated both by GABAA and glycine receptors ( Figure 3 ) , in line with the immunohistochemical evidence that iNC terminals contain both GABA and glycine ( Figure 2D–E ) . The average synaptic conductance at the iNC synapses ( estimated to be 1 . 9 and 8 . 5 nS in physiological and symmetrical chloride , respectively ) is about six times the conductance reported at unitary Golgi–Golgi synapses ( 0 . 33 nS in physiological chloride; Hull and Regehr , 2011 ) and similar to the conductance at Lugaro to Golgi cell synapses in the juvenile animal ( Dumoulin et al . , 2001 ) . As our spatially restricted light stimulation likely activated a single or only a few iNC axons , the evoked IPSCs likely represent unitary responses through the multiple contacts made by single axons on Golgi cells . Most of the current physiological work on Golgi cells tends to assume a homogeneous neuronal population while simultaneously using various , partly contradictory , identification criteria to target them for experiments ( Schulman and Bloom , 1981; Holtzman et al . , 2006; Xu and Edgley , 2008; Hull and Regehr , 2011; Hull et al . , 2013 ) . However , accumulating evidence indicates that Golgi cells can be divided into different groups based on their neurotransmitter content ( GABA , glycine or both; Ottersen et al . , 1988 ) as well as specific molecular marker expression ( Simat et al . , 2007; Dugué et al . , 2009; Pietrajtis and Dieudonné , 2013 ) . By demonstrating that the Golgi cells contacted by the iNC axons are characterized by none or low level of GlyT2-eGFP staining as well as high-neurogranin expression ( Figure 5; compare with ‘type 4’ Golgi cells in Simat et al . , 2007 ) , we present the first evidence that neurochemical subtypes of Golgi cells may participate in specific microcircuits . Our work further supports the functional specialization of purely GABAergic Golgi cells by showing that they can be distinguished from other Golgi cells based on their electrophysiological properties ( Figure 4 ) . A previous work showed that glycinergic ( GlyT2-eGFP positive ) Golgi cells are not spontaneously active in vitro ( Dugué et al . , 2009 ) . Here , we show that spontaneously spiking Golgi cells ( s-Golgi cells ) , unlike the not-spontaneously spiking Golgi cells ( ns-Golgi cells ) , receive functional synaptic contacts from the iNC fibers ( Figure 3 ) . s-Golgi cells had relatively uniform properties ( Figure 4 ) , confirming their identification as a distinctive functional group . In contrast , ns-Golgi cells varied in all of the examined features ( Figure 4 and Table 1 ) suggesting that this population may be further divided into several functional subgroups . The functional microcircuit of the GrCL , thus , needs to be re-examined in the light of the existence of multiple Golgi cell subtypes . Optogenetic activation of iNC axons was found to modulate Golgi cell discharge in vitro and in vivo ( Figures 3 and 6 ) . The most common effect was a short-latency inhibition of spiking ( Figure 3E–H ) . In vivo inhibition of spiking could last for tens of milliseconds , most likely due to the iNC neurons’ propensity for high-frequency burst firing ( Figure 6 ) and to the slow kinetics of the relatively large synaptic conductances at iNC to Golgi cell synapses ( Figure 3 ) . Although the iNC axons may represent less than 5% of the afferent fibers in the GrCL ( Hámori et al . , 1980; Legendre and Courville , 1986 ) , the ramification of iNC axons , the high divergence of the Golgi cell axon , and the electrical coupling between Golgi cells ( Dugué et al . , 2009 ) will amplify the potency of the iNC effects on the GrCL network . The anatomical and electrophysiological evidence presented here suggests that the iNC pathway is likely to induce a period of disinhibition in the GrCs . This disinhibition could influence the time-window for MF input integration in GrCs , enhancing GrC excitability and thereby facilitate the activation of PNs ( Chadderton et al . , 2004; Kanichay and Silver , 2008; D’Angelo and De Zeeuw , 2009 ) . Confirmation of this functional significance will require experimentation in awake animals , as the MF/parallel fiber ( PF ) pathway is known to be quiescent in anaesthetized animals ( Bengtsson and Jörntell , 2007; Wilms and Häusser , 2015 ) and no modulation of PN spiking is thus expected by the disinhibition of GrCs ( Figure 6C2 ) . No iNC axons were found outside the cerebellar structures in the GlyT2-cre model , and no evidence for iNC contacts on cerebellar GrCs was seen , excluding the possibility of extracerebellar or PF effects on Golgi spiking . However , the glutamatergic projection neurons of the CN collateralising as an ‘eNC’ pathway that contact the neurogranin-positive Golgi cells ( Tolbert et al . , 1976 , 1977 , 1978; Hámori et al . , 1980; Payne , 1983; Houck and Person , 2013 , 2015 ) may be contacted by local iNC axons . The iNC synapses on the eNC neurons constitute only of a tiny fraction of their synaptic inhibition ( Husson et al . , 2014 ) , but we cannot completely exclude that some facet of the iNC-mediated depression of Golgi cell spiking in vivo may reflect a decrease in excitatory synaptic drive from CN-originating MFs . However , it is unlikely that the short bursts of iNC spikes evoked by our stimulation protocol would result in a long pause in spiking of CN projection neurons as they are extremely resistant to inhibition ( Person and Raman , 2012; Chaumont et al . , 2013; Najac and Raman , 2015 ) . A short delay in eNC spikes is unlikely to be the principal source of the observed Golgi cell inhibition , unless the Golgi cell activity would be to a large extent determined by CN . Thus , while further work elucidating the functional role of the eNC projection to the cerebellar GrC layer is sorely needed , we conclude here that the inhibition of Golgi cell spiking observed in vivo ( Figure 6 ) is mainly caused by a direct inhibition by the iNC axon terminals impinging on Golgi cell dendrites and cell bodies . A major feature of the iNC projection to the cortex is its divergence: upon relatively localized viral injection in the CN entire lobules may be innervated . Furthermore , single iNC axons traverse long distances along the medio-lateral plane , similarly to the PFs , making numerous contacts on individual Golgi cell dendrites . Such an arrangement could partly explain the synchronization of Golgi cell activity observed along the axis of the lobules ( Vos et al . , 1999 ) . Furthermore , disinhibition of GrCs by iNC axons could enhance and synchronize MF-PF transmission specifically along medio-lateral stripes , possibly contributing to on-beam synchronization of PNs ( Heck and Thach , 2007 ) . iNC neurons may thus implement refined temporal binding of parasagittal cerebellar modules within a lobule . Intriguingly , while the iNC neurons , like all other CN neurons examined so far , are contacted by PN axons ( Figure 1—figure supplement 1; compare with Bagnall et al . , 2009 ) , their intrinsic quiescence ( Uusisaari and Knöpfel , 2010; Figure 6 ) calls for identification of the sources of synaptic excitation , as they will determine the context within which Golgi cells would be inhibited . Possible candidates include the collaterals of CFs , MFs , and the local axons of CN neurons . While there is no direct evidence either for or against any of these sources , a few earlier works have described inhibition of Golgi cells in response to electrical stimulation of the IO or to sensory stimulation ( Schulman and Bloom , 1981; Xu and Edgley , 2008 ) , suggesting that the IO might be the source of excitatory drive for iNC neurons . Cerebellar GrC layer gating by Golgi cell network has been postulated for a long time to play a critical role in cerebellar function; however , the absence of experimental tools allowing specific control of the Golgi network during behavior has prevented investigation of this hypothesis . The novel , inhibitory pathway from the CN to the Golgi cells revealed in our present work opens a way for targeted manipulation and analysis of the information gating in the cerebellar granule layer . Furthermore , it is now reasonable to assert that through iNC neurons as well as the collateralization of glutamatergic projection neurons to the GrCL ( Houck and Person , 2015 ) , the CN hold a key position to control the activity of the cerebellar cortex . Experiments were performed on adult mice ( p > 30 days; both males and females ) of two mouse lines: the GAD-ires-cre ( Taniguchi et al . , 2011 ) and the GlyT2-cre ( Husson et al . , 2014 ) . These cre-lines , combined with floxed adeno-associated viral ( AAVs; see Table 2 for details ) injections into the CN , allowed specific transfection of either GABAergic or glycinergic CN neurons , respectively . In addition , for immunostaining experiments , heterozygous GlyT2-cre mice were bred with GlyT2-eGFP transgenic mice ( Zeilhofer et al . , 2005 ) and the offspring carrying both GlyT2-cre and GlyT2-GFP genes were transfected as above . Retrograde labeling of the NO neurons ( Figure 1C–D ) was performed in adult wild-type C57BL/6 mice via non-floxed viral injection into the IO . For Figure 1—figure supplement 1 , adult L7-CHR2-YFP mice ( Chaumont et al . , 2013 ) were bred with GlyT2-eGFP mouse and double-positive offspring were used for optogenetic experiments . All animal manipulations were made in accordance with guidelines of the Centre national de la recherche scientifique and the Hebrew University's Animal Care and Use Committee . 10 . 7554/eLife . 06262 . 012Table 2 . Summary of the viral constructs usedDOI: http://dx . doi . org/10 . 7554/eLife . 06262 . 012VirusConstructsMouse line and injection siteUsed in FiguresAAV2/9 . EF1 . dflox . hChR2 ( H134R ) -mCherryAddgene 20 , 297GAD-cre ( CN ) 1A , 1E1 , 2A , 3A , 3E–I , 4A–H , 6BAAV2/9 . EF1a . DIO . eNpHR3 . 0-EYFP . WPRE . hGHAddgene 26 , 966GAD-Cre ( CN ) 2C , 5DAAV2 . 1 . EF1á . DIO . hChR2 ( H134R ) . eYFPAddgene 20 , 298GlyT2-cre ( CN ) 1B , F , 2B , D , 3B–D , 6A , CAAV2 . 1 . CAG-Flex . tdTomatoAllen Institute #864GlyT2-Cre x GlyT2-eGFP ( CN ) 5A–C , E–FAAV2/9 . hSynapsin . EGFP . WPRE . bGHUPenn AV-9-PV1696GAD-cre ( CN + Cortex ) 1C , 3AAAV2/9 . CAG . Flex . EGFP . WPRE . bGHAllen Institute #854GAD-cre ( CCTX ) not shown Mice were deeply anesthetized with a mixture of ketamine and xylazine ( 106 mg/kg and 7 . 5 mg/kg , respectively ) and placed in a stereotaxic frame . Small craniotomies were performed above the CN . The target regions were mostly in the lateral and interpositus nuclei , and the injections were performed unilaterally for immunohistochemical and anatomical protocols , and bilaterally for electrophysiological experiments . A quartz capillary pipette ( 35- to 40-µm tip diameter ) was positioned in the brain at the proper coordinates for CN ( 1 . 8–2 . 2 mm lateral from midline , 3 . 2–3 . 4 mm deep , 6 . 0–6 . 2 mm from Bregma ) , and small amount ( 50–300 nl for electrophysiological experiments , 50–100 nl for immunohistochemical protocols ) of viral suspension ( summarized in Table 2 ) was slowly pressure-injected either by a hand-held syringe or using a Picospritzer II ( General Valve Corporation ) . In some experiments , additional virus ( either non-specific or cre-dependent GFP reporter ) was injected in several locations in the cerebellar cortex or into the IO . After the entire volume was injected , pipettes were held at the same position for 10 to 15 min and were then carefully and slowly removed from the tissue , in order to avoid backflow of the viral suspension and unwanted contamination in the cerebellar cortex along the pipette tracts . Animals were closely monitored for 3 days until recovery from surgery and then housed for at least 3 to 4 weeks before being used in experiments , as described below . Throughout this report , we refer to data obtained using the GAD-cre mouse cerebella injected with floxed AAV2/9 virus with mCherry and ChR2 in the CN as ‘GAD-cre’ , and that from GlyT2-cre mouse cerebella injected with floxed AAV2/1 virus with EYFP and ChR2 as ‘GlyT2-cre’ , unless otherwise specified . 300-μm thick cerebellar slices were cut from the GAD-cre or GlyT2-cre cerebella using the Campden 7000smz oscillating blade microtome and ceramic blades ( Campden Instruments , UK ) . For the experiments performed at HUJI ( GAD-cre animals ) , horizontal slices were prepared at physiological temperature as described previously ( Huang and Uusisaari , 2013; Ankri et al . , 2014 ) and incubated in Solution 1 . For the experiments performed at IBENS ( GlyT2-cre animals ) , sagittal slices were prepared using ice-cold Solution 2 . After cutting , the slices were rinsed in warm Solution 3 for few seconds before being transferred to a recovery chamber with Solution 4 . Table 3 summarizes the ionic compositions of all experimental solutions . Acute slices of L7-ChR2-YFP × GlyT2-eGFP animals used for Figure 1—figure supplement 1 were prepared similarly . Notably , in all experiments , the extent of viral transfection was carefully examined in all slices to make sure no unwanted cerebellar structures were labeled . 10 . 7554/eLife . 06262 . 013Table 3 . Composition of solutions used for slice preparation and experimentsDOI: http://dx . doi . org/10 . 7554/eLife . 06262 . 013Solution 1 ( in MilliQ water ) Solution 2 ( in Volvic water ) Solution 3 ( in Volvic water ) Solution 4 ( in Volvic water ) Intracellular solutionExperimental detailsHUJI Cutting ( 33°C ) and Chamber-perfusion ( 25–28°C ) solutionsIBENS Cutting solution Ice-coldIBENS Recovery solution 33°CIBENS Chamber-perfusion solution ( 33°C ) IBENS HUJINaCl124125 . 74K-gluconate130140D-mannitol225KCl314 . 62 . 33 . 3Glucose20252525KH2PO41 . 2NaH2PO41 . 251 . 25MgSO43 . 5MgCl27 . 71 . 5NaHCO3262524 . 8CaCl220 . 511 . 60 . 5EGTA25HEPES2010D-APV ( µM ) 20 [in chamber-perfusion solution]505020Minocycline ( nM ) 505050Mg-ATP3 The slices were incubated for at least 30 min to an hour in physiological temperature ( Huang and Uusisaari , 2013; Ankri et al . , 2014 ) before being transferred to a recording chamber , mounted on an Olympus ( BX51WI or BX61WI; Tokyo , Japan ) microscope equipped with an epifluorescence illumination pathway ( Roper Scientific , Photometrics , Tucson , AZ ) and a camera ( Vx45; Optronix , Goleta , CA ) . During experiments , the GAD-cre slices were perfused with room temperature Solution 1 ( 25–28 °C; flow rate 3 ml/min ) , and the GlyT2-cre slices were perfused with Solution 4 at physiological temperature ( 33°C; flow rate 3 . 5 ml/min ) . The bicarbonate-buffered solutions ( 1 and 4 ) were continuously gassed with 5% O2/95% CO2 . Borosilicate glass patch electrodes ( resistance 3–12 MΩ ) were filled with intracellular solution ( see Table 3; pH 7 . 3 , 280 mOsm ) . For selecting region for patch-clamp experiments , as well as ascertaining that there was no transfection of any cortical neurons , slices were visualized with arc lamp illumination and appropriate filters ( for mCherry fluorescence in GAD-cre brains , emission: 605–685 nm , excitation: 530–588; GFP fluorescence: excitation: 473–508 nm; emission: 518–566 nm , for YFP fluorescence in GlyT2-Cre brains , emission: 523–563 nm; excitation: 500–523 nm ) . Whole-cell patch-clamp recordings , both current clamp and voltage clamp , were acquired using a Multiclamp 700B amplifier ( Molecular Devices , Sunnyvale , CA ) , digitized at 10 kHz ( current-clamp experiments ) or 50 kHz ( voltage-clamp experiments ) with USB-6229 acquisition board ( National Instruments , Austin , Texas ) and low-pass filtered at 2 kHz . Golgi cells were unambiguously identified from other cells in the cerebellar granular cell layer by the size of their soma and their bi-exponential capacitive current ( Dieudonné , 1995 ) ; furthermore , in some experiments using the GAD-cre mice , additional cre-dependent reporter virus was used to label GABAergic Golgi cells in the cortex and was used to guide neuronal selection . Thus , the percentage of s-Golgi cells out of all Golgi cells recorded ( Figure 4 ) was biased towards GABAergic Golgi cells . In the current-clamp experiments , intrinsic electrophysiological properties and synaptic inputs were assessed in Golgi cells either with zero holding current or with negative current injection so that the spontaneously spiking Golgi cells were hyperpolarized to subthreshold voltage values ( −55 mV to −60 mV ) , similar to the resting membrane potential of the not-spontaneous Golgi cells . During voltage-clamp experiments , Golgi cells' holding potential was −50 mV . All experiments were performed in the presence of 20 µM D-2-amino-5-phosphonopentoate ( D-APV , Abcam or Sigma Aldrich ) and 10 µM 6-cyano-7-nitroquinoxaline-2 , 3-dione ( CNQX , Sigma Aldrich ) or 10 µM 2 , 3-dihydroxy-6-nitro-7-sulfamoyl-benzo[f]quinoxaline-2 , 3-dione ( NBQX , Abcam ) to block N-Methyl-D-aspartic acid ( NMDA ) and α-Amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid ( AMPA ) receptors , respectively . In some experiments , strychnine ( Abcam or Sigma Aldrich ) and SR 95531 ( ‘gabazine’; Abcam or Sigma Aldrich ) were added to the bath . Regarding Figure 1—figure supplement 1 , slice perfusion system was similar to what described above . GlyT2-eGFP positive neurons were identified in the CN by epifluorescence and recorded ( holding potential -60 mV , intracellular solution as described in Husson et al . , 2014 ) . For optogenetic activation of ChR2s in acute GAD-cre slices , whole-field band-pass-filtered Hg-lamp ( Oregon Green filter 473–508 nm; ∼5 mW/mm2 ) was used , while for the GlyT2-cre slices an optical system combining low-numerical aperture ( NA ) Gaussian beam illumination and fast acousto–optic focusing system with a 473-nm continuous-wave diode-pumped solid-state laser ( LRS 0473-00100-03 , Laserglow Technologies ) was used as a one-photon light source . Small field of view ( 1 . 35 µm × 1 . 08 µm ) around the ChR2-expressing fibers was stimulated ( stimulation duration 5 ms; inter-stimulation interval 20 s ) . PN terminals in the CN of L7-ChR2-YFP × GlyT2-eGFP mouse ( Figure 1—figure supplement 1 ) were stimulated with 470 nm LED whole-field illumination ( Thorlabs , Newton , NJ ) with one millisecond duration . Throughout the work , special care was taken to prevent inadvertent transfection and stimulation of other inhibitory cerebellar interneurons expressing GAD and GlyT2 in our two mouse models . In addition to the precautions taken during the stereotaxic injection procedures , during acute experiments , the slices were carefully and systematically examined before being used for electrophysiological experiments; if unintended labeling was present , the slices were discarded . Finally , before patch-clamping a Golgi cell , the morphology and location of the fibers was carefully examined in order to exclude the possibility of activating parasagittal long-range Lugaro axons . Electrophysiological data were analyzed with Igor Pro 6 . 1 ( Wavemetrics , Portland , OR ) and MATLAB R2009b ( MathWorks , Natick , MA ) . Statistical analysis was performed using R GNU and MATLAB R2009b or R2012b . Data are presented in the text as mean ± S . D , unless otherwise specified . For statistical significance , Wilcoxon rank-sum , two-tailed Student's t-test ( paired or unpaired ) , F-test , K-S test , and signed-rank tests were used , as applicable , taking into account possible assumptions of normality as mentioned in the results . Spike delay analysis was performed by aligning the last spikes before light stimulation in each trace for each cell and measuring the time to the next spike after light stimulation; the measurements were normalized to the cell's average ISI . For comparing electrophysiological properties of different types of Golgi cells , grand average AP waveforms were generated for each cell by averaging peak-aligned APs obtained during 1-s voltage sweeps while adjusting the firing frequency to ∼25 Hz with current injection as necessary , and then by averaging these peak-aligned mean waveforms across experiments . Cm was defined as the ratio of membrane resistance ( Rm ) and time constant ( τ ) , estimated from voltage responses ( <5 mV ) to small hyperpolarizing current steps that did not activate voltage-gated conductances ( evidenced by the good single-exponential fits to the voltage responses ) . AP threshold was defined as the voltage at the time of the main peak in the second derivative of the voltage trace; AP amplitude was measured as the voltage difference between the spike threshold and peak voltage . Spike half-width was measured spike duration at half-amplitude . Spike AHP voltage was measured at the post-spike minimum voltage . AHP time was measured as the time of AHP voltage after threshold , and AHP amplitude was measured as the difference of the AHP voltage and spike threshold voltage . For fair comparison of current-to-firing frequency ratios in different sized cells , the current injection values were normalized to the Cm of each cell . Spike frequency adaptation index was quantified as the relative decrease in instantaneous firing frequency during a 1-s long depolarizing current step ( from −65 mV holding level ) during which the mean firing frequency was ∼25 Hz . In the pharmacological voltage-clamp experiments ( Figure 3 ) , the responses were recorded after at least 6 min from the beginning of the perfusion of the drug into the recording chamber to provide the time for the steady-state effect . Time-locking of spikes ( Figure 3—figure supplement 1 ) was quantified as the decrease in the normalized Vm variability between subsequent stimulation trials . Decrease in variability , as an index of spike-time locking , was considered statistically significant at the level of 3*SD ( see also Schneider et al . , 2014 ) . Animals were placed in a stereotaxic apparatus ( Harvard Apparatus , Halliston , MA ) . A scalp incision was made along the midline; the skull was cleaned by scraping and by application of hydrogen peroxide . Crus I and II were exposed with a craniotomy , but the dura was not removed to enhance mechanical stability . Commercial tetrodes embedded in a quartz tube ( Thomas Recording , Giessen , Germany ) , gold-plated to reach a 100–200 kΩ impedance , were lowered into the CN . Signals were referenced against a tungsten electrode positioned in saline at the surface of the cerebellar cortex . The light was delivered immediately above the CN via an optical fiber ( diameter 200-μm core ) connected to CrystaLaser at 473 nm and inserted in a cannula placed above the injection sites in the CN ( see ‘Stereotaxic injections’ paragraph for coordinates ) . 25- to 50-ms light pulses ( 45 mW ) were delivered at 4 Hz in 3-s bouts separated by 7-s recovery periods; post hoc inspection showed no indication of decreased inhibition during the 3-s bouts , suggesting that these stimulation parameters did not induce cumulative ChR2 inactivation . Signals were acquired using a custom-made headstage and amplifier and a custom-written LabVIEW software ( National Instruments , Austin , TX ) allowing real-time monitoring of cellular activity . To isolate spikes , continuous wide-band extracellular recordings were filtered off-line with a Butterworth 1 kHz high-pass filter . Spikes were then extracted by thresholding the filtered trace and the main parameters of their waveform extracted ( width and amplitude on the 4 channels ) . The data were hand-clustered by polygon-cutting in 2-dimensional projections of the parameter space using Xclust ( Matt Wilson , MIT ) . The quality of clustering was evaluated by inspecting the auto-correlograms of the units ( Gao et al . , 2011 ) . Golgi cells and Purkinje cells were identified according to most recently published criteria ( Van Dijck et al . , 2013 ) . These criteria provide a simple approach to classify cerebellar units using only a few statistical parameters describing the firing frequency and irregularity of discharge: the mean spike frequency ( MSF ) , the coefficient of variation of the log of the ISI ( LCV ) , and entropy of the ISIs ( ENT ) . We used the boundary values on these parameters defined by Van Dijck et al . ( 2013 ) to identify Golgi cells ( MSF<20 Hz , 0 . 5<ENT<7 . 5 and 0 . 02<LCV<0 . 25 ) . Golgi cells exhibited an average firing rate ( FR ) of 8 . 7 ± 4 . 2 Hz . To assess the presence of a response to optogenetic stimulations , peri-stimulus time histograms ( PSTH ) were constructed . Each PSTH was normalized by subtracting the average baseline spike count ( before stimulation ) and dividing by the baseline standard deviation yielding a Z-score . A modulation of FR in response to optogenetic stimulation was considered significant when the absolute Z-score of a 3-ms bin was higher than 2 . 5 in at least two time bins in the 50-ms time window following the stimulus . The total time during which the Z-score was significant defined the duration of the inhibition . The latencies ( onset and peak ) were calculated starting from the beginning of the light-pulse , the peak latency being the point where the Z-score was maximal and the onset latency being defined as the first time point with a significant Z-score . Animals were deeply anesthetized with intra-peritoneal injection of sodium pentobarbital ( 50 mg/kg ) and perfused through the aorta with ice-cold solution of phosphate buffer saline ( PBS; pH 7 . 4; Sigma , Saint Louis , MO , USA ) followed by 50–75 ml of 4% wt/vol paraformaldehyde ( PFA; VWR International , Radnor , PA ) in PBS . The entire brain was then dissected and post-fixated ( 3 hr for immunohistochemistry , overnight for anatomical examination ) in 4% PFA at 4°C before rinsing in PBS . For anatomical confocal imaging , 80-µm sections were cut with a Leica 1000 TS vibratome ( Leica Microsystems , Wetzlar , Germany ) , placed on objective glass , mounted with Immu-Mount ( Thermo Scientific , Waltham , MA ) , and coverslipped with #1 . 5 glass . For immunohistochemistry experiments , the brains were cryoprotected by equilibration in 30% sucrose wt/vol PBS at 4°C and then cut at −20°C with a Leica CM3050S cryostat ( Leica Microsystems , Wetzlar , Germany ) . Free-floating 80-µm thick parasagittal sections were rinsed in PBS and permeabilized 2 hr at room temperature in 0 . 4% vol/vol Triton 100-X ( Sigma ) in PBS . Non-specific sites were saturated by incubation in 0 . 4% Triton 100-X—1 . 5% cold fish skin gelatin ( Sigma ) in PBS at room temperature for 3 hr . Primary antibodies were applied overnight at 4°C in a PBS solution containing 0 . 1% Triton 100-X—1 . 5% fish gelatin ( mouse GAD65-67 antibody mAB 9A6 ( Enzo Life Sciences , Farmingdale , NY ) at 1/500 final dilution; chicken GFP antibody ( Avès , Oregon , USA ) at 1/1000 final dilution; guinea pig VIAAT antibody ( Synaptic Systems , Göttingen , Germany ) at 1/1500 final dilution; guinea pig GlyT2 antibody ( Millipore , Darmstadt , Germany ) at 1/1500 final dilution , rabbit Neurogranin antibody ( Millipore ) at 1/500 final dilution ) . After rinsing in 0 . 1% Triton 100-X in PBS , slices were incubated overnight at 4°C with secondary antibodies coupled to 488 , 549 , or 649 DyLight fluorophores ( Jackson ImmunoResearch , West Grove , PA ) or Alexa Fluor 555 IgG ( Invitrogen , Carlsbad , CA ) at 1/500 final dilution in PBS—0 . 1% Triton 100-X −1 . 5% cold fish skin gelatin . Slices were finally rinsed with PBS and mounted in Prolong Gold Antifade Reagent ( Sigma ) . For Figure 1—figure supplement 1 , immunostaining against GFP and VIAAT ( same antibodies as above ) was performed on paraffin-embedded sections as previously described ( Husson et al . , 2014 ) . Confocal stacks from immunolabeled cerebellar slices were acquired using an inverted confocal microscope ( Leica , SP8 ) using a 63× oil-immersion objective ( NA 1 . 3 ) . Confocal stacks for anatomical visualizations were acquired with Leica SP5 microscope using 40× ( NA 1 . 25 ) and 63× ( NA 1 . 5 ) oil-immersion objectives , with 8- or 12-bit color depth , and with 0 . 1 μm z-step . The images were acquired for mCherry , EYFP , and GFP fluorescence with excitation lasers and emission filters set to: 561 DPSS laser , 587–655 nm; 488 argon laser , 520–580 nm; GFP , 500–550 nm . Wide-view images ( in Figure 2A1 , B1 ) were composed by merging tiles of confocal stacks ( with 10% overlap and 1 μm z-step ) . The grayscale background images in 2A1 and B1 were obtained from autofluorescence signals acquired at the same time as the specific fluorescence signals . Morphological features ( NC cell body sizes , iNC axonal bouton sizes ) were measured using the Fiji image analysis software ( Schindelin et al . , 2012 ) . The soma sizes are given as the major length axis; for axon bouton sizes , as the area of the cross section of each bouton in maximal projection image . To quantify neurogranin and GlyT2-eGFP staining intensities at Golgi cell bodies , z-stacks containing the somata were projected and averaged ( z-projection thickness: 6 . 8 µm ) . Intensities for each channel were normalized according to the slope of the fit to the logarithmic distribution of their pixel intensity before being retrieved and the ratios were calculated . As the neuropil in cerebellar granule layer is densely labeled in the GlyT2-eGFP mice preventing backtracking individual distal dendrites to their somata in order to attribute them a ratio value , the iNC varicosities in the granular cell layer were included in the statistics only when contacting proximal dendrites and cell bodies . Using GNU R , K-means 2D clustering was performed on mean GlyT2-eGFP vs mean neurogranin data set to cluster the GoC subpopulation .	The cerebellum is a region in the brain that plays a central role in controlling posture and movement . The cerebellum is composed of a cortex and several nuclei . The nuclei are thought to ‘compute’ the signals that are sent from the cerebellum to other parts of the brain to control posture and movement . They do this under the supervision of the cortex . The main interaction between the cortex and the nuclei involves cortical neurons called Purkinje cells inhibiting the activity of the nuclei . Ankri , Husson et al . have now used various genetic techniques and mutant mice to identify a new population of neurons in the nuclei of the cerebellum and to express fluorescent markers into these cells . This approach reveals that the axons of these neurons ‘climb’ from the nuclei to the cortex to form a new circuit called the inhibitory nucleo-cortical ( iNC ) pathway . Moreover , activating the iNC axons with light reveals that they selectively target and silence a population of neurons called the Golgi cells , which control the transmission of information in the cerebellar cortex . Ankri , Husson et al . go on to show that the Golgi cells silenced by the iNC pathway differ from other Golgi cells in a number of ways: in particular , these Golgi cells use a chemical called GABA to communicate with neurons . The next challenge is to explore how the iNC pathway fine-tunes how sensory inputs are processed in the cerebellum , and to better understand its role in the execution of complex movements , including in people with conditions that affect motor function , such as ataxias .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	A novel inhibitory nucleo-cortical circuit controls cerebellar Golgi cell activity
The SWI/SNF-family remodelers regulate chromatin structure by coupling the free energy from ATP hydrolysis to the repositioning and restructuring of nucleosomes , but how the ATPase activity of these enzymes drives the motion of DNA across the nucleosome remains unclear . Here , we used single-molecule FRET to monitor the remodeling of mononucleosomes by the yeast SWI/SNF remodeler , RSC . We observed that RSC primarily translocates DNA around the nucleosome without substantial displacement of the H2A-H2B dimer . At the sites where DNA enters and exits the nucleosome , the DNA moves largely along or near its canonical wrapping path . The translocation of DNA occurs in a stepwise manner , and at both sites where DNA enters and exits the nucleosome , the step size distributions exhibit a peak at approximately 1–2 bp . These results suggest that the movement of DNA across the nucleosome is likely coupled directly to DNA translocation by the ATPase at its binding site inside the nucleosome . The SWI/SNF-family chromatin remodelers are important regulators of chromatin structure in transcriptional activation and in the creation and maintenance of nucleosome-free regions ( Clapier and Cairns , 2009; Bowman , 2010; Becker and Workman , 2013; Narlikar et al . , 2013; Bartholomew , 2014; Lorch and Kornberg , 2015 ) . Consistent with these roles , SWI/SNF remodelers are capable of disrupting nucleosomes in a number of ways , including repositioning the histone octamer along the DNA , ejecting H2A-H2B dimers from the octamer , and ejecting the histone octamer from the DNA ( Clapier and Cairns , 2009; Bowman , 2010; Becker and Workman , 2013; Narlikar et al . , 2013; Bartholomew , 2014; Lorch and Kornberg , 2015 ) . All SWI/SNF complexes are composed of a catalytic subunit , which harbors a superfamily 2 ( SF2 ) ATPase domain , and a variety of accessory subunits ( Clapier and Cairns , 2009 ) . During remodeling , the ATPase domain contacts the nucleosome at an internal site 20 bp from the nucleosome dyad , referred to as the super-helical 2 ( SHL2 ) site , and translocates DNA ( Saha et al . , 2002; Saha et al . , 2005; Zofall et al . , 2006; Dechassa et al . , 2008 ) . However , how DNA translocation by the ATPase domain is coupled to the various activities of the remodeling complexes remains unclear . SWI/SNF remodelers expose substantial amounts of intra-nucleosomal DNA to nucleases during remodeling , suggesting that the remodelers disrupt the wrapping of DNA around the nucleosome ( Narlikar et al . , 2001; Aoyagi et al . , 2002; Lorch et al . , 2010; Shukla et al . , 2010 ) . The nature of the disrupted intermediates remains unclear , and previous studies have suggested unpeeling of DNA from the edges of the nucleosome ( Bazett-Jones et al . , 1999; Floer et al . , 2010 ) as well as the formation of loops or bulges of DNA inside the nucleosome ( Narlikar et al . , 2001; Kassabov et al . , 2003; Zhang et al . , 2006; Chaban et al . , 2008; Shukla et al . , 2010; Liu et al . , 2011; Ramachandran et al . , 2015 ) . Given that the ATPase domain of the enzymes acts at the SHL2 site , approximately 50 and 100 bp away from the sites where the DNA enters and exits the nucleosome , respectively , how ATPase activity drives the overall movement of DNA around the nucleosome remains unclear . The isolated catalytic subunit of a SWI/SNF remodeler has been shown to translocate naked DNA ( in the absence of a histone octamer ) in ~2 bp steps ( Sirinakis et al . , 2011 ) . However , DNA translocation around the nucleosome by SWI/SNF holoenzyme complexes has been reported to occur in large steps of ~50 bp in size ( Zofall et al . , 2006 ) . It is unknown whether this 50-bp step represents the fundamental step size of nucleosome translocation by SWI/SNF complexes . The related ISWI-family of chromatin remodelers move DNA in multi-bp compound steps that are composed of 1-bp fundamental steps ( Blosser et al . , 2009; Deindl et al . , 2013 ) . Although the ISWI-family remodelers share a homologous ATPase domain with the SWI/SNF-family remodelers , they differ in both the domains flanking the ATPase domain on their catalytic subunits and the accessory subunits that associate with the catalytic subunits ( Clapier and Cairns , 2009 ) , and they bind the nucleosome differently ( Hota and Bartholomew , 2011; Dechassa et al . , 2012 ) . Thus , it is unclear whether SWI/SNF remodelers share a similar nucleosome translocation mechanism . In this study , we used single-molecule FRET ( Ha et al . , 1996 ) with a variety of labeling schemes to monitor nucleosome remodeling by yeast RSC , a prototypical SWI/SNF-family enzyme ( Cairns et al . , 1996 ) , in real time . These experiments allowed us to observe transient intermediates of the remodeling process and characterize the motion of the DNA at the nucleosomal edges where the DNA moves into and out of the nucleosome . At these locations , we found that DNA was translocated largely along or near its canonical path in a stepwise manner and that the distribution of the step sizes showed a peak at a step size of approximately 1–2 bp . We reconstituted dye-labeled mononucleosomes on a double-stranded DNA containing the 601 positioning sequence ( Lowary and Widom , 1998 ) to ensure reproducible positioning of the histone octamer . We used several labeling schemes with the FRET donor dye on various sites of the histone octamer and the FRET acceptor dye on various locations of the DNA . In the first labeling scheme , the mononucleosome was flanked by a shorter 6 bp linker on one side and a longer 78 bp linker on the other side . The FRET acceptor dye , Cy5 , was attached to the end of the shorter linker and a biotin moiety was attached to the end of the longer linker for immobilization on the microscope slide ( Figure 1A ) . The histone octamer was labeled with the FRET donor dye , Cy3 , on the C-terminal tail of H2A ( position 119 ) . We refer to this construct as the H2A/[end , +6] construct to indicate the position of the Cy3 label ( on histone H2A ) and Cy5 label ( at the end of the linker DNA , 6 bp from the edge of the nucleosome ) . This labeling scheme allowed us to monitor the movement of DNA relative to the histone octamer in real time during nucleosome remodeling . 10 . 7554/eLife . 10051 . 003Figure 1 . Single-molecule FRET assay for monitoring nucleosome translocation by RSC . ( A ) Diagram depicting the nucleosome substrates before and after remodeling by RSC . The top row depicts the nucleosomes in cartoon form while the bottom row shows the footprint of the histone octamer ( tan oval ) on the DNA ( black lines ) . ( B and C ) Representative traces showing the Cy3 intensity ( green ) , Cy5 intensity ( red ) , and FRET value ( blue ) for exit-side movement ( B ) and entry-side movement ( C ) . [RSC] = 5 nM , [ATP] = 5 µM . ( D ) The fraction of exit-side movement ( cyan ) and entry-side movement ( orange ) traces observed with 1 nM RSC and 20 µM ATP . Error bars represent the standard error from > 200 nucleosomes . ( E–G ) Histograms of the distribution of initial ( before remodeling , blue ) and final ( after remodeling , red ) FRET values of remodeling traces from nucleosomes lacking any ssDNA gap ( E ) , nucleosomes with a 2-nt ssDNA gap at the SHL–2 site ( F ) , and nucleosomes with a 2-nt ssDNA gap at the SHL+2 site ( G ) . [RSC] = 1 nM , [ATP] = 20 µM . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 003 10 . 7554/eLife . 10051 . 004Figure 1—figure supplement 1 . Remodeling of nucleosomes with a reversed 601 positioning sequence . ( A ) Cartoon of the H2A/[end , +6] nucleosome construct ( B ) Sequence of the top strand of the DNA molecule used for the H2A/[end , +6] nucleosomes with the 601 positioning sequence shown in orange . ( C ) Sequence of the top strand of the DNA molecule in which the orientation of the 601 positioning sequence ( orange ) has been reversed . ( D ) The fraction of exit-side movement ( cyan ) and entry-side movement ( orange ) traces observed when remodeling nucleosomes with the DNA containing the 601 sequence from panel B ( reproduced from Figure 1D for comparison ) or the DNA sequence from panel C containing the reversed 601 positioning sequence . [RSC] = 1 nM , [ATP] = 20 µM . Error bars represent the standard error from > 100 nucleosomes . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 004 10 . 7554/eLife . 10051 . 005Figure 1—figure supplement 2 . Surface anchoring of the nucleosomes does not affect the kinetics of remodeling . Comparison of the average remodeling kinetics for freely diffusing nucleosomes in the absence ( grey , ensemble assay ) or presence of ATP ( black , ensemble assay ) , and surface-tethered nucleosomes in the presence of ATP ( red , single-molecule assay ) . The remodeling kinetics were monitored by reading the Cy5 intensity over time and normalizing the initial and steady-state values to 1 and 0 , respectively . The Cy5 intensities in the single-molecule assay were determined by summing the Cy5 signals from ~2000 single-nucleosome traces . [RSC] = 6 nM , [ATP] = 20 µM . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 005 10 . 7554/eLife . 10051 . 006Figure 1—figure supplement 3 . Locations of the 2-nt ssDNA gaps in the DNA sequence . The sequence of the DNA used for the top and bottom strands of the H2A/[end , +6] and H3/[end , +6] nucleosomes is shown with the 601 positioning sequence colored orange . To create a gap at the SHL–2 site , the underlined nucleotides in the top strand were replaced with a 2-nt gap . To create a gap at the SHL+2 site , the underlined nucleotides in the bottom strand were replaced with a 2-nt gap . RSC translocates on DNA with a 3’→5’ polarity ( Saha et al . , 2005 ) , and the gaps are positioned to prevent translocation in that direction . However , it has been previously shown that gaps in either strand prevent DNA translocation by RSC ( Saha et al . , 2005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 006 10 . 7554/eLife . 10051 . 007Figure 1—figure supplement 4 . Effects of 2-nt ssDNA gaps at the SHL ± 2 sites on the direction of nucleosome translocation . ( A and C ) Top: cartoon of the H2A/[end+6] nucleosome with a 2-nt ssDNA gap at the SHL–2 site ( A ) or SHL+2 site ( C ) . Bottom: Representative FRET traces from the remodeling of constructs with a gap at the SHL–2 site ( A ) and SHL+2 site ( C ) . [RSC] = 1 nM , [ATP] = 5 µM . ( B and D ) The fraction of traces showing exit-side movement ( cyan ) and entry-side movement ( orange ) observed during the remodeling of constructs with a gap at the SHL–2 ( B ) or SHL+2 ( D ) sites . Error bars represent the standard error from > 100 nucleosomes per construct . [RSC] = 1nM , [ATP] = 20µM . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 007 We anticipate that remodeling of the dye-labeled mononucleosomes by RSC will generate two distinct products , depending on which side of the nucleosome the ATPase engages ( Figure 1A ) . If RSC engages the SHL+2 site , which we define to be the SHL2 site near the longer linker DNA , RSC will translocate DNA toward the shorter linker . Previous studies of SWI/SNF-family remodelers have shown that the enzyme can translocate DNA around the nucleosome until the end of the DNA reaches the SHL2 site , which is ~50 bp past the edge of the nucleosome ( Flaus and Owen-Hughes , 2003; Kassabov et al . , 2003 ) . Therefore , this type of action should generate a ~130 bp movement of the DNA toward the shorter linker , moving the Cy5 dye away from the Cy3 dye on the octamer and causing a monotonic decrease in FRET . This action will eventually position the Cy5-labeled DNA end >40 nm from the Cy3 label on the H2A , resulting in zero FRET . On the other hand , if RSC engages the SHL–2 site , the SHL2 site near the shorter linker DNA , RSC will translocate DNA toward the longer linker , first moving the Cy5 dye closer to the Cy3 dye on the octamer and then further away from the Cy3 , causing an initial increase in FRET followed by a decrease . This action will generate a final product where the labeled DNA end resides at the SHL–2 site . Based on the crystal structure of the nucleosome ( Luger et al . , 1997 ) , this should place the Cy5 ~6 . 8 nm from the Cy3 labeling site , giving a low but non-zero FRET value . Consistent with these expectations , we observed two major classes of single-molecule traces upon addition of RSC and ATP to the nucleosomes . One class of traces showed a monotonic decrease in FRET to zero FRET ( Figure 1B ) . We assigned these traces to the case where the ATPase domain of RSC bound to the SHL+2 site and translocated the DNA toward the shorter linker . Because the dye labels monitored the dynamics of the DNA end moving away from the octamer , we refer to these traces as monitoring exit-side movement . The second class of traces showed an initial increase in FRET followed by a decrease to a final FRET of ~0 . 17 ( Figure 1C ) . We assigned these traces to the case where the ATPase domain bound to the SHL–2 site and translocated DNA toward the longer linker , which is expected to first bring the FRET donor and acceptor dyes closer and then move them farther apart . Because the dye labels in these cases monitored the dynamics of the DNA end moving into the nucleosome , we refer to these traces as monitoring entry-side movement . We classified the traces as reflecting entry-side or exit-side movement based on the presence or absence of an initial FRET increase during remodeling and identified a roughly equal number traces showing entry-side and exit-side movement when RSC and ATP were added to the H2A/[end , +6] nucleosome construct ( Figure 1D ) . This result is consistent with previous results indicating that translocation by SWI/SNF enzymes is bidirectional and does not depend on linker DNA length ( Flaus and Owen-Hughes , 2003; Kassabov et al . , 2003; Shundrovsky et al . , 2006 ) . It has been shown previously that the strength of the histone-DNA contacts between the 601 positioning sequence and the histone octamer is asymmetric with respect to the nucleosomal dyad ( Ngo et al . , 2015 ) , so we tested whether the fraction of nucleosomes undergoing exit-side or entry-side movement depends on the orientation of the 601 positioning sequence by reversing the 601 sequence on the H2A/[end , +6] nucleosome . We found that the fractions of traces exhibiting entry-side and exit-side movement were essentially identical for the two sequence orientations ( Figure 1—figure supplement 1 ) , suggesting that the asymmetry of the 601 positioning sequence does not influence the directionality of remodeling by RSC . The overall kinetics of the FRET changes observed in the single-molecule assay were similar to the kinetics observed in solution-based ensemble FRET measurements , indicating that surface attachment did not substantially affect remodeling by RSC ( Figure 1—figure supplement 2 ) . As expected , no major FRET changes were observed when RSC was added to the nucleosomes in the absence of ATP ( Figure 1—figure supplement 2 ) . In order to confirm our assignment of entry-side and exit-side movement traces , we made use of the fact that the ATPase of RSC is incapable of translocating past a 2-nt single-stranded ( ss ) gap in the DNA ( Saha et al . , 2005; Zofall et al . , 2006 ) . Therefore , placing such a gap at the SHL+2 or SHL–2 site ( Figure 1—figure supplement 3 ) should allow us to control the direction of DNA translocation around the nucleosome . After remodeling by RSC , nucleosomes without any ssDNA gap showed two distinct populations of nucleosomes with FRET values of ~0 and ~0 . 17 , corresponding to the products of exit-side and entry-side movement , respectively ( Figure 1E ) . As expected , remodeling of a construct containing a gap at the SHL–2 site showed only a single peak at zero FRET after remodeling ( Figure 1F ) , consistent with the gap preventing RSC from engaging the SHL–2 site to generate entry-side movement traces . Furthermore , >95% of the remodeling traces from this construct were classified as exit-side movement traces and showed a monotonic FRET decrease without an initial FRET increase ( Figure 1—figure supplement 4A , B ) . Similarly , a gap at the SHL+2 site largely eliminated the zero FRET population and resulted in essentially a single population with a FRET of ~0 . 14 after remodeling ( Figure 1G ) . Only a small fraction ( ~8% ) of nucleosomes showed a final FRET of zero , which is most likely due to photobleaching of the Cy5 dye during remodeling . Moreover , >95% of the traces observed were classified as entry-side movement traces and showed an initial FRET increase during remodeling ( Figure 1—figure supplement 4C , D ) , consistent with the gap preventing RSC from engaging the SHL+2 site and generating exit-side movement traces . These results confirm that the traces classified as reflecting entry-side movement resulted from ATPase action at the SHL–2 site and the traces classified as reflecting exit-side movement resulted from ATPase action at the SHL+2 site . In addition to nucleosome sliding , RSC has been reported to facilitate a number of other changes to the nucleosome , such as ejecting H2A-H2B dimers or the entire histone octamer , but we do not expect to observe these activities in our assay because these activities require free nucleosomes or free DNA as acceptors , the activity of histone chaperones , or a dinucleosome construct ( Clapier and Cairns , 2009; Bowman , 2010; Becker and Workman , 2013; Narlikar et al . , 2013; Bartholomew , 2014; Lorch and Kornberg , 2015 ) . Indeed , after incubating the H2A/[end , +6] nucleosomes with 5 nM RSC and 5 µM ATP for 20 min , >90% of the nucleosomes remained present with the H2A subunit attached , similar to the fraction of nucleosomes remaining intact after incubation in the absence of RSC and ATP ( 91% ) . However , because RSC could destabilize the H2A-H2B dimer ( Lorch et al . , 2010 ) , it is possible that RSC may cause transient H2A-H2B dimer displacement during remodeling that contributes to the FRET changes . To investigate whether movement of the H2A-H2B dimer ( and hence the Cy3-label on H2A ) relative to the rest of the octamer contributed to the FRET changes we observed during RSC remodeling , we took three approaches . First , we created a H3/[end , +6] construct ( Figure 2A ) by moving the Cy3 label from the C-terminal tail of histone H2A to the N-terminal tail of histone H3 ( position 33 ) . If movement of the H2A-H2B dimer relative to the rest of the octamer was responsible for the FRET changes that we observed on the H2A/[end , +6] construct , we expected that moving the Cy3 dye to histone H3 would eliminate these FRET changes . However , if H2A-H2B dimer dynamics did not contribute the observed FRET changes , the H3-labeled nucleosomes would produce FRET dynamics similar to those observed with the H2A-labeled nucleosomes because the N-terminal tail of histone H3 lies near the C-terminal tail of histone H2A ( Luger et al . , 1997 ) . Similar to H2A-labeled nucleosomes , remodeling of the H3-labeled nucleosomes also produced two classes of FRET traces , one class showing a monotonic decrease in FRET ( Figure 2B ) and the other class showing an initial FRET increase followed by a FRET decrease ( Figure 2C ) . As in the case of the H2A/[end , +6] constructs , placing a gap at the SHL–2 site of the H3/[end , +6] nucleosomes eliminated nearly all of the traces showing an initial FRET increase , and placing a gap at the SHL+2 site of the H3/[end , +6] nucleosomes eliminated most of the traces showing a monotonic FRET decrease ( Figure 2D ) , indicating that these traces reflect entry-side and exit-side movement , respectively . The observation that the FRET changes were similar for the H2A-labeled and H3-labeled nucleosomes suggests that these observed FRET changes were not specific to the dynamics of the H2A-H2B dimer . 10 . 7554/eLife . 10051 . 008Figure 2 . Single-molecule FRET assay for probing the displacement of the H2A-H2B dimer during RSC-mediated nucleosome remodeling . ( A ) Diagram of the H3/[end , +6] nucleosome . ( B and C ) Representative FRET traces reflecting the exit-side movement ( B ) and entry-side movement ( C ) of the H3/[end , +6] nucleosomes . [RSC] = 1 nM , [ATP] = 2 µM . ( D ) Fraction of traces showing entry-side movement ( orange ) and exit-side movement ( cyan ) observed with 1 nM RSC and 20 µM ATP for nucleosomes lacking any ssDNA gap , nucleosomes with a 2-nt ssDNA gap at the SHL–2 site , and nucleosomes with a 2-nt ssDNA gap at the SHL+2 site . Error bars represent the standard error from > 100 nucleosomes per construct . ( E ) Diagram of the H2A/H2A nucleosomes . ( F ) Histogram of the FRET values from the H2A/H2A nucleosomes in the absence of RSC ( black ) , after the addition of 6 nM RSC ( red ) , or after the addition of 6 nM RSC and 5 µM ATP ( cyan ) . Histograms were constructed from >200 nucleosomes per condition . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 008 10 . 7554/eLife . 10051 . 009Figure 2—figure supplement 1 . Remodeling of the H2A/H2A construct by RSC . ( A ) Top: Cartoon of the H2A/H2A nucleosome construct . Bottom: Representative Cy3 intensity , Cy5 intensity , and FRET traces when H2A/H2A nucleosomes are incubated with 6 nM RSC and 5 µM ATP . ( B ) The mean ratio of the Cy3 signals before and after remodeling ( green ) and the mean ratio of the Cy5 signals before and after remodeling ( red ) . Error bars represent the SEM from >100 nucleosomes . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 009 10 . 7554/eLife . 10051 . 010Figure 2—figure supplement 2 . Remodeling of nucleosomes labeled in the globular domain of histone H2B . ( A ) Left: diagram of the H2B/H2B nucleosomes . Right: histogram of the FRET values from H2B/H2B nucleosomes before ( blue ) and after ( red ) the addition of 6 nM RSC and 5 µM ATP . ( B ) Cartoon of the H2B/[end , +0] nucleosome construct . ( C and D ) Representative FRET traces showing H2B/[end , +0] nucleosomes undergoing exit-side movement ( C ) and entry-side movement ( D ) . [RSC] = 1 nM , [ATP] = 5 µM . ( E ) Fraction of traces showing entry-side and exit-side movement observed with 1 nM RSC , and 20 µM ATP for nucleosomes lacking any ssDNA gap , nucleosomes with a 2-nt ssDNA gap at the SHL–2 site , and nucleosomes with a 2-nt ssDNA gap at the SHL+2 site . Error bars represent the standard error from > 90 nucleosomes per condition . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 010 Second , to directly probe for potential movement of the H2A-H2B dimer during remodeling , we reconstituted histone octamers with a 1:1 mixture of Cy3-labeled H2A and Cy5-labeled H2A and assembled mononucleosomes using these octamers and unlabeled DNA . We referred to these nucleosomes as H2A/H2A to indicate that both donor and acceptor dyes are on the H2A subunits ( Figure 2E ) . Nucleosomes with a single Cy3 on one of the H2A subunits and a single Cy5 on the other H2A subunit could be readily identified at the single-molecule level and gave FRET ~ 0 . 6 ( Figure 2F ) , whereas nucleosomes lacking a Cy5 showed zero FRET and nucleosomes lacking a Cy3 were not visible under green laser illumination . Addition of RSC and ATP to this H2A/H2A construct resulted in only a small decrease in FRET , ΔFRET ~0 . 1 ( Figure 2F and Figure 2—figure supplement 1A ) . This FRET decrease was partially recapitulated by the addition of enzyme in the absence of ATP ( Figure 2F ) , suggesting that this FRET change was associated , at least in part , with RSC binding . Furthermore , this apparent FRET change was due almost entirely to an increase in Cy3 fluorescence without a corresponding decrease in Cy5 fluorescence ( Figure 2—figure supplement 1A , B ) , suggesting that this apparent FRET change was probably due to a change in the photophysical properties of the Cy3 dye upon RSC binding rather than a bona fide change in distance between the dyes . Even if the FRET change was in part due to a dye-to-dye distance change , the small magnitude of this change was not sufficient to explain the much larger changes in FRET observed with the H2A/[end , +6] and H3/[end , +6] constructs during remodeling . Finally , because the H2A labeling site resides on the flexible , basic tail of this histone subunit , it is possible that this region could interact with the negatively-charged DNA and mask the detection of H2A-H2B motion . To address this possibility , we moved the dye label to histone H2B at position 49 , which resides within the globular domain of the histone . We first reconstituted a H2B/H2B nucleosomal construct containing a mixture of Cy3- and Cy5-labeled H2B on unlabeled DNA , as we did for the H2A/H2A construct . We identified nucleosomes containing a Cy3 dye on one H2B subunit and a Cy5 dye on the other H2B subunit by selecting those nucleosomes with a finite FRET value ( FRET~0 . 2 ) , again because nucleosomes lacking a Cy5 showed zero FRET and nucleosomes lacking a Cy3 were not visible under green laser illumination . The FRET values of these H2B/H2B nucleosomes did not change appreciably after addition of RSC and ATP ( Figure 2—figure supplement 2A ) . Next , to observe the remodeling behavior of the nucleosomes with H2B-labeled octamer , we constructed a H2B/[end , +0] nucleosome , containing Cy3 on the H2B subunit ( position 49 ) and a Cy5 at the 5’ end of the DNA , positioned at the edge of the nucleosome ( the +0 position ) ( Figure 2—figure supplement 2B ) . Upon addition of RSC and ATP to the H2B/[end , +0] nucleosomes , we again observed both exit-side and entry-side movement traces ( Figure 2—figure supplement 2C , D ) resembling those obtained with the H2A- and H3-labeled nucleosomes . The H2B/[end , +0] nucleosomes showed a preference for entry-side remodeling ( Figure 2—figure supplement 2E ) , perhaps because the label biased binding of RSC in the orientation supporting entry-side remodeling . However , control constructs with gaps at the SHL–2 or SHL+2 site eliminated nearly all of the entry-side or exit-side movement traces , respectively ( Figure 2—figure supplement 2E ) , confirming our assignment of entry-side and exit-side movement traces . Thus , in our assays using nucleosomes labeled with Cy3 on the octamer and Cy5 on the DNA , we saw similar entry-side and exit-side movement traces when using three different positions of the Cy3 on the octamer ( on H2A , H3 or H2B ) , suggesting that the FRET dynamics we observed were due primarily to movement of the DNA relative to the octamer . Furthermore , when we placed both Cy3 and Cy5 dyes on the H2A-H2B dimer , we did not observe substantial motion of the H2A-H2B dimer with two different positions of the dye ( on H2A or H2B ) . However , our data cannot exclude the possibility of smaller-scale movement of the H2A-H2B dimer , not detectable by our FRET assay , that could disrupt important histone-histone or histone-DNA contacts during remodeling . Next , we asked if RSC induces large-amplitude unwrapping of DNA at the edges of the nucleosome . We separately considered this possibility for the two edges of the nucleosome where DNA enters or exits the nucleosome . At the nucleosomal edge where DNA enters the nucleosome , translocation of the DNA along its canonical path on the nucleosome would be expected to produce a FRET increase as the Cy5 at the end of linker DNA moves toward the edge of nucleosome , followed by a FRET decrease as the Cy5 moves along the nucleosome surface toward the SHL–2 site . On the other hand , if RSC were to unwrap a substantial amount of DNA and lift the DNA off the nucleosomal surface by a large distance , as expected if the unwrapped DNA extended from the nucleosome in a unbent fashion , we would expect the FRET traces to exhibit a FRET decrease as the Cy5 dye on the linker DNA moves away from the nucleosome , where the Cy3 dye resides . The entry-side movement traces that we observed showed a substantial increase in FRET followed by a FRET decrease for both H2A/[end , +6] and H3/[end , +6] constructs ( Figures 1C , 2C , Figure 1—figure supplement 4C and Figure 3—figure supplement 1A ) , which was consistent with movement largely along or near the canonical path at the nucleosome edge where DNA enters the nucleosome . This phenomenon was also seen for constructs with increased linker DNA length , such as the H2A/[end , +11] and H3/[end , +9] constructs , where the Cy5 dye-labeled DNA end was initially 11 bp and 9 bp away from the nucleosome edge ( Figure 3—figure supplement 1B , C ) . As described above , similar entry-side movement traces were also observed for the H2B/[end , +0] construct , in which the Cy3 label resides inside the globular domain of histone H2B instead of on the flexible tail of histone H2A ( Figure 2—figure supplement 2D ) . Some of these entry-side movement traces also showed a small apparent decrease in FRET ( ΔFRET ~ 0 . 1 ) prior to the FRET increase , but this apparent FRET decrease resulted primarily from an increase in Cy3 intensity without a corresponding decrease in Cy5 intensity ( Figure 3—figure supplement 1A-F ) . The magnitude of this Cy3 intensity change was consistent with the changes seen when we labeled the two H2A subunits on the octamer with Cy3 and Cy5 ( i . e . the H2A/H2A construct ) ( Figure 3—figure supplement 1G ) , suggesting that this FRET change likely resulted from a change in the photophysical properties of the Cy3 dye upon RSC binding . Even if these initial FRET decreases reflected a real change in distance between the Cy3 and Cy5 dyes , the magnitude of the FRET change , ΔFRET ~ 0 . 1 ( Figure 3—figure supplement 1D-F ) , was much smaller than both the FRET changes that we observed for RSC-induced DNA translocation around the nucleosome and the FRET changes previously observed for transcription factor-mediated DNA unwrapping from the edge of the nucleosome ( ΔFRET ~ 0 . 6 ) on a similarly labeled nucleosome construct ( Li and Widom , 2004 ) . Assuming a previously measured Förster radius of 6 nm for the Cy3-Cy5 pair ( Murphy et al . , 2004 ) , these small initial FRET decreases that we observed would correspond to a change in the Cy3-Cy5 distance of only ~0 . 5 nm . Thus , our data suggest that under our remodeling conditions , the DNA was not lifted by a large distance away from the nucleosomal surface and did not deviate substantially from its canonical wrapping path as it entered the nucleosome . In order to determine whether the DNA was unwrapped and lifted by a large distance off the nucleosomal surface at the nucleosomal edge where DNA exits , our current labeling scheme was not adequate since both lifting of DNA off the nucleosomal surface and translocation of DNA along the canonical path on the nucleosome would generate a decrease in FRET . We therefore moved the Cy5 label from the end of the DNA to a site 15 bp inside the edge of nucleosomes by incorporating the dye into the sugar-phosphate backbone to generate H2A/[backbone , –15] nucleosomes , where H2A again indicates the position of the Cy3 label and [backbone , –15] indicates the position of the Cy5 label ( Figure 3A ) . If RSC translocates the DNA around the nucleosome along its canonical path , the exit-side movement traces should show an initial FRET increase as the Cy5 moves toward the nucleosome edge , followed by a FRET decrease as the Cy5 exits the nucleosome . In contrast , lifting the DNA by a large distance off the nucleosome surface should produce a substantial FRET decrease , as this structural change would move the Cy5 dye on the DNA farther away from the Cy3 dye on the octamer , assuming that the unwrapped DNA remains largely unbent . Addition of RSC and ATP to this construct again generated two classes of traces: exit-side movement traces showing a transient increase in FRET followed by a decrease to zero FRET ( Figure 3B ) and entry-side movement traces showing a decrease in FRET to ~0 . 23 ( Figure 3C ) . Without any ssDNA gap on the nucleosome , the FRET traces of the H2A/[backbone , –15] construct preferentially showed entry-side movement ( Figure 3D ) , likely because placing the Cy5 inside the nucleosome biased RSC to bind in the orientation that positioned the ATPase at the SHL–2 site . Nevertheless , when we introduced a 2-nt ssDNA gap at the SHL–2 site , nearly all of the traces showed the exit-side movement behavior , displaying an initial increase in FRET followed the decrease to zero FRET; when a 2-nt ssDNA gap was introduced at the SHL+2 site , nearly all of the traces showed the entry-side movement behavior , displaying only a monotonic decrease in FRET ( Figure 3D ) . These experiments confirm that the traces showing the initial FRET increase followed by a FRET decrease resulted from ATPase action at the SHL+2 site and thus represented exit-side movement . Because lifting of the DNA by a large distance away from the nucleosomal surface would have caused a FRET decrease instead of an initial FRET increase , our data suggest that the DNA was not lifted away from the nucleosome surface by a large distance and did not deviate substantially from its canonical path as it exited the nucleosome . 10 . 7554/eLife . 10051 . 011Figure 3 . Assay for probing DNA unwrapping at the nucleosomal edge where DNA exits the nucleosome . ( A ) Diagram of the H2A/[backbone , –15] nucleosome construct . ( B and C ) Representative FRET traces showing exit-side movement ( B ) and entry-side movement ( C ) . [RSC] = 1 nM , [ATP] = 5 µM . ( D ) The fraction of entry-side movement ( orange ) and exit-side movement ( cyan ) traces observed with 1 nM RSC and 20 µM ATP for nucleosomes lacking any ssDNA gap , nucleosomes with a 2-nt ssDNA gap at the SHL–2 site , and nucleosomes with a 2-nt ssDNA gap at the SHL+2 site . Error bars represent the standard error from >100 nucleosomes per construct . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 011 10 . 7554/eLife . 10051 . 012Figure 3—figure supplement 1 . Assays for probing DNA unwrapping at the nucleosomal edge where DNA enters the nucleosome . ( A–C ) Top: cartoons representing the H2A/[end , +6] ( A ) , H2A/[end , +11] ( B ) , and H3/[end , +9] ( C ) constructs used in each experiment . Bottom: Representative Cy3 intensity , Cy5 intensity , and FRET time traces showing entry-side movement . The region showing the transient FRET decrease is highlighted in yellow . [RSC] = 1 nM , [ATP] = 5 µM . ( D–F ) Histograms showing the starting FRET ( blue ) and FRET during the transient FRET decrease ( orange ) for the H2A/[end , +6] ( D ) , H2A/[end , +11] ( E ) , and H3/[end , +9] ( F ) constructs . ( G ) The mean ratio of the Cy3 signals before and after the transient FRET decrease ( green ) and the mean ratio of the Cy5 signals before and after the transient FRET decrease ( red ) for the H2A/[end , +6] construct . The data for the H2A/H2A constructs from Figure 2—figure supplement 1B are replotted for comparison . Error bars represent the SEM from >100 nucleosomes per construct . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 012 Although our data are not consistent with RSC moving the DNA at the nucleosomal edges by a large distance away from the octamer surface , as would be expected if RSC were to unwrap a substantial amount of DNA from the edge and allow the DNA to follow its unconstrained path , our data do not exclude the possibility that the enzyme disrupts many of the histone-DNA contacts simultaneously during remodeling ( Lorch et al . , 2010 ) , but still holds the unwrapped or lifted DNA near the surface of the nucleosome . Next , we characterized the step size with which RSC translocated DNA into and out of the nucleosome . To determine the step size of DNA translocation at the nucleosomal edge where DNA exits the nucleosome , we first monitored remodeling of the H2A/[end , +6] nucleosomes at 20°C ( as opposed to 30°C in the previous experiments ) with 2 µM ATP in order to slow the remodeling reaction and better resolve the translocation steps . The exit-side movement traces exhibited intermittent pauses interrupting the monotonic decrease to zero FRET ( Figure 4A ) , and we applied a step-finding algorithm based on chi-square minimization ( Kerssemakers et al . , 2006 ) to identify the location of the pauses and the sizes of the steps . The step size histogram showed a distribution of step sizes with a major peak centered at a ΔFRET of 0 . 1 and a tail extending to larger ΔFRET values ( Figure 4B ) . Using an alternative , hidden Markov model ( HMM ) -based algorithm ( McKinney et al . , 2006 ) , we identified similar steps in the FRET traces and produced a similar step size histogram , again with the major peak centered at a ΔFRET of 0 . 1 ( Figure 4—figure supplement 1 ) . We note that our results cannot completely exclude the possibility that a small fraction of nucleosomes undergo large-step movement that brings the FRET to zero in a single step as these traces would not be distinguishable from Cy5 photobleaching . However , during remodeling of the H2A/[backbone , –15] construct ( Figure 3 ) , where exit-side movement led to an increase in FRET that can be distinguished from Cy5 photobleaching , the vast majority ( 85% ) of the exit-side movement traces showed gradual FRET changes inconsistent with such large step sizes , suggesting that at most a small fraction of nucleosomes could remodel with such large-step movement . 10 . 7554/eLife . 10051 . 013Figure 4 . DNA exits the nucleosome in a stepwise manner , exhibiting a step size distribution peaked at ~1–2 bp during RSC-mediated remodeling . Remodeling was monitored for H2A/[end , +6] ( A and B ) , H3/[end , +6] ( C and D ) , and H2A/[backbone , +6] ( E and F ) nucleosome constructs . ( A , C , and E ) Top: diagram of the nucleosome construct used . Bottom: Representative exit-side movement traces in the presence of 5 nM RSC and 2 µM ATP at 20°C . Light grey , raw FRET data; blue , 5-point median-filtered data; red , fit by a step-finding algorithm based on Chi-square minimization . ( B , D , and F ) Histograms of the measured step sizes in FRET change ( blue bars ) and the fit to the modeled step size distribution shown in Equation 1 ( black line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 013 10 . 7554/eLife . 10051 . 014Figure 4—figure supplement 1 . Step size determination using a hidden Markov model ( HMM ) -based step-finding algorithm . ( A ) The exit-side movement trace from Figure 4A is reproduced showing the steps identified by the HMM-based step-finding algorithm . Light grey , raw FRET data; blue , 5-point median-filtered data; Red , HMM fit . ( B ) Histogram of the measured step sizes in FRET change ( blue bars ) identified by the HMM algorithm and the fit to the modeled step size distribution shown in Equation 1 ( black line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 014 10 . 7554/eLife . 10051 . 015Figure 4—figure supplement 2 . Calibration of FRET values as a function of the linker DNA length for constructs monitoring exit-side movement . ( A , C , and E ) Distribution of FRET values measured with 5 nM RSC and 2 µM ATP at 20°C for the H2A/[end , +n] ( A ) , H3/[end , +n] ( C ) , and H2A/[backbone , +n] ( E ) constructs before ( blue ) and after ( red ) the first observed step of FRET change . ( B , D , and F ) Plots of the mean FRET value versus linker DNA length ( n ) for the H2A/[end , +n] ( B ) , H3/[end , +n] ( D ) , and H2A/[backbone , +n] ( F ) constructs before ( blue ) and after ( red ) the first observed step of FRET change . The mean and standard deviation of each point were obtained by fitting the FRET distributions to a Gaussian curve . The slopes were obtained by linear regression and the error is the standard error . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 015 10 . 7554/eLife . 10051 . 016Figure 4—figure supplement 3 . Analysis of DNA translocation step sizes of exit-side movement . ( A ) Table showing the estimated step size ( ± standard error ) and fraction of steps missed ( f ) from fits of the ΔFRET step size histograms in Figure 4 and Figure 4—figure supplement 4 to the stepping model in Equation 1 . These ΔFRET values are converted to step sizes in bp by dividing the FRET step sizes by the slopes of the calibration curves in Figure 4—figure supplement 2 . ( B ) Histograms of the observed pause lifetimes for the H2A/[end , +6] construct during RSC-induced remodeling at 2 µM , 20 µM and 80 µM ATP , collected with camera frame rates of 1 Hz , 8 Hz , and 16 Hz , respectively . ( C ) Histograms of the observed pause lifetimes identified for the H3/[end , +6] and H2A/[backbone , +6] constructs during RSC-induced remodeling at 2 µM ATP . The mean pause lifetime ( τ ) is estimated by fitting the distribution to an exponential , and the expected fraction of missed events ( f ) is the fraction of pauses expected to have a pause lifetime < 5 frames ( the threshold for detecting a step in the step-size analysis ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 016 10 . 7554/eLife . 10051 . 017Figure 4—figure supplement 4 . Monitoring exit-side DNA motion at higher ATP concentrations . ( A ) Cartoon of the H2A/[end , +6] construct used in the experiments . ( B and C ) Analysis of exit-side movement from remodeling of H2A/[end , +6] nucleosomes . Reactions were performed at 20°C in the presence of 5 nM RSC and 20 µM ATP ( B ) or 5 nM RSC and 80 µM ATP ( C ) . Data were collected at 8 Hz ( B ) or 16 Hz ( C ) , as opposed to the 1 Hz rate used in the 2 µM ATP experiments . Left: Representative FRET traces . Light grey , raw FRET data; blue , 5-point median-filtered data; red , fit by the step-finding algorithm . Right: Histograms of the measured step sizes in FRET change ( blue bars ) and the fit to the modeled step size distribution in Equation 1 ( black line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 017 Next , we determined the step sizes of DNA translocation assuming that the DNA moved along its canonical path around the nucleosome . To calibrate the FRET changes associated with moving the DNA along its canonical path , we positioned Cy5 at the end of the linker DNA and varied the linker DNA length ( Figure 4—figure supplement 2A , blue histograms ) . Consistent with previous results ( Blosser et al . , 2009 ) , these measurements showed an approximately linear relationship between the linker DNA length and the observed FRET value over the measured range with a slope of 0 . 053 ± 0 . 004 bp−1 ( Figure 4—figure supplement 2B , blue ) . Because the photophysical effects associated with enzyme binding may affect this relationship , we also performed the calibration on enzyme-bound nucleosomes by measuring the FRET value of the first observed step during remodeling of these constructs ( Figure 4—figure supplement 2A , red histograms ) as a function of the initial linker DNA length . These experiments also showed a linear relationship with a similar slope of 0 . 047 ± 0 . 005 bp−1 ( Figure 4—figure supplement 2B , red ) . From these slopes , we determined that the major step size of DNA movement , corresponding to the peak of the ΔFRET distribution in Figure 4B ( ΔFRET ~ 0 . 1 ) , to be ~2 bp . The step size histogram , however , showed many steps that were larger than expected for ~2 bp of DNA motion ( Figure 4B ) . Although these data could reflect a heterogeneous step size , the distribution could also result from movement with a single step size , where the larger steps represented the sum of two or more steps where the intervening pause ( s ) were too short to detect . To test whether our data were consistent with such a model , we took advantage of the linear dependence of FRET on the linker DNA length and fit our data to a series of evenly spaced Gaussian peaks whose amplitude decreased by a factor of f from the previous peak , representing the probability of missing a step ( see Materials and methods ) . The data derived from the step-finding algorithm based on chi-square minimization are consistent with a step size of ΔFRET = 0 . 104 ± 0 . 002 with a probability of missing a step being f = 34% ( Figure 4B and Figure 4—figure supplement 3A ) . Analyzing the step size distribution generated from the alternative HMM-based step-finding algorithm produced a step size of ΔFRET of 0 . 102 ± 0 . 003 with a missed fraction of f = 41% ( Figure 4—figure supplement 1B ) . Given the similar results from the two analysis methods , we used only the chi-square-minimization-based step-finding algorithm in subsequent analyses . The probabilities of missing a step obtained from these fits are reasonable given that the steps occur stochastically , and many pauses could be too short to be observed . Based on the distribution of observed pause durations , we would expect a 25% probability of missing a step due to our limited time resolution ( Figure 4—figure supplement 3B ) . The observed probability of missing a step was moderately larger than that expected from our time resolution limitation , suggesting the possibility of additional mechanisms for missing steps , for example , successive DNA translocation steps generated by the ATPase at the SHL2 site merging together while transiting to the edge of the nucleosome . To test whether the step sizes are sensitive to ATP concentration , we determined the step size of the enzyme at two additional concentrations of ATP , 20 µM and 80 µM by performing the experiments with a faster camera frame rate and increased laser intensities ( Figure 4—figure supplement 4 ) . The FRET from the exit-side movement traces again decreased in a stepwise manner , and the step size distributions exhibited a major peak at a ΔFRET of ~0 . 1 with a tail extending to larger values ( Figure 4—figure supplement 4 ) . Fitting the resulting step size histograms to multiple Gaussian peaks as described earlier indicated that the distributions are consistent with a step size of 1 . 7 ± 0 . 2 bp with a missed step probability of f = 26% for 20 µM ATP and 2 . 0 ± 0 . 2 bp with a missed step probability of f = 41% for 80 µM ATP ( Figure 4—figure supplement 3A , B ) . These step size estimates at higher ATP concentrations are similar to the results observed at 2 µM ATP . To test whether the step size determination was sensitive to the labeling scheme , we monitored remodeling ( at 2 µM ATP ) with a second nucleosome construct where we moved the Cy3 dye from histone H2A to histone H3 ( the H3/[end , +6] construct , Figure 4C , D ) and a third construct where we moved the Cy5 to the middle of the linker DNA on the DNA backbone ( instead of at the end of the linker DNA ) 6 bp from the nucleosome edge ( the H2A/[backbone , +6] construct , Figure 4E , F ) . Calibration curves for these two labeling schemes showed that the slopes of the FRET versus linker DNA length were 0 . 059 ± 0 . 005 bp−1 and 0 . 057 ± 0 . 002 bp−1 , respectively ( Figure 4—figure supplement 2C-F , blue ) . Like for the H2A/[end , +6] construct , these values did not change appreciably in the presence of a bound RSC enzyme ( Figure 4—figure supplement 2C-F , red ) . The exit-side movement traces of these new constructs also showed stepwise DNA translocation ( Figure 4C , E ) , and the step size distributions ( Figure 4D , F ) showed a major peak around a ΔFRET of 0 . 1 with a tail extending to larger values . Fitting these histograms to multiple Gaussian peaks as described earlier indicated that the distributions were consistent with a step size of ΔFRET = 0 . 113 ± 0 . 003 with a missed step probability of f = 34% for H3/[end , +6] and ΔFRET = 0 . 092 ± 0 . 004 and f = 42% for H2A/[backbone , +6] ( Figure 4—figure supplement 3A , C ) . Comparing the ΔFRET values to the slopes of the calibration curves gave step size estimates of 1 . 9 ± 0 . 2 bp and 1 . 6 ± 0 . 1 bp for the H3/[end , +6] and H2A/[backbone , +6] constructs , respectively , consistent with the 2 . 0 ± 0 . 2 bp step size estimate for the H2A/[end , +6] construct . Again , because some of the estimated missed step probabilities were moderately larger than would be expected based on our time resolution ( Figure 4—figure supplement 3 ) , the DNA may have occasionally exited the nucleosome with larger step sizes . Next , we characterized the step size of DNA translocation at the nucleosomal edge where DNA enters the nucleosome . In the experiments that characterized the step sizes of exit-side motion , the dye-labeled linker DNA was “upstream” ( to the left ) of the 601 sequence ( Figure 4A , C , E ) , so exit-side movement traces resulted from ATPase action at the SHL+2 site . To maintain the same direction of DNA translocation around the nucleosome but move the dye locations to monitor the movement of DNA into the nucleosome , we made H2A/[end , +12] and H3/[end , +9] constructs where the dye-labeled linker DNA lied “downstream” ( to the right ) of the 601 sequence ( Figure 5A , B ) . The entry-side movement traces measured at 2 µM ATP primarily showed an increase in FRET as expected from the movement of the short linker DNA into the nucleosome , bringing the Cy5 dye closer to Cy3 . As before , the entry-side movement traces showed an initial , small apparent decrease in FRET before the major FRET increase , likely due to the photophysical effect on the dye upon enzyme binding ( Figure 3—figure supplement 1 ) . The FRET increase occurred in a stepwise manner , indicating stepwise DNA translocation into the nucleosome ( Figure 5A , B ) . The distributions of the ΔFRET step sizes again showed a major peak around ~0 . 1 with a tail extending to larger values ( Figure 5C , D ) . Fitting to multiple evenly spaced Gaussian peaks as described earlier yielded a step size of ΔFRET = 0 . 088 ± 0 . 002 with a missed step probability of f = 22% for the H2A/[end , +12] construct and ΔFRET = 0 . 084 ± 0 . 001 and f = 17% for the H3/[end , +9] construct ( Figure 5C , D and Figure 5—figure supplement 2A , B ) . Assuming that the DNA moved along the canonical nucleosomal wrapping path as it entered the nucleosome , we constructed calibration curves for the two labeling schemes by varying the initial linker DNA lengths and measured slopes of the FRET change versus linker DNA length to be 0 . 055 ± 0 . 002 bp−1 for the H2A/[end , +12] construct and 0 . 048 ± 0 . 004 bp−1 for the H3/[end , +9] construct , which again did not change appreciably in the presence of bound enzyme ( Figure 5—figure supplement 1 ) . Based on these slopes , the step size distributions were consistent with step sizes of 1 . 6 ± 0 . 1 and 1 . 7 ± 0 . 2 bp , respectively , for these constructs . 10 . 7554/eLife . 10051 . 018Figure 5 . DNA enters the nucleosome in a stepwise manner , exhibiting a step size distribution peaked at ~1–2 bp during RSC-mediated remodeling . Remodeling was monitored for H2A/[end , +12] ( A and C ) , and H3/[end , +9] ( B and D ) constructs . ( A and B ) Top: diagram of the nucleosome constructs used . Bottom: Representative entry-side movement traces in the presence of 5 nM RSC and 2 µM ATP at 20°C . Light grey , raw FRET data; blue , 5-point median-filtered data; red , fit by the step-finding algorithm . ( C and D ) Histograms of the measured step sizes in FRET change ( blue bars ) and the fit to the modeled step size distribution shown in Equation 1 ( black line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 018 10 . 7554/eLife . 10051 . 019Figure 5—figure supplement 1 . Calibration of FRET values as a function of the linker DNA length for constructs monitoring entry-side movement . ( A and B ) Distribution of FRET values before ( blue ) and after ( red ) the first observed step of FRET change in the presence of 5 nM RSC and 2µM ATP at 20°C for the H2A/[end , +n] ( A ) and H3/[end , +n] ( B ) constructs . ( C and D ) Plots of the mean FRET versus linker DNA length ( n ) for the H2A/[end , +n] ( C ) and H3/[end , +n] ( D ) constructs before ( blue ) and after ( red ) the first observed step of FRET change . The mean and standard deviation for each point were obtained by fitting the FRET distributions to a Gaussian curve . The slopes were obtained by linear regression and the error is the standard error . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 019 10 . 7554/eLife . 10051 . 020Figure 5—figure supplement 2 . Analysis of DNA translocation step sizes of entry-side movement . ( A ) Table showing the estimated step size ( ± standard error ) and fraction of steps missed ( f ) from fits of the ΔFRET step size histograms in Figure 5 and Figure 5—figure supplement 3 to the stepping model in Equation 1 . These ΔFRET values are converted to step sizes in bp by dividing the FRET step sizes by the slopes of the calibration curves in Figure 5—figure supplement 1 . ( B ) Histograms of the observed pause lifetimes identified by the step-finding algorithm for the H2A/[end , +12] and H3/[end , +9] constructs during RSC-induced remodeling at 2 µM ATP . ( C ) Histograms of the observed pause lifetimes identified for the H2A/[end , +12] construct during RSC-induced remodeling at 20 and 80 µM ATP . Experiments with 2 , 20 , and 80 µM ATP were collected with camera frame rates of 1 Hz , 8 Hz , or 16 Hz , respectively . The mean pause lifetime ( τ ) is estimated by fitting the distribution to an exponential , and the expected fraction of missed events ( f ) is the fraction of pauses expected to have a pause lifetime < 5 frames ( the threshold for detecting a step in the step-size analysis ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 020 10 . 7554/eLife . 10051 . 021Figure 5—figure supplement 3 . Monitoring entry-side DNA motion at higher ATP concentrations . ( A ) Cartoon of the H2A/[end , +12] nucleosomes used in the experiments . ( B and C ) Analysis of entry-side movement from remodeling of H2A/[end , +12] nucleosomes . Reactions were performed at 20°C in the presence of 5 nM RSC and 20 µM ATP ( B ) or 80 µM ATP ( C ) . Data were collected at 8 Hz ( B ) or 16 Hz ( C ) , as opposed to the 1 Hz rate used in the 2 µM ATP experiments . Left: Representative FRET traces . Light grey , raw FRET data; blue , 5-point median-filtered data; red , fit by the step-finding algorithm . Right: Histograms of the measured step sizes in FRET change ( blue bars ) and the fit to the modeled step size distribution in Equation 1 ( black line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10051 . 021 Finally , to test the dependence of the entry-side step size on the ATP concentration , we monitored remodeling of the H2A/[end , +12] construct at 20 µM and 80 µM ATP , again by increasing the time resolution of our measurements . The entry-side movement traces from these experiments also showed a stepwise increase in FRET ( Figure 5—figure supplement 3 ) . The resulting step size distributions were consistent with step sizes of 2 . 0 ± 0 . 1 bp at 20 µM ATP and 1 . 9 ± 0 . 2 bp at 80 µM ATP ( Figure 5—figure supplement 2A ) , again similar to the results measured at 2 µM ATP . Again , the observation that some of the estimated missed step probabilities were larger than would be expected based on our time resolution ( Figure 5—figure supplement 2 ) suggests that the DNA may occasionally enter the nucleosome with larger step sizes . Therefore , our results from experiments using a number of different labeling schemes and ATP concentrations suggested that RSC translocated DNA primarily in ~1–2 bp increments at both the entry side and exit side of the nucleosome . In this study , we used single-molecule FRET to monitor nucleosome remodeling by a prototypical SWI/SNF-family remodeler , RSC . This approach enabled us to track the motion of DNA across individual nucleosomes in real time , providing new insight into the mechanisms by which RSC repositions nucleosomes along DNA . Our results showed that RSC remodeled mononucleosomes primarily by translocating DNA around the nucleosome under our remodeling conditions ( without acceptor nucleosomes , acceptor DNA , transcription factors , chaperones , or other remodeling factors ) . At the nucleosome edges where DNA enters and exits the nucleosome , RSC did not lift the DNA by a large distance away from the nucleosomal surface , and our data are consistent with translocation of the DNA largely along or close to the canonical nucleosomal wrapping path . In some cases , we observed a small FRET decrease preceding DNA translocation ( Figure 3—figure supplement 1 ) that was likely due to changes in the photophysical properties of Cy3 upon enzyme binding . Similarly , constructs designed to report on the dynamics of the H2A-H2B dimer showed no FRET change or a small apparent FRET change likely due to photophysical changes of Cy3 upon enzyme binding . Even if these FRET changes were due to actual distance changes , rather than photophysical changes of the dye , they would still represent minor deviations ( ~0 . 5 nm ) of DNA from the canonical wrapping path around the octamer or slight repositioning of the H2A-H2B dimer relative to the rest of the octamer . However , because lifting the DNA from the surface of the octamer or displacing the H2A-H2B dimer even by this small distance could disrupt many histone-DNA or histone-histone contacts , we cannot exclude the possibility that the observed translocation of DNA around the nucleosome is associated with the disruption of a substantial number of histone-DNA or histone-histone contacts ( Lorch et al . , 2010 ) . In other words , it is possible that RSC causes DNA unwrapping or lifting of the DNA off the nucleosome , but still holds the unwrapped/lifted DNA near the surface of the nucleosome . However , such a small-distance motion would have only a minor effect on our measured FRET signal . These results are also consistent with an electron microscopy structure of the RSC complex , which suggests a tight fit of the nucleosome in the central cavity of the remodeler that would not seem to accommodate large-scale displacement of DNA from the nucleosome surface ( Leschziner et al . , 2007 ) . Thus , we characterized the step size of DNA translocation into and out of the nucleosome by assuming that the DNA moved along the canonical path around the nucleosome during RSC-mediated remodeling . We observed a step size distribution that peaked at a step size of ~1–2 bp both for entry-side and exit-side movement though movements with larger step sizes were also observed . Even if the motion of the DNA is not exactly along the canonical path , the small ΔFRET step sizes of 0 . 08–0 . 11 that we observed in the single-nucleosome remodeling traces likely correspond to a distance change of ~0 . 3–0 . 6 nm ( assuming a previously measured Förster radius of 6 nm for the Cy3-Cy5 FRET pair ( Murphy et al . , 2004 ) ) . Based on the geometry of our labeling schemes and the crystal structure of nucleosomes containing the 601 positioning sequence ( Chua et al . , 2012 ) , translocation by 1 bp should be associated with changes to the Cy3-Cy5 distance of ~0 . 3 nm on average , so our data are again consistent with a DNA translocation step size of ~1–2 bp . Although changes to the photophysical properties of the dyes or the relative orientation of the dyes could potentially affect our interpretation of the FRET changes , many of these effects would be expected to depend on the specific labeling scheme . The fact that we observed similar step-size histograms with several different attachment sites of the Cy5 dye on the DNA and different attachment sites of the Cy3 dye on the octamer suggests that the measured step sizes likely reflect the true translocation step sizes of DNA around the nucleosome during RSC-catalyzed remodeling . It has been shown previously that some SF2 ATPase-containing DNA translocases or helicases translocate DNA in 1 bp increments ( Myong et al . , 2007; Gu and Rice , 2010; Rajagopal et al . , 2010; Cheng et al . , 2011; Deindl et al . , 2013 ) , so the SF2 ATPase domain of RSC might also translocate DNA at the SHL2 site of the nucleosome in 1 bp steps . The reason that the step sizes that we observed for RSC-mediated nucleosome translocation at the nucleosomal edges were not exactly 1 bp could potentially be due to the following reasons: 1 ) missing steps due to our limited time resolution could cause two consecutive 1 bp steps to appear as a 2 bp step , 2 ) two 1 bp DNA distortions generated at the SHL2 site could merge together while transiting to the entry/exit site , and 3 ) the DNA translocation path could deviate slightly from the canonical wrapping path , making our calibration , which was determined based on canonically wrapped nucleosomes , slightly off . It has been shown previously that RSC translocates DNA at the SHL2 site where the enzyme’s ATPase domain engages the nucleosome ( Saha et al . , 2002; Saha et al . , 2005; Zofall et al . , 2006; Dechassa et al . , 2008 ) and that the isolated RSC catalytic subunit translocates naked DNA in ~2 bp increments ( Sirinakis et al . , 2011 ) . Based on these previous findings , it is reasonable to expect that the RSC complex may translocate DNA around the nucleosome in 1–2 bp steps , as has been proposed before ( Saha et al . , 2005 ) . However , such small steps had not been previously observed during nucleosome remodeling by SWI/SNF enzymes , so it was unclear whether the small DNA translocation step size by the ATPase at the SHL2 site would result in small DNA movement steps around the nucleosome , especially given that the entry and exit sites of nucleosome reside ~50 and 100 bp away from the SHL2 site . Indeed , in the context of remodeling intact nucleosomes , a previous biochemical study showed that SWI/SNF complexes translocate DNA around the nucleosome in large , ~50 bp steps ( Zofall et al . , 2006 ) . Because our current observations suggest that RSC translocates DNA around the nucleosome primarily in steps of 1–2 bp , the previously observed ~50 bp translocation steps likely do not reflect the fundamental step size of nucleosome translocation by RSC . Rather , they may be compound steps composed of many ~1–2 bp steps and may reflect relatively long kinetic pauses of DNA translocation imposed by the energy landscape of the nucleosomal substrates . However , we note that because the cavities in the RSC and SWI/SNF complexes that accommodate nucleosomes appear somewhat different ( Leschziner et al . , 2007; Chaban et al . , 2008; Dechassa et al . , 2008 ) , it is possible that SWI/SNF and RSC do not translocate DNA around the nucleosome in exactly the same manner . Furthermore , it is also worth noting that our measurements were performed at limiting ATP concentrations , up to roughly the Km value for ATP ( ~80 µM ) of RSC ( Cairns et al . , 1996 ) , and it is possible that at higher ATP concentrations , other processes could become rate limiting and alter the behavior of the complex . Nevertheless , the behavior at limiting ATP concentrations likely reflects the response of the nucleosome to the individual translocation events catalyzed by the ATPase at the SHL2 site . Because the SF2 ATPase of RSC likely translocates DNA 1 bp at a time at the SHL2 site of the nucleosome , our observation that RSC moves the DNA into and out of the nucleosome primarily in ~1–2 bp increments suggests that DNA motion across the nucleosome from the nucleosomal entry site to the nucleosomal exit site is likely coupled directly to DNA translocation at the SHL2 site by the ATPase domain . Because we tracked the motion of the DNA only at the nucleosomal edges , our data cannot determine whether an intranucleosomal loop or bulge was formed inside the nucleosome . If such loops exist , our observations of the small step sizes of DNA translocation at both entry and exit sites of the nucleosome place constraints on such internal DNA loops/bulges: these loops/bulges are either small and most often contain 1–2 bp of extra DNA; or if they are not so small , they must accumulate and release gradually , primarily 1–2 bp at a time at the nucleosome entry and exit sites , which would require a specific mechanism to extrude a large internal loop to the exit side in small steps . Finally , we note that although the nucleosome translocation step size observed here for RSC is similar to the nucleosome translocation step size previously observed for ISWI-family enzymes ( Deindl et al . , 2013 ) , these results do not necessarily suggest that the remodeling mechanisms by these two families of chromatin remodelers are entirely similar . For both enzyme families , the observed nucleosome translocation step sizes likely resulted from the intrinsic DNA translocation step size of their ATPase domains , which both belong to the SF2 helicase family . However , it has been shown previously that the SWI/SNF-family remodelers disrupt the DNA-octamer contacts to a much greater extent than the ISWI-family remodelers ( Fan et al . , 2003; Dechassa et al . , 2012 ) . Our results suggest that , despite such disruption , the RSC enzyme likely still holds the DNA close to the nucleosome and hence , such disruption does not strongly perturb the step sizes with which DNA is translocated across the nucleosome . Recombinant X . laevis histones were expressed in BL21 ( DE3 ) pLysS cells ( Promega , Madison , WI ) and purified under denaturing conditions . Briefly , we isolated inclusion bodies and extracted the histones as described previously ( Luger et al . , 1999 ) , then dialyzed the histones into buffer A ( 7 M urea , 20 mM Na-HEPES pH 8 . 0 , 100 mM NaCl , 1 mM DTT and 1 mM Na-EDTA ) . This solution was then loaded onto HiTrap-Q cation exchange and ResourceS anion exchange columns ( GE Healthcare , Pittsburgh PA ) connected in series ( Wittmeyer et al . , 2004 ) . After washing with buffer A , the HiTrap-Q column was then removed before the protein was eluted by gradually increasing the concentration of NaCl . To generate histones site-specifically labeled with Cy3 , plasmids for the expression of H2A K119C , H2B T49C , and double mutant H3 ( G33C+C110A ) were created by site-directed mutagenesis . These constructs were purified and labeled with sulfo-Cy3 maleimide or sulfo-Cy5 maleimide ( GE Healthcare ) under denaturing conditions ( Hwang et al . , 2014 ) . The histone H2A and H3 mutants have been previously used for FRET-based studies of nucleosome remodeling ( Yang et al . , 2006; Rowe and Narlikar , 2010 ) . Histone octamer was reconstituted with an ~1:1 ratio of labeled and unlabeled histone in order to maximize the yield of singly-labeled octamer and purified by gel filtration chromatography as described previously ( Luger et al . , 1999; Hwang et al . , 2014 ) . For the reconstitution of the H2A/H2A and H2B/H2B constructs , histone octamer was reconstituted with an ~1:1 ratio of Cy3-labeled and Cy5-labeled histone . DNA constructs were made by PCR from a plasmid containing a modified 601 positioning sequence ( Lowary and Widom , 1998; Partensky and Narlikar , 2009 ) . The PCR primers contained 5’ Cy5 or biotin-TEG modifications ( IDT , Coralville , IA ) to install these modifications at the indicated locations . For constructs containing gaps or backbone Cy5 labels , the DNA was constructed by annealing a set of overlapping oligonucleotides ( IDT ) and ligating them into a single double-stranded DNA construct ( Hwang et al . , 2014 ) . Backbone Cy5 labels were inserted opposite guanosine residues . DNAs were purified by PAGE . Mononucleosomes were assembled by the salt dialysis method after mixing labeled octamer and DNA at a 1 . 2:1 ratio and purified by glycerol gradient centrifugation as previously described ( Lee and Narlikar , 2001 ) . RSC was purified from S . cerevisiae strain BCY211 that expresses a TAP-tagged Rsc2 subunit following published protocols ( Saha et al . , 2002; Wittmeyer et al . , 2004 ) . Protein concentration was quantified by Sypro Red staining and comparison with BSA standards . Quartz slides were cleaned , functionalized with a 1:100 mixture of biotin-PEG:PEG ( Laysan Bio , Arab , AL ) , and assembled into flow chambers as previously described ( Joo and Ha , 2008 ) . Nucleosomes were immobilized on the biotin-PEG on the slide via streptavidin . The sample was excited with a 532 nm Nd:YAG laser ( CrystalLaser , Reno , NV ) on a custom built total internal reflection fluorescence ( TIRF ) microscope ( Deindl and Zhuang , 2012 ) . Fluorescence emission was collected with a 60x water immersion objective ( Olympus , Tokyo , Japan ) , filtered with a 550 nm long-pass filter ( Chroma Technology , Bellows Falls , VT ) , split with a 630 nm dichroic mirror ( Chroma Technology ) , and imaged onto two halves of an Andor iXon+888 EM-CCD camera ( Andor Technology , Belfast , UK ) . Imaging was done at 30°C ( unless otherwise specified ) in imaging buffer ( 40 mM Tris , 12 mM HEPES , pH 7 . 5 , 60 mM KCl , 3 mM MgCl2 , 10% ( v/v ) glycerol , 0 . 02% ( v/v ) Igepal CA-630 , 10% ( w/v ) glucose , 2 mM trolox , 0 . 1 mg/mL acetylated BSA ) , supplemented with an oxygen scavenging system ( 800 µg/mL glucose oxidase , 50 µg/mL catalase ) ( Rasnik et al . , 2006; Blosser et al . , 2009 ) . Images were collected at 1 Hz , except for the data in Figure 4—figure supplement 4 and Figure 5—figure supplement 3 which were taken at 8 Hz ( for the experiments with 20 µM ATP ) or 16 Hz ( for the experiments with 80 µM ATP ) . The field of view was illuminated with ~25–50 W/cm2 of laser light for data collected at 1 Hz . For experiments done at 8 Hz and 16 Hz , ~400 and 800 W/cm2 , respectively , was used . In experiments where the FRET acceptor dye is on the DNA and the FRET donor is on the histone ( H2A , H2B or H3 ) , we observed three populations of labeled nucleosomes: nucleosomes with a single donor dye residing on the H2A , H2B or H3 subunit proximal to the short linker , nucleosomes with a single donor dye residing on the H2A , H2B or H3 subunit distal to the short linker , and nucleosomes with two donor dye molecules ( Blosser et al . , 2009 ) . These nucleosomes can be distinguished on the basis of their FRET value , and we select only those nucleosomes with the Cy3 on the proximal H2A , H2B or H3 subunit for analysis , which gave the highest FRET values . For the H2A/H2A and H2B/H2B constructs , we selected those showing FRET > 0 . 2 and FRET > 0 . 1 , respectively , as nucleosomes lacking Cy5 show zero FRET and nucleosomes lacking Cy3 are not visible during 532 nm illumination . Single-molecule FRET traces were generated and analyzed with IDL ( ITT Visual Information Solutions , Boulder , CO ) as described previously ( Deindl and Zhuang , 2012 ) ( code available at http://zhuang . harvard . edu/smFRET . html ) . Single nucleosomes were identified by selecting traces that showed one-step photobleaching . The fluorescence intensity after photobleaching was used for background subtraction . To classify traces as entry-side or exit-side movement traces , we manually identified traces exhibiting remodeling , smoothed the data with a 3 pt median filter , then classified them in an automated manner using the following criteria: If a trace showed 3 consecutive points that were greater than the initial FRET by 0 . 1 ( 0 . 07 for the H2A/[backbone , –15] data ) before reaching the final FRET value , indicating an increase in FRET followed by FRET decrease , it was classified as an entry-side movement trace for the H2A/[end , +6] and H3/[end , +6] constructs or an exit-side movement trace for the H2A/[backbone , –15] construct . If the trace reached the final FRET value without such an initial increase , indicating a monotonic FRET decrease , the trace was classified as an exit-side movement traces for the H2A/[end , +6] and H3/[end , +6] constructs or an entry-side movement trace for the H2A/[backbone , –15] construct . In the experiments to determine the step size of remodeling , pauses in the FRET traces were identified using a previously developed step-finding algorithm based on Chi-square minimization ( Kerssemakers et al . , 2006 ) . We also identified steps using an alternative , hidden Markov model-based method ( McKinney et al . , 2006 ) . Only the steps identified by these algorithms that satisfy the following criteria were included for further analysis: 1 ) the pauses are not shorter than 5 frames , and 2 ) the pauses occurred at FRET values that are inside the range defined by the calibration curves in Figure 4—figure supplement 2 and Figure 5—figure supplement 1 where FRET varies linearly with distance from the nucleosome . Occasionally , the step finding algorithm identified steps in the backward direction . Because these events were rare ( accounting for less than 5% of steps ) , we did not include these steps in our analysis . To simulate the step size distribution resulting from a uniform step size and a fixed probability of missing short pauses , causing larger steps whose size is an integer multiple of the step size , the observed step size histograms were fit to the function: ( 1 ) y=∑n=16A fn-1 exp - ( x-n c ) 22 s2 where f is the probability that a pause is missed , c is the step size , and s is the standard deviation of the step size . The s parameter was fit globally across all step size histograms showing the same type of motion ( i . e . exit-side movement or entry-side movement ) . Ensemble FRET measurements were performed by monitoring the Cy5 intensity at 670 nm under 532 nm excitation in a Cary Eclipse Fluorescence Spectrophotometer ( Varian , Palo Alto , CA ) . Reactions were performed in imaging buffer , and initiated by the addition of RSC and ATP to a solution containing nucleosomes . Data were normalized by scaling the initial Cy5 intensity to 1 . 0 and the final steady state Cy5 intensity to 0 .	Cells package their genetic information in a "complex” of proteins and DNA called chromatin . This complex is made of units called nucleosomes , each of which consist of a short stretch of DNA wrapped around proteins known as histones . These nucleosomes restrict access to the DNA wrapped around the histone proteins , and thus serve to regulate whether genes are activated and a variety of other cellular processes . Certain enzymes regulate the structure of chromatin by altering the position and structure of nucleosomes . However , it is not clear exactly how these “chromatin remodeling” enzymes alter the contacts between the DNA and histone proteins to move DNA around the nucleosome . RSC is a chromatin-remodeling enzyme that typically helps to activate genes . Harada et al . used a technique called single molecule fluorescence resonance energy transfer ( or single molecule FRET for short ) to observe the movement of DNA around the histone proteins . The technique involves placing a green fluorescent dye on the histone proteins and a red fluorescent dye on the DNA . If the red dye is close to the green dye , some of the energy can be transferred from the green dye to the red dye when the green dye is excited by a laser . By looking at the ratio of green and red light emitted , it is possible to tell how far apart they are , and how this changes over time . The experiments show that the RSC enzyme moves the DNA into and out of the nucleosome in small steps . These steps match the expected step size of DNA movements by a section of the enzyme called the ATPase domain . This suggests that the ATPase domain drives the motion of DNA across the entire nucleosome . A future challenge is to better understand how chromatin remodeling enzymes cooperate with other molecules in cells to remodel nucleosomes and chromatin .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2016	Stepwise nucleosome translocation by RSC remodeling complexes
Rapid and stable control of pupil size in response to light is critical for vision , but the neural coding mechanisms remain unclear . Here , we investigated the neural basis of pupil control by monitoring pupil size across time while manipulating each photoreceptor input or neurotransmitter output of intrinsically photosensitive retinal ganglion cells ( ipRGCs ) , a critical relay in the control of pupil size . We show that transient and sustained pupil responses are mediated by distinct photoreceptors and neurotransmitters . Transient responses utilize input from rod photoreceptors and output by the classical neurotransmitter glutamate , but adapt within minutes . In contrast , sustained responses are dominated by non-conventional signaling mechanisms: melanopsin phototransduction in ipRGCs and output by the neuropeptide PACAP , which provide stable pupil maintenance across the day . These results highlight a temporal switch in the coding mechanisms of a neural circuit to support proper behavioral dynamics . Environmental light influences a variety of subconscious physiological functions , including circadian photoentrainment , light modulation of sleep/mood , and the pupillary light response ( PLR ) . These diverse effects of light are all mediated by a small subpopulation of retinal output neurons called intrinsically photosensitive retinal ganglion cells ( ipRGCs ) ( Altimus et al . , 2008; Göz et al . , 2008; Güler et al . , 2008; Hatori et al . , 2008; LeGates et al . , 2012; Lupi et al . , 2008; Tsai et al . , 2009 ) . Even in the vast array of environmental light conditions , subconscious visual behaviors are remarkable for their rapid induction and stable maintenance throughout the day . However , how the ipRGC circuit achieves rapid and stable control of visual behaviors remains uncertain . Multiple photoreceptive systems participate in the ipRGC circuit , including their endogenous melanopsin-based phototransduction and indirect synaptic input from the classical rod and cone photoreceptors ( Hattar et al . , 2003; Panda et al . , 2003 ) . Each photoreceptive system presumably encodes a unique aspect of the light environment , but to date no consensus exists on the photoreceptive mechanisms supporting ipRGC-dependent behaviors . Several studies using a variety of methods have proposed competing models arguing for the predominance of cone-based ( Allen et al . , 2011; Butler and Silver , 2011; Dkhissi-Benyahya et al . , 2007; Lall et al . , 2010 ) or rod-based ( Altimus et al . , 2010; McDougal and Gamlin , 2010 ) synaptic input to ipRGCs and their behavioral responses . Additionally , it has been suggested that melanopsin mediates persistent light detection in ipRGCs ( Altimus et al . , 2008; Gooley et al . , 2012; Lupi et al . , 2008; Mrosovsky and Hattar , 2003; Zhu et al . , 2007 ) because melanopsin phototransduction is relatively slow to initiate but stable for minutes to hours ( Berson et al . , 2002; Gooley et al . , 2012; Wong , 2012 ) . However , animals lacking melanopsin still retain sustained light responses in ipRGCs and their central targets ( Schmidt et al . , 2014; van Diepen et al . , 2013; Wong , 2012 ) and relatively normal circadian photoentrainment ( Panda et al . , 2002; Ruby et al . , 2002 ) and PLR ( Lucas et al . , 2003; Xue et al . , 2011 ) . In total , it remains unclear how ipRGCs utilize each distinct photoreceptive input , especially across the environmental range of light intensities and durations . ipRGCs must faithfully relay information about the light environment to the brain . Many neurons , including ipRGCs , release multiple neurotransmitters , a classical neurotransmitter and one or more neuropeptides ( Vaaga et al . , 2014 ) . However , methods to evaluate mammalian cotransmitter systems in vivo in real time are lacking . ipRGCs contain the principal excitatory neurotransmitter glutamate and the neuropeptide PACAP ( pituitary adenylyl cyclase-activating polypeptide ) ( Engelund et al . , 2010; Hannibal et al . , 2002 ) . Recent studies have suggested that glutamate is the predominant regulator of ipRGC-dependent behaviors , including circadian photoentrainment and the PLR ( Delwig et al . , 2013; Gompf et al . , 2015; Purrier et al . , 2014 ) . By comparison , animals lacking PACAP or its receptors show at best minor deficits in circadian photoentrainment and the PLR ( Beaulé et al . , 2009; Colwell et al . , 2004; Engelund et al . , 2012; Kawaguchi et al . , 2010 , 2003 ) . This difference in outcomes between glutamate and PACAP has led to the conclusion that PACAP is dispensable and serves primarily as a modulator of glutamatergic signaling ( Chen et al . , 1999 ) . It remains puzzling why ipRGCs , like many other neuronal cell types , would possess two distinct neurotransmitters . To date , the precise behavioral contributions of rod , cone , and melanopsin input or their output neurotransmitters glutamate and PACAP to visual behaviors across time are essentially unknown . Here , we have systematically addressed the behavioral contributions of all three photoreceptive inputs and both neurotransmitter outputs of ipRGCs , and how these change with time . To do so , we have silenced each individual photoreceptor or neurotransmitter component of ipRGCs , and in multiple combinations , while measuring pupil size across environmental light intensities and time domains . We have taken advantage of the fact that the PLR provides the unique opportunity to dissect the precise temporal dynamics of inputs and outputs of the ipRGC circuit in a behaving animal . This study reveals how ipRGC circuit dynamics in vivo support pupil regulation across time and provides insights into ipRGC regulation of other subconscious visual behaviors . To measure ipRGC responses in real time , we measured the pupillary light response ( PLR ) . Importantly , we used a novel experimental setup that mimics environmental light using overhead light with spectral composition similar to daylight in an unanesthetized mouse ( Figure 1A and Figure 1—figure supplement 1 ) , unlike previous studies that used monochromatic light delivered to a single eye ( Delwig et al . , 2013; Gooley et al . , 2012; Güler et al . , 2008; Kawaguchi et al . , 2010; Lall et al . , 2010; Lucas et al . , 2003 ) . 10 . 7554/eLife . 15392 . 003Figure 1 . The pupillary light response contains two phases: transient and sustained . ( A ) Approximate light intensity ranges ( lux ) at different times of day . ( B ) Transient constriction in response to a 10 lux overhead stimulus ( mean ± SD ) . Boxes contain representative pupil images at time 0 and 30 s . ( C ) Continued monitoring of pupil constriction from b for 60 min of continuous light at 5 min intervals with representative images . ( D ) Intensity-response curve for transient and sustained constriction ( 30 s and 60 min , respectively ) . Data fit with a sigmoidal curve ( n = 5 , mean ± SD ) . ( E ) Light intensity required for half-maximal constriction ( EC50 ) determined for both transient and sustained phases of the PLR . EC50 extracted from the sigmoidal curve fits for each mouse ( points are individual mice , line is mean ) . Statistical significance determined with a student’s t test . ( F ) Half-life of PLR decay at 1 , 10 , and 100 lux . Statistical significance determined by main effect of light intensity from one-way ANOVA . See also Figure 1—figure supplement 1 , Figure 1—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 00310 . 7554/eLife . 15392 . 004Figure 1—figure supplement 1 . Experimental setup and light stimulus details . ( A ) Environmental light intensity measured in lux across one day ( April 2 , 2015 ) in Baltimore , Maryland , USA . The light meter used is unable to measure light intensities below 1 lux , indicated with the gray box . Dotted lines refer to the meteorological sunrise and sunset . Data is fit with a hand-drawn curve for ease of visualization . ( B ) Mice are unanesthetized and restrained by hand under a light bulb with a broad spectrum similar to sunlight ( C ) . Spectral power is normalized to the most highly represented wavelength in sunlight . Breaking down the fraction of light into 50 nm bins for each light source , the daylight bulbs are very similar to sunlight across all wavelengths ( D ) , while incandescent bulbs lack short wavelengths and are enriched in long wavelengths . Pupils are continuously recorded in darkness and light using an infrared video camera . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 00410 . 7554/eLife . 15392 . 005Figure 1—figure supplement 2 . Negative-feedback model of PLR decay . ( A ) Diagram displaying how the negative feedback model works ( 7 s light in example ) ( See Online Methods for step-by-step explanation ) . The model assumes that packets of light information are discrete and are relayed to the PLR circuit to result in pupil constriction at later timepoints . We determined the kinetics of light information relay using a 1-s light pulse-chase . Then , we simply modulate the relative light intensity reaching the retina based on assuming continuous 1-s packets of information . At each new 1-s interval , the model samples the assumed pupil sizes currently driven by each previous packet of light information , uses the maximum value as the current pupil size , and then reduces the stimulus intensity using that pupil size . We then use this new intensity to determine constriction caused at that time . This iterates every second . ( B ) Putative kinetics of feedback’s impact on PLR at several light intensities ( 0 . 0001 , 0 . 001 , 0 . 01 , 0 . 1 , 1 , 10 , 100 , 1000 , and 10 , 000 lux ) . ( C ) Magnitude of PLR decay caused by feedback as modeled with ( D ) EC50 . Note that our model predicts minor PLR decay as a result of PLR feedback . ( E ) Experimental investigation of feedback’s role in PLR decay . Atropine was applied to the left eye to inhibit pupil constriction and thus feedback . No enhancement of sustained PLR of the right eye was observed ( paired two-tailed t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 005 Following light onset , we observed rapid pupil constriction that is maintained for the duration of the 30-s recording ( Figure 1B ) , with greater constriction under higher light intensities ( Figure 1D ) . Previous studies have noted a PLR decay during a sustained light stimulus lasting minutes ( Gooley et al . , 2012; Loewenfeld , 1993; McDougal and Gamlin , 2010 ) , prompting us to systematically monitor the pupil across a range of times and light intensities . We observed a decay in pupil constriction over time that reached a new steady state ( Figure 1C ) , resulting in two phases in the PLR: transient and sustained ( mean intensity to reach 50% constriction ( EC50 ) for transient PLR = 0 . 53 lux , sustained PLR = 7 . 9 lux ) ( Figure 1D , E ) . Because pupil constriction itself lowers the amount of light reaching the retina and therefore limits the drive to continued pupil constriction , the PLR is a form of negative feedback . To test if PLR decay is a consequence of negative feedback , we measured the effect of negative feedback both computationally and experimentally , and found that it has little role in PLR decay ( Figure 1—figure supplement 2 ) . Furthermore , we observed full PLR decay at dim light intensities ( ≤1 lux ) within the first 5 min of light stimulation ( Figure 1C , F ) , but full maintenance of pupil constriction at high light intensities ( ≥1000 lux ) , with apparently slower decay rates at higher light intensities ( half-life: ~2–5 min , Figure 1F ) . These results suggest that ipRGCs possess temporally distinct inputs and/or outputs for transient and sustained signaling . To identify the photoreceptor ( s ) inputs that contribute to transient ipRGC responses ( Figure 2A ) , we tested the PLR in mutant mouse lines that lack the function of a single photoreceptor type , leaving the function of the other photoreceptors intact ( Table 1 , for references on production and initial characterization of each line ) ; we refer to these lines as cone knockout , rod knockout , and melanopsin knockout mice . To corroborate our findings , we tested a variety of mutant mouse lines that silence each photoreceptor type in unique ways ( Table 1 ) . 10 . 7554/eLife . 15392 . 006Figure 2 . Transient input to ipRGCs is mediated by rods . ( A ) Diagram of ipRGC behavioral circuit . ( B ) Intensity-response curves of the PLR in each of the photoreceptor mutant mouse lines ( mean ± SD ) : wildtype ( n = 6 ) , Rod KO ( Gnat1-/- n = 6 ) , Melanopsin KO ( Opn4-/- n = 8 ) , and Cone KO ( Gnat2-/- n = 7 ) . Representative pupil images for each mouse line at 10 lux . ( C ) Gene schematic comparison of endogenous mouse M-cone allele and human red cone knock-in allele as well as the spectral sensitivity shift observed . Notice that cones are more sensitive to red light in Red cone KI line . ( D ) The PLR to red light ( 626-nm LED ) is identical in mice with cones that are more sensitive to red light ( Red cone KI , n = 6 ) compared to littermate WT ( n = 5 ) , mean ± SD . ( E ) Removing rod function abolishes the PLR in response to red light ( 626-nm LED ) , even in mice with cones with enhanced sensitivity to red light . WT n = 7 , Red cone KI ( Opn1mwred ) n = 8 , Rod KO ( Gnat1-/- ) - n = 8 , Red cone KI; Rod KO ( Gnat1-/-; Opn1mwred ) n = 4 . Light intensity is 14 . 3 log photons/cm2/s . ( F ) Intensity-response curves in mutant mice with each photoreceptor isolated ( Rod-only: Cnga3-/-; Opn4-/- n = 6 ) ( Cone-only: ( Gnat1-/-; Opn4-/- n = 6 ) ( Mel . -only: Gnat1-/-; Gnat2-/- n = 7 ) Data is mean ± SD , statistical significance determined using a one-way ANOVA with Sidak’s post-test . ( right ) Representative pupil images at 100 lux . ( G ) Kinetics of transient pupil constriction ( 100 lux ) in mice with only rod , cone , or melanopsin function , same genotypes and number of animals as in F . Traces of individual mice are shown behind curve-fits . One-phase decays were fit to all except cone-only which was fit with a two-phase decay due to its rapid pupil decay within 30 s . Melanopsin-only kinetic fit was offset from 0 by 3 s to account for delay in constriction . See also Figure 2—figure supplements 1–5 . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 00610 . 7554/eLife . 15392 . 007Figure 2—figure supplement 1 . Dark-adapted pupil sizes of photoreceptor mutant mouse lines used . Dark-adapted pupil sizes of all mouse lines used for photoreceptor investigation . Pupil size was recorded before light onset and pupil area ( mm2 ) is reported . No statistical difference was found for any genotype compared to wildtype ( p>0 . 05 for all comparisons ) . Statistical significance was determined by one-way ANOVA followed by Sidak’s post-test . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 00710 . 7554/eLife . 15392 . 008Figure 2—figure supplement 2 . Rods are required for the transient phase of the PLR . ( A ) Diagram of the retina labeling the photoreceptors . For experiments in B–D , WT n = 14 , Opn4-/- n = 8 , Cnga3-/- n = 4 , Gnat2-/- n = 7 , Cone-DTA n = 7 , Gnat1-/- n = 6 , Rod-DTA n = 9 . ( B ) Kinetics of rapid constriction in response to dim light ( 10 lux ) . Rod KO mice are the only photoreceptor mutants to display a deficit . Cone and Mel . KO mice are identical to wildtype . ( C ) Intensity-response curves of the PLR in each of the photoreceptor mutant mouse lines ( mean ± SD ) . The bar on top of the figure denotes the estimated sensitivities of rods and cones . ( D ) Rod mutant animals are the only mutants that display a sensitivity ( EC50 ) deficit compared to WT ( p<0 . 0001 ) . In fact , Cone-DTA mice are moderately more sensitive than WT ( p=0 . 011 ) . Points indicate individual mice , line indicates mean . Statistical significance determined using a one-way ANOVA with Sidak’s post-test . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 00810 . 7554/eLife . 15392 . 009Figure 2—figure supplement 3 . Melanopsin is not required for transient PLR in response to environmentally relevant overhead light . Transient PLR determined under 3 different experimental light conditions . ( Left ) Blue ( 474-nm ) LED light presented to contralateral eye ( 1 . 9 × 1016 photons/cm2/s ) . ( Middle ) White halogen light presented to contralateral eye ( 27 . 58 W/m2 ) . ( Right ) 1000 lux white compact fluorescent light presented overhead to both eyes ( 4 . 4 W/m2 ) . Line represents mean and points are individual mice . Statistical significance determined by one-way ANOVA followed by Sidak’s post-test . No difference observed when light presented overhead . Control ( Opn4+/- ) n = 7 and Opn4-/- n = 9 . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 00910 . 7554/eLife . 15392 . 010Figure 2—figure supplement 4 . Rod input to the transient PLR is influenced by cones . ( A ) Cartoon representation of a cone and a diagram of its phototransduction cascade . Different aspects of this cascade are disrupted in the various ‘rod-only’ lines we use . ( B ) Multiple mouse lines with rods as the only functional photoreceptors . For the experiments in C and D: WT n = 6 , Rod-only type 1 ( RO1: Cnga3-/-; Opn4-/- ) n = 6 , Rod-only type 2 ( RO2: Gnat2-/-; Opn4-/- ) n = 8 , Rod-only type 3 ( RO3: Cone-DTA; Opn4-/- ) n = 5 . ( C ) Intensity-response curve of the PLR in all of the rod-only lines , which are all similar to wild-type at all light intensities ( mean ± SD ) . At 1000 lux , only RO2s are statistically different from wildtype ( p=0 . 006 by one-way ANOVA with Sidak‘s post-test ) . ( D ) Sensitivity ( EC50 ) in each of the mutant lines . No statistical differences were observed between the mouse lines ( compared to WT , RO1 p=0 . 956 , RO2 p=0 . 340 , RO3 p=0 . 141 using a one-way ANOVA with Sidak’s post-test ) , although the RO2 line had more variability and trended toward lower sensitivity . ( E and F ) Kinetic comparison of rod-only lines at dim ( E ) and bright ( F ) light intensities . RO1 and RO3 lines are identical to wildtype under both light intensities , however , RO2 mice display PLR decay within 30s . All statistics are one-way ANOVA with Sidak’s post-test , line indicates mean . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 01010 . 7554/eLife . 15392 . 011Figure 2—figure supplement 5 . Melanopsin can drive rapid constriction at high light intensities . Multiple mouse lines with ipRGCs as the only functional photoreceptors ( melanopsin-only ) or a mouse line with cones as the only functional photoreceptors ( cone-only ) were tested . For the experiments in A and B: WT n = 9 , Gnat1-/-n = 10 , Melanopsin-only type 1 ( MO1: Gnat1-/-; Cnga3-/- ) n = 7 , Melanopsin-only type 2 ( MO2: Gnat1-/-; Gnat2-/- ) n = 9 , Melanopsin-only type 3 ( MO3: Rod-DTA; Cone-DTA ) n = 6 , Cone-only ( Gnat1-/-; Opn4-/- ) n = 6 . ( A ) Intensity-response curve of the PLR in all of the melanopsin-only lines and in the cone-only mouse line ( mean ± SD ) . ( B ) EC50 in each of the lines . All mutant lines are less sensitive than WT ( p<0 . 0001 ) by >2 log units . Cone-only mice are additionally less sensitive than Rod KO mice ( p<0 . 0001 ) , but no melanopsin-only line is significantly different from Rod KO ( Compared to RKO: MO1 p=0 . 201 , MO2 p=0 . 625 , MO3 p=0 . 591 ) . All statistics are one-way ANOVA with Sidak’s post-test , line indicates mean . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 01110 . 7554/eLife . 15392 . 012Table 1 . Description of photoreceptor mutant mouse lines used . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 012Mouse lineGenotypeEffect on retinal functionCitationsRod KOGnat1-/-No rod phototransduction ( Calvert et al . , 2000 ) Rod-DTArdtaNo rod cell bodies; cones present early in lifeCone KO1Cnga3-/-No cone phototransduction ( Biel et al . , 1999 ) Cone KO2Gnat2cpfl3/cpfl3No cone phototransduction ( Chang et al . , 2006 ) Cone-DTAh . red DT-AAblation of all M cones; >95% loss of S cones ( Soucy et al . , 1998 ) Melanopsin KOOpn4-/-No melanopsin phototransduction ( Lucas et al . , 2003 ) Cone-onlyGnat1-/-; Opn4-/-No rod/melanopsin phototransductionRod-only 1Cnga3-/-; Opn4-/-No cone/melanopsin phototransductionRod-only 2Gnat2-/-; Opn4-/-No cone/melanopsin phototransductionRod-only 3h . red DT-A; Opn4-/-No cone cells nor melanopsin phototransductionMelanopsin-only 1Gnat1-/-; Cnga3-/-No rod/cone phototransductionMelanopsin-only 2Gnat1-/-; Gnat2-/-No rod/cone phototransductionMelanopsin-only 3rdta; h . red DT-ANo rod or cone cell bodiesRed cone KIOpn1mwredCones have shifted sensitivity to red ( Smallwood et al . , 2003 ) Red cone KI; Rod KOOpn1mwred;Gnat1-/-Cones have shifted sensitivity to red , no rod phototransduction Importantly , these mutant mouse lines have been extensively tested for visual function ( Alam et al . , 2015; Altimus et al . , 2010; Biel et al . , 1999; Cahill and Nathans , 2008; Calvert et al . , 2000; Naarendorp et al . , 2010; Nathan et al . , 2006; Zhao et al . , 2014 ) . Rod sensitivity and function is unchanged in cone mutant animals and cone sensitivity and function is unchanged in rod mutant animals ( Alam et al . , 2015; Altimus et al . , 2010; Biel et al . , 1999; Cahill and Nathans , 2008; Calvert et al . , 2000; Naarendorp et al . , 2010; Nathan et al . , 2006 ) . Electrophysiological recordings of ipRGCs show functional rod input in cone mutants and functional cone input in rod mutants ( Zhao et al . , 2014 ) . Additionally , all of the photoreceptor mutant lines we used have similar pupil sizes in darkness ( Figure 2—figure supplement 1 ) . Therefore , these mouse lines allow precise separation of rod , cone , and melanopsin activation while leaving the function of the other photoreceptors intact . When we tested the transient PLR of rod , cone , and melanopsin mutant mice , we found that both cone and melanopsin knockout mice were identical to wildtype in both sensitivity and kinetics ( Figure 2B and Figure 2—figure supplement 2B ) . Despite previous reports of melanopsin requirement for the transient PLR ( Lucas et al . , 2003 ) , we find that melanopsin is dispensable for the PLR when using more environmentally relevant stimuli ( Figure 2—figure supplement 3 ) . In contrast , rod knockout mice displayed no pupil constriction until the light intensity becomes relatively bright ( i . e . >10 lux , Figure 2B ) , despite the normal spatial vision in rod knockout mice at these moderate light intensities ( Alam et al . , 2015 ) . To corroborate these results , we tested three different cone mutant lines and two different rod mutant lines with distinct mutations and observed virtually identical results: cone mutants are similar to wildtype and rod mutants have severe transient sensitivity deficits ( Figure 2—figure supplement 2C , D ) . These results are surprising given previous proposals that cones are important for transient ipRGC responses , including acute changes in pupil size ( Allen et al . , 2011; Dkhissi-Benyahya et al . , 2007; Gooley et al . , 2012 , 2010; Ho Mien et al . , 2014; Kimura and Young , 2010 , 1999; Lall et al . , 2010; Spitschan et al . , 2014; van Oosterhout et al . , 2012 ) . Therefore , we sought to acutely modulate cone activity using a previously characterized mouse line that expresses the human ‘red’ opsin ( OPN1LW ) in place of the mouse ‘green’ opsin ( Opn1mw ) ( Red cone KI ) , making cones the only photoreceptors with enhanced sensitivity to red light ( Lall et al . , 2010 ) ( Figure 2C ) . We found that these mice have identical transient PLR in response to red light as wildtype ( Figure 2D ) , indicating that acute cone modulation does not affect the overall magnitude of the PLR . Furthermore , crossing this line to a rod knockout line abolishes the PLR in response to red light ( Figure 2E ) . These results show that rods are the predominant photoreceptor inputs for transient PLR at low to moderate light intensities , even in a mouse line with sensitized cones . To evaluate the inputs contributed by each photoreceptor in isolation to the PLR , we generated double mutants lacking the function of two photoreceptor types , resulting in mice with only rods ( Rods alone ) , only cones ( Cones alone ) or only melanopsin ( Melanopsin alone ) ( Table 1 ) . We found that the only photoreceptors capable of recapitulating the wildtype PLR are rods . Mice with only rod function had identical light sensitivity as wildtype and a similar rapid induction of pupil constriction ( Figure 2F , G ) , though their ability to maintain stable pupil sizes in bright light was slightly diminished ( Figure 2G ) . We corroborated the sufficiency of rods using three different mouse lines ( Figure 2—figure supplement 4 ) . Interestingly , while two of the lines were nearly identical to wildtype , one line had similar sensitivity , but altered kinetics , suggesting that cones might regulate rod signaling dynamics . In marked contrast to rod input , cone and melanopsin inputs were severely deficient in mediating the transient PLR ( Figure 2F , G ) . Animals with melanopsin alone retained a normal PLR at bright light intensities ( Figure 2F ) , as seen previously ( Gooley et al . , 2012; Lucas et al . , 2001; Xue et al . , 2011 ) , with sensitivity that is indistinguishable from rod knockouts ( Figure 2—figure supplement 5 ) , though they had relatively sluggish kinetics ( Figure 2G ) . In contrast , cone-only animals had minimal PLR ( Figure 2F ) , resulting in a further sensitivity deficit compared to rod knockout and melanopsin-only animals ( Figure 2—figure supplement 5 ) . Additionally , cone input decayed rapidly ( Figure 2G ) , presumably due their robust light adaptation properties . Collectively , these results show that rods serve as the primary input to ipRGCs for transient PLR responses , especially at low to moderate light intensities . At bright light intensities , additional input originates predominantly from melanopsin phototransduction . To investigate how ipRGCs relay transient light detection to the brain , we tested the transient PLR in mice lacking glutamatergic neurotransmission in ipRGCs ( Opn4Cre/+ ; Slc17a6fl/fl , also known as Vglut2fl/fl ) or mice lacking PACAP in ipRGCs ( Opn4Cre/+ ; Adcyap1fl/- ) ( Figure 3A and Table 2 ) . See Figure 3—figure supplement 2 for details on design of the conditional PACAP allele ( Adcyap1fl ) . 10 . 7554/eLife . 15392 . 013Figure 3 . Glutamaterigic output provides precise and rapid transient signaling . ( A ) Diagram of ipRGC behavioral circuit . ( B ) Intensity-response curves of the PLR in each of the neurotransmitter mutant mouse lines ( Wildtype n = 6 ) ( ipRGC glu . KO: Opn4Cre/+ ; Slc17a6fl/fl n = 4 ) ( ipRGC PACAP KO: Opn4Cre/+ ;Adcyap1fl/- n = 6 ) ( mean ± SD ) . ( C ) Sensitivity ( EC50 ) in each of the mutant lines . Statistical significance determined by one-way ANOVA with Sidak’s post-test . ( D ) Kinetics of transient pupil constriction ( 1000 lux ) in mice lacking glutamatergic or PACAPergic neurotransmission . Traces of individual mice are shown behind one-phase decay curve-fits . Half-lives: Wildtype ( 1 . 1 s ) , ipRGC glu . KO ( 4 . 8 s ) , ipRGC PACAP KO ( 1 . 1 s ) . ( E ) Representative pupil images at 5 s and 30 s post-illumination ( 1000 lux ) . Figure 3—figure supplements 1–3 . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 01310 . 7554/eLife . 15392 . 014Figure 3—figure supplement 1 . Dark-adapted pupil sizes of neurotransmitter mutant lines used . Dark-adapted pupil sizes of all mouse lines used for neurotransmitter investigation . Pupil size was recorded before light onset and pupil area ( mm2 ) is reported . ipRGC glutamate KO mice are the only line used which display a significant difference in dark-adapted pupil size suggesting that glutamatergic signaling is important for setting pupil size in darkness ( p=0 . 0001 ) . Statistical significance was determined by one-way ANOVA followed by Sidak’s post-test . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 01410 . 7554/eLife . 15392 . 015Figure 3—figure supplement 2 . Description of conditional PACAP allele . Schematic of the conditional PACAP allele ( Adcyap1lox ) . Boxes indicate exons ( 1–5 ) . Grey indicates UTR while black indicates protein coding sequence . A single FRT site remains after removal of selection cassette . LoxP sites flank exon 2 . Cre-mediated excision results in a frameshift and production of a truncated protein . See Materials and methods for further information of allele generation and confirmation . A more detailed description of the generation and use of the allele will appear in a manuscript that is in preparation ( Ross and Lowell , unpublished ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 01510 . 7554/eLife . 15392 . 016Figure 3—figure supplement 3 . PACAP can drive significant constriction within 30s of high light onset . Transient constriction was monitored in neurotransmitter mutant mice under high light ( 5000 lux ) . Data from each mouse is shown with the mean ( black bar ) . ipRGC glutamate KO mice ( Opn4Cre/+ ; Slc17a6fl/fl: n = 4 ) display a significant reduction in transient phase pupil constriction compared to wildtype ( n = 6 ) ( p<0 . 0001 ) while ipRGC PACAP KO ( Opn4Cre/+;Adcyap1fl/-: n = 6 ) and PACAP KO ( n = 4 ) mice are indistinguishable from wildtype ( p>0 . 999 ) . Statistical signficance determined via one-way ANOVA followed by Sidak’s post-test . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 01610 . 7554/eLife . 15392 . 017Table 2 . Description of neurotransmitter mutant mouse lines used . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 017Mouse lineGenotypeEffect on retinal functionCitationsMelanopsin-CreOpn4Cre/+ Cre expression in ipRGCs ( Ecker et al . , 2010 ) Slc17a6-floxSlc17a6fl/flExon 2 flanked by loxP sites ( Hnasko et al . , 2010 ) ipRGC glutamate KOOpn4Cre/+; Slc17a6fl/flSilences ipRGC glutamatergic releasePACAP KOAdcyap1-/-Whole animal PACAP removal ( Hamelink et al . , 2002 ) PACAP-floxAdcyap1fl/flExon 2 flanked by loxP sitesSee Figure 3—figure supplement 2ipRGC PACAP KOOpn4Cre/+ ; Adcyap1fl/-Silences ipRGC PACAP release Though ipRGC glutamate knockout mice ( Opn4Cre/+ ; Slc17a6fl/fl ) exhibited a small decrease in resting pupil size ( Figure 3—figure supplement 1 ) ( Delwig et al . , 2013 ) , we observed that they had minimal transient PLR at all light intensities ( Figure 3B–E ) , with more robust PLR at very bright light intensities ( Figure 3—figure supplement 3 ) , in agreement with previous studies ( Delwig et al . , 2013; Purrier et al . , 2014 ) . This indicates that ipRGC glutamatergic neurotransmission is a critical transient signal for the PLR . Presumably , the residual transient response is PACAPergic . In contrast to ipRGC glutamate knockout mice , ipRGC PACAP knockout mice had no deficits in transient PLR sensitivity or kinetics ( Figure 3B–E ) , as observed previously ( Kawaguchi et al . , 2010 ) , suggesting that glutamate is sufficient for the entirety of the transient PLR . Additionally , these results show that any potential modulation of glutamatergic signaling by PACAP ( Chen et al . , 1999; Toda and Huganir , 2015 ) is dispensable for the transient PLR . Together , these data derived from retinal mutants for photoreceptors and neurotransmitters identify rods as the principal input and glutamate as the principal output of ipRGC-mediated transient PLR signaling . Since wildtype responses decay over time ( Figure 1 ) , we next asked how ipRGC inputs and outputs drive the PLR across longer times ( Figure 4A ) . Strikingly , when we measured the sustained PLR in melanopsin knockout mice , which have a normal transient PLR ( Figure 2B ) , there was virtually no pupil constriction ( Figure 4B ) , even at bright light intensities ( up to 10 , 000 lux , Figure 4—figure supplement 1A ) . We observed that melanopsin knockout mice lose pupil constriction in minutes ( half-life: ~4 min , Figure 4C ) , similar to the wildtype PLR decay rate at lower light intensities ( WT half-life range: ~2–4 min at 1–100 lux , Figure 1F ) . This suggests that melanopsin phototransduction maintains robust light input in ipRGCs during the day ( Figure 4—figure supplement 1B ) , after rods adapt to background light . 10 . 7554/eLife . 15392 . 018Figure 4 . Melanopsin/rod synergy supports PLR under persistent conditions . ( A ) Diagram of ipRGC behavioral circuit . ( B ) Intensity-response curves for wildtype and melanopsin knockout mice ( Opn4-/- ) : transient ( dotted lines for reference ) and sustained ( 60 min: solid lines ) ( WT n = 6 , Opn4-/- n = 12 ) . ( right ) Representative pupil images under 1000 lux persistent light . ( C ) 60-min time course of pupil constriction under constant light ( 1000 lux ) . Data fit with a one-phase association curve ( WT n = 5 , Opn4-/- n = 7 ) . ( D ) Sustained pupil constriction monitored every 5 min for 1 hr in melanopsin knockout mice ( Opn4Cre/Cre ) expressing the Gq-coupled DREADD ( hM3D ) specifically in ipRGCs ( AAV2-hSyn-DIO-hM3D ( Gq ) -mCherry ) . CNO injection IP ( blue ) caused robust constriction within 5–10 min that was sustained for 60 min , whereas PBS injection ( black ) did not . CNO data is fit with a one-phase association curve and PBS data is fit with a linear regression ( n = 6 , mean ± SD ) . ( E ) ( top ) Diagram showing viral eye injection in only one eye . ( bottom ) Confocal microscope images of an Opn4Cre/Cre retina injected with AAV2-CMV-DIO-mRuby-P2A-Melanopsin-FLAG showing infection and expression ( mRuby , top; anti-OPN4 , bottom ) . Scale bar = 50 µm . ( F ) Successful rescue of pupil constriction by virally restored melanopsin expression in a single eye of adult mice ( WT n = 6 , Mel . KO n = 12 , Mel . -Rescue n = 4 ) . ( right ) Representative pupil images of Mel . KO and Mel . -Rescue mice at 1000 lux . ( G ) PLR intensity-response curves of Wildtype ( n = 6 ) , Mel . -only ( Rod-DTA; Cone-DTA n = 8 ) , Cone KO ( Cnga3-/-n = 4 ) , and Rod KO ( Rod-DTA n = 5 ) mice ( mean ± SD ) . Melanopsin is sufficient at high light ( ≥1000 lux ) , however , rods are required at lower light intensities . Cone KO mice are similar to wildtype . ( top ) Representative pupil images at 1000 lux . See also Figure 4—figure supplement 1–4 . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 01810 . 7554/eLife . 15392 . 019Figure 4—figure supplement 1 . Melanopsin is required for sustained constriction across the day . ( A ) Sustained constriction at 10 , 000 lux ( WT n = 6 , only 1 is plotted due to inability to see extremely small pupils in very bright light , Mel . KO n = 6 ) . ( B ) Time course of pupil constriction under 12 hr of constant light corresponding to circadian day ( room lighting = 350 lux ) using wildtype ( n = 3 ) and melanopsin knockout mice ( n = 4 ) ( line is smoothed mean ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 01910 . 7554/eLife . 15392 . 020Figure 4—figure supplement 2 . Viral infection and expression is specific to ipRGCs . ( A ) Schematic of the method to activate exclusively ipRGCs using an exogenous GPCR ( hM3D ( Gq ) ) and its ligand ( CNO ) . ( B ) Confocal microscope image showing infection of ipRGCs observed by mCherry expression following administration of a Cre-dependent AAV injected into the vitreous of melanopsin-Cre knockout mice ( Opn4Cre/Cre ) . ( C–E ) Confirmation of ipRGC-specific expression of melanopsin from AAV-DIO-mRuby-P2A-Melanopsin-FLAG viral injections . Opn4Cre/+ mice were used to colocalize viral ( C ) mRuby with ( D ) endogenous and exogenous melanopsin expression . ( E ) We observe specific expression of mRuby in a significant portion of ipRGCs , although some ipRGCs lack mRuby staining , presumably due to lack of infectivity ( arrows show mRuby-negative ipRGCs ) . Scale bars = 50 μm . ( F ) Quantification of fraction of ipRGCs ( melanopsin-antibody ) which are mRuby-positive . Quantification shown for three mice ( A single 20x field was quantified for each mouse ) . Approximately 90% of melanopsin-positive cells express mRuby . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 02010 . 7554/eLife . 15392 . 021Figure 4—figure supplement 3 . Rods , but not cones , contribute to sustained PLR sensitivity . ( A ) PLR intensity-response curves of wildtype and mice with only melanopsin phototransduction intact ( ‘melanopsin-only’: Gnat1-/-; Gnat2-/- n = 4 , Gnat1-/-; Cnga3-/- n = 4 , Rod-DTA; Cone-DTA n = 8 ) ( mean ± SD ) . ( B ) Sustained PLR intensity-response curves of wildtype ( n = 11 ) and rod mutant mice ( Gnat1-/- n = 5 , Rod-DTA n = 5 ) ( mean ± SD ) . ( C ) Sustained PLR intensity-responses of wildtype and cone mutant mice ( ‘cone mutants’: Gnat2-/- ( n = 4 ) , Cnga3-/- ( n = 4 ) , Cone-DTA ( n = 4 ) ) . ( D ) Sustained EC50 for wildtype and cone mutant , rod mutant and melanopsin-only mice ( line = mean ) . All rod mutant and melanopsin-only mouse lines display significnt loss of sensitivity ( p<0 . 0001 ) . Two of three cone mutant mouse lines were not significantly different from wildtype ( Cnga3-/-p=0 . 57 , Cone-DTA p>0 . 999 ) , though Gnat2-/- displayed a 0 . 69 log-unit decrease in sustained PLR EC50 ( Gnat2-/-P = 0 . 004 ) . Additionally , all rod mutant lines were similar to their corresponding melanopsin-only line ( p>0 . 706 ) while all cone mutant lines were significantly more sensitive than their corresponding melanopsin-only line ( p≤0 . 0001 ) . Statistical significance determined via one-way ANOVA with Sidak’s post-test . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 02110 . 7554/eLife . 15392 . 022Figure 4—figure supplement 4 . Rods drive the residual sustained pupil constriction observed in the absence of melanopsin . ( A ) Sustained PLR dose-responses for wildtype ( n = 11 ) , melanopsin knockout ( Opn4-/- , n = 12 ) and mice with only rod phototransduction intact ( ‘rod-only’: Cnga3-/-; Opn4-/- n = 4 ) ( mean ± SD ) . ( right ) Scatter plot of 1000 lux sustained PLR . Melanopsin knockout and ‘rod-only’ mice not statistically different by one-way ANOVA with Sidak’s post-test ( p=0 . 983 ) ( line indicates mean ) . ( B ) Sustained PLR intensity-responses for wildtype ( n = 11 ) , melanopsin knockout ( Opn4-/- n = 12 ) and mice with only cone phototransduction intact ( ‘cone-only’: Opn4-/-; Gnat1-/- , n = 6 , mean ± SD ) . ( right ) Scatter plot of 1000 lux sustained PLR . Melanopsin knockout and ‘cone-only’ mice are statistically different by one-way ANOVA followed by Sidak’s post-test . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 022 The severe deficits we observed in the sustained PLR in melanopsin knockout mice raised the possibility that these animals may have developmental deficits that affect their signaling ( Rao et al . , 2013; Renna et al . , 2011 ) . To directly address this issue , we rescued ipRGC function in adult melanopsin knockout mice using either chemogenetics or restoration of melanopsin expression . Using our mouse line with Cre introduced into the melanopsin locus ( Opn4Cre/Cre ) and a Cre-dependent chemogenetic DREADD virus ( AAV2-hSyn-DIO-hM3D ( Gq ) -mCherry ) ( Figure 4—figure supplement 2 ) , we administered the selective DREADD agonist CNO ( Armbruster et al . , 2007 ) and observed robust and sustained pupil constriction for at least one hour ( Figure 4D ) . This result demonstrates that ipRGCs and their downstream circuits remain competent for sustained signaling in melanopsin knockout mice . Furthermore , we acutely restored melanopsin in the majority of ipRGCs of melanopsin-Cre knockout mice ( Opn4Cre/Cre ) using a virus that expresses melanopsin in a Cre-dependent manner ( Figure 4E and Figure 4—figure supplement 2C–E , AAV2-CMV-DIO-mRuby-P2A-Melanopsin-FLAG ) . Following melanopsin restoration , we observed a rescue of the sustained PLR ( Figure 4F ) . These results demonstrate for the first time that the effect of melanopsin loss can be rescued in adulthood , indicating that melanopsin-based light detection is directly required for ipRGCs to signal sustained PLR . Surprisingly , although melanopsin is required for sustained signaling , we found that melanopsin signaling could not fully recapitulate the sustained PLR . Despite the observation that the sustained PLR is normal at bright light intensities in melanopsin-only mice , these mice had a sensitivity deficit compared to wildtype ( Figure 4G ) . Notably , we observed that rod knockout mice display an identical sensitivity deficit as melanopsin-only ( Figure 4G and Figure 4—figure supplement 3 ) , indicating that rods contribute to sustained ipRGC signaling . This indicates that at intermediate intensities , both rod and melanopsin signaling cooperate to sustain the PLR . As with the transient PLR , we found that cone knockout mice had no deficit in sustained PLR ( Figure 4G ) . Again , multiple independent mouse lines corroborate these conclusions ( Figure 4—figure supplement 3 ) . Furthermore , we found that rods alone could drive the remainder of the sustained PLR in melanopsin knockout mice ( Figure 4—figure supplement 4A ) , whereas cone-only mice had no sustained PLR ( Figure 4—figure supplement 4B ) . These results show that melanopsin signaling dominates sustained light input to ipRGCs , but rods , which are thought to be nonfunctional under continuous bright light , are intimately involved in supporting the sustained PLR . Notably , rod contributions to the sustained PLR occur predominantly at light intensities above their presumed saturation ( ~40 lux ) , showing that rods are indeed capable of contributing to visual function above previously defined limits ( Alam et al . , 2015; Altimus et al . , 2010; Naarendorp et al . , 2010 ) . Therefore , sustained ipRGC responses are not a simple consequence of a single photoreceptive system , but instead require rod/melanopsin synergy for highest sensitivity . Studies of ipRGC neurotransmitters , in combination with our transient PLR results presented here , suggest that glutamate is the primary ipRGC neurotransmitter , and that PACAP plays a minor , or modulatory , role ( Beaulé et al . , 2009; Colwell et al . , 2004; Delwig et al . , 2013; Gompf et al . , 2015; Kawaguchi et al . , 2010 , 2003; Purrier et al . , 2014 ) . However , when we tested the sustained PLR in ipRGC glutamate knockout mice , we found that their pupil constriction improved over time compared to their transient PLR sensitivity ( Figure 5B , C ) . In contrast , PLR sensitivity either stays the same or declines in all other mutant lines , suggesting that the remaining signal in glutamate knockout mice , presumably PACAP , becomes more effective with longer stimulus duration . Intriguingly , ipRGC glutamate knockout mice showed pulsatile or periodic pupil constriction over time , potentially due to waves of neuropeptide vesicle delivery and release from ipRGC axons ( Video 1 ) . 10 . 7554/eLife . 15392 . 023Figure 5 . PACAP is essential for the sustained PLR . ( A ) Diagram of ipRGC behavioral circuit . ( B ) PLR intensity-response curves of sustained constriction in mice lacking glutamatergic or PACAPergic neurotransmission ( WT n = 6 , ipRGC glu . KO n = 4 , ipRGC PACAP KO n = 6 ) ( mean ± SD ) . Both mutants display deficits at 10 , 100 , and 1000 lux as compared to wildtype ( wildtype v . ipRGC Glu . KO: 10 and 100 lux p<0 . 0001 , 1000 lux p=0 . 0004 by two-way ANOVA with Sidak’s post-test ) ( wildtype v . ipRGC PACAP KO: 10 , 100 , and 1000 lux p<0 . 0001 by two-way ANOVA with Sidak’s post-test ) . ( C ) Representative pupil images of sustained constriction at 1000 lux . ( D ) Comparison of transient and sustained constriction under high light ( 1000 lux ) . ipRGC glu . KO mice ( red ) show an increase in pupil constriction with time whereas ipRGC PACAP KOs ( blue ) display a significant loss of constriction over time ( ipRGC glu . KO transient v . sustained p<0 . 0001 , ipRGC PACAP KO transient v . sustained p=0 . 0003 , wildtype transient v . sustained p=0 . 9921 by one-way ANOVA with Sidak’s post-test ) . ( E ) Pupil constriction of neurotransmitter mutant mice after sustained 5000 lux light . Data from individual mice shown with mean ( black bar ) . Statistical significance determined by one-way ANOVA with Sidak’s post-test . See also Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 02310 . 7554/eLife . 15392 . 024Figure 5—figure supplement 1 . PACAP KO mice display similar PLR phenotypes to ipRGC-specific PACAP KO mice . ( A ) Intensity-response curves of the transient PLR ( 30s light ) in each of the neurotransmitter mutant mouse lines ( Wildtype n = 6 ) ( ipRGC glu . KO: Opn4Cre/+ ; Slc17a6fl/fl n = 4 ) ( PACAP KO: Adcyap1-/- n = 4 ) ( mean ± SD ) . ( B ) Sensitivity ( EC50 ) in each of the mutant lines . Statistical significance determined by one-way ANOVA with Sidak’s post-test . ( C ) Comparison of transient and sustained ( 60 min . light ) constriction under high light ( 1000 lux ) . ipRGC glu . KO mice ( red ) show an increase in pupil constriction with time whereas PACAP KOs ( blue ) display a significant loss of constriction over time . ( D ) PLR intensity-response curves of sustained constriction in mice lacking glutamatergic or PACAPergic neurotransmission ( WT n = 6 , ipRGC glu . KO n = 4 , PACAP KO n = 4 ) ( mean ± SD ) . Both mutants display similar deficits until 1000 lux where PACAP KO mice show a further deficit ( PACAP KO v . ipRGC Glu . KO: p=0 . 0019 by one-way ANOVA with Sidak’s post-test ) . ( right ) Representative pupil images of sustained constriction at 1000 lux . ( E ) Pupil constriction of neurotransmitter mutant mice after sustained 5000 lux light . Data from individual mice shown with mean ( black bar ) . Statistical significance determined by one-way ANOVA with Sidak’s post-test . ( F ) 60-min . time course of pupil constriction under constant light ( 1000 lux ) . Data fit with a one-phase association curve ( WT n = 5 , PACAP KO n = 4 ) . ( mean ± SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 02410 . 7554/eLife . 15392 . 025Video 1 . Pulsatile pupil constriction in the absence of glutamatergic neurotransmission . This video is at 5x speed . 1000 lux white light ( 6500K ) turns on at approximately 1s . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 025 Neuropeptides have been shown to require high frequency neuronal activity for release and have relatively slow signaling kinetics compared to classical neurotransmitters ( Vaaga et al . , 2014 ) , suggesting that PACAP may be involved in sustained ipRGC signaling at bright light intensities . In support of a role for PACAP in sustained PLR signaling , we find that even though ipRGC PACAP knockout mice show normal transient PLR , they have an attenuated sustained PLR ( Figure 5B–E ) . This deficit in ipRGC PACAP knockout mice occurs even at moderate light intensities ( 10 and 100 lux ) . ipRGC PACAP KO mice display decaying constriction over time at 1000 lux as opposed to maintained constriction in wildtype mice and enhanced constriction in ipRGC glutamate KOs ( Figure 5D ) . At the brightest light intensity tested , 5000 lux , ipRGC PACAP KO mice display significantly worse sustained constriction than ipRGC glutamate KO mice ( Figure 5E ) , suggesting that PACAP is more important than glutamate for maintained responses under daylight conditions ( 1000–100 , 000+ lux ) . Additionally , we observed similar yet more pronounced deficits in full body PACAP KO mice ( Adcyap1-/-; Figure 5—figure supplement 1 ) . They display wildtype transient responses ( Figure 5—figure supplement 1A , B ) and severely attenuated sustained responses ( Figure 5—figure supplement 1C–E ) . Interestingly , these PACAP knockout mice exhibit PLR decay on a similar timescale as melanopsin knockout mice ( half-life: ~5 min , Figure 4C and Figure 5—figure supplement 1F ) . These results provide evidence that PACAP allows ipRGCs to communicate sustained input to downstream neurons . As observed with the photoreceptor contributions , the highest sensitivity of sustained PLR requires PACAP/glutamate synergy . Based on our results , we generated a quantitative representation of the distinct roles played by each photoreceptor input and neurotransmitter output of ipRGCs for the PLR over a range of light intensities and light stimulus durations ( Figure 6 , see Materials and methods for detailed explanation ) . We integrated individual necessity ( i . e . from knockout lines ) and sufficiency ( i . e . from ‘–only’ lines ) of rods , cones , and melanopsin in driving the PLR ( Figure 6—figure supplement 1 ) to generate a merged heat map representing each photoreceptor’s input to the PLR ( Figure 6A , B ) . We then performed the same technique to represent the neurotransmitter outputs of ipRGCs for the PLR ( Figure 6C , D and Figure 6—figure supplement 1 ) using only the necessity heat maps because we cannot rule out the possibility that other neurotransmitters contribute to ipRGC function . These heat maps provide a comprehensive visualization of the contribution made by each photoreceptor’s input and each neurotransmitter’s output for ipRGC signaling at any particular time or environmental light intensity . ipRGC transient signaling for the PLR is dominated by input from rods ( Figure 6A , red ) and output by glutamate ( Figure 6C , green ) . In contrast , sustained PLR signaling is dominated by melanopsin ( Figure 6B , blue ) and PACAP ( Figure 6D , blue ) . Together , these experiments and our model highlight a mechanistic transition in the ipRGC circuit supporting transient and sustained behavioral outputs . 10 . 7554/eLife . 15392 . 026Figure 6 . Model of ipRGC circuit transitions . ( A and B ) Heat maps of ( A ) transient and ( B ) sustained PLR as duration and intensity vary . Night , dawn/dusk , and daytime light intensities indicated by ticks on right side of plot . ( top ) Heat maps of individual photoreceptor contributions ( grayscale ) . Black represents no contribution and degree of white represents increasing contribution . Each photoreceptor contribution heat map is a combination of necessity ( individual photoreceptor transduction knockouts ) and sufficiency ( ‘photoreceptor-only’ ) heat maps ( for example: Input Contributionrod = Max ( Necessityrod , Sufficiencyrod ) ) . ( middle ) Rod ( red ) , cone ( green ) , melanopsin ( blue ) contributions are combined into a single heat map . ( bottom ) Color combination guide for reference when viewing heat map . ( C and D ) Same as above for neurotransmitter contributions to transient ( C ) and sustained ( D ) ipRGC signaling . Glutamatergic contribution is in green and PACAPergic contribution is in blue . See the Materials and methods section for details on heat map generation . Note that the axes are the same for the individual and combined heatmaps . See also Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 02610 . 7554/eLife . 15392 . 027Figure 6—figure supplement 1 . Necessity/Sufficiency heat maps for photoreceptor input to pupil constriction . Heat maps of necessity and sufficiency of each input ( top: rods , cones , melanopsin ) as stimulus duration and intensity vary . The necessity/sufficiency heat maps for a particular component were subsequently used to generate a photoreceptor contribution heat map ( See Figure 6 ) . Black indicates no necessity/sufficiency and white indicates full necessity/sufficiency . DOI: http://dx . doi . org/10 . 7554/eLife . 15392 . 027 We show here how inputs and outputs for a specific circuit change across time to support a behavioral response . Remarkably , the mechanisms supporting transient and sustained responses are distinct , suggesting stimulus duration as a critical determinant of circuit state . Transient PLR responses predominantly utilize classical , well-characterized visual system synaptic mechanisms: rod phototransduction and signal relay to ipRGCs , followed by ipRGC glutamatergic output . However , as conventional signaling mechanisms adapt , non-conventional mechanisms are recruited to maintain persistent signaling , including endogenous melanopsin phototransduction and peptidergic neurotransmission through PACAP . Our findings highlight fundamental circuit changes in the light-adapted retina that are relatively unexplored ( Tikidji-Hamburyan et al . , 2015 ) . Our results reveal the roles of distinct photoreceptors and neurotransmitters in the PLR and probably other ipRGC-dependent behaviors . We show how ipRGC inputs and outputs can contribute to the PLR through changes in their relative contribution across stimulus intensity and duration . Our ability to decipher these elaborate dynamic changes stems from the fact that we used a large array of environmental light intensities and durations , coupled with genetic means to silence individual circuit components . Ultimately , our quantitative model makes testable predictions about the role of each photoreceptor and neurotransmitter for other ipRGC-dependent behaviors . We show that in contrast to many proposed models , rods provide the exclusive transient input to ipRGCs for the PLR at dim ( scotopic ) and moderate ( mesopic ) light intensities . That rods are capable of rapid and sensitive input to ipRGCs is not surprising given electrophysiological evidence of sensitive rod input to ipRGCs ( Weng et al . , 2013; Zhao et al . , 2014 ) and the fact that rods are widely appreciated as the mediators of dim light vision . However , their exclusive input at mesopic light intensities suggests that cone input to ipRGCs is relatively weak , consistent with the inability of cones to drive circadian photoentrainment ( Lall et al . , 2010; Mrosovsky and Hattar , 2005 ) . Furthermore , we report here that in addition to their role in high-sensitivity transient signaling , rods are capable of driving sustained signaling at bright light intensities well above their saturation level ( ~40 lux , Figure 4—figure supplement 4 ) . This agrees with previous findings that rods are capable of supporting circadian photoentrainment at bright light intensities ( Altimus et al . , 2010 ) but also provides more precise temporal kinetics of rod input to subconscious behaviors . It has been proposed that rods never fully saturate ( Blakemore and Rushton , 1965 ) , and here we provide a physiological role for rod activity at daylight intensities . In contrast to previous data that melanopsin is largely dispensable for the PLR ( Lucas et al . , 2003 ) , we find that it is the dominant determinant of pupil size during the day . This is likely due to the fact that rod and cone inputs adapt to background light , while we find no evidence of behavioral light adaptation in melanopsin phototransduction ( i . e . identical sensitivity of melanopsin-only mice in transient and sustained PLR ) . While melanopsin phototransduction adapts in vitro ( Do and Yau , 2013; Wong et al . , 2005 ) , it has been proposed that only the adapted state is able to influence downstream behaviors ( Do and Yau , 2013 ) . We predict that melanopsin will be required in other visual functions throughout the day , for example as in more natural photoentrainment conditions that need to precisely measure changing light intensity under bright conditions or measuring day length ( Gooley et al . , 2010; Mrosovsky and Hattar , 2003; VanderLeest et al . , 2007 ) . This requirement for melanopsin in sustained light detection is likely the main reason melanopsin has been conserved in vertebrates . To date , glutamatergic neurotransmission is the only retina-brain signaling mechanism that has been robustly characterized . We confirm previous data that ipRGCs predominantly rely on glutamatergic output for the transient PLR ( Delwig et al . , 2013; Gompf et al . , 2015; Purrier et al . , 2014 ) . However , we show that the stimulus durations in which glutamate predominates over PACAP is relatively restricted ( <5 min ) , revealing the first critical role for a neuropeptide in retinal signaling to the brain . Further , we find that PACAP appears sufficient to drive the PLR independent of its potential to modulate glutamate . There have been discrepancies in the literature about the role of PACAP in the PLR ( Engelund et al . , 2012; Kawaguchi et al . , 2010 ) , which we believe is likely due to differences in light stimulus duration . Intriguingly , PACAPergic neurotransmission appears to be pulsatile , potentially reflecting the imprecision of slow vesicle delivery from the soma and suggesting why ipRGCs also require a fast and reliable glutamatergic signal . Glutamate and PACAP are the only known ipRGC neurotransmitters , but it remains possible there are neurotransmitters which remain undiscovered . An ipRGC-specific glutamate/PACAP double knockout is a crucial next step in understanding ipRGC neurotransmission . Given the expression of other neuropeptides in many RGCs , including ipRGCs ( Brecha et al . , 1987; Djeridane , 1994; Kay et al . , 2011; Liu et al . , 2011 ) , it remains possible that neuropeptides have a broader role in visual function than previously appreciated . The complementary arrangement of inputs and outputs for the PLR we describe here demonstrates how the visual system accomplishes high sensitivity , transient responses as well as integrative , long-term responses . Many other signaling systems may employ discrete methods for signaling robustly through time . While melanopsin is specific to the ipRGC circuit , PACAP and other neuropeptides may play similar roles in long-term signaling in other circuits , such as hypothalamic feeding circuits ( Krashes et al . , 2013 ) . Expanding the timescales over which we investigate these systems is likely to reveal entirely new aspects of cell signaling . C57Bl/6 × Sv129 hybrid mice were used in all experiments except PACAP KO mice which were C57Bl/6 . All mice were housed according to guidelines from the Animal Care and Use Committee of Johns Hopkins University . Male and female mice age 2–8 months were housed in plastic translucent cages with steel-lined lids in an open room . Ambient room temperature and humidity were monitored daily and tightly controlled . Food and water were available ad libitum . All mice were maintained in a 12 hr:12 hr light-dark cycle with light intensity around 100 lux for the entirety of their lives . All mice were dark-adapted for at least 30 min prior to any experiments and all PLR experiments were performed between Zeitgeber times ( ZT ) 2 and 10 . For all experiments , mice were unanesthetized and restrained by hand . Because stress can affect pupil size , we ensured that the mice were not stressed during these experiments . To do so , we handled the mice for several days prior to the experiments to get them accustomed to the researchers and to being scruffed . Any mice that showed signs of stress , including vocalizations and wriggling during the experiments , were not used and were subjected to more handling sessions before use in experiments . Mice were restrained manually under a 10- , 13- , or 23-Watt compact fluorescent light bulb ( GE Daylight FLE10HT3/2/D or Sylvania Daylight CF13EL and CF23EL ) with a color temperature of 6500 K to simulate natural sunlight . The light intensity was measured using a light meter ( EXTECH Foot Candle/Lux Light Meter , 401025 ) at the surface on which the mouse was held . The light meter was initially calibrated by EXTECH using a Tungsten 2856 K light source; because our experiments used a fluorescent bulb of 6500 K , all measured light intensities reported here may vary by 0 . 92–1 . 12 times the actual light intensity . Light intensity was adjusted by a combination of altering the distance of the light bulb ( s ) from the mouse and/or applying neutral density filters ( Roscolux ) . The light meter is incapable of detecting light intensities below 1 lux , so one neutral density filter cutting the light intensity by 12 . 5% was applied to the bulb to estimate 1-log unit decreases in illumination below 1 lux . Light intensities above 500 lux required the use of multiple light bulbs . For the monochromatic light PLR experiments , an LED light ( SuperBrightLEDs ) was housed in a microscope light source with fiber optic gooseneck arms to direct the light source to the mouse eye . For the experiments involving the Opn1mwred mice , we used a 626-nm LED in this setup and directed light to both eyes simultaneously or to just one eye and measured the PLR in the illuminated eye ( see figure legends ) . The photon flux was measured using a luminometer ( SolarLight ) and converted from W/m2 to photons/cm2/s . The light intensity was decreased by 12 . 5% using neutral density filters ( Rosco ) . Videos of the eye were taken using a Sony Handycam ( DCR-HC96 ) mounted on a tripod a fixed distance from the mouse . Manual focus was maintained on the camera to ensure that only one focal plane existed for each mouse and that therefore variable distance from the camera should not contribute to differences in relative pupil area throughout the video . Pupil size was first recorded under dim red light and the endogenous infrared light source of the camera to capture the dark-adapted pupil size . Following at least 5 s of recording in dark , the pupil was continuously recorded for at least 30 s of a light step stimulus . For all sustained PLR , animals were kept in a cage for 60 min under the light stimulus . Animals were removed from the cage after 60 min and held in front of the camera for 30 s as for the transient PLR . All pupil images presented in the paper were cropped to a fixed square area ( generally 100 × 100 pixels ) surrounding the eye using GNU Image Manipulation Program ( GIMP ) . The images were made grayscale and then brightness and contrast were adjusted to enhance visibility of the pupil and exported as PNG files . Videos were transferred from the camera to a computer as Audio Video Interleave ( AVI ) files and individual frames were taken using VLC media player ( www . videolan . org/vlc/ ) and saved in portable network graphics format ( PNG ) . Images were taken in the dark , at 5 s , and 30 s following stimulus onset . Pupil area was then quantified manually in ImageJ ( http://rsbweb . nih . gov/ij/ ) software . The pupil area was measured in pixels using the oval tool in which the 4 cardinal points of the oval were touching their respective edges of the pupil . The relative pupil area was calculated using LibreOffice Calc or Microsoft Excel in which the area during the light stimulus was divided by the area prior to lights onset . For the transient PLR , the minimum relative pupil size of either 5 s or 30 s after stimulus was used for all genotypes . The intensity-response curve was fit using a variable slope sigmoidal dose-response curve in Graphpad Prism 6 . The top and bottom of the fit were constrained to 1 . 0 and between 0 and 0 . 10 , respectively , to ensure the EC50 for each genotype was represented by similar curves . For genotypes that never showed evidence of reaching between 0 and 0 . 10 relative pupil size , the bottom was not constrained . The sensitivity for each genotype was calculated using the same process of fitting each individual animal’s data points with a sigmoidal dose-response curve to generate EC50 . The lox-modified PACAP ( Adcyap1 ) targeting construct was made by recombineering technology . To engineer the targeting vector , 5’ homology arm , 3’ homology arm and CKO region were amplified from mouse Sv129 BAC genomic DNA and confirmed by end sequencing ( Cyagen biosciences , Santa Clara , CA ) . The two loxP sites flank the second exon and when recombined , create a frameshift mutation and truncated protein . The plasmid was electroporated into W4 ES cells and cells expanded from targeted ES clones were injected into C57BL6 blastocysts . Germline transmitting chimeric animals were obtained and then mated with flpE mice to delete the frt-site flanked neomycin selection cassette . The resulting heterozygous offspring were crossed to generate homozygous PACAPlox/lox study subjects . All mice are thus on a mixed C57Bl6/J and 129Sv background . Offspring were genotyped by PCR using 2 primers ( F: CCGATTGATTGACTACAGGCTCC and R: GTGTTAAACACCAGTTAGCCACGC ) which detect the presence or absence of the 5’ loxP site and a 3rd primer was used in conjunction with the forward primer ( CKO-R GGGCTTTGATCTGGGAACTGAAG ) to detect the recombination event . By generating mice homozygous for a germline deleted cre-deleted allele , we have established that the cre-deleted allele does not express intact PACAP mRNA ( by PCR and by ISH ) . A more detailed description of the generation and use of the allele will appear in a manuscript that is in preparation ( Ross and Lowell , unpublished ) . Mice were anesthetized by intraperitoneal injection of avertin ( 2 , 2 , 2-Tribromoethanol ) and placed under a stereo microscope . 1 μl of AAV2-hSyn-DIO-hM3DGq-mCherry ( 4 . 6 × 1012 viral particles/ml , Roth lab , UNC Vector Core ) or AAV2-CMV-DIO-mRuby-P2A-Melanopsin-FLAG ( Robinson lab , UMBC ) was placed on a piece of Parafilm and drawn into a 10-μl microcapillary tube ( Sigma P0674 ) that had been pulled to a needle ( Sutter Instruments , Model P-2000 ) . The loaded needle was then placed in the holster of a pico-injector ( Harvard Apparatus PLI-90 ) . The needle punctured the eye posterior to the ora serrata and air pressure was used to drive the viral solution into the vitreous chamber of the eye to ensure delivery specifically to the retina . Mice recovered from surgery on a heating pad until they woke from anesthesia . All PLR experiments and confocal imaging were done at least 3 weeks following viral injection . Mice that had been infected with the AAVs were anesthetized with avertin and then euthanized using cervical dislocation . The eyes were removed and the retinas were dissected in PBS and then fixed in 4% paraformaldehyde for 1–2 hr on ice . The retinas were then washed in PBS at least three times before mounting on a microscope slide ( Fisher , Hampton , NH ) in Fluoromount ( Sigma , St . Louis , MO ) with DAPI ( 2- ( 4-amidinophenyl ) -1H -indole-6-carboxamidine ) . Some retinas were co-stained for melanopsin using rabbit anti-OPN4 ( Advanced Targeting Systems , San Diego , CA , AB-N38 , 1:1000 ) in 4% goat serum with secondary antibody Alexa Fluor 488 goat anti-rabbit ( Life Technologies , Carlsbad , CA , A11008 , 1:1000 ) . Images were taken on a Zeiss LSM 710 confocal microscope using a 20× objective . After imaging , images were made grayscale , background subtracted , and brightness and contrast were adjusted in FIJI ( http://fiji . sc ) for the image presented in the paper . All statistical tests were performed in Graphpad Prism 6 . Specific statistical comparisons are listed in the figure captions . Because the EC50 data appears to be a normal distribution on a log scale ( log-normal distribution ) , all statistical tests and data analysis involving EC50 were performed on the log transformed data set . The photoreceptor contribution heat map was generated by first creating estimated pupil size matrices for the both the rapid and sustained PLR at every light intensity and time for wildtype mice ( x axis = time , y axis = intensity ) . To do so , we applied the equation for a one-phase association:Y=Y0+ ( Plataeu−Y0 ) ∗ ( 1−e ( −K∗x ) ) In our case , Y is the relative pupil area generated at time , x . For the WT rapid PLR heat map , Y0rapid is set to 1 for every light intensity and the Krapid was extracted from the wildtype rapid constriction kinetics curve at 100 lux . The Plateaurapid value at each light intensity is the rapid PLR value extracted from the WT rapid intensity-response curve fit . This allows us to generate a full matrix of WT pupil sizes at every intensity and time by knowing the final pupil size ( Plateau ) and the rate of constriction ( K ) . This then generates a full matrix of values for every time and intensity for WT mice . The same method was applied to make the sustained PLR heat map . However , in this case , Y0sustained was set to the value of the rapid PLR at each light intensity ( e . g . the same value as Plateaurapid ) . The Plateausustained value is extracted from the sustained intensity-response curve fit at each intensity . The Ksustained was extracted from our wildtype sustained time courses ( Figure 1c ) . Because the decay rate for sustained constriction appeared to change with intensity ( Figure 1f ) we used a sigmoidal curve fit to our experimentally determined decay rates ( 1 , 10 , 100 lux ) to generate decay rates for a range of light intensities . We constrained the top and bottom of this curve to the decay rates determined for 1 and 100 lux respectively . This process was used to generate two matrices of relative pupil areas with the y-axis being light intensity varying logarithmically ( 0 . 001–100 , 000 lux ) and the x-axis being time varying linearly from 0 to 30 s for the rapid and 30 s to 60 min for the sustained . This was done using a custom MATLAB script . The matrices generated for the wildtype mice were also done to the photoreceptor mutants . In order to determine necessity of a photoreceptor we subtracted rod ( average of Gnat1-/- and Rod-DTA ) , cone ( average of Cnga3-/- , Gnat2-/- and Cone-DTA ) , or melanopsin ( Opn4-/- ) knockout matrices from the wildtype matrix . This yields larger values for genotypes that are more required and also normalizes for the overall constriction in wildtype mice at that intensity ( i . e . because rods are fully necessary at some dim intensities at which WT mice have minimal constriction , the necessity value attributed to rods is small despite their absolute necessity at that intensity ) . To determine sufficiency we used ‘rod-only’ ( Cnga3-/-; Opn4-/- ) , ‘cone-only’ ( Gnat1-/-;Opn4-/- ) and ‘melanopsin-only’ ( average of Gnat1-/-;Gnat2-/- , Gnat1-/-; Cnga3-/- and Rod-DTA;Cone-DTA ) matrices . Additionally , we applied the decay rate of pupil constriction from the ‘cone-only’ mouse line transient PLR at 100 lux for all light intensities . Finally , matrices generated above were exported as heat map images with MATLAB . In order to isolate negative feedback’s impact on the PLR , we generated a computational model . Computational modeling was performed with MATLAB using two experimentally determined parameters . First , the relative pupil area ( RPA ) values for the wildtype intensity-response curve ( Figure 1d ) . These values give us the response driven when the pupil starts fully open . We will later multiply the environmental intensity by the new relative pupil area to determine the new retinal intensity . We will use this new retinal intensity to extract the pupil size from the rapid intensity-response curve to find the constriction driven by that new intensity under baseline conditions . The model does this recalculation of retinal intensity and the PLR driven by it every second for 956 s . The second experiment integrated into the model is a 1 s light pulse-chase experiment . Here , we dark-adapted the mouse , gave a single second of light and then followed subsequent constriction for 30 s . These constriction values were normalized to the maximum constriction achieved , in this case the 6-s time point . This gives us the ability to weight the contribution of light at a particular time to constriction at subsequent times . As you can see , light does not instantly constrict the pupil . It takes several seconds for the signal to maximally impact pupil size , which is then followed by signal decay . Importantly , this temporal weighting , while not required for the model , does give us a rough estimate of the potential kinetics of feedback’s impact on PLR decay . With these pieces of experimental data in hand , the model does the following at every light intensity ( 0 . 0001–100 , 000 lux ) : ( 1 ) it extracts the RPA in response to a particular light intensity from the wildtype intensity-response curve . ( 2 ) The model uses the temporal weighting values from the pulse-chase experiment to weight that RPA across subsequent times ( 0–30 s ) . This gives us a 30-s constriction time course for the light detected at time zero . ( 3 ) The model next moves to time 1 s . Now it takes into account the maximum constriction caused by light at previous times ( time 0 in this case ) . The model uses that constriction to reduce the light intensity and calculate a new retinal light intensity: RPA Light intensity = Retinal intensity . ( 4 ) Next , it determines the RPA driven by this new retinal intensity using the DRC once again . ( 5 ) Repeats step ( 2 ) for this RPA giving another time course of constriction ( 1–31 s ) . ( 6 ) The model repeats steps ( 3–5 ) moving up in 1s increments each time . Importantly , at each new time point it finds the maximum constriction value in response to all previous time points in order to calculate the new retinal intensity . ( 7 ) Finally , it finds the maximum constriction at each time point in order to produce a negative feedback PLR decay time course . See graphical representation of the negative feedback model ( Figure 1—figure supplement 2A ) The primary assumption the model makes is that the PLR system has zero summation of signal . This is probably unlikely . However , this assumption was made to maximize the impact of feedback on pupil constriction . This model provides us with an upper bound on negative feedback’s contribution to PLR decay . Source code and materials used are available on Github ( https://github . com/keenanw27/PLR-Decay-Model ) . At a given environmental light intensity: luxo . The effect of pupillary negative-feedback during a 956s stimulation is modeled as follows: ( 1 ) for time t=1 , 2 , 3…956 max ( RPA→ ( : , t ) ) ×luxo=luxt In equation ( 1 ) above , we determine the retinal light intensity , luxt , that is , the intensity of light after modulation by pupil size at time t . At t = 1 there is no pupil constriction and therefore no light intensity modulation ( luxo=luxt ) . RPA→ is a 956 × 956 matrix which stores subsequent pupil constriction values . With luxt we determine the constriction driven by light sensed at time , t: ( 2 ) α→ ( luxt ) ×ω→=RPA→ ( t , t:t+30 ) In equation ( 2 ) , we calculate the amount of constriction driven by luxt , α→ ( luxt ) , and approximate the temporal characteristics of that constriction with ω→ . ω→ is based on a 1s light pulse-chase experiment where we followed the constriction driven by 1 s of light for 30 s . Again , we store calculated constriction values: RPA→ ( t , t:t+30 ) . Finally , we extract the highest constriction value at t: ( 3 ) max ( RPA→ ( : , t ) ) =Model→luxo ( 1 , t ) After completing t = 956 , Model→luxo is a vector containing the model-predicted timecourse of pupil constriction when negative-feedback is the only source of PLR decay .	The retina is the part of our eye that detects light and sends visual information to the brain . There are several different types of light-sensitive cell in the retina that perform different roles . For example , retinal cells called intrinsically photosensitive retinal ganglion cells ( or ipRGCs for short ) rapidly respond to the intensity of background light and regulate the size of the pupils to control how much light enters the eyes . These cells receive information from other light-sensitive cells in the retina called rods and cones . There are at least two mechanisms that ipRGCs may use to relay information to the brain: one uses a protein called PACAP , while the other involves a molecule called glutamate . However , it is still not clear which mechanisms are actually used by ipRGCs , or when they might use them . Like other mammals , mice can rapidly reduce the size of their pupils when they are suddenly exposed to a bright light . Keenan , Rupp et al . investigated how ipRGCs control the size of the pupils in mice that had been genetically engineered to lack different components of the visual system . Mutant mice that lacked rod cells or were unable to produce glutamate in their ipRGCs failed to reduce the size of their pupils when a bright light was switched on . In contrast , other mutant mice that were unable to produce a light-sensitive pigment in their ipRGCs showed a normal response initially , but had trouble keeping their pupils small if the light stayed on for a longer period of time . The same was true for mice that were missing the PACAP protein in their ipRGCs . These findings show that ipRGCs use different systems to quickly alter the size of the pupil in response to sudden changes in light level and then to maintain the size of the pupil over a longer period of time . Further work is needed to find out if ipRGCs use the same mechanisms to control the other behaviors they influence , such as mood and sleep patterns .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	A visual circuit uses complementary mechanisms to support transient and sustained pupil constriction
The circadian transcriptional repressors cryptochrome 1 ( Cry1 ) and 2 ( Cry2 ) evolved from photolyases , bacterial light-activated DNA repair enzymes . In this study , we report that while they have lost DNA repair activity , Cry1/2 adapted to protect genomic integrity by responding to DNA damage through posttranslational modification and coordinating the downstream transcriptional response . We demonstrate that genotoxic stress stimulates Cry1 phosphorylation and its deubiquitination by Herpes virus associated ubiquitin-specific protease ( Hausp , a . k . a Usp7 ) , stabilizing Cry1 and shifting circadian clock time . DNA damage also increases Cry2 interaction with Fbxl3 , destabilizing Cry2 . Thus , genotoxic stress increases the Cry1/Cry2 ratio , suggesting distinct functions for Cry1 and Cry2 following DNA damage . Indeed , the transcriptional response to genotoxic stress is enhanced in Cry1−/− and blunted in Cry2−/− cells . Furthermore , Cry2−/− cells accumulate damaged DNA . These results suggest that Cry1 and Cry2 , which evolved from DNA repair enzymes , protect genomic integrity via coordinated transcriptional regulation . Mammalian circadian clocks involve transcriptional feedback loops ( Green et al . , 2008 ) : Brain and muscle ARNT-like protein 1 ( BMAL1 ) and ‘circadian locomotor output cycles kaput’ ( CLOCK ) activate expression of many transcripts including the period ( Per1 , Per2 , and Per3 ) and cryptochrome ( Cry1 and Cry2 ) genes , whose protein products ( PERs and CRYs ) inhibit CLOCK and BMAL1 , resulting in rhythmic expression . Posttranslational modifications reset the clock ( Green et al . , 2008 ) , including ubiquitination and subsequent degradation of CRYs by Skp-Cullin-Fbox ( SCF ) E3 ligases in which substrates are recruited by F-box and leucine-rich repeat proteins 3 ( FBXL3 ) ( Busino et al . , 2007; Siepka et al . , 2007 ) and 21 ( FBXL21 ) ( Dardente et al . , 2008; Hirano et al . , 2013; Yoo et al . , 2013 ) . Phosphorylation of CRY1 by AMP-activated protein kinase ( AMPK ) increases its association with FBXL3 ( Lamia et al . , 2009 ) by disrupting interaction with PER ( Xing et al . , 2013 ) . CRY stability seems to be a key factor in circadian period determination: several mutants identified in forward genetic screens selected by robust changes in circadian period have been alleles of FBXL3 or FBXL21 ( Godinho et al . , 2007; Siepka et al . , 2007; Yoo et al . , 2013 ) . In addition to their roles in circadian clock negative feedback , Cry1 and Cry2 are key effectors of a variety of physiological pathways . In mammals , Cry1 and Cry2 modulate glucose homeostasis by repressing the transcriptional activity of the glucocorticoid receptor ( Lamia et al . , 2011 ) and the CRE-responsive element binding protein ( CREB ) ( Zhang et al . , 2010 ) . Consistent with these results , small molecules that stabilize Cry1/2 depress glucose production in hepatocytes and may be useful in the treatment of hyperglycemia ( Hirota et al . , 2012 ) . Genetic disruption of both Cry1 and Cry2 also alters the expression of proinflammatory cytokines ( Narasimamurthy et al . , 2012 ) , the severity of arthritis ( Hashiramoto et al . , 2010 ) , and salt-induced blood pressure elevation ( Doi et al . , 2010 ) . Genetic inactivation of Cry1 and/or Cry2 has also been reported to alter rates of tumor formation ( Ozturk et al . , 2009 ) , though the reported effects have varied ( Fu and Kettner , 2013 ) . In addition , Cry-deficient mice are resistant to genotoxic stress in the context of cyclophosphamide treatment ( Gorbacheva et al . , 2005 ) . Consistent with the idea that Cry1/2 may be promiscuous transcriptional repressors involved in a wide variety of physiological pathways , a recent study found that Cry1 and Cry2 each bound thousands of chromatin sites independently of other clock transcription factors in mouse liver ( Koike et al . , 2012 ) . Though Cry1 and Cry2 are mostly believed to associate with chromatin via binding a variety of transcription factors , they can also interact directly with DNA . Cry1 and Cry2 evolved from prokaryotic light-activated DNA repair enzymes , known as photolyases . While they seem to have lost the [6-4]photolyase catalytic activity characteristic of their ancestral homologs ( Ozturk et al . , 2007 ) , they retain the ability to bind preferentially to UV-damaged DNA containing a [6-4]photoproduct ( Ozgur and Sancar , 2003 ) . The three-dimensional structures of Cry1 and Cry2 resemble those of photolyases , including the DNA binding surfaces ( Maul et al . , 2008; Czarna et al . , 2013; Xing et al . , 2013 ) . Together , these properties suggest that Cry1 and Cry2 could retain a residual role in sensing or responding to damaged DNA . Such conservation of function by divergent molecular mechanisms has been seen previously between cryptochromes derived from different species ( Yuan et al . , 2007; Lamia et al . , 2009; Kim et al . , 2014 ) . Ubiquitination of substrate proteins by E3 ligases , like SCFFbxl3 and SCFFbxl21 , is reversed by ubiquitin-specific proteases ( USPs ) ( Eletr and Wilkinson , 2014 ) . Herpes virus associated ubiquitin-specific protease ( Hausp; a . k . a . Usp7 ) was first identified as the cellular partner of the herpes virus protein Vmw110 ( Everett et al . , 1997 ) . Hausp modulates proliferation by catalyzing the removal of polyubiquitin chains from the tumor suppressor p53 and from the p53-destabilizing E3 ligases Mdm2 and MdmX ( Li et al . , 2002 , 2004 ) . The affinity of Hausp for p53 is increased and for Mdm2/MdmX is decreased in response to DNA damage ( Khoronenkova et al . , 2012 ) , contributing to stabilization of p53 . Knockout of Hausp in mice is lethal ( Kon et al . , 2010 ) , probably due to disrupted cell proliferation . A growing list of Hausp substrates has been identified recently , including several components of DNA damage response and DNA repair pathways ( Nicholson and Suresh Kumar , 2011; Schwertman et al . , 2012; Jacq et al . , 2013; Eletr and Wilkinson , 2014 ) . In this study , we demonstrate that Hausp participates in DNA damage-induced resetting of circadian clock time by stabilizing Cry1 . In an ongoing effort to understand the molecular determinants of cryptochrome stability , we used mass spectrometry to identify novel protein partners of mammalian CRYs and found Hausp to be the most highly enriched protein in Cry1-containing complexes ( Figure 1A–B , Supplementary file 1 ) . Co-immunoprecipitation of endogenous ( Figure 1C ) and overexpressed ( Figure 1D–E ) Cry1 and Hausp confirmed the specificity of this interaction . Interestingly , Hausp interacts much more strongly with Cry1 than with the closely related Cry2 ( Figure 1D ) . Indeed , the divergent Cry1 C-terminus is necessary and sufficient for strong interaction with Hausp ( Figure 1D–E ) . 10 . 7554/eLife . 04883 . 003Figure 1 . Hausp interacts with Cry1 . ( A and B ) Lysates from 293T cells expressing pcDNA3-2xFLAG with no insert ( − ) , Cry1 , or Cry2 after the FLAG tag with ( + ) or without ( − ) co-expression of Per2 were used to purify control , Cry1 , or Cry2-containing complexes by immunoprecipitation ( IP ) of the FLAG tag . 5% of each purification was analyzed by SDS-PAGE and silver stain ( A ) and components of the resulting complexes were identified by mass spectrometry performed on the remaining 95% of the sample . The experiment was performed in triplicate and Pattern Lab for Proteomics ( Carvalho et al . , 2012 ) was used to identify statistically enriched partners . In ( B ) Enrichment is the ratio of spectral counts in Cry1 vs control samples for all statistically enriched partners over three experiments ( e . g . , lane 1 vs lane 5 from [A] ) . Arrows depict several established partners for Cry1 as well as the observed 37-fold enrichment for Hausp in Cry1-containing samples . ( C ) Endogenous Hausp bound to endogenous Cry1 was detected by immunoblot ( IB ) following IP from nuclear and cytoplasmic fractions of mouse embryonic fibroblasts ( MEFs ) harvested at the indicated times ( CT , hours ) following circadian synchronization by dexamethasone . ( D ) Top: Hausp-V5 bound to FLAG-Cry1/2 hybrids was detected by IB following IP from 293T cells . Bottom: schematic diagram showing the composition of the Cry1/2 hybrids and domains used in D and E . ( E ) Hausp-V5 bound to FLAG-Cry1 full length or isolated domains was detected by IB following IP . DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 00310 . 7554/eLife . 04883 . 004Figure 1—figure supplement 1 . Circadian measurement of Hausp mRNA expression in mouse tissues . Hausp expression was measured by quantitative RT-PCR in RNA prepared from mouse liver , quadriceps , spleen , and kidneys harvested at the indicated zeitgeber times ( ZT , hours after lights on ) from wild-type mice housed under normal 12:12 light:dark conditions . Data represent the mean ± s . d . for three samples at each ZT . DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 00410 . 7554/eLife . 04883 . 005Figure 1—figure supplement 2 . Circadian measurement of Hausp protein expression in mouse tissues . Hausp , Per2 , Cry1 , Cry2 , Tubulin , and Lamin were measured by IB in whole cell lysates or nuclei prepared from mouse liver or quadriceps harvested at the indicated ZTs . Each lane on the gel represents a sample collected from a unique animal . DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 005 The regulation of Cry1/2 protein stability is complex and involves differential expression and localization of the E3 ligase subunits Fbxl3 and Fbxl21 that compete for Cry binding and have different rates of ubiquitin conjugation ( Hirano et al . , 2013; Yoo et al . , 2013 ) . Similar to what has been described for Fbxl3 and Fbxl21 , we found no significant tissue specificity or circadian rhythm of expression or localization for Hausp ( Figure 1—figure supplements 1 , 2 ) . However , while both Hausp and Cry1 are more abundant in the cytoplasm , their interaction is stronger in the nucleus , regardless of circadian phase ( Figure 1C , Circadian Time , CT , denotes hours after dexamethasone-induced synchronization of circadian cycles ) . Because Hausp is an ubiquitin-specific protease , its interaction with Cry1 in the nucleus seemed likely to stabilize nuclear Cry1 by removing polyubiquitin chains . We used small hairpin RNA ( shRNA ) -expressing viruses to demonstrate that Hausp depletion led to decreased Cry1 protein primarily in the nuclear compartment in mouse embryonic fibroblasts ( MEFs ) independent of circadian phase , as expected from the ubiquity of Hausp expression ( Figure 2A , Figure 2—figure supplement 1 ) . Treatment of cells with pharmacological inhibitors of Hausp ( Nicholson and Suresh Kumar , 2011; Weinstock et al . , 2012 ) also decreases Cry1 protein , especially in the nucleus ( Figure 2B ) , consistent with the hypothesis that Hausp stabilizes nuclear Cry1 in vivo . ( Note that compound 7 also inhibits Usp47 . ) 10 . 7554/eLife . 04883 . 006Figure 2 . Hausp stabilizes Cry1 via deubiquitination and alters circadian rhythms . ( A ) Wild-type or Cry1−/−;Cry2−/− ( Cry−/− ) MEFs stably expressing a control sequence ( − ) or shRNA targeting Hausp ( #1 ) were subjected to nuclear and cytoplasmic fractionation . Cry1 , Hausp , Lamin , and Tubulin were analyzed by IB from fractions harvested at the indicated times following circadian synchronization with dexamethasone ( CT , hours ) . ( B ) Cry1 , Hausp , Lamin and Tubulin were detected by IB in nuclear and cytoplasmic fractions from MEFs treated with vehicle ( DMSO , − ) or Compound 7 ( + ) . ( C ) Wild-type MEFs stably expressing control or Hausp-targeting shRNA or Cry−/− MEFs were treated with vehicle ( DMSO , − ) or MG132 ( + ) for 6 hr , and lysed in RIPA buffer containing iodoacetamide . 6 mg of RIPA lysates from each condition was subjected to IP with 5 μg of anti-Cry1 antibody . Ubiquitinated Cry1 ( Cry1− ( Ub ) N ) , Cry1 , and Hausp were detected by IB in IPs and whole cell lysates ( WCL ) . ( D , F , H ) Typical results of continuous monitoring of luciferase activity from MEFs expressing Per2-luciferase fusion protein from a knock-in allele ( D and F ) or from U2OS cells stably expressing luciferase under the control of the Bmal1 promoter ( H ) with stable expression of control or either of two shRNA sequences targeting Hausp ( D ) or treated with Compound 7 and/or AICAR ( F and H ) . ( E , G , I ) Quantitation of the circadian period of luciferase activity from experiments performed as described in ( D , F , H ) . Data represent the mean ± s . d . for 4–8 samples per condition . p < 0 . 01 , p < 0 . 001 vs control samples ( control shRNA for E or DMSO-treated cells for G and I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 00610 . 7554/eLife . 04883 . 007Figure 2—figure supplement 1 . Validation of shRNA targeting Hausp . Hausp expression was measured by quantitative RT-PCR in RNA prepared from MEFs stably expressing the indicated shRNA . Data represent the mean ± s . d . for three samples per cell line measured in triplicate . DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 00710 . 7554/eLife . 04883 . 008Figure 2—figure supplement 2 . In vitro deubiquitination of Cry1 by recombinant Hausp . Full-length and ubiquitylated Cry1 were measured by IB following in vitro exposure of purified ubiquitylated Cry1 to the indicated amounts of recombinant USP7 ( Hausp ) or USP8 . Right: quantitation of the western blots shown at left . Data represent a typical result of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 00810 . 7554/eLife . 04883 . 009Figure 2—figure supplement 3 . Quantitation of in vivo Cry1 ubiquitination . Quantitation of ubiquitinated Cry1 immunoprecipitated from MEFs expressing shRNA targeting Hausp ( blue ) or a control sequence ( black ) . Left , western blot from Figure 2C with boxes used for quantitation . The average signal detected in the first two lanes ( background nonspecific signal from Cry1−/−;Cry2−/− cells ) was subtracted from each of the other lanes to generate the data show on the right . DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 009 Recombinant Hausp can deubiquitinate Cry1 in vitro ( Figure 2—figure supplement 2 ) . To examine Cry1 ubiquitination in vivo , we measured ubiquitinated Cry1 in MEFs expressing control or Hausp-targeting shRNA in the presence or absence of the proteasome inhibitor MG132 to stabilize ubiquitinated proteins . Cry1 from Hausp-depleted cells was much more highly ubiquitinated than Cry1 from control cells ( Figure 2C , Figure 2—figure supplement 3 ) as expected if Hausp catalyzes the removal of polyubiquitin chains from Cry1 in vivo . ( Note that while Cry1 is decreased in Hausp-depleted cells , we used a limiting amount of anti-Cry1 antibody for immunoprecipitation to normalize the amount of Cry1 and enable comparison between samples . Cry−/− cells were used as a control to subtract the non-specific ubiquitin signal and enable quantitative comparison of control and Hausp-depleted cells . ) Cryptochrome stability is a critical determinant of circadian period length , though the direction and magnitude of the period change associated with altered expression or stability of Cry1 and/or Cry2 seems to depend on the mechanism and context of altered stability ( Vitaterna et al . , 1999; Hirota et al . , 2012; St John et al . , 2014 ) . Nonetheless , if Hausp stabilizes Cry1 by removing ubiquitin chains , reducing Hausp expression or activity is expected to alter circadian rhythms . In immortalized fibroblasts expressing a Per2-Luciferase fusion from the endogenous Per2 locus ( Per2::Luc [Yoo et al . , 2004] ) , shRNA-mediated depletion of Hausp increased circadian period ( Figure 2D , E ) . We also observed period lengthening in immortalized Per2::Luc MEFs when Hausp activity was inhibited pharmacologically ( Figure 2F , G ) . Because our data suggest that Hausp inhibition and AMPK activation each destabilizes nuclear Cry1 , we examined whether they could synergistically increase circadian period . Using cells stably expressing luciferase under a circadian promoter ( U2OS-B6 [Vollmers et al . , 2008] ) , we observed that activation of AMPK increased the circadian period as expected ( Lamia et al . , 2009 ) , inhibition of Hausp also increased period , and combined activation of AMPK and inhibition of Hausp led to a dramatic increase in period , perhaps reflecting synergistic destabilization of nuclear Cry1 ( Figure 2H , I ) . Given that the Cry1–Hausp interaction occurs primarily in the nucleus and that Hausp interaction with other partners is regulated by DNA damage , we examined the impact of DNA damage on the Hausp–Cry1 association and found that it increases the interaction ( Figure 3A , Figure 3—figure supplements 1 , 2 ) . Because Hausp catalyzes the removal of polyubiquitin chains from Cry1 thereby decreasing its proteasomal degradation ( Figure 2 ) , increased Cry1–Hausp association in response to genotoxic stress leads to a prediction that DNA damage should increase Cry1 protein levels . Consistent with this hypothesis , we found that exposure to DNA damage transiently stabilized endogenous Cry1 in primary MEFs ( Figure 3A–C ) . Intriguingly , Cry2 was destabilized following exposure to DNA damage , demonstrating that the increase in Cry1 does not merely reflect a change or synchronization of the circadian rhythm and suggesting differential regulation of these highly homologous family members , consistent with our observation that Hausp preferentially interacts with Cry1 . Because Cry1 and Cry2 each can repress the other's expression , Cry2 protein could decrease in response to damage secondary to stabilization of Cry1 . However , Cry2 protein decreases and Cry1 protein increases in response to DNA damage in MEFs expressing only a single Cry paralog ( i . e . , Cry2 in Cry1−/− MEFs and vice versa; Figure 3B ) . Thus , DNA damage acutely regulates Cry1 and Cry2 protein levels independently . 10 . 7554/eLife . 04883 . 010Figure 3 . DNA damage resets the clock via Hausp-dependent stabilization of nuclear Cry1 . ( A ) Endogenous Hausp , Cry1 , Cry2 , phospho-P53 ( Ser15 ) , P53 , Lamin , and Tubulin were detected by IB in Cry1 immunoprecipitates or input samples from nuclear and cytoplasmic fractions of primary MEFs treated with vehicle ( − ) or doxorubicin ( + ) . ( B ) Cry1 , Cry2 , phospho-P53 ( Ser15 ) , and Actin were detected by IB in lysates from wildtype ( WT ) , Cry1−/− or Cry2−/− MEFs treated with doxorubicin for the indicated times . ( C ) Cry1 , Cry2 , phospho-P53 ( Ser15 ) , Hausp , Lamin , and Tubulin were detected by IB in nuclear and cytoplasmic fractions from MEFs expressing control or Hausp-targeting shRNA and treated with doxorubicin for the indicated times . ( D and E ) Typical results of continuous monitoring of luciferase activity from primary adult ear fibroblasts expressing Bmal1-luciferase and control or Hausp-targeting shRNA and treated with 0 ( black ) or 10 Gy ( red ) irradiation 3 hr after circadian synchronization with dexamethasone . Data represent the mean luciferase counts of eight samples per condition from one of four independent experiments . ( F ) Quantitation of the differences in initial circadian phase of luciferase activity caused by irradiation calculated from experiments performed as described in ( D and E ) . Data in ( D–F ) represent the mean ± s . d . of phase shifts observed in four independent experiments . p < 0 . 01 vs control samples . DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 01010 . 7554/eLife . 04883 . 011Figure 3—figure supplement 1 . Effect of DNA damage on Cry1-Hausp interaction in transfected 293T cells . HAUSP-V5 , FLAG-Cry1 , and Actin were detected by IB in IPs or whole cell lysates ( WCL ) from 293T cells transfected with the indicated plasmids ( by calcium phosphate method ) and treated with doxorubicin . DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 01110 . 7554/eLife . 04883 . 012Figure 3—figure supplement 2 . Proteostasis and/or membrane stress increase the Cry1-Hausp interaction . HAUSP-V5 , FLAG-Cry1 , and Actin were detected by IB in IPs or whole cell lysates ( WCL ) from 293T cells transfected with the indicated plasmids ( by calcium phosphate CaP , or PEI method ) and treated with ( A ) Mg132 or ( B ) 10 Gy ionizing radiation ( IR ) . Note that the Cry1-Hausp interaction , which can be increased by DNA damage stimuli , is independently induced by other stress signals including proteostasis stress and membrane stress . The degree to which Cry1 and/or Hausp are overexpressed in transfected cells influences their affinity of interaction . The method of transfection also altered their interaction . We consistently observed induced interaction between Cry1 and Hausp after DNA damage in cells transfected with low concentrations of plasmids using standard calcium phosphate transfection protocols . However , cells transfected with PEI or otherwise expressing excessive amounts of exogenous proteins or treated with the proteasome inhibitor Mg132 displayed high basal interaction between Cry1 and Hausp that was not further increased upon treatment with DNA damaging agents . DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 01210 . 7554/eLife . 04883 . 013Figure 3—figure supplement 3 . Circadian time of exposure determines phase shift in response to DNA damage . Typical results of continuous monitoring of luciferase activity from mouse embryonic fibroblasts expressing Per2::Luciferase treated with 0 ( black curves ) or 10 Gy ( red curves ) ionizing radiation 4 ( CT4 ) or 18 ( CT18 ) hours after circadian synchronization with 1 μM dexamethasone . Data represent the mean luciferase counts of six samples per condition from one of two independent experiments . Right: quantitation of the differences in initial circadian phase of luciferase activity caused by irradiation . Data represent the mean ± propagated s . d . difference between initial phase in Mock vs irradiated samples for six samples per condition . *p < 0 . 001 for a significant interaction between CT and irradiation by 2-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 013 To determine the contribution of Hausp to the damage-induced stabilization of nuclear Cry1 , we examined nuclear Cry1 protein levels following DNA damage in MEFs expressing either control sequences or Hausp-targeting shRNA . Indeed , depletion of Hausp prevents damage-induced stabilization of nuclear Cry1 , similar to the effect of Hausp depletion on p53 accumulation ( Figure 3C ) . Note that the weaker interactions that we observed between Hausp and Cry2 or hybrid constructs containing the C-terminus of Cry2 compared to those containing the Cry1 C-terminus ( Figure 1D ) are likely artefacts of overexpression in 293T cells since we did not observe destabilization of Cry2 upon Hausp depletion ( Figure 3C ) . It has been reported that DNA damage causes phase shifts of circadian rhythms ( Oklejewicz et al . , 2008; Engelen et al . , 2013 ) . Consistently , we observed phase shifts in primary MEFs with a peak shift following irradiation at CT2-4 , ( Figure 3—figure supplement 3 ) . The requirement for Hausp in stabilization of nuclear Cry1 after DNA damage suggested Hausp could contribute to phase shifts in response to DNA damage . By examining the circadian phase of control and Hausp-depleted fibroblasts after exposure to irradiation at CT3 , we found that although the circadian phase of the non-irradiated cells is similar ( Figure 2D ) , DNA damage-induced phase shifts were greatly diminished in Hausp-deficient fibroblasts ( Figure 3D–F ) . ATM- and PPM1G-dependent dephosphorylation of serine 18 in the N-terminus of Hausp has been reported to drive the DNA damage dependent disruption of Hausp interaction with Mdm2 and MdmX ( Khoronenkova et al . , 2012 ) . Conversely , S18 de-phosphorylation may increase Hausp–Cry1 association because mutation of S18 to the non-phosphorylatable amino acid alanine ( S18A ) increases interaction and mutation to aspartic acid , which is chemically similar to phospho-serine , decreases the interaction ( Figure 4A ) . However , S18 dephosphorylation cannot fully explain DNA damage induction of Cry1–Hausp interaction as evidenced by persistent stimulated association between Cry1 and Hausp S18A after DNA damage . Intriguingly , we ( Figure 4—figure supplement 1 , Supplementary file 2 ) and others ( Gao et al . , 2013 ) find that Cry1 and Cry2 interact with kinases that are activated by DNA damage and phosphorylate serine or threonine followed by glutamine , ( S/T ) -Q ( Kim et al . , 1999; O'Neill et al . , 2000 ) . Cry1 and Cry2 contain several such sequences ( Figure 4—figure supplement 2 ) , including three serines in the Cry1 C-terminal tail that are not conserved in Cry2 . 10 . 7554/eLife . 04883 . 014Figure 4 . DNA damage induced signaling modulates interactions of Cry1/2 , Hausp , and Fbxl3 . Hausp-V5 , FLAG-Cry1/2 , phospho-P53 ( Ser15 ) , Phospho-SQ/TQ , Phospho-Cry1S588 ( P-S588 ) , Fbxl3-V5 , and Actin were detected by IB in IPs and lysates ( WCL ) from 293T cells transfected with the indicated plasmids and lysed at the indicated times following treatment with doxorubicin ( doxo ) or irradiation ( IR ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 01410 . 7554/eLife . 04883 . 015Figure 4—figure supplement 1 . Composition of Cry1- and Cry2-associated protein complexes . ( A and B ) Lysates from 293T cells expressing pcDNA3-2xFLAG with no insert , Cry1 , or Cry2 were used to purify control , Cry1 , or Cry2-containing complexes by immunoprecipitation ( IP ) of the FLAG tag . Components of the resulting complexes were identified by mass spectrometry . The experiment was performed in triplicate and PatternLab for Proteomics ( Carvalho et al . ) was used to identify statistically enriched partners . Enrichment is the ratio of spectral counts in Cry1 or Cry2 vs control samples for the top 150 statistically enriched partners over three experiments . Arrows depict several established partners for Cry1 and Cry2 as well as the observed 18-fold and 27-fold enrichment for DNA-PKcs in Cry1- and Cry2-containing samples , respectively . ( C and D ) Biological processes identified by Gene Set Enrichment Analysis to be statistically enriched in Cry1-associated complexes ( C ) or Cry2-associated complexes ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 01510 . 7554/eLife . 04883 . 016Figure 4—figure supplement 2 . Conserved SQ/TQ motifs present in Cry1 and/or Cry2 . Sequence alignment of mouse and human Cry1 and Cry2 indicating the positions and conservation of several SQ/TQ motifs . ( Numbers correspond to the amino acid positions in mouse Cry1 . ) DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 016 Using an antibody that recognizes phospho- ( S/T ) Q , we determined that Cry1 is rapidly phosphorylated in response to either chemically or radiation-induced DNA damage , while damage-induced phosphorylation of Cry2 on ( S/T ) Q was reduced and delayed compared to that of Cry1 ( Figure 4B and not shown ) , indicating that the rapid phosphorylation of Cry1 in response to DNA damage likely occurs on the non-conserved C-terminal tail . Mutating each or all of the Cry1 C-terminal SQ serines to alanine decreased or abolished , respectively , the phospho- ( S/T ) Q signal after DNA damage , indicating that these are the sites in Cry1 that are rapidly phosphorylated in response to DNA damage ( Figure 4C , and not shown ) . Notably , S588 is the only one of these sites on which phosphorylation has been directly detected in vivo ( Lamia et al . , 2009; Hegemann et al . , 2011 ) . We generated an antibody that specifically recognizes Cry1 phosphorylated on S588 and measured a rapid increase in the presence of this phosphorylated species after exposure to DNA damage ( Figure 4D ) . Consistent with the reported stabilization of Cry1 by mimicking phosphorylation at S588 , mutation of this site to aspartic acid increased its association with Hausp ( Figure 4E , left ) . Because the effects of DNA damage on Cry1 and Cry2 stability are not fully explained by regulated interaction with Hausp , we examined the effect of DNA damage and subsequent phosphorylation events on the interactions between Cry1/2 and Fbxl3 . To our surprise , prolonged exposure to DNA damage increases the interactions of Cry1 and especially Cry2 with Fbxl3 ( Figure 4F ) , probably contributing to the transient nature of the Cry1 stabilization and to Cry2 destabilization following damage . Notably , mutation of Cry1 S588 to aspartic acid , which increases Cry1 interaction with Hausp , decreases the association of Cry1 with Fbxl3 ( Figure 4E , right ) suggesting that phosphorylation of the unique Cry1 C-terminus may oppose the increase in Fbxl3 binding to Cry1 , possibly explaining the preferential induction of Fbxl3 binding to Cry2 compared to Cry1 . Given that Cry1 and Cry2 are transcriptional repressors and that we found a robust regulation of their stability by DNA damage , we asked whether the transcriptional response to DNA damage is altered by genetic disruption of Cry1 or Cry2 . By measuring the induction of transcripts activated by DNA damage in fibroblasts ( Kenzelmann Broz et al . , 2013 ) , we found that genetic loss of Cry1 or Cry2 enhances or suppresses , respectively , the induction of Cdkn1a ( p21 ) by genotoxic stress and alters the dynamic response of other established damage responsive transcripts as well ( Figure 5A–E and Figure 5—figure supplement 1 ) . Although the chromatin association of cryptochromes may be different in different cell types , both Cry1 and Cry2 bind some of these loci in mouse liver ( Koike et al . , 2012 ) ( Figure 5—figure supplement 2 ) . In addition , Cry1 and Cry2 bind to chromatin regions near several genes encoding proteins that participate in DNA repair ( Supplementary file 3 ) . Interestingly , the expression of several of those genes in response to DNA damage is also altered by genetic loss of Cry1 and/or Cry2 ( Figure 5F–I ) , suggesting that cryptochromes may modulate the activation of DNA repair in response to damage . The regulation of some transcripts in Cry1−/−;Cry2−/− cells resembles that in Cry1−/− cells ( e . g . , Rrm2b , Gadd45a , p16ink4a ) , suggesting that Cry1 is more relevant to their regulation than is Cry2 . For other transcripts ( e . g . , p21 , Puma , Xrcc1 ) , the response to DNA damage in Cry1−/−;Cry2−/− cells is closer to the response in Cry2−/− cells suggesting that Cry2 is more important for regulation of those targets . A full understanding of how Cry1 and Cry2 influence gene expression following DNA damage will require further study . 10 . 7554/eLife . 04883 . 017Figure 5 . Cry1/2 deficiency alters transcriptional response to DNA damage . Expression of the indicated transcripts was measured by quantitative PCR ( qPCR ) in cDNA from wildtype ( black ) , Cry1−/− ( blue ) , Cry2−/− ( red ) , and Cry1−/−;Cry2−/− ( gray ) fibroblasts treated with doxorubicin for the indicated times . p < 0 . 05 , p < 0 . 01 , p < 0 . 001 for effect of genotype by repeated measures ANOVA analysis ( blue—WT vs Cry1−/−; red—WT vs Cry2−/−; gray—WT vs Cry1−/−;Cry2−/− ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 01710 . 7554/eLife . 04883 . 018Figure 5—figure supplement 1 . Transcriptional response to irradiation-induced DNA damage . Expression of the indicated transcripts was measured by quantitative PCR ( qPCR ) in cDNA prepared from wild-type ( black ) , Cry1−/− ( blue ) , and Cry2−/− ( red ) fibroblasts at the indicated times following exposure to 5 Gy irradiation . DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 01810 . 7554/eLife . 04883 . 019Figure 5—figure supplement 2 . Circadian pattern of Cry1 and Cry2 binding to selected chromatin sites . Association of Cry1 ( blue ) or Cry2 ( red ) with chromatin at the indicated locations in ChIP sequencing data set published by Koike et al . ( 2012 ) . Data represent the reported Cry1 or Cry2 signal normalized to the reported ‘input’ signal for each locus . DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 019 We next asked whether disruption of Cry1/2-dependent transcriptional regulation causes a functional deficit in the cellular response to DNA damage in cryptochrome-deficient cells . Indeed , Cry2−/− and Cry1−/−;Cry2−/− fibroblasts accumulate damaged DNA , reflected by an increased percentage of non-dividing cells containing multiple 53BP1-positive foci ( Figure 6A , B ) . Accumulation of damaged DNA in cells lacking Cry2 was surprising given that Cry2−/− mice are viable and fertile and that genetic disruption of Cry1 and Cry2 decreases tumor formation in p53-deficient animals ( Ozturk et al . , 2009 ) . To determine whether the increased accumulation of DNA damage that we observed in cells could possibly be relevant in vivo , we analyzed breeding records of a large number of progeny from Cry1+/−;Cry2+/− mice over several years: while Cry1 genotypes segregate in the expected Mendelian ratios , the Cry2−/− genotype is significantly underrepresented ( Figure 6C ) . This is similar to reduced survival observed in mice with genetic defects in established components of the DNA damage response or DNA repair pathways ( Tsai et al . , 2005; Mukherjee et al . , 2010; Crossan et al . , 2011 ) and is consistent with the possibility that animals lacking Cry2 are prone to genetic instability , though we cannot exclude other possible explanations for the reduced viability of Cry2−/− mice . 10 . 7554/eLife . 04883 . 020Figure 6 . Cry2−/− cells accumulate damaged DNA . ( A ) Representative early passage ( P3–4 ) primary wildtype ( WT ) , Cry1−/− , Cry2−/− , and Cry1−/−;Cry2−/− adult ear fibroblasts stained with anti-53BP1 antibody ( green ) and DAPI ( blue ) . Insets show enlarged view of indicated cells . ( B ) Quantitation of 53BP1-positive cells prepared as described in ( A ) . Nuclei containing more than five 53BP1 puncta and negative for BrdU labeling were considered positive for DNA damage . Data represent the mean ± s . d . for at least 200 cells per genotype . ( C ) Chi-squared analysis of the distributions of Cry1 and Cry2 wildtype ( black ) , heterozygous ( gray ) , and homozygous knockout ( red ) genotypes establishes a significantly reduced survival of Cry2−/− mice . *p < 0 . 01 by chi-squared analysis with 2 degrees of freedom ( χ2 = 10 . 39 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 020 Diverse organisms use circadian clocks to optimize the timing of physiological processes in relation to predictable diurnal changes in the external environment ( Dodd et al . , 2005; Lamia et al . , 2008; Marcheva et al . , 2010; Sadacca et al . , 2011 ) . It has long been suspected that clocks influence the timing of cell division to temporally separate DNA replication from predictably recurring exposure to DNA damage ( Sahar and Sassone-Corsi , 2009; Sancar et al . , 2010 ) . This hypothesis is supported by the non-random distribution of cell division within circadian cycles ( Nagoshi et al . , 2004 ) . Theoretically , in order for clocks to enable such a separation , their timing must be responsive to genotoxic stress , analogous to entrainment by metabolic signals , which enables clocks to optimally coordinate cellular metabolism with externally fluctuating metabolic demands ( Ramsey and Bass , 2011; Jordan and Lamia , 2013 ) . Indeed , others have shown that DNA damage shifts circadian clock time ( Oklejewicz et al . , 2008 ) . In this study , we confirm that phenomenon and describe a molecular mechanism by which Hausp-mediated deubiquitination and stabilization of Cry1 contributes to it . Furthermore , we provide evidence for specific and divergent roles of the circadian transcriptional repressors Cry1 and Cry2 in modulating the transcriptional response to DNA damage , thus addressing the longstanding question of whether circadian rhythmicity per se is sufficient to minimize the occurrence or accumulation of DNA damage . Indeed , our results suggest that circadian rhythmicity as such is not protective because Cry2−/− cells maintain robust circadian rhythms ( Khan et al . , 2012 ) , but they accumulate DNA damage at rates comparable to arrhythmic Cry1−/−;Cry2−/− cells . Therefore , it appears that the genome-protective function of normal circadian rhythms is dependent on the expression of specific clock components , including Cry2 . This distinction may underlie some of the controversy over the importance of circadian clocks for maintaining genomic integrity ( Fu and Kettner , 2013 ) . Increased accumulation of DNA damage in Cry2−/− cells would be expected to lead to an increased mutation rate; consistent with that hypothesis , we observed sub-Mendelian inheritance of the Cry2−/− genotype . Though the potential for circadian clocks to influence the cellular response to DNA damage has been controversial ( Gaddameedhi et al . , 2012 ) , our results are also consistent with several lines of evidence supporting a conserved role for clocks in modulating the DNA damage response and/or DNA repair ( Kang et al . , 2009 , 2010; Cotta-Ramusino et al . , 2011; Gaddameedhi et al . , 2011 ) . Interestingly , we identified several proteins that specifically bind damaged DNA or participate in DNA repair in Cry1- and/or Cry2-associated complexes ( Supplementary files 1 , 2 ) , suggesting that cryptochromes may influence DNA repair by non-transcriptional mechanisms as well . While Cry1 and Cry2 have long been established as repressors of Clock:Bmal1-driven gene expression , and we observe altered expression of several transcripts in response to DNA damage in Cry1/2-deficient cells , it remains unclear how Cry1/2 regulate gene expression . Though it is not the focus of this study , the composition of the Cry1- and Cry2-associated protein complexes suggests that regulation of mRNA processing may be an important function for Cry1 and Cry2: a large number of RNA binding and RNA processing factors were found associated with Cry1 and Cry2 ( Figure 4—figure supplement 2; Supplementary files 1 , 2 ) . This is consistent with other recent literature describing the association of RNA processing machinery in complex with Per proteins ( Padmanabhan et al . , 2012 ) and the importance of post-transcriptional regulation in circadian clock function generally ( Kojima et al . , 2011 ) . It has long been hypothesized that the C-termini of cryptochromes are important for distinguishing their species-specific functions ( Green , 2004 ) and for enabling regulated interactions with protein and possible nucleic acid partners ( Czarna et al . , 2011; Zoltowski et al . , 2011; Engelen et al . , 2013 ) . Here , we identify a Cry1-specific partner ( Hausp ) that interacts with the C-terminus in isolation . Furthermore , we describe phosphorylation events in Cry1 , Cry2 , and Hausp that are influenced by DNA damage and contribute to their regulated interactions . It appears that DNA damage initiates a complex cascade of signal transduction that alters circadian clock dynamics in a complicated manner . The large number of phosphorylation events in Cry1 , Cry2 , and Hausp induced by genotoxic stress supports an important role for these proteins in sensing or responding to such stress . Understanding the specific functions of each of these modifications will require further study . These results suggest that mammalian cryptochromes , Cry1 and Cry2 , represent a node of interaction between circadian clocks and the cellular response to genotoxic stress . Cry1 and Cry2 play non-redundant roles in this pathway , highlighting the importance of analyzing their roles independently rather than relying on the exclusive use of doubly deficient cells and animals to understand their functions . Further study will determine whether Cry1 and Cry2 modulate the transcriptional response to DNA damage via sequence-specific DNA binding transcription factors or through direct binding of damaged DNA as has been observed in vitro ( Ozgur and Sancar , 2003 ) . Finally , accumulation of damaged DNA in Cry2−/− cells suggests that Cry2 is a particularly important integrator of circadian rhythms and genomic integrity . Taking all of these data together , we conclude that Cry1 and Cry2 cooperatively regulate the transcriptional response to genotoxic stress and the inverse relationship of their stability in response to DNA damage contributes to transient activation of gene networks that protect genome integrity ( Figure 7 ) . 10 . 7554/eLife . 04883 . 021Figure 7 . Model depicting a novel mechanism by which the regulation of Cry1 and Cry2 enables coordination of the transcriptional response to genotoxic stress . In quiescent cells , Cry1 and Cry2 repress transcription of target genes . Upon DNA damage , Cry2 is degraded , relieving repression . As Cry1 accumulates , it replaces Cry2 and returns gene expression to normal levels resulting in transient activation . In Cry1−/− cells , gene expression is enhanced , while in Cry2−/− cells , damage-induced transcription is suppressed . Note that this model does not explain the dynamics of altered response observed for all transcripts but may apply to the average change in the transcriptional response to DNA damage in Cry1/2-deficient cells . DOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 021 For mass spectrometry analysis , samples were denatured , reduced , and alkylated before an overnight digestion with trypsin . Peptide mixtures were analyzed by liquid chromatography mass spectrometry using an Accela pump and LTQ mass spectrometer ( Thermo Fisher Scientific , Waltham , MA ) using a four-step multidimensional protein identification technology ( MudPIT ) separation ( MacCoss et al . , 2002 ) . Tandem mass spectrometry spectra were collected in a data-dependent fashion and resulting spectra were extracted using RawXtract . Protein identification was done with Integrated Proteomics Pipeline ( IP2 ) by searching against the UniProt Human database and filtering to <1% false positive at the protein level using DTASelect . Statistically enriched partners for Cry1 were identified by Pattern Lab ( Carvalho et al . , 2012 ) . All cells were grown in complete Dulbecco's Modified Eagle Medium ( DMEM ) ( cat #10569; Invitrogen ) unless otherwise indicated . 293T and U2OSB6 cell media were supplemented with 10% fetal bovine serum , and 1% penicillin and streptomycin . MEF media were supplemented with 15% fetal bovine serum , and 1% penicillin and streptomycin . 293T cells were grown in a 37°C incubator maintained at 5% CO2 and 20% O2 ( high oxygen ) . MEF cells were grown in high oxygen conditions as above ( Figure 2C , D , F ) or a 37°C incubator maintained at 5% CO2 and 3% O2 ( low oxygen ) . Note that in most cases , the MEFs were initially cultured in high oxygen at the time of harvest even when they were later grown in low oxygen . MG-132 was used at a concentration of 10 μM for 6–8 hr or overnight as indicated . AICAR ( cat #A61170010; Toronto Research Chemicals , Canada ) was used at a concentration of 1–2 mM as indicated . Hausp inhibitor Compound 7 ( Progenra , Malvern , PA ) was used at a concentration of 10 μM for 6–8 hr prior to cell lysis or as indicated . Doxorubicin ( cat #ICN15910101; Thermo Fisher Scientific ) was used at a concentration of 0 . 5 μg/ml for 16–24 hr or as indicated . Ionizing radiation exposure was performed using a 137Cs γ-radiation source at the indicated times after dexamethasone synchronization . Transfections were carried out using calcium phosphate or polyethylenimine ( cat #23966-2; PEI; Polysciences Inc , Warrington , PA ) by standard protocols . pcDNA3-2xFlag-mCRY1 , pcDNA3-2xFlag-mCRY2 , and pcDNA3-Fbxl3-v5 are as described previously ( Lamia et al . , 2009 ) . pCl-neo Flag HAUSP deposited by Dr Bert Vogelstein was purchased from Addgene ( Addgene plasmid 16655 ) ( Cummins and Vogelstein , 2004 ) and cloned into pcDNA 3 . 2/V5/GW-CAT purchased from Invitrogen ( cat #K244020 ) using standard protocols . Lentiviruses expressing Bmal1-luciferase and Per2-luciferase were from Dr Satchidananda Panda . Five shRNAs against Hausp and one shRNA against Gapdh were purchased from Open Biosystems . pLKO . 1 sh_scramble deposited by Dr David Sabatini was purchased from Addgene ( Addgene plasmid 1864 ) ( Sarbassov et al . , 2005 ) . Either sh_Scramble or sh_Gapdh was used as controls for sh_Hausp . pLenti-lox-GFP shRNA p19-2 for immortalizations deposited by Dr Tyler Jacks was purchased from Addgene ( Addgene plasmid 14091 ) ( Sage et al . , 2003 ) . psPAX plasmid ( Addgene plasmid 12 , 260 ) and pMD2 . G plasmid ( Addgene plasmid 12259 ) deposited by Dr Didier Trono used for infection also purchased from Addgene . Cry hybrid constructs were a gift from Dr Andrew C Liu ( Khan et al . , 2012 ) ; the hybrid coding sequences were transferred to pcDNA3-2xFlag using standard protocols; several observed mutations in the hybrid coding sequences were corrected by site-directed mutagenesis . All mutations were generated using Agilent Site-Directed Mutagenesis kit and protocols ( cat #200521 ) . HEK 293T cells were from the American Type Culture Collection ( ATCC , Manassas , VA ) . U2OS-B6 cells were a gift from Dr Satchidananda Panda . MEFs were isolated from embryos of the indicated genotypes at E15 . 5 and were used as primary ( passaged no more than 10 times and grown in 3% oxygen ) , immortalized with pLenti-lox-GFP shRNA p19-2 , or spontaneously immortalized . Ear Fibroblasts were isolated from 3-month-old littermates . Ear punches were put in 70% ethanol for 2 min , washed in PBS , cut into small pieces using a scalpel and transferred to a 15-ml tube . 2 ml of trypsin 0 . 25% was added and samples were incubated for 1 hr at 37°C in a water bath , vortexing briefly every 10 min . The trypsin was inactivated with 8 ml of EF media ( DMEM/15%FBS/PS1% ) . Cells were spun down 5 min at 1000 rpm , re-suspended in 3 ml of EF media and transferred into a well of a 6-well plate . The medium was changed the next day . Fibroblasts grew after 3–5 days . Lentiviral shRNA were transfected into HEK 293T cells using psPAX and pMD2 . G packaging plasmids for virus generation . Viral supernatants were collected 48 hr after transfection , filtered through a 0 . 45-μm filter , supplemented with 6 μg/ml polybrene and added to parental cell lines . After 4 hr , additional media were added to dilute the polybrene to <3 μg/ml . 48 hr after viral transduction , the infected cells were split into selection media containing 5 μg/ml puromycin . Selection media were replaced every 2–3 days until selection was complete ( as judged by death of mock-infected cells; typically 1–2 weeks ) . 293T whole cell extracts and mouse liver lysates were prepared using Lysis buffer containing 1%TX-100 as previously described ( Lamia et al . , 2004 ) . MEF cell extracts were prepared from RIPA buffer containing 1% TX-100 , 147 mM NaCl , 12 mM sodium deoxycolate , 0 . 1% SDS , 50 mM Tris pH 8 . 0 , 10 mM EDTA , 50 μM PMSF , phosphatase inhibitors ( cat #P5266 and cat #P0044; Sigma , St . Louis , MO ) and protease inhibitor ( cat #11697498001; Roche , Switzerland ) . For ubiquitination experiments , iodoacetamide was added to the buffer to a final concentration of 5 mM ( Fisher AC122270050 ) . Antibodies were anti-Flag M2 agarose beads , anti-Flag polyclonal , anti-v5 polyclonal , anti-Lamin A , anti-aTubulin , and anti-βactin from Sigma , anti-Hausp and anti-V5 from Bethyl Labs ( Montgomery , TX , cat #A300-033A and cat #A190-120A ) , anti-53BP1 from Novus Biologicals ( Littleton , CO , NB100-304 ) , Cry1-CT and Cry2-CT as described ( Lamia et al . , 2011 ) , anti-p21 from Santa Cruz Biotechnologies ( Dallas , TX , cat #sc-6246 ) , anti-p53 as previously described ( Pasini et al . , 2004 ) , and anti-Ubiquitin , anti-phospho-P53 ( S15 ) , and anti-phosphoATM/ATR substrate ( phospho-SQ/TQ ) from Cell Signaling Technology ( Danvers , MA ) . Anti-Cry1-phosphoS588 antibody was affinity purified from rabbit antisera raised against a phospho-S588 containing peptide . Cells were grown on glass coverslips and pulse-labeled for 30 min by adding 10 µM of BrdU to the cell culture medium , washed three times with PBS before fixation with 4% ( wt/vol ) paraformaldehyde in PBS for 15 min at room temperature ( RT ) and permeabilized with 0 . 5% ( vol/vol ) Triton X-100 in PBS for 10 min at RT . Coverslips were blocked with 1% BSA in PBS for 30 min at RT . For BrdU co-staining , cells were subjected to a DNase I treatment ( Sigma; 200 U/ml in 30 mM Tris HCl pH 8 . 1 , 0 . 33 mM MgCl2 , 0 . 5 mM Mercaptoethanol , 1% BSA , and 0 . 5% Glycerol ) for 1 hr at 37°C in the presence of anti-BrdU 1/50 ( BD Pharmingen ) . Then , coverslips were washed three times with PBS prior to incubation with primary antibodies ( anti-53BP1; 1/3000 ) for 2 hr at RT in blocking buffer . Cells were washed with PBS and incubated with secondary antibodies ( Alexa Fluor 488 goat anti-rabbit 1/150 and Alexa Fluor 594 goat anti-mouse 1/150 ) for 1 hr at RT in blocking buffer . Cells were then washed three times with PBS and stained 15 min with DAPI ( 0 . 4 μg/ml in PBS1X ) to visualize DNA . The coverslips were mounted onto glass slides with Fluoromount G ( Electron microscopy Science ) . For quantification , at least 200 cells were counted following IF analysis . Cells with at least five 53BP1 foci and negative for BrdU labeling were considered positive for DNA damage . Images were processed using Image J software . Cells were washed once with ice cold PBS , fresh cold PBS was added and the cells were transferred to a 5-ml tube and centrifuged 5 min at 2000 rpm . The resulting pellets were washed with cold PBS and transferred to 1 . 5-ml eppendorf tubes and centrifuged 5 min at 2000 rpm . The resulting pellets were resuspended in Solution A ( 10 mM Hepes pH 8 , 1 . 5 mM MgCl2 , 10 mM KCl , plus protease and phosphatase inhibitors ) , and incubated for 15 min at 4°C . An equal volume of Solution B ( solution A + 1% NP40 ) was added and the samples were further incubated for 5 min at 4°C and centrifuged 5 min at 3000 rpm . Supernatants from this step represent the cytoplasmic fraction . The remaining nuclear pellets were then washed twice with cold PBS , lysed in RIPA buffer and either used directly ( nuclear lysates ) or diluted sixfold into IP buffer for immunoprecipitation . U2OSB6 cells , MEFs , or adult ear fibroblasts were plated at 100% confluency in 35-mm dishes ( cat #82050-538; VWR , Radnor , PA ) . The next day , cells were treated for 1–2 hr in normal growth medium containing 1 mM dexamethasone and 100 μM D-luciferin . Media were removed and replaced with media containing DMEM , 5% FBS , 1% penicillin-streptomycin , 15 mM Hepes , pH 7 . 6 , and 100 μM D-luciferin . Plates were sealed with vacuum grease ( Dow Corning high vacuum grease; cat #59344-055; VWR ) and glass cover slips ( cat #22038999; 40CIR-1 , Fisher Scientific ) and placed into the Lumicycle 32 from Actimetrics , Inc . ( Wilmette , IL ) . Data were recorded using Actimetrics Lumicycle Data Collection software and analyzed using Actimetrics Lumicycle Analysis program . Background subtraction of the recorded data was performed with Running Average setting , and fit by least mean squares calculation to a damped sine wave to calculate the period , amplitude , and phase of the curves . Only data with a goodness of fit percentage of 80 or above was included in the analysis . RNA was extracted from mouse tissues or cells with Qiazol reagent using standard protocols ( cat #799306; Qiagen , Germany ) . cDNA was prepared using QScript cDNA Supermix ( cat #101414-106; VWR ) and analyzed for gene expression using quantitative real-time PCR with iQ SYBR Green Supermix ( cat #1708885; Biorad , Hercules , CA ) . For analysis of transcriptional response to DNA damage ( doxorubicin or irradiation ) , cells were used at approximately 70% confluency ( Table 1 ) . 10 . 7554/eLife . 04883 . 022Table 1 . Primers used for qPCRDOI: http://dx . doi . org/10 . 7554/eLife . 04883 . 022Cdkn2a ( p21 ) :Fwd: CCAGGCCAAGATGGTGTCTTRev: TGAGAAAGGATCAGCCATTGCMdm2:Fwd: CTGTGTCTACCGAGGGTGCTRev: CGCTCCAACGGACTTTAACARrm2b:Fwd: GACAGCAGAGGAGGTTGACTTGRev: AAAACGCTCCACCAAGTTTTCAPuma:Fwd: GTACGGGCGGCGGAGACGAGRev: GCACCTAGTTGGGCTCCATTTCTGGadd45a:Fwd: AAGACCGAAAGGATGGACACGRev: CAGGCACAGTACCACGTTATCRad23b:Fwd: ACCTTCAAGATCGACATCGACCRev: ACTTCTGACCTGCTACCGGAARad51l3:Fwd: GGAGCTTTGTGCCCAGTACCRev: TCCCCAATGTCCCAATGTCTATXrcc1:Fwd: AGCCAGGACTCGACCCATTRev: CCTTCTCCAACTGTAGGACCAp16ink4a:Fwd: GTGTGCATGACGTGCGGGRev: GCAGTTCGAATCTGCACCGTAGRad51:Fwd: TCACCAGCGCCGGTCAGAGARev: CCGGCCTAAAGGTGCCCTCG 293T cells transiently expressing Flag-tagged Cry1 were treated with 10 μM MG132 for 18 hr and lysed in RIPA buffer containing Roche complete protease inhibitors , 1 mg/ml iodoacetamide , and 50 μM PMSF . Flag-Cry1 was immunoprecipitated for 2 hr with M2-agarose ( Sigma A2220 ) , washed five times in RIPA buffer and three times in reaction buffer , and eluted for 1 hr in reaction buffer ( 60 mM Hepes pH 7 . 4 , 5 mM MgCl2 , 4% glycerol , 2 μg/ml aprotinin , 50 μM PMSF , 2 mg/ml BSA ) containing 3XFLAG peptide . Equal volumes of eluted Flag-Cry1 were combined with the indicated amounts of recombinant Hausp ( cat #E-519; USP7 , Boston Biochem ) or USP8 ( cat #E-520; Boston Biochem , Cambridge , MA ) and the reactions were incubated for 30 min at 30°C before adding SDS sample buffer and boiling for 5 min . The resulting samples were separated by 8% SDS-PAGE and Cry1 was detected by immunoblot . Cry1−/−;Cry2−/− mice were from Dr Aziz Sancar ( Thresher et al . , 1998 ) ; Per2::Luciferase mice ( Yoo et al . , 2004 ) were purchased from Jackson laboratories ( Bar Harbor , ME ) . All animal care and treatments were in accordance with The Scripps Research Institute guidelines for the care and use of animals under protocol #10-0019 .	Many aspects of our physiology and behavior , most notably our patterns of sleep and wakefulness , are synchronized with the day–night cycle . These circadian rhythms are generated and maintained by the circadian clock , which consists of positive and negative feedback loops formed by a large number of genes and proteins . The end result is that the rates at which thousands of proteins are produced varies rhythmically over the course of the day–night cycle . It has long been suspected that one of the functions of this circadian clock is to control the timing of cell division . Moreover , since UV radiation can give rise to genetic mutations when cells divide , it is thought that the circadian clock limits the amount of DNA damage that occurs during daytime . Papp , Huber et al . have now confirmed that the circadian clock does indeed participate in the DNA damage response and have revealed that two proteins known to be involved in the circadian clock—Cryptochrome 1 and 2—have a central role in protecting the integrity of the genetic information in the cell . These proteins evolved from light-activated enzymes that repair DNA in bacteria . While mammalian cryptochromes have lost their ability to repair DNA , they still prefer to bind to genetic material that has been damaged by UV radiation . Papp , Huber et al . show that DNA damage triggers cryptochrome 1 to bind to a protein called Hausp , which stabilizes the cryptochrome and prevents it from being broken down . By contrast , DNA damage triggers cryptochrome 2 to bind to a protein called Fbxl3 , which has a destabilizing effect on the cryptochrome and promotes its degradation . Since the cryptochromes regulate the activity of BMAL1 and CLOCK , the proteins associated with the two master clock genes , these changes can have a significant effect on the circadian clock of an organism . Further experiments are needed to work out how these proteins influence the activity of BMAL1 and CLOCK , and to investigate the seemingly conflicting roles of the two cryptochromes and the interactions between them .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "neuroscience" ]	2015	DNA damage shifts circadian clock time via Hausp-dependent Cry1 stabilization
Starch , as the major nutritional component of our staple crops and a feedstock for industry , is a vital plant product . It is composed of glucose polymers that form massive semi-crystalline granules . Its precise structure and composition determine its functionality and thus applications; however , there is no versatile model system allowing the relationships between the biosynthetic apparatus , glucan structure and properties to be explored . Here , we expressed the core Arabidopsis starch-biosynthesis pathway in Saccharomyces cerevisiae purged of its endogenous glycogen-metabolic enzymes . Systematic variation of the set of biosynthetic enzymes illustrated how each affects glucan structure and solubility . Expression of the complete set resulted in dense , insoluble granules with a starch-like semi-crystalline organization , demonstrating that this system indeed simulates starch biosynthesis . Thus , the yeast system has the potential to accelerate starch research and help create a holistic understanding of starch granule biosynthesis , providing a basis for the targeted biotechnological improvement of crops . Starch is an agricultural raw material coveted for its nutritional value , but also for its functional properties , which have extensive applications in the food industry , in paper manufacturing , and in the production of biodegradable materials , amongst others ( Ellis et al . , 1998; Blennow et al . , 2003; Zhang et al . , 2014 ) . Furthermore , special starches with reduced digestibility ( resistant starch ) are considered to be health-promoting , having a lower glycemic index and serving as a form of dietary fiber ( Raigond et al . , 2014 ) . Both industrial use and nutritional benefits depend on distinct biophysical characteristics and therefore on the underlying structure of starch . Starch is comprised of two α-polyglucan components ( here named glucans ) : large moderately-branched amylopectin molecules and interspersing small , near-linear amylose molecules ( Manners , 1989 ) . Both are built from chains of α-1 , 4 linked glucose units that are connected via α-1 , 6-linkages ( branch points ) . In amylopectin , the major and essential component , the branching pattern is non-random , giving rise to clusters of unbranched chain segments ( Pérez and Bertoft , 2010 ) . Neighboring chains within these clusters form double helices that pack with distinct inter-helical spacing . The resulting crystalline lamellae , alternating with amorphous lamellae containing the branch points , stack with a periodicity of ~9–10 . 5 nm ( Jenkins et al . , 1993 ) . This unique semi-crystalline nature of amylopectin renders starch insoluble . Amylopectin biosynthesis involves multiple enzyme activities , often composed of several isoforms with distinct specificities: four classes of starch synthases ( SSs ) elongate glucan chains using ADPglucose as glucosyl donor , creating α-1 , 4 glucose linkages; two classes of branching enzymes ( BEs ) introduce branches in the form of α-1 , 6 linked chains; and at least one isoamylase-type debranching enzyme hydrolyzes some branches again , probably tailoring the glucan for crystallization ( Pfister and Zeeman , 2016; Tetlow and Emes , 2011; Zeeman et al . , 2010; Jeon et al . , 2010 ) . In contrast , amylose is synthesized by a single class of granule-bound starch synthases ( Denyer et al . , 2001 ) . According to the classification of carbohydrate-active enzymes ( CAZy; Lombard et al . , 2014 ) , all SSs belong to the glycosyltransferase family 5 ( GT5 ) , while BEs and isoamylases belong to distinct subfamilies of the glycoside hydrolase family 13 ( GH13 ) . Starches from different plants vary in terms of amylose content , amylopectin structure and glucan modifications ( e . g . level of phosphorylation ) ( Santelia and Zeeman , 2011 ) . However , to date , this variation cannot fully meet industrial needs , necessitating costly physicochemical modifications post-extraction to yield the desired properties ( Tharanathan , 2005 ) . Although the starch-biosynthetic enzymes are highly conserved , we still do not have sufficient knowledge required for the rational modification of starch structure and properties in crops . This deficiency has several roots; first , the difficulty in generating plant mutants hinders a systematic analysis in most species . Second , the results obtained from comparable mutants in different plant systems are not always identical ( e . g . due to variation in genetic , environmental and/or developmental backgrounds ) . Third , enzyme kinetics obtained from soluble substrates in vitro may not be representative of how an enzyme acts on a crystallizing surface in vivo ( O'Neill and Field , 2015 ) . These problems are further compounded by the fact that the glucans that make up starch are polydisperse , necessitating multiple biochemical and biophysical techniques for their structural characterization . Collectively , these limitations impair our ability to define enzyme functions unambiguously and to derive an overall model of starch biosynthesis . As a result , progress in engineering new starches with enhanced functionalities in planta has been empirical and slow , and the full potential of this important renewable resource has not been realized . Here we created a model system in yeast that allows the expression of multiple starch-biosynthetic enzyme combinations in a targeted , controlled and fast manner . Using a suite of molecular and biochemical analyses , we show that the enzymes are functional and , in the right combinations , capable of producing semi-crystalline starch-like granules . This shows that our yeast system is a highly tractable experimental system in which starch biosynthesis can be simulated and the capacities of different interdependent enzymatic combinations can be evaluated , bringing us much closer to a holistic view of this critical process . Our yeast system employs an expression platform in haploid Saccharomyces cerevisiae CEN . PK113-11C , which had been developed for stable heterologous gene expression ( Mikkelsen et al . , 2012 ) . To eliminate interfering effects from the yeast’s endogenous glycogen metabolic pathway , we purged it of the five genes involved in glycogen biosynthesis ( two glycogenins , two glycogen synthases and glycogen branching enzyme ) and two involved in glycogen degradation ( glycogen debranching enzyme and α-glucan phosphorylase; Figure 1 ) . We simultaneously introduced all known amylopectin-biosynthetic genes from Arabidopsis thaliana via stable integration of coding sequences driven by galactose-inducible promoters . These are the starch synthases SS1 ( AGI locus code At5g24300 ) , SS2 ( At3g01180 ) , SS3 ( At1g11720 ) and SS4 ( At4g18240 ) , the branching enzymes BE2 ( At5g03650 ) and BE3 ( At2g36390 ) and the isoamylase encoded by ISA1 ( At2g39930 ) and ISA2 ( At1g03310 ) ( see Figure 1—figure supplement 1 for details on constructs ) . ISA1 and ISA2 together form the active heteromultimer that here we name ISA . Furthermore , we tested BE1 ( At3g20440 ) , a third Arabidopsis gene annotated as branching enzyme ( Han et al . , 2007 ) . This gene is distantly related to known branching enzymes , but studies to date have not revealed any starch-related function ( Dumez et al . , 2006; Wang et al . , 2010 ) . Arabidopsis was chosen as the donor for amylopectin-biosynthetic genes because it has been intensively used for starch research , resulting in well-characterized mutants that enable us to make direct comparisons between yeast- and plant-derived glucans . In addition , we introduced a non-regulated form of ADPglucose pyrophosphorylase from Escherichia coli ( GlgC-TM; Sakulsingharoj et al . , 2004 ) , as yeast glycogen biosynthesis is UDPglucose dependent , while the plant starch biosynthesis is ADPglucose dependent . 10 . 7554/eLife . 15552 . 003Figure 1 . Workflow of the S . cerevisiae system . The yeast’s endogenous glycogen-metabolic pathway ( grey box ) was removed and varying components of the starch-biosynthetic pathway from Arabidopsis ( green box ) was introduced . The deregulated mutant of ADPglucose pyrophosphorylase ( AGPase ) from E . coli was used for the supply of ADPglucose . For details of the constructs see Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 00310 . 7554/eLife . 15552 . 004Figure 1—figure supplement 1 . Constructs for heterologous gene expression and/or deletion of endogenous genes . ( A ) Expression constructs contain the coding sequence ( less chloroplast transit peptide ) of Arabidopsis branching enzymes BE2 or BE3 , the putative branching enzyme BE1 , starch synthases SS1 to SS4 or bacterial glgC-TM ( encoding an allosterically insensitive ADPglucose pyrophosphorylase ) fused to galactose-inducible PGAL1 promoters . SS2 is tagged with a C-terminal HA tag and SS4 and BE1 with a C-terminal FLAG tag , as their activities are not observed in zymograms from plant extracts . Similarly , glgC-TM carries a C-terminal HA tag . ( B ) ISA1 and ISA2 were combined within a single construct using the bidirectional PGAL10-PGAL1 promoter . ( C ) Example of an empty yeast integration vector . The insertion site for the expression construct is indicated with a green arrowhead . The vectors additionally contain terminator sequences ( CYC1 [C] and ADH1 [A] terminators , grey arrows ) , a URA3 gene for selection ( flanked by direct repeats [DR] for later marker recycling ) and regions homologous to genome sequences of yeast ( here GSY1 up or GSY1 down , yellow arrows ) . The homologous regions target the insert either to glycogen-metabolic genes ( replacing the majority of the gene; here the GSY1 gene ) or between essential genes ( Mikkelsen et al . , 2012 ) . Restriction sites used for the creation of the vectors are depicted in blue . Plasmid-based primers for genotyping of transgenic yeast strains are shown as red flags ( described in Supplementary file 1D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 004 We first established that the heterologously expressed enzymes are functional . Zymograms of soluble native protein extracts from yeast strains expressing ISA and single isoforms of SS and BE showed that each enzyme was expressed and active ( SS1: lines 0 and 1; SS2: lines 2 and 3; SS3: line 5; SS4: line 7; BE2: line T; BE3: line S; ISA: all lines with uneven numbers; Figure 2A ) . The enzymes migrated similarly as in extracts of the wild-type ( WT ) Arabidopsis plant controls , even though we noted additional activity bands in some cases , possibly due to small alterations in post-translational modifications . Occasionally , we observed that protein activity varied in strength between the lines ( e . g . SS2 activity between lines 2 and 3 ) . We reasoned that the enzyme’s binding to insoluble glucans potentially accumulating in these lines may affect its extractability as a soluble protein . Indeed , protein abundance detected in Western blots using total protein extracts – where proteins in the insoluble fractions had been solubilized by boiling in SDS-containing buffer – was more uniform ( Figure 2B ) , suggesting that expression levels between individual lines are in fact comparable . 10 . 7554/eLife . 15552 . 005Figure 2 . Native PAGE and immunoblots of heterologously expressed proteins . Total proteins or native soluble proteins were extracted from yeast lines grown in liquid cultures with complex medium . Native soluble proteins were subjected to native PAGE for detection of enzyme activity ( A ) and SDS-PAGE followed by Western blotting ( B , upper panels ) . Total proteins were also subjected to Western blots ( B , lower panels ) . Whether the yeast main cultures contained galactose ( + ) or glucose ( - ) is indicated . Arabidopsis genes present in the yeast lines are given above the strain number ( see Supplementary file 1A for complete genotypes ) , and all yeast lines except for the wild type ( WT ) also contain the glgC-TM gene . ( A ) Native PAGE of soluble proteins in glycogen-containing gels that were incubated either with glucose-1-phosphate and phosphorylase ( for visualization of branching enzyme and isoamylase activity , top panel ) or ADPglucose ( for visualization of starch synthase activity , bottom panel ) . The left- and right-hand panels show separate gels that were processed at the same time . Glucan-modifying activities were revealed by iodine staining . Strains expressing only one branching enzyme ( lines T and S ) and WT Arabidopsis plant extracts ( WS or Col-0 ecotype ) are shown for comparison . Enzyme activities in plant extracts , deduced from earlier mutant analyses , are indicated on the right hand side in green ( for a representative summary refer to Supplemental Fig . S1 in Pfister et al . [2014] ) . Enzyme activities in yeast extracts , deduced from strain comparisons , are indicated on the left side in black . BE2 migrates as three activities; two very close bands ( indicated with the double arrow ) and a third slower band that overlaps with BE3 activity ( compare strains T and S ) . SS4 gives a weak but distinct band , as indicated . The branching activity in WT yeast is the endogenous branching enzyme , Glc3p ( red asterisk ) . PHS1 , plastidial phosphorylase; PHS2 , cytosolic phosphorylase . ( B ) Immunoblots of yeast soluble protein extracts ( upper two panels ) and yeast total protein extracts ( lower two panels ) , respectively , after separation by SDS-PAGE . The plant sample is the same soluble protein extract in both cases . SS2 , GlgC-TM ( both carrying a C-terminal HA tag ) and SS4 ( carrying a C-terminal FLAG tag ) were visualized using α-HA or α-FLAG antibodies , as indicated . The expected molecular weights are 83 kDa for SS2-HA , 50 kDa for GlgC-HA and 116 kDa for SS4-FLAG . Soluble and total proteins are shown because each protein’s abundance in the soluble extract is influenced by its binding to glucans and the resultant partitioning between the soluble and insoluble fraction . The signal intensities from the soluble proteins should not be compared with those from total proteins as they arise from separate blots . The following figure supplement is available for Figure 2:DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 00510 . 7554/eLife . 15552 . 006Figure 2—source data 1 . Glucan contents and λmax values of individual replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 00610 . 7554/eLife . 15552 . 007Figure 2—figure supplement 1 . Enzyme function of ISA1/ISA2 , BE1 , BE2 and BE3 in yeast . The relevant genotypes are indicated; the Arabidopsis genes are highlighted in yellow and the endogenous yeast genes in grey . The WT and the strains A , B , U to Y also contain the endogenous glycogenin ( GLG1 , GLG2 ) and phosphorylase ( GPH1 ) genes . Lines V and W are individual isolates from the same transformation event . Genotypes are listed in Supplementary file 1A . ( A ) Quantification of insoluble ( black bars ) and methanol-precipitable soluble ( grey bars ) glucans and their wavelengths of maximum absorption after complexion with iodine ( λmax; values above the bars , in nm ) . Yeast lines were grown for 6 hr in liquid main cultures with galactose . Values are means ± S . D . from 3 replicate cultures for the quantifications ( except for strain 30 where n = 4 ) and from 2 replicate cultures for the λmax measurements ( except for strain 29 where n = 4 ) . Malto-oligosaccharide content in strains U , V , W , X and Y was <0 . 7 mg g−1 wet weight ( WW ) including S . D . Note that the glucans made with BE1 ( lines V and W ) or in the absence of any BE ( line U ) have a λmax typical of long linear chains . The glucans from strain 29 ( expressing all core amylopectin-biosynthetic genes ) were also very similar to those from strain 30 ( expressing also BE1 ) in terms of chain-length distribution , granule morphology and X-ray scattering patterns ( not shown; see results section 'Yeast glucans have semi-crystalline properties of starch' for a description of these techniques ) . WW , wet weight; BE , branching enzyme; DBE , debranching enzyme; AGPase , ADPglucose pyrophosphorylase . ( B ) Light micrographs ( LM ) of the indicated yeast lines grown as in A . Cells were stained with iodine . ( C ) Anti-FLAG immunoblots of yeast total protein extracts from cultures grown as in A . The expected molecular weights are 99 kDa for BE1-FLAG and 116 kDa for SS4-FLAG . The weak band slightly above BE1-FLAG reflects unspecific binding of the antibody ( marked by a green asterisk ) . ( D ) Yeast carrying ISA1/ISA2 or BE3 ( replacing the endogenous branching enzyme GLC3 ) under galactose-inducible promoters were first grown in complex medium and then transferred to medium lacking nitrogen to trigger glycogen synthesis . Media contained either galactose ( strongly inducing heterologous gene expression; black bars ) or glucose ( strongly repressing expression , grey bars ) as the sugar source . Induction of ISA1/ISA2 suppressed glucan accumulation ( line A ) . Induction of BE3 ( line B ) in an otherwise BE-deficient strain restored glycogen synthesis . Shown are mean values ± S . D . of 3 replicate cultures . WW , wet weight; BE , branching enzyme; DBEs , debranching enzymes . DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 007 We tested the in-vivo functionality of the enzymes by assessing glucan production of yeast strains expressing only single enzyme isoforms . Either plant BE2 or BE3 enabled substantial glucan production in yeast cells lacking their endogenous BE ( strains X and Y; Figure 2—figure supplement 1A ) . This was , however , not the case for strains expressing BE1 . Here , only low amounts of amylose-like glucans could be detected ( strains V and W ) , comparable to the strain lacking any branching enzyme ( strain U; Figure 2—figure supplement 1A , B ) . This suggests that BE1 in fact is not a starch branching enzyme . Since the presence of BE1 also did not alter the glucans made by the other amylopectin biosynthetic enzymes , it was excluded from subsequent strain sets ( see Figure 2—figure supplement 1A , B and associated text ) . In contrast , every plant SS produced glucans when expressed as the sole synthase activity ( lines 0 , 2 , 4 and 6; Figure 3B ) . Expression of ISA in wild-type yeast suppressed the accumulation of glycogen ( Figure 2—figure supplement 1D ) . This result was consistent with that observed when expressing ISA in E . coli , where it also reduced glycogen accumulation ( Sundberg et al . , 2013 ) . Branch-point hydrolysis by ISA may have a degrading effect when it is provided with glycogen as a substrate rather than an amylopectin ‘precursor’ . The latter presumably crystallizes upon debranching , rendering it inaccessible to further modification ( Sundberg et al . , 2013 ) . 10 . 7554/eLife . 15552 . 008Figure 3 . Yeast strains synthesize high amounts of glucans . ( A ) Iodine staining of 30 yeast strains ( numbered 0–29 ) which vary the complement of SS isoforms and ISA1/ISA2 in the glycogen-metabolism free background . Cells were grown for 24 hr on plates with galactose and then stained with iodine vapor . Wild-type ( WT ) yeast , potato amylose ( AM ) dissolved in DMSO and native , non-gelatinized plant amylopectins from Arabidopsis ( aAP ) and potato ( pAP ) are shown for comparison . Indicative genotypes are given in B and Figure 3—figure supplement 1 . Full genotypes are listed in Supplementary file 1A . ( B ) Quantification of insoluble ( black bars ) and methanol-precipitable soluble ( grey bars ) glucans of yeast lines grown for 6 hr in liquid main cultures with galactose . The genotypes are indicated; the varying SS and ISA1/ISA2 genes are highlighted in yellow and the endogenous yeast genes in grey . For the glucose-grown cultures in B-D , galactose was replaced by glucose to repress heterologous gene expression . Values are means ± S . D . from 4 replicate cultures ( except for the glucose-grown samples , and line 6 , where n = 3 ) . WW , wet weight; BE , branching enzyme; DBE , debranching enzyme; AGPase , ADPglucose pyrophosphorylase . ( C ) Light micrographs ( LM ) from cells of the indicated yeast lines grown as in B . Cells were stained with iodine . ( D ) Transmission electron micrographs ( TEM ) of the indicated yeast lines ( grown as in B ) after chemical fixation . While only particulate , putatively soluble glucans ( s ) were observed in line 28 , line 29 also contained uniform , putatively insoluble glucans ( i ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 00810 . 7554/eLife . 15552 . 009Figure 3—source data 1 . Glucan contents of individual replicates from strains 0 to 29 . DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 00910 . 7554/eLife . 15552 . 010Figure 3—figure supplement 1 . Quantification of glucans and light micrographs of yeast strains 8 to 29 . ( A , C ) Quantification of insoluble and soluble glucans produced in the indicated yeast strains . Growth of yeast and fractionation of glucans was performed as described in Figure 3B . Values are means ± S . D . from 4 replicate cultures . The data for lines 28 and 29 are the same as presented in Figure 3B ( all yeast strains shown here and in Figure 3B were grown and analyzed together ) and can be found in Figure 3—source data 1 . WW , wet weight; BE , branching enzyme; DBE , debranching enzyme; AGPase , ADPglucose pyrophosphorylase . ( B , D ) Light micrographs ( LM ) of cells from A and C , respectively , after staining with iodine . Cells with iodine-stained glucans were obtained only rarely in line 9 . The scale bar ( 10 µm ) applies to all pictures . Source data: Figure 3—source data 1 . Glucan contents of individual replicates from strains 0 to 29 . DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 01010 . 7554/eLife . 15552 . 011Figure 3—figure supplement 2 . Dependence of starch synthases on glycogenins . ( A ) Quantification of insoluble and methanol-precipitable soluble glucans produced in the indicated yeast strains ( grown as described in Figure 3B ) . The relevant genotypes are indicated; the varying Arabidopsis genes are highlighted in yellow and the endogenous yeast genes in grey . Genotypes are listed in Supplementary file 1A . Values are means ± S . D . from 4 replicate cultures . Numbers indicate the percentage of change in glucan content ( sum of insoluble and methanol-precipitable soluble glucans ) upon loss of glycogenins . Statistical comparisons were performed using two-sided t-tests as described in Materials and methods . **p value < 0 . 0001; p value < 0 . 001; p value < 0 . 01; p value < 0 . 05; n . s . , not significant , p value ≥ 0 . 05 . WW , wet weight; BE , branching enzyme; DBE , debranching enzyme; AGPase , ADPglucose pyrophosphorylase . ( B ) Light micrographs of cells from A after staining with iodine . Cells with iodine-stained glucans were obtained only rarely in line K . In case of SS1 , deletion of both endogenous glycogenins GLG1 and GLG2 resulted in a marked reduction in glucan content with only few ( strain E ) or no ( strain G ) cells remaining capable of glucan production . This effect was additive to that of ISA1/ISA2 expression ( which also repeatedly reduced glucan accumulation in yeasts where only SS1 and/or SS2 were present ) . Strains with SS2 produced fewer glucans when glycogenins were absent , but the difference was small when ISA was absent . SS3 showed no or little dependence on glycogenins and no dependence was observed in case of SS4 . DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 01110 . 7554/eLife . 15552 . 012Figure 3—figure supplement 3 . Accumulation of linear malto-oligosaccharides ( MOS ) in the presence of isoamylase . ( A ) Quantification of malto-oligosaccharides ( MOS ) of the yeast samples presented in Figure 3B and Figure 3—figure supplement 1 . All lines also contain BE2 , BE3 and glgC-TM . Values are means ± S . D . from 4 replicate cultures ( except for the glucose-grown samples and line 6 , where n = 3 ) . The MOS contents of individual replicates can be found in Figure 3—source data 1 . ( B ) Chain-length distribution of malto-oligosaccharides from line 29 that were not enzymatically debranched during sample preparation . Numbers indicate the degree of polymerization ( DP ) of the corresponding linear chains . Branched chains typically elute slightly earlier than linear chains of the same molecular weight and were absent in the chromatograms . A representative chromatogram from one replicate culture is shown . Similar results were obtained in a second replicate from line 29 and for malto-oligosaccharides from the three other lines tested ( lines 7 , 11 , and 17 ) . Source data: Figure 3—source data 1 . Glucan contents of individual replicates from strains 0 to 29 . DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 012 Having established the functionality of the heterologous starch-biosynthetic enzymes , we created a library of over 100 yeast lines . This collection included 30 lines in the mutant background deficient in the endogenous glycogen-metabolic genes ( lines 0–29 , Supplementary file 1A ) . In these lines , we systematically varied the complement of the four SS isoforms and ISA while keeping GlgC-TM , BE2 and BE3 constant . No loss of heterologous genes could be detected by PCR in line 29 ( carrying the highest number of modifications ) during strain maintenance and growth in liquid cultures , indicating genetic stability ( described in Materials and methods , not shown ) . We assessed glucan production in a qualitative manner by growing the 30 yeast lines on plates containing galactose and staining them with iodine vapor ( Figure 3A ) . The color and intensity after iodine staining depend on the type of complex formed between iodine and the glucan’s secondary structure: Predominantely linear , single-helical glucans such as amylose stain deep blue; double-helical amylopectin chains stain brown , and glycogen ( lacking secondary structures ) stains weakly red-brown ( Saenger , 1984; Manners , 1991; Streb et al . , 2008 ) . The engineered yeast lines stained in a variety of shades , suggesting that they produced glucans with different structures . Notably , yeast strains expressing ISA ( lines with odd numbers; + ISA ) tended to stain darker than their ISA-free counterparts ( lines with even numbers; – ISA; placed above ) . To quantify glucan accumulation , we cultured yeasts in shake flasks using galactose as the sugar source in the main cultures . The cells were harvested after 6 hr , as growth of yeast strains producing glucans was reduced afterwards ( not shown ) . After homogenization of the cells and centrifugation at 6000 g for 5 min , water-insoluble glucans were collected in the pellet . The supernatant containing soluble glucans was further fractionated by addition of methanol ( 80% v/v ) . This yielded a second fraction of water-soluble but methanol-precipitable glucans ( i . e . polysaccharides; here named soluble glucans ) and a third fraction of methanol-soluble glucans ( i . e . short malto-oligosaccharides and free sugars ) . Most lines without ISA synthesized high amounts of soluble glucans , which reached around 10% of the wet weight ( Figure 3B and Figure 3—figure supplement 1 ) . Light microscopy ( LM ) of these cells revealed fairly uniform iodine staining ( Figure 3C and Figure 3—figure supplement 1 ) . Transmission electron micrographs ( TEM ) from line 28 indicated numerous small particles reminiscent of glycogen ( Figure 3D ) . Interestingly , cells expressing just SS1 as a synthase ( line 0 ) accumulated far fewer glucans than lines expressing the other SS isoforms ( lines 2 , 4 and 6; Figure 3B ) . However , in yeast lines still containing the endogenous glycogenins , all SS isoforms synthesized large amounts of glucan ( Figure 3—figure supplement 2; see figure legend for details ) . This suggests that , rather than having low activity , SS1 is less efficient than the other SSs in utilizing available primers or in generating its own . Most lines expressing ISA produced water-insoluble glucans as well as soluble glucans ( Figure 3B and Figure 3—figure supplement 1 ) . These lines also accumulated linear malto-oligosaccharides , most likely representing chains liberated by ISA ( Figure 3—figure supplement 3 ) . The presence of insoluble glucans was typically associated with discrete , patchy staining of cells in LM ( Figure 3C and Figure 3—figure supplement 1 ) . This staining showed some variation within single cells and from cell to cell , probably reflecting different stages of yeast cell age and glucan maturation . TEMs from line 29 showed solid continuous glucan particles which often occupied a significant fraction of the yeast cell volume in addition to small , presumably soluble glucan particles ( Figure 3D ) . Interestingly , the lines which expressed ISA but neither SS3 nor SS4 ( lines 1 , 3 and 9 ) had very low glucan levels when grown in liquid culture ( lines 1 and 3 , Figure 3B; line 9 , Figure 3—figure supplement 1; see also lines F , G , J and K , Figure 3—figure supplement 2 ) . This again indicates that hydrolysis by ISA may limit glucan accumulation under some conditions ( as described above ) . Having a library of strains allows systematic comparisons of glucan accumulation , structure and partitioning between the soluble and insoluble fractions . This enables functional analyses of individual enzymes and an exploration of their interdependencies . Such comparisons furthermore allow us to test whether the specificities evident from plant and in-vitro studies are retained in the yeast . To monitor the effect of ISA and SSs on the presence of insoluble glucans , we re-calculated these as percentages of total glucans ( Figure 4A ) . As mentioned , ISA enabled the synthesis of insoluble glucans ( also see Figure 3B and Figure 3—figure supplement 1 ) . Moreover , we observed a clear effect of SS isoforms on the synthesis of insoluble glucans when ISA was present . Comparing the percentages of insoluble glucans of strains with and without SS4 ( e . g . , strain 25 vs . strain 11 ) revealed that SS4 consistently promoted the synthesis of insoluble glucans , both in absolute and relative terms ( Figure 4B ) . In contrast , SS1 activity rendered the glucans less insoluble and consistently promoted the synthesis of soluble glucans . The presence of SS2 had variable effects , but significantly increased the percentage of insoluble glucans in the absence of SS4 ( line 21 vs . 11: + 46% , line 15 vs . line 5: + 30% ) . Reciprocally , the increase in the proportion of insoluble glucans caused by SS4 was strongest in the lines without SS2 ( line 25 vs . 11: + 108%; line 19 vs . 5: + 95% ) . 10 . 7554/eLife . 15552 . 013Figure 4 . Isoamylase , starch synthases and branching enzymes have distinct effects on the formation of insoluble glucans . ( A ) Percentage of total glucans that were insoluble glucans of lines without isoamylase ( grey bars ) and their corresponding partners with isoamylase ( ISA , black bars ) . Only strain pairs in which both lines have glucan levels > 5 mg g−1 wet weight ( WW ) were included . All lines furthermore contain BE2 , BE3 and glgC-TM . Values are means ± S . D . from 4 replicate cultures , except for line 6 ( n = 3 ) . Statistical comparisons were performed using two-sided t-tests as described in Materials and methods . The underlying data is available in Figure 3—source data 1 . **p value < 0 . 0001; p value < 0 . 001 . ( B ) Fold changes ( FC ) of percentage of insoluble glucans depending on individual SS , using data from lines with ISA presented in A . The compared lines are indicated . The comparisons of SS2 in the absence of SS4 and vice versa are shown in bold . Given values for each SS are mean percentage changes from all comparisons ± S . E . M . ( n = 6 for SS1 and SS2 , n = 4 for SS3 and SS4; see Figure 4—source data 1 for the calculations ) . Statistical comparisons were performed using ANOVA as described in Materials and methods . *p value < 0 . 0001; p value < 0 . 001; p value < 0 . 01; p value < 0 . 05; n . s . , not significant , p value ≥ 0 . 05 . ( C ) Quantification of insoluble ( black bars ) and methanol-precipitable soluble ( grey bars ) glucans and their wavelengths of maximum absorption after complexion with iodine ( λmax; values above/within the bars , in nm ) . Yeast lines were grown as described in Figure 3B . Values are means ± S . D . from 4 replicate cultures for quantifications and from 2 replicate cultures for the λmax measurements ( except for insoluble glucans from strains I , 5 and 7 where n = 4 ) . The quantification data for lines E , G , I , K , M , O , Q and S are the same as presented in Figure 3—figure supplement 2 ( all yeast strains shown here and in Figure 3—figure supplement 2 were grown and analyzed together ) . WW , wet weight; BEs , branching enzymes; DBE , debranching enzyme; AGPase , ADPglucose pyrophosphorylase . ( D ) Light micrographs of cells from A after staining with iodine . Bar = 10 µm . Source data:DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 01310 . 7554/eLife . 15552 . 014Figure 4—source data 2 . Glucan contents and λmax values of individual replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 01410 . 7554/eLife . 15552 . 015Figure 4—source data 1 . Glucan contents of individual replicates from strains 0 to 29 . DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 015 During the creation of the library of strains described above , we observed lines that produced insoluble glucans despite lacking ISA: Strain I and strain M , containing BE3 as sole BE and either SS2 or SS3 , respectively , contained up to 19% of their glucans in an insoluble form ( Figure 4C ) . The equivalent strains with both BEs contained only soluble glucans ( strains 2 and 4 ) . An effect of BE on the partitioning between soluble and insoluble glucans could also be observed when ISA was present: When together with SS3 and ISA , having BE3 alone promoted insoluble glucan formation ( compare strain O vs . 5 ) . In contrast , when together with SS4 and ISA , having BE3 alone reduced both the total glucan content and the percentage that was insoluble . In all instances , however , having BE3 alone rather than both BEs , was accompanied by a shift towards longer wavelengths of maximum absorption [λmax] after iodine staining . Together , these data show that each enzyme activity influences insoluble glucan formation in a distinct manner . They also highlight the complex functional interplay between both enzyme classes and isoenzymes of each class that are collectively responsible for starch synthesis . Studies of plant mutants and of individual enzymes in in-vitro have suggested that SS classes have distinct preferences in terms of glucan chain elon gation ( Zeeman et al . , 2010; Tetlow and Emes , 2011; Pfister and Zeeman , 2016 ) . To investigate whether these are preserved in the yeast system , we obtained chain-length distributions ( CLDs ) from the glucans from strains 0–29 . We debranched the glucans , then separated and quantified the resulting linear chains by HPAEC-PAD ( high-performance anion-exchange chromatography with pulsed amperometric detection; see Figure 5A for examples ) . We then compared the CLDs from yeast strains either having or not having individual SS isoforms via difference plots . 10 . 7554/eLife . 15552 . 016Figure 5 . Starch synthases retain their chain-elongation specificities . ( A ) Chain length distributions ( CLDs ) of debranched insoluble glucans from line 29 and of wild-type Arabidopsis starch ( WS ecotype ) . Values are means ± S . D . from 4 replicate yeast cultures or 3 plants , respectively . The CLD of gbss mutant Arabidopsis starch is reported to be identical to that from wild-type starch ( Seung et al . , 2015 ) . ( B-D ) Representative graphs illustrating the changes in glucan fine structure upon the addition of SS1 ( B ) , SS2 ( C ) or SS3 ( D ) in yeast and in planta . Comparisons were done by subtracting the CLDs as indicated ( means ± S . E . M . , n = 4 , except for ss2isa , ss1isa , ss2 , yeast line 13 ( soluble ) and ss2 [insoluble shown in B and D] , where n = 3 ) . Data from plants in B and C are recalculations from Pfister et al . , ( 2014 ) . Horizontal comparisons in yeast and in planta involve the same set of amylopectin-synthesizing genes in each case ( i . e . line 24 has the whole known complement except for SS2 and ISA1/ISA2 , corresponding to an ss2isa mutant; line 18 the same enzymes except for SS1 , corresponding to an ss1ss2isa mutant ) . Subtracting the CLD of line 18 from that of line 24 thus reveals the effect of the presence of SS1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 01610 . 7554/eLife . 15552 . 017Figure 5—source data 1 . Numerical data of chain-length distributions from the plant and yeast glucans . DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 017 In all instances , the presence of SS1 increased the relative abundance of shorter chains ( degree of polymerization or DP = 6–10 ) at the expense of intermediate-length ones ( DP = 11–25 ) ( Figure 5B ) . This result is in accordance with that from Arabidopsis ( Figure 5B; Delvallé et al . , 2005; Szydlowski et al . , 2011 ) and cereal mutants ( Fujita et al . , 2006; McMaugh et al . , 2014 ) . Similarly , the presence of SS2 had a consistent effect on the CLDs of yeast-derived glucans , increasing the relative abundance of intermediate-length chains ( Figure 5C ) . Again , these alterations were consistent with previous data of ss2 mutants from Arabidopsis and other species ( Zhang et al . , 2008; Edwards et al . , 1999; Craig et al . , 1998; Morell et al . , 2003; Zhang et al . , 2004 ) . The magnitudes of these isoform-specific CLD alterations were smaller among the yeast glucans than among the plant glucans . This is probably due to the overall relative increase in longer chains in the yeast glucans compared to Arabidopsis starch ( Figure 5A ) . SS3 and SS4 did not influence the lengths of short and intermediate chains in a distinct manner ( not shown ) , which is in-line with reports from Arabidopsis mutants ( Zhang et al . , 2005; Roldán et al . , 2007 ) . However , the tendency of SS3 to increase the abundance of intermediate-length chains in the absence of SS2 ( Zhang et al . , 2008 ) was also apparent in yeast ( Figure 5D ) . The presence of water-insoluble glucans prompted us to test whether these possessed the characteristics of genuine starch granules . We therefore performed a series of microscopic and biophysical analyses , focusing on the insoluble glucans from line 29 ( the line expressing all amylopectin-biosynthetic enzymes ) . The insoluble glucans were laid down as distinct spherical particles that were several micrometers in diameter , as determined by scanning electron microscopy ( Figure 6A ) . The size of these particles was within the range observed for plant starch granules ( Pérez and Bertoft , 2010 ) , including Arabidopsis leaf starch ( Figure 6A ) . Similar , albeit not identical , particles could be purified from other yeast strains producing insoluble glucans , but not from those lacking measurable insoluble glucans ( Figure 6—figure supplement 1 and Figure 6A ) . The presence of massive insoluble particles is consistent with the observation of dense glucans in TEMs from line 29 ( Figure 3D ) . 10 . 7554/eLife . 15552 . 018Figure 6 . Structure of starch granules made in S . cerevisiae compared with amylose-free Arabidopsis starch . ( A ) Scanning electron micrographs ( SEMs ) of purified insoluble particles from line 29 and of amylose-free Arabidopsis starch . No insoluble particles could be purified from WT yeast or line 28 . The granules of WT Arabidopsis starch are similar to those of amylose-free starch ( Figure 6—figure supplement 1 ) . Bars = 2 µm . ( B ) 2D slices through 3D electron density maps obtained by cryo X-ray ptychographic tomography of intact yeast cells ( grown as in Figure 3B ) and purified amylose-free Arabidopsis starch granules . i , putative insoluble glucans; s , putative soluble glucans; bars = 3 µm . ( C ) Absorption spectra of glucans from the indicated yeast lines ( WT , line 28 and line 29 ) and various plant glucans after complexion with iodine . Wavelengths of maximum absorption are indicated . a . u . , arbitrary units . ( D ) Wide-angle and ( E ) small-angle X-ray scattering intensity profiles of insoluble glucans from line 29 and amylose-free Arabidopsis starch ( i . e . amylopectin ) . The profiles of WT Arabidopsis starch are similar to those of amylose-free starch ( Figure 6—figure supplement 3 ) . q , scattering vector; d , lamellar periodicity; a . u . , arbitrary units . The following figure supplements are available for Figure 6:DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 01810 . 7554/eLife . 15552 . 019Figure 6—figure supplement 1 . Scanning electron micrographs ( SEMs ) of purified insoluble particles from the indicated yeast strains and from wild-type Arabidopsis ( WS ecotype ) leaf starch . The micrographs from lines 28 , 29 and WT yeast are the same as shown in Figure 5A . DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 01910 . 7554/eLife . 15552 . 020Figure 6—figure supplement 2 . Absorption spectra of insoluble glucans from the indicated yeast lines . Shown is one representative spectrum from one replicate . Wavelengths of maximum absorption ( means ± S . D . from 4 biological replicates ) are indicated in brackets . a . u . , arbitrary units . DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 02010 . 7554/eLife . 15552 . 021Figure 6—figure supplement 3 . X-ray scattering wild-type Arabidopsis starch . Wide-angle ( left ) and small-angle ( right ) X-ray scattering intensity profiles of purified WT Arabidopsis starch ( Col-0 ecotype ) . q , scattering vector; d , lamellar periodicity; a . u . , arbitrary units . DOI: http://dx . doi . org/10 . 7554/eLife . 15552 . 021 To investigate the internal glucan structure , we obtained electron density maps of intact yeast cells using cryo X-ray ptychographic tomography ( Dierolf et al . , 2010; Diaz et al . , 2015 ) . Cells of line 29 contained regions of a high density - putatively insoluble glucans - as well as regions of intermediate density , probably reflecting soluble glucans . Remarkably , the high-density regions had a density identical to that of Arabidopsis amylopectin , pointing towards a comparable internal organization of the glucans ( mean electron density [± S . D . , n = 6] from line 29 = 0 . 44 ± 0 . 01 e−Å−3 , corresponding to a mass density of 1 . 36 ± 0 . 04 g mL−1; Arabidopsis amylopectin density = 0 . 441 ± 0 . 004 e- Å−3 , corresponding to 1 . 36 ± 0 . 01 g mL−1; see Materials and methods for details ) ( Figure 6B ) . Further , the absorption characteristics of the insoluble glucans after iodine complexion from line 29 were almost identical to those of amylopectins from wild-type and gbss mutant plants ( λmax after iodine-complexion of ~548 nm; Figure 6C; [Zeeman et al . , 2002b] ) . In addition , they were distinct from that of highly-branched , soluble glycogen from wild-type yeast ( λmax445 nm ) , from that of soluble glucans from the corresponding line without ISA ( line 28 , λmax505 nm ) and from that of linear glucans such as potato amylose ( λmax634 nm ) . The similarity in absorption spectra indicates that the insoluble glucans from line 29 form comparable secondary structures to those in plant amylopectin , i . e . double helices between adjacent chains . Notably , also the iodine absorption spectra of insoluble glucans from yeast strains expressing only subsets of SSs ( e . g . SS3 , line 5 , or SS4 , line 7 , or SS1 , SS2 and SS3 , line 21 ) were dissimilar to that from line 29 , re-affirming the influence of SSs on glucan structure ( Figure 6—figure supplement 2 ) . To directly investigate the formation of secondary and tertiary structures in the insoluble glucans from line 29 , we acquired X-ray scattering intensity profiles . Wide-angle X-ray scattering ( WAXS ) of purified and dried glucans resulted in a diffraction pattern typical of many starches , where glucan chains form double helices that align with an inter-helical spacing of ~1 . 6 nm , indicative of a B-type allomorph ( Kong et al . , 2014; Figure 6D ) . Small-angle X-ray scattering ( SAXS ) indicated that the helices were stacked in a weakly ordered lamellar structure with a periodicity of 13 . 6 nm ( Figure 6E ) , slightly longer than typically reported for plant starches ( Pérez and Bertoft , 2010 ) . Here , Arabidopsis starch had a ~10 . 5 nm periodicity ( Figure 6E and Figure 6—figure supplement 3 ) . Chain-length distribution ( CLD ) analyses revealed that the insoluble glucans had relatively more long chains ( DP > 18 ) than wild-type Arabidopsis amylopectin . Nevertheless , the key characteristics of amylopectin were preserved as they grouped in sub-populations of chains with DPs of 6–8 , 9–18 and >18 ( Figure 5A ) . Collectively , these analyses show that the insoluble glucans from line 29 form granular particles which display the typical internal organization of semi-crystalline plant starch . Our work confirms that debranching by ISA is the major driving force for the formation of insoluble glucans: In most ( but not all ) cases , insoluble glucans were only obtained when ISA was present ( Figure 4A ) . The promotion of insoluble glucans by ISA is in-line with plant mutants lacking this enzyme , which typically accumulate water-soluble polysaccharides , so-called phytoglycogen ( Nakamura et al . , 1996; Mouille et al . , 1996; Zeeman et al . , 1998; Dinges et al . , 2001; Burton et al . , 2002; Bustos et al . , 2004 ) . Presumably , ISA selectively hydrolyzes excessive or misplaced branches that would otherwise interfere with the formation of double helices ( Myers et al . , 2000 ) . Consistent with this , we could detect a pool of linear malto-oligosaccharides in strains where ISA promoted insoluble glucan formation ( Figure 3—figure supplement 3 ) . In wild-type plants , such a malto-oligosaccharide pool is probably rapidly metabolized and therefore not detectable ( Streb et al . , 2008 ) . In the present study , we also dissected the contribution of SSs and BEs to insoluble glucan formation . When ISA was present , SS1 always decreased the production of insoluble glucans ( Figure 4B ) . The decreased crystallinity of glucans in the presence of SS1 probably stems from the increased abundance of glucan chains of DP 6–10 ( Figure 5B ) that are too short to efficiently contribute to the formation of higher-order structures underlying starch crystallinity ( Pfannemüller , 1987; Gidley and Bulpin , 1987 ) . We have recently described a comparable effect of SS1 in Arabidopsis , where the lack of SS1 in the isa mutant background restored granule formation in the mesophyll cells ( Pfister et al . , 2014 ) . In contrast , SS4 enhanced the production of insoluble glucans . This effect cannot be traced back to distinct structural modifications – no consistent alterations in glucan structure by SS4 were observed in our comparisons or in studies of Arabidopsis ss4 mutants ( data not shown; Roldán et al . , 2007; Szydlowski et al . , 2009 ) . One has to bear in mind , though , that the methods currently available for the analysis of starch structure give an incomplete picture , and structural features introduced by SS4 may not have been revealed . We did not observe uniform effects upon the expression of SS2 and SS3 ( Figure 4B ) . For SS2 , this was surprising; it has been shown that SS2 specifically elongates chains from around DP 8 to around DP 13 and suggested that this promotes the formation of higher-order structures ( Nakamura et al . , 2005; Fujita et al . , 2012; Pfister et al . , 2014 ) . Indeed , mutation of SS2 in Arabidopsis results in small amounts of phytoglycogen and enhances the accumulation of phytoglycogen in isa mutants of both rice and Arabidopsis ( Fujita et al . , 2012; Pfister et al . , 2014 ) . Although the alterations in glucan structure introduced by SS2 are preserved in the yeast system ( Figure 5C ) , SS2 promoted the synthesis of insoluble glucans only when SS4 was absent ( Figure 4B , comparisons in bold ) . It is possible that the strong effect of SS4 in promoting insoluble glucan formation masks this effect of SS2 in yeast , potentially due to differences in relative expression levels compared to Arabidopsis . Our data also add new insights into how branching level influences the synthesis of insoluble glucans: In some lines expressing only BE3 as a branching enzyme , insoluble glucans were obtained despite the absence of any debranching activity ( Figure 4C ) , confirming that the latter is not an absolute requirement for the insoluble-glucan production ( Streb et al . , 2008 ) . BE2 and BE3 are closely related class II BEs and largely redundant in Arabidopsis ( Dumez et al . , 2006 ) . Therefore , the impact of having only BE3 is probably due to overall branching activity rather than to a difference in specificity . It is striking that in combination with some , but not all SS this low BE activity resulted in formation of insoluble glucans . This was despite the fact that in all cases the λmax of the iodine-stained glucans increased , suggesting that the structure was altered in a similar way , i . e . more chains long enough to form single or double helices were present . Clearly , such secondary structures are not sufficient for insoluble glucan formation , presumably because they need also to align in parallel to form a semi-crystalline matrix . Further investigations of the insoluble glucans made by different strains in our yeast collection will deepen our understanding of the structural features that allow crystallization . Glycogen biosynthesis in yeast and animals begins with the self-glycosylation of glycogenins on tyrosine residues , yielding short glucan chains that are then elongated by glycogen synthases ( Rodriguez and Whelan , 1985; Lomako et al . , 1988; Pitcher et al . , 1988; Cheng et al . , 1995 ) . In contrast , the mechanism of starch initiation is not known . In Arabidopsis , genetic evidence suggests a specialized role for SS4 in this process since ss4 mutants display a strong reduction in starch granule number ( Roldán et al . , 2007 ) . Although ss3 mutants have normal number of starch granules ( Seung et al . , 2016 ) , further loss of SS3 from the ss4 mutant background resulted in a nearly complete failure of glucan synthesis ( Szydlowski et al . , 2009 ) , indicating that SS3 becomes indispensable for glucan priming when SS4 is absent . In our yeast strains , it is unclear what primes glucan synthesis . Every SS was capable of glucan synthesis on its own ( Figure 3B ) , but to varying extents . Particularly , when SS1 was the sole synthase , the overall glucan content of the whole population was low and iodine staining revealed that only a fraction of the cells accumulated glucans ( strains 0 and E , Figure 3C , D and Figure 3—figure supplement 2 ) . However , glucan synthesis by SS1 was strongly improved in a corresponding strain with the endogenous glycogenins ( GLG1 and GLG2; strain D , Figure 3—figure supplement 2 ) . This phenotype is reminiscent of the yeast glycogenin mutant ( glg1glg2 ) which synthesizes only little glycogen overall but accumulates it in single colonies , presumably due to stochastic initiation events ( Torija et al . , 2005 ) . This suggests that , similar to yeast glycogen synthases , SS1 has limited capacity to initiate glucan synthesis , but can use primers provided by glycogenins . It is worth noting that SS2 also synthesized slightly fewer glucans when glycogenins were absent ( strain H vs . I , Figure 3—figure supplement 2 ) . Our study also suggests that degradation influences glucan initiation . When only SS1 and/or SS2 were present , glucan accumulation was strongly diminished by ISA ( e . g . , in strains 1 , 3 and 9; Figure 3and Figure 3—figure supplement 1; also see Figure 3—figure supplement 2 ) . We first reasoned that SS1 and SS2 alone may generate a polymer that is excessively debranched by ISA , possibly because the glucan fails to become crystallization-competent upon debranching and hence remains accessible to further ISA action . However , the absence of malto-oligosaccharides - the expected products of ISA action - in these lines ( Figure 3—figure supplement 3 ) argues against such a scenario . Rather , the degradation may happen at a very early time point and could impair the glucan synthesis altogether by limiting the availability of primers . This observation is particularly interesting since Arabidopsis ss3ss4 mutants are virtually glucan-free due to an apparent failure in the priming of glucan synthesis ( Szydlowski et al . , 2009 ) . Indeed , it was recently shown that the additional knock-out of the α-amylase AMY3 in ss4 Arabidopsis mutants increases granule number and overall starch content , and loss of AMY3 in ss3ss4 Arabidopsis mutants ( amy3ss3ss4 mutants ) even restored some granule formation ( Seung et al . , 2016 ) . These phenotypes were interpreted to mean that at least part of the priming function of SS4 lies in the protection of primers from premature degradation . Our observation that ISA may also diminish primers could explain why glucan accumulation in amy3ss3ss4 mutants is only partially restored . Notably , ISA has been implicated as a negative regulator of granule initiation before , since the remaining starch of isa mutants is typically deposited as numerous small granules ( Burton et al . , 2002; Bustos et al . , 2004 ) . The generation and analysis of plant mutants deficient in SS3 , SS4 and different starch degrading enzymes is ongoing in order to test this hypothesis . It is interesting - but perhaps not surprising - that the granules made in yeast by the Arabidopsis enzymes are not identical to those made by Arabidopsis itself . For instance , long chains ( DP > 18 ) were overrepresented in the yeast glucans , in particular when ISA was present ( Figure 5A and data not shown ) . In plant amylopectin , the group of chains with DP < 25 is believed to make up a single 9- to 10 . 5-nm-repeat , being part of the double helices that underlie starch crystallinity ( so-called A and B1 chains; Hizukuri [1986] ) . Since the length of these chains probably defines the width of the lamellar repeat , the relative abundance of long chains in yeast glucans may explain the increased width of the repeat we observed ( 13 . 6 nm in the insoluble glucans from line 29; Figure 6E ) . The molecular factors determining the width of the lamellar repeat are not known , but were suggested to involve the specific placement of branches by BEs ( Nakamura , 2002 ) . Nonetheless , we did not obtain a wild-type repeat distance , despite having expressed both known starch BEs from Arabidopsis . Most plants contain an additional BE of class I , which displays distinct specificities when assessed in vitro ( Rydberg et al . , 2001; Nakamura et al . , 2010; Sawada et al . , 2014 ) , but the Arabidopsis’ genome does not encode such a BE ( Dumez et al . , 2006 ) . In-line with earlier reports ( Dumez et al . , 2006; Wang et al . , 2010 ) , our data also do not support the idea that the gene annotated as BE1 in Arabidopsis represents a third BE activity ( Figure 2—figure supplement 1 ) . While we believe that we have expressed the right enzyme combinations , differences in the relative amounts of each enzyme between the yeast lines and Arabidopsis are likely . Indeed , although the BE activities according to our zymograms appear comparable in yeast and in Arabidopsis leaves , some SS activities differ ( Figure 2A ) . In particular , SS3 activity was elevated in yeast compared with Arabidopsis . The same may be true for SS4 , which was proposed to contribute only little to chain elongation in Arabidopsis ( Szydlowski et al . , 2009 ) . It is interesting to note that the repression or mutation of class II BEs causes the so-called amylose-extender phenotype , where amylopectin is characterized by longer chains ( Boyer et al . , 1980; Hedman and Boyer , 1982; Mizuno et al . , 1993; Sestili et al . , 2010; Regina et al . , 2010; Klucinec and Thompson , 2002 ) . In barley , this was shown to be accompanied by a longer lamellar repeat ( 12 . 5 nm instead of 10 . 4 nm; Regina et al . , [2012] ) . However , it is unclear whether this phenotype reflects a limitation of overall BE activity ( creating an imbalance with chain elongation ) or the specificity of the remaining class I BE , or both . In a recent molecular genetic approach , an impact on amylopectin structure either by BE specificity or by the amount of BE activity relative to the other starch-biosynthetic enzymes was shown directly . Lu et al . , ( 2015 ) complemented the be2be3 Arabidopsis mutant with class I or class II BEs . Analysis of different transgenic lines established that when either BE activity was limiting , it resulted in decreased amounts of starch and amylopectin with a higher proportion of chains of DP > 18 , compared with when the same BE was expressed at higher levels . It would be interesting to see if , in such transgenic Arabidopsis , there is variation in the length of the lamellar repeat . Another difference between the glucans made in yeast and in plant cells concerns the presence of soluble polysaccharides: while virtually all plant glucans are normally laid down as insoluble starch granules , every yeast strain accumulated also substantial amounts of soluble glucans . The percentages of insoluble glucans from total glucans varied depending on the complement of enzymes present ( as discussed above ) . Still , it did not exceed ~70% in the best cases ( strains O and N , Figure 4C and Figure 3—figure supplement 2 ) and ranged between 17 and 58% when ISA and both BEs were present ( Figure 4A ) . This incomplete formation of insoluble glucans in yeast can have several reasons . First , it could again be a matter of enzyme balance , although we do not know of a comparable instance from plants here . Second , the efficient crystallization of glucans may require additional factors that we have not yet introduced into yeast . An interesting candidate is Early Starvation1 ( or ESV1 ) , a non-catalytic starch-binding protein whose absence causes altered granule morphology and starch turnover in Arabidopsis ( Feike et al . , 2016 ) . ESV1 was proposed to facilitate the correct alignment of glucans within the granule matrix , which in turn could indirectly control the accessibility and thus degradation rate of glucans . However , its precise action remains unclear . Third , it is possible that soluble glucans that fail to crystallize efficiently are also synthesized in plant cells , but are rapidly removed by the starch-degrading enzymes described above . In yeast , such soluble glucans would accumulate as soluble glucans , since the degrading machinery is absent . Finally , it is also important to note that our definition of insoluble and soluble glucans in this context is based on the fractionation by centrifugation at 6000 g for 5 min . Insoluble glucans with a tiny particle size may remain in the supernatant after such a low-speed centrifugation . Indeed , preliminary experiments suggest that half of the glucans in the soluble fraction from line 29 sediments upon extending the centrifugation to 20 min ( data not shown ) , indicating that at least part of it may actually be insoluble . Further work should allow us to dissect the causes of the structural differences described above and simultaneously take us closer to a holistic view of starch biosynthesis . For instance , by applying quantitative proteomics , we could assess the relative differences in enzyme levels between yeast and Arabidopsis . By exerting further control over the relative expression level of the starch-biosynthetic enzymes in yeast , we would expect to obtain insoluble glucans with an even closer match to genuine Arabidopsis leaf starch . Furthermore , the systematic variation of the relative activities could allow as to determine the impact of enzyme level as well as enzyme specificity . This could be achieved using well-defined modules for heterologous gene expression ( Lee et al . , 2015 ) , benefitting from the recent large-scale characterization of yeast promoters , 5’UTRs and terminator sequences ( Keren et al . , 2013; Dvir et al . , 2013; Yamanishi et al . , 2013 ) . Our system may also be optimal for exploring the special characteristics of each enzyme in detail by studying the effects of random or site-directed mutations , domain swaps etc . Further , it provides exciting new opportunities to explore other important , but less-well understood aspects of starch granule synthesis . For instance , a number of starch-biosynthetic enzymes , have been found to be phosphorylated ( Tetlow et al . , 2004 , 2008; Grimaud et al . , 2008; Makhmoudova et al . , 2014; Kötting et al . , 2010 ) and , in case of cereals , to assemble into protein complexes ( Tetlow et al . , 2008; Hennen-Bierwagen et al . , 2008 , 2009; Liu et al . , 2009; Crofts et al . , 2015; Ahmed et al . , 2015 ) . Still , whether and how these protein modifications and associations affect the enzyme activities has mostly remained elusive . In yeast , these questions could be readily addressed by expressing putative regulatory kinase/phosphatase pairs or enzymes with mutated phosphorylation sites or interaction domains . Our understanding of industrially important starch traits , such as starch phosphorylation or amylose content could also be greatly improved . Enzymes controlling starch phosphorylation ( i . e . , glucan , water dikinases and glucan phosphatases; Silver et al . , [2014] ) could be expressed and studied in yeast , as could factors affecting amylose content . For the latter , one could integrate GBSS with its recently-identified granule-targeting protein PTST ( Protein Targeting to Starch , Seung et al . , 2015 ) , with or without enzymes altering the levels of malto-oligosaccharides that may serve as primers of amylose synthesis ( e . g . , starch-degrading enzymes; Zeeman et al . , 2002a ) . There is also the potential to study non-enzymatic proteins , such as ESV1 ( Feike , 2016 ) or the rice protein Floury Endosperm6 ( FLO6; Peng et al . , 2014 ) . These proteins are of particular interest since they appear to exert their function through binding to starch itself or to starch-biosynthetic enzymes , thus providing an additional level of enzyme regulation . Furthermore , our system could be used to investigate the potential of non-plant enzymes , for instance 4 , 6-α-glucanotransferases ( Bai et al . , 2015 ) , to modify starch polymers and thereby improve their properties . Such functional aspects could also be explored given that yeast culture is scalable . Having illustrated the feasibility of this approach using the Arabidopsis genes , we suggest that it could easily be applied to study starch-biosynthetic pathways in crop species where knock-out mutants are rarer and/or more tedious to create ( due to generation times , polyploidy , etc . ) . Even with advanced genome editing and transgenesis methods , the creation of a set of homozygous plant lines may take substantially longer than recreating and analyzing the pathway in detail in yeast , particularly in the light of the emerging systems for combinatorial gene assembly ( Weber et al . , 2011; Lee et al . , 2015 ) . The yeast platform may thus be used to perform comprehensive pioneering studies to identify promising enzyme combinations for subsequent testing in plants . Overall , this system could usher in a new era in starch biosynthesis research and inspire work ranging from theoretical modelling of the biosynthetic process to the strategic analysis of functional properties of novel glucans for industrial use . Unless otherwise noted , chemicals were purchased from Sigma-Aldrich and restriction enzymes from Fermentas ( Thermo Fisher Scientific ) . Complex medium consisted of 1% ( w/v ) Bacto yeast extract ( BD ) and 2% ( w/v ) Bacto peptone ( BD ) , supplemented with either 2% ( w/v ) glucose ( medium named YPD ) or 2% ( w/v ) galactose ( Acros Organics; medium named YPGal ) . Sugars were added after autoclaving; 20% ( w/v ) glucose stock was separately autoclaved and 20% ( w/v ) galactose stock was filter-sterilized through a 0 . 22 µm filter unit . Minimal medium lacking nitrogen ( MIN-N medium ) was prepared as described ( Johnston et al . , 1977 ) , except for the replacement of FeCl33 × 6 H2O ( 10 µg L−1 ) by NH4Fe ( SO4 ) 2 × 10 H2O ( 16 . 5 µg L−1 ) and the addition of histidine and uracil ( 20 mg L−1each ) . The medium was supplemented with either 2% ( w/v ) glucose ( named MIN-N-Glu ) or 2% ( w/v ) galactose ( Acros Organics; medium named MIN-N-Gal ) . YPD plates were prepared as for liquid medium , but included 2% ( w/v ) bacto agar ( BD ) . Plates containing synthetic complete ( SC ) or complex medium with glycerol ( YPG ) were prepared as described ( Sherman et al . , 1986 ) , except for a higher concentration of leucine in SC media ( 60 mg L−1 ) . Uracil was omitted in SC-ura plates . Plates with synthetic medium and galactose ( SCgal ) , used for iodine staining of yeast colonies were prepared like SC plates but with replacement of glucose by 2% ( w/v ) galactose . Plates with 5-fluoroorotic acid ( FOA plates ) are SC plates supplemented with 0 . 1% ( w/v ) 5-fluoroorotic acid ( Fermentas ) . The underlying alleles of the homozygous Arabidopsis thaliana mutants are listed in Supplementary file 1E . Wild-type ( WT ) Arabidopsis ( either of Wassilewskija [WS] or Columbia-0 [Col-0] ecotype ) and mutant plants were grown as described ( Streb et al . , 2008 ) . Unless otherwise noted , plants were grown for four weeks and harvested at the end of the light period for maximum leaf starch content . Purified WT Arabidopsis leaf starch from Col-0 ecotype ( used for SEM ) and amylose-free Arabidopsis leaf starch from gbss ptst mutants ( Seung et al . , 2015 ) ( used for X-ray scattering , X-ray tomography , iodine staining and iodine absorption spectra ) were kindly donated by David Seung . Potato amylopectin is tuber starch isolated from the amf amylose-free variety ( Visser et al . , 1991 ) ( kindly provided by Prof . Richard Visser ) and potato amylose is type III from Sigma . CEN . PK113-11C ( MATa MAL2-8C SUC2 his3Δ ura3-52 ) was kindly provided by Prof . Barbara A . Halkier , Department of Plant and Environmental Sciences , University of Copenhagen , Denmark . Yeast strains were generated by sequential integration of constructs either at loci of the yeast expression platform ( Mikkelsen et al . , 2012 ) ( designed for stable multiple gene expression ) or at loci of the glycogen-metabolic pathway ( described below ) . The genotypes of S . cerevisiae strains used in the present study are presented in Supplementary file 1A . Yeasts were grown at 30°C with 260 rpm shaking . Unless otherwise specified , all yeasts were grown in complex medium . Yeast lines from YPD plates were first inoculated in YPD and grown overnight . Cells from these pre-cultures were inoculated in flask main cultures with either YPGal ( medium with galactose; inducing condition ) to a starting optical density at 600 nm ( OD600 ) of 0 . 3 or YPD ( medium with glucose; repressing condition ) to a starting OD600 of 0 . 1 . After growing for 6 hr , cells were pelleted , washed twice with water and the cell pellet flash frozen in liquid nitrogen . The wet weight is the weight of the cell pellet after carefully removing the supernatant prior to freezing . Samples were stored at −80°C . Replicate cultures arise from independent pre-cultures of a single yeast line . Yeasts were grouped according to replicate number ( instead of line/genotype ) in a non-random way during growth and processing . For growth of yeast lines that express BE3 or ISA1/ISA2 only ( Figure 2—figure supplement 1D ) , a specific protocol was used to induce the glycogen biosynthesis pathway ( e . g . , glycogen synthases ) . Yeasts from YPD plates were first inoculated in YPD for overnight cultures , then inoculated in YPgal ( inducing condition ) or YPD ( repressing condition ) to a starting OD600 of 0 . 6 and grown for 6 hr . To promote glycogen biosynthesis , yeasts were transferred to minimal medium lacking nitrogen ( MIN-N-gal or MIN-N-glu medium , respectively ) . Therefore , the cells were pelleted , once washed with the respective MIN-N medium and then resuspended in the MIN-N medium to reach the same volume as in YPgal/YPD . After additional shaking for 3 hr , cells were harvested as described for cultures in complex medium . Replicate cultures arise from independent main cultures . Unless otherwise noted , all steps were performed at 4°C . Yeast cell pellets from shake-flask experiments were resuspended in 3 volumes 1 . 12 M perchloric acid and 4 volumes glass beads ( acid-washed , 425–600 µm diameter ) . To calculate the volume of the cell pellets , the wet weight [mg] was multiplied with the factor 0 . 9 [µL mg−1] . The cells were then homogenized by vigorous vortexing for 45 to 50 min , and complete disruption of cells was confirmed at random using a light microscope . Samples were subject to centrifugation for 5 min at 6000 g , thereby fractionating the soluble fraction ( supernatant; soluble glucans , sugars , water-soluble cell components ) and the insoluble fraction ( pellet; cell debris , insoluble glucans ) . For isolation of soluble proteins , yeast cell pellets from liquid cultures with complex medium were resuspended in 3 . 3 volumes of native extraction buffer ( 100 mM MOPS , pH 7 . 5 , 1 mM EDTA , 5 mM DTT , 10% [v/v] glycerol , protease inhibitor [Complete EDTA-free; Roche] ) and 4 . 3 volumes of glass beads ( acid-washed , 425–600 µm diameter ) and homogenized by vortexing for 6 min in total at 4°C with cooling in between . Rosette material from two pooled four-week-old Arabidopsis plants ( harvested in the middle of a 12 hr day ) was homogenized in the same extraction buffer as above in an all-glass homogenizer . Soluble proteins were separated from cell debris by two sequential centrifugation steps ( each 8 min , 16 , 000 g at 4°C ) , frozen in liquid nitrogen and stored at −80°C until use . Protein amounts were determined using a Bradford-based protein assay ( Bio-Rad ) with bovine serum albumin ( BSA ) as standard . Equal loading was confirmed in SDS-PAGE . For total protein extracts , yeasts were grown and homogenized as above , but with non-native extraction buffer ( 100 mM Tris , pH 7 . 0 , 2% [w/v] SDS , protease inhibitor [Complete EDTA-free; Roche] ) . The extracts were heated to 99°C for 12 min , then clarified by centrifugation for 5 min at 16 , 000 g at 25°C . The supernatant , constituting the total protein extract , was frozen in liquid nitrogen and stored at −80°C until use . Protein concentrations were measured using the bicinchoninic acid ( BCA ) protein assay ( Pierce ) with BSA as standard . To monitor branching enzyme and isoamylase activity , 15 µg of native soluble proteins were loaded on a 7 . 5% ( w/v ) native polyacrylamide gel containing 0 . 015% ( w/v ) oyster glycogen and run at 10 V cm−1for 3 . 5 hr at 4°C . After washing for 30 min at 4°C in 50 mM Hepes-NaOH , pH 7 . 0 , and 10% ( v/v ) glycerol , the gel was incubated overnight at 25°C with gentle shaking in 50 mM Hepes-NaOH , 10% ( v/v ) glycerol , 2 . 5 mM AMP , 50 mM glucose 1-phosphate and 28 U ( per gel ) phosphorylase a ( from rabbit muscle ) . For starch synthase activity , 22 . 5 µg of proteins were separated on a 7 . 5% ( w/v ) native polyacrylamide gel containing 0 . 3% ( w/v ) oyster glycogen using 10 V cm−1for 3 . 5 hr at 4°C . The gels were incubated for 16 hr at 25°C with gentle shaking in 100 mM HEPES-NaOH , pH 7 . 5 , 2 mM DTT , 10% ( v/v ) glycerol , 0 . 5 mM EDTA , 0 . 5 M trisodium citrate ( tribasic ) and 0 . 8 mM ADPglucose . Both types of gels were stained with ~1:3 diluted Lugol's solution to visualize glucan-modifying enzyme activities . For the immunoblots shown in Figures 2B 15 µg of either soluble or total proteins were loaded on 12 . 5% ( w/v ) ( for SS2 and GlgC-TM ) or 7 . 5% ( w/v ) ( for SS4 ) SDS-PAGE gels , blotted on polyvinylidene difluoride membranes ( Immobilon-P , Millipore ) and probed with monoclonal anti-HA-peroxidase ( clone HA-7 , H6533; RRID:AB_439706 ) ( Egan et al . , 2014 ) or anti-FLAG ( M2 , F1804; RRID:AB_262044 ) primary antibody ( an antibody validation profile can be found in 1DegreeBio under http://1degreebio . org/reagents/product/751433/ ? qid=1013436; accessed 1 . 11 . 2015 ) and goat anti-mouse IgG ( H + L ) -HRP secondary antibody ( BIO-RAD , #1706516 ) , respectively . Chemiluminescence was visualized using the WesternBright ECL HRP substrate kit ( Advansta ) . The immunoblot shown in Figure 2—figure supplement 1C was performed using total protein extracts as the described above for SS4 , but using 7 . 5 µg of total proteins . Data analyzed by t tests or ANOVA ( Figure 3—figure supplement 2 and Figure 4A , B ) passed the Shapiro-Wilk test for normality ( p value ≥ 0 . 05 ) . Equality of variances between compared samples was assessed using F-tests ( α = 0 . 05 ) . Pair-wise comparisons of glucan content ( Figure 3—figure supplement 2 ) and of untransformed percentages of insoluble glucans ( Figure 4A ) were performed using two-sided t-tests ( with Welch’s correction when equal variances could not be assumed ) . Multiple comparisons of untransformed percentages of insoluble glucans ( Figure 4B ) were performed using one-way ANOVA with Tukey multiple comparison test as equal variances among compared samples could be assumed . The Arabidopsis Genome Initiative gene codes for the Arabidopsis genes used in this study are the following: ISA1 , At2g39930; ISA2 , At1g03310; SS1 , At5g24300; SS2 , At3g01180; SS3 , At1g11720; SS4 , At4g18240; BE1 , At3g20440; BE2 , At5g03650; BE3 , At2g36390 . The GenBank accession number for glgC , the AGPase from Escherichia coli , is V00281 . 1 . The gene IDs of the S . cerevisiae loci other than the loci of the yeast expression platform ( Mikkelsen et al . , 2012 ) are as follows: GSY1 , YFR015C; GSY2 , YLR258W; GLC3 , YEL011W; GLG1 , YKR058W; GLG2 , YJL137C; GDB1 , YPR184W; GPH1 , YPR160W ( CENPK113-7D database; www . sysbio . se/cenpk ) .	Most plants and algae produce a carbohydrate called starch , which provides the plant with a dense store of energy . Starch is also the main carbohydrate in our diet and its unusual physical properties mean that it has many industrial uses . It is made of two different sugar-based molecules known as glucans and forms large , partially crystalline granules inside plant cells . Several enzymes are known to be involved in making starch , yet it is not clear exactly how the process works . Animals and fungi cannot make starch but they do make another type of carbohydrate called glycogen , which is also a glucan . Yeast is a single-celled fungus that is often used in research because it is easy to genetically engineer and quick to grow . To study the plant enzymes that make starch in more detail , Pfister et al . aimed to genetically engineer yeast to make their own starch . For the experiments , different combinations of enzymes involved in starch production in a plant called Arabidopsis thaliana were inserted into mutant yeast cells that were unable to make glycogen . The experiments show that all the plant enzymes are active in yeast and retain the roles that they perform in plants . Some of the enzyme combinations yielded glucan granules that occupied a large part of the yeast cell . These granules had many of the physical characteristics of plant starch , showing that yeast can be used as a system to better understand how starch is made . Important next steps will be to insert more plant proteins into the yeast and to fine-tune the production of these proteins . This should help researchers to design starches with desired properties in yeast and ultimately engineer crop plants to produce them .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology", "biochemistry", "and", "chemical", "biology" ]	2016	Recreating the synthesis of starch granules in yeast
Human cytomegalovirus ( HCMV ) is a highly prevalent pathogen that induces life-long infections notably through the establishment of latency in hematopoietic stem cells ( HSC ) . Bouts of reactivation are normally controlled by the immune system , but can be fatal in immuno-compromised individuals such as organ transplant recipients . Here , we reveal that HCMV latency in human CD34+ HSC reflects the recruitment on the viral genome of KAP1 , a master co-repressor , together with HP1 and the SETDB1 histone methyltransferase , which results in transcriptional silencing . During lytic infection , KAP1 is still associated with the viral genome , but its heterochromatin-inducing activity is suppressed by mTOR-mediated phosphorylation . Correspondingly , HCMV can be forced out of latency by KAP1 knockdown or pharmacological induction of KAP1 phosphorylation , and this process can be potentiated by activating NFkB with TNF-α . These results suggest new approaches both to curtail CMV infection and to purge the virus from organ transplants . Human cytomegalovirus ( HCMV ) , a member of the β-herpes virus family , is a highly prevalent pathogen that induces life-long infections notably through the establishment of latency in hematopoietic stem cells ( HSC ) ( Sinclair , 2008 ) . Bouts of reactivation are normally asymptomatic because rapidly controlled by the immune system , but can be fatal in immuno-compromised individuals such as AIDS patients and transplant recipients , in particular when grafts from HCMV-positive donors are given to HCMV-negative individuals ( Sissons and Carmichael , 2002 ) . Furthermore , primary HCMV infection during pregnancy is a leading cause of congenital malformations of the central nervous system ( Britt , 2008 ) . The approximately 250 kb genome of HCMV encodes several hundred proteins , some 14 miRNAs and a few long-noncoding RNAs ( Stern-Ginossar et al . , 2012 ) . Infection of permissive targets such as epithelial cells or fibroblasts leads to a lytic cycle , with a highly orderly transcriptional cascade that first expresses the viral immediate early ( IE ) genes , the products of which set the cellular stage for the virus and activate the viral early ( E ) genes . These yield the effectors of viral genome replication , before proteins encoded by the viral late ( L ) genes finally trigger the formation of new particles . In contrast , when HCMV infects a HSC , the expression of lytic genes is rapidly suppressed and only a few latency-associated transcripts are detected ( Goodrum et al . , 2012 ) . The viral genome is then maintained as a stable episome , without replicating or producing new virions ( Goodrum et al . , 2004 ) . When HCMV-infected HSC are differentiated in vitro into dendritic cells ( DCs ) and these are induced to mature , the virus resumes a lytic cycle . Accordingly , blood CD14+ monocytes , which are circulating precursors of DCs , harbor latent HCMV . HCMV reactivation is thus tightly coupled with myeloid differentiation . However , the molecular mechanisms of its persistence in latently infected cells are incompletely understood , albeit known to be associated with epigenetic modifications of the viral chromatin ( Avdic et al . , 2011; Umashankar et al . , 2011; Mason et al . , 2012; Petrucelli et al . , 2012 ) . In latently infected CD34+ HSC or circulating monocytes isolated from seropositive individuals , chromatin at the major immediate early promoter ( MIEP ) bears histone 3 trimethylated on lysine 9 ( H3K9me3 ) , a repressive mark , and heterochromatin protein 1 ( HP1 ) ( Sinclair , 2010; Reeves and Sinclair , 2013 ) . Upon differentiation of latently infected precursors into dendritic cells , H3K9me3 is replaced by high levels of histone acetylation , HP1 is lost , and a lytic cycle is triggered ( Taylor-Wiedeman et al . , 1994; Mendelson et al . , 1996; Hahn et al . , 1998; Reeves et al . , 2005 ) . KAP1 ( KRAB-associated protein 1 ) , also known as TRIM28 ( tripartite motif protein 28 ) or TIF1β ( transcription intermediary factor 1 beta ) is a transcriptional co-repressor essential for the early embryonic silencing of endogenous retroelements ( Rowe et al . , 2010; Castro-Diaz et al . , 2014; Turelli et al . , 2014 ) and involved in regulating multiple aspects of mammalian homeostasis , including in the hematopoietic system ( Jakobsson et al . , 2008; Bojkowska et al . , 2012; Chikuma et al . , 2012; Santoni de Sio et al . , 2012a , 2012b; Barde et al . , 2013 ) . KAP1 binds to the Krüppel-associated box ( KRAB ) domain present at the N-terminus of KRAB-containing zinc finger proteins ( KRAB-ZFPs ) ( Friedman et al . , 1996 ) . These constitute the single largest family of transcriptional repressors encoded by the genomes of higher organisms , with close to four-hundred members in either human or mouse , and are endowed with sequence-specific DNA binding ability via a C-terminal array of zinc fingers ( Urrutia , 2003 ) . KAP1 harbors from its N- to its C-terminus a RING finger , B-boxes , a coiled-coil region , a HP1-binding motif , a PHD finger and a bromodomain ( Iyengar and Farnham , 2011 ) . The first three of these motifs define the so-called RBCC or TRIM ( tripartite motif ) region , which is both necessary and sufficient for homo-oligomerization and direct binding to KRAB . The C-terminal effector end of the protein recognizes the backbone of histone tails , and interacts with two histone-modifying enzymes: Mi2α , an isoform of the Mi2 protein found in the NuRD ( nucleosome remodeling and histone deacetylation ) complex ( Schultz et al . , 2001 ) , and SETDB1 ( SET domain , bifurcated 1 ) , an H3K9me3-specific histone methyltransferase ( Schultz et al . , 2002 ) . The H3K9me3 mark in turn creates high affinity genomic binding sites for HP1 , bringing more KAP1 complex , which likely explains that KAP1-induced heterochromatin formation can spread several tens of kilobases away from an initial KAP1 docking site ( Groner et al . , 2010 ) . SETDB1 recruitment is stimulated by sumoylation of the KAP1 bromodomain , the last step of which is mediated intra-molecularly by an E3 ligase activity contained in the PHD domain ( Ivanov et al . , 2007 ) . Within the context of DNA damage , the ATM ( ataxia telangiectasia mutated ) kinase phosphorylates KAP1 at position 824 , resulting in decreased KAP1 sumoylation and SETDB1 recruitment , with secondary loss of KAP1 repressor activity ( White et al . , 2006 , 2012; Noon et al . , 2010 ) . Histone 3 lysine 9 trimethylation and HP1 recruitment are thus signatures of KAP1 action . Furthermore , we previously demonstrated that KRAB/KAP-mediated repression is functional within the context of episomal DNA ( Barde et al . , 2009 ) . Finally , the KRAB/KAP pathway has been found to influence the replication of two members of the herpes virus family , Epstein–Barr virus ( EBV ) ( Liao et al . , 2005 ) and Kaposi's sarcoma-associated herpes virus ( KSHV ) ( Chang et al . , 2009; Cai et al . , 2013 ) . Based on these premises , we investigated a possible role for KAP1 in the control of HCMV latency . We found that KAP1 is a key mediator of this process , and that it is targeted by an mTOR-mediated phosphorylation switch that can be exploited to force the virus out of latency . Our results suggest new avenues not only for controlling HCMV infection but also for purging the virus from organ transplants . We first infected either MRC5 fibroblasts or cord blood-derived CD34+ cells with the TB40-E HCMV clinical strain . Monitoring the levels of viral DNA and various RNA transcripts during the following days confirmed that lytic replication occurred in MRC5 cells , known to be fully permissive , whereas latency was established in HSC after an initial outburst of viral gene expression ( Figure 1—figure supplement 1A , B ) , as previously reported ( Goodrum et al . , 2007 ) . To examine the possible impact of KAP1 on this process , we depleted the corepressor by lentivector-mediated RNA interference , complementing Kap1 knockdown cells by transduction with a vector expressing an shRNA-resistant Kap1 allele when relevant . We then infected these cells with TB40-E HCMV . In MRC5 cells , where KAP1 depletion was around 98% ( Figure 1—figure supplement 2A ) , levels of viral transcripts indicative of a lytic cycle ( the IE genes UL123 , UL122 , the E gene UL54 and the L gene UL94 ) were comparable in control and KAP1-depleted cells ( Figure 1A ) and production of viral proteins IE1 and IE2 was unchanged ( Figure 1—figure supplement 2B ) , indicating that the regulator is not essential for productive HCMV replication . In unsorted population of HSC exposed to the knockdown lentivector , KAP1 reduction was only 85% but in sorted CD34+ GFP-positive , that is , transduced cells , it reached 95% ( Figure 1—figure supplement 2C , D ) . In this KAP1-depleted subpopulation , expression of immediate early , early and late viral genes at 7 days post-infection was increased between 10- and 35-fold compared with control cells ( Figure 1B ) , and massive amounts of infectious HCMV particles were released in the supernatant ( Figure 1C ) . Furthermore , complementation of KAP1-depleted cells with an shRNA-resistant Kap1 allele restored HCMV latency ( Figure 1—figure supplement 2E–I ) . These results thus indicated that KAP1 is necessary for the establishment of HCMV latency in HSC . We then inverted the sequence of our manipulations to induce Kap1 knockdown in CD34+ cells that had been infected 7 days earlier with the TB40-E virus ( Figure 1—figure supplement 2J ) . 7 days after transduction , we measured the expression of the UL122 , UL123 , UL54 and UL94 mRNAs , which are found only during a lytic cycle ( Reeves and Sinclair , 2013 ) ( Figure 1D ) . All transcripts were significantly upregulated in knockdown cells , which accordingly released infectious viral particles ( Figure 1E ) . Furthermore , the NF-κB activator TNFα increased viral gene expression and virion production from Kap1 knockdown but not from control TB40-E-infected CD34+ cells ( Figure 1D , E ) . These data indicate that ( i ) KAP1 is necessary both for the establishment and for the maintenance of HCMV latency in human HSC , and ( ii ) the relief of KAP1-mediated repression is not in itself sufficient for full HCMV reactivation in CD34+ cells , but provides the ground for stimulation of viral gene expression by HCMV activators such as NF-κB . 10 . 7554/eLife . 06068 . 003Figure 1 . KAP1 is required for HCMV latency . RT-qPCR analysis of indicated HCMV genes expression in MRC-5 ( A ) or cord blood CD34+ ( B , C , D , E ) cells infected with the TB40-E strain 3 days after ( A , B , C ) or 7 days prior to ( D , E ) being transduced or not ( NT ) with lentivectors expressing ( shKAP1 ) or not ( empty ) a small hairpin RNA against Kap1 , sorting CD34+ cells for GFP and CD34 expression . RT-qPCRs were performed 7 days after HCMV infection ( A and B ) or 7 days after lentiviral transduction ( D and E ) . Results are presented as average of fold change expression vs NT after GAPDH and β-2M normalization ( n = 4 , p < 0 . 05 , p < 0 . 01 , error bars as s . d . ) . ( C and E ) HCMV production in CD34+ cells was quantified by plaque assay on MRC-5 cells . Results are presented as average of PFU/ml ( n = 3 , p < 0 . 05 , p < 0 . 01 , p < 0 . 001 , error bars as s . d . ) . See Figure 1—figure supplements 3 , 4 for all individual experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 00310 . 7554/eLife . 06068 . 004Figure 1—figure supplement 1 . KAP1 is necessary for both establishment and maintenance of HCMV latency in HSC . HCMV kinetics analysis performed on MRC-5 fibroblasts ( A ) or CD34+ HSC ( B ) . HCMV mRNA and DNA expression were quantified at day 1-3-5-7 and 10 post-infection by RT-qPCR . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 00410 . 7554/eLife . 06068 . 005Figure 1—figure supplement 2 . KAP1 is necessary for both establishment and maintenance of HCMV latency in HSC . KAP1 knockdown and re-complementation efficiency were tested by RT-qPCR ( A , C , E , J ) and Western-blot analysis ( D ) . MRC-5 fibroblasts ( A and B ) or CD34+ cells ( C–J ) were transduced or not ( NT ) with lentiviral vectors expressing ( shKAP1 ) or not ( empty , scramble ) an shRNA against Kap1 , and co-transduced with a vector expressing ( KAP1-optim ) or not ( control ) an shRNA-resistant Kap1 allele , 3 days before ( A–H ) or 7 days after ( J ) infection with the TB40-E HCMV strain . Indicated HCMV transcripts were quantified by RT-qPCR for CD34+ cells ( F–I ) and by western blot analysis for MRC-5 ( B ) . All results are presented as average of fold change expression vs NT after GAPDH and β-2M normalization for mRNA and GAPDH and Albumin for DNA ( n = 4 , p < 0 . 05 , p < 0 . 01 , error bars as s . d . ) . Western-Blots are representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 00510 . 7554/eLife . 06068 . 006Figure 1—figure supplement 3 . KAP1 is necessary for establishment of HCMV latency in HSC . Figure 1A , B , C source datas . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 00610 . 7554/eLife . 06068 . 007Figure 1—figure supplement 4 . KAP1 is necessary for maintenance of HCMV latency in HSC . Figure 1D , E source datas . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 007 Cells in which HCMV expression was induced following KAP1 depletion kept expressing the CD34 stem cell marker , indicating that viral activation was not simply due to their differentiation into normally permissive cells such as activated DCs or macrophages . Still , as KAP1 exerts pleomorphic effect on hematopoiesis ( Barde et al . , 2013 ) , we could not exclude that its depletion induced HCMV lytic genes expression through indirect effects . We thus asked whether the co-repressor associates with the HCMV genome by performing KAP1-specific chromatin immunoprecipitation-deep sequencing ( ChIP-seq ) analyses on material isolated from human CD34+ HSC at 7 days post-TB40-E infection . Using a stringent peak-calling algorithm and validating its results by ChIP-PCR , we identified 28 major KAP1-enriched regions on the HCMV genome ( Figure 2A , B and Supplementary file 1 ) . Supporting the functional significance of KAP1 recruitment to latent HCMV genomes , ChIP-PCR analyses also detected SETDB1 , H3K9me3 and HP1α at these KAP1-enriched regions ( Figure 2C and Figure 2—figure supplement 1A ) . Furthermore , in line with our former demonstration that heterochromatin formation can spread several tens of kilobases away from primary KAP1 docking sites ( Groner et al . , 2010 ) , H3K9me3 and HP1α were found at significant distances from these major KAP1 peaks , and particularly covered the major immediate early promoter ( MIEP ) , the viral origin of lytic replication ( OriLyt ) and the UL112 gene . In contrast , the repressive mark was not found at genes known to be expressed during latency such as UL138 or LUNA ( Reeves and Sinclair , 2013 ) ( Figure 2D and Figure 2—figure supplement 1B ) . Importantly , SETDB1 and HP1α recruitment as well as H3K9 trimethylation were lost when KAP1 was depleted by RNA interference prior to HSC infection , supporting a model whereby the corepressor recruits these heterochromatin-inducing factors ( Figure 2—figure supplement 1A–D ) . 10 . 7554/eLife . 06068 . 008Figure 2 . KAP1 , SetDB1 and H3K9Me3 are enriched on the HCMV genome in latently infected HSC . ( A ) KAP1 binding sites on HCMV genome in TB40-E-infected CD34+ cells , mapped by ChIP-Seq performed at day 7 post-infection . Results are presented as hit point upper 120 on the reference sequence GenBank EF999921 . 1 , indicating all KAP1 peaks ( 1–28 ) . ( B and C ) ChIP-PCR with anti-KAP1 ( B ) , anti-SetDB1 and anti-H3K9Me3 ( C ) antibodies were performed on ChIP-Seq-mapped KAP1 peaks ( 1–28 ) plus two KAP1-negative HCMV regions ( Neg 1 and 2 ) , using EVX-1 and GAPDH as negative cellular gene controls , and ZNF180 and RP11-517P14 . 7 as positive controls . ( D ) H3K9me3-specific ChIP-PCR of two latency genes ( UL138 , LUNA ) and three regions active during lytic replication ( OriLyt , UL112 and MIEP ) , with the same controls as in ( C ) . Results are presented as total input fold enrichment ( EVX-1 normalized ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 00810 . 7554/eLife . 06068 . 009Figure 2—figure supplement 1 . KAP1-dependent recruitment of HP1α and SETDB1 with H3K9 trimethylation of HCMV genome in latently infected HSC . ChIP-PCR analysis of TB40-E-infected WT MRC-5 ( A and B ) KAP1-depleted CD34+ ( A–D ) , WT CD34+ ( A and B ) or mDC ( A and B ) with antibodies against HP1α ( A and B ) SETDB1 ( C ) or H3K9me3 ( C and D ) , as described in Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 009 We then examined the epigenetic status of HCMV genome when TB40-E-infected CD34+ cells were differentiated in mature dendritic cells ( mDCs ) , a procedure previously demonstrated to result in HCMV activation . As expected , the TB40-E DNA was no longer associated with SETDB1 or HP1α and did not bear the H3K9me3 repressive mark in the CD34-derived mDCs ( Figure 3A and Figure 2—figure supplement 1A ) . However , KAP1 was surprisingly still associated with the viral genome in these targets ( Figure 3B ) . The same pattern was recorded in TB40-E infected MRC-5 cells , where the virus achieves a complete lytic cycle ( Figure 3—figure supplement 1A , B ) . It suggested that HCMV reactivation and replication could occur in spite of corepressor binding . To probe this issue further , we infected CD34+ cord blood and MRC5 cells with the HCMV AD169 laboratory strain , which is incapable of inducing latency ( Goodrum et al . , 2007; Saffert et al . , 2010 ) . At day 7 post-infection , while the productively transcribed AD169 genome carried as expected neither SETDB1 nor H3K9Me3 , it was still bound by KAP1 in both cell types as robustly as the latent TB40-E strain in HSC ( Figure 3—figure supplement 1C , D ) . Therefore , it is not KAP1 recognition but rather the secondary recruitment of its SETDB1 effector and other KAP1-associated heterochromatin inducers such as HP1 , which is responsible for HCMV latency . 10 . 7554/eLife . 06068 . 010Figure 3 . A KAP1 phosphorylation switch governs HCMV progression from latency to lytic replication . ChIP-PCR of indicated HCMV genomic regions was performed with anti-SetDB1 and anti-H3K9Me3 ( A ) , anti-KAP1 ( B ) or anti- S824 phosphoKAP1 antibodies ( C and D ) on material extracted from TB40-E-infected cells CD34+ cells ( C ) or mature dendritic cells derived therefrom ( A , B , D ) , using same controls as in Figure 2 . Results are presented as total input fold enrichment after EVX-1 normalization . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 01010 . 7554/eLife . 06068 . 011Figure 3—figure supplement 1 . A KAP1 phosphorylation switch distinguishes HCMV latency and lytic replication . ChIP-PCR with SETDB1- , H3K9Me3- , KAP1- , pS824KAP1-specific antibodies were performed on MRC-5 fibroblasts ( A , B , F , G ) or CD34+ HSC ( C , D , E ) infected with the HCMV TB40-E ( A , B , G ) or AD169 ( C , D , E , F ) strains , as described in Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 011 Within the context of DNA damage , ATM phosphorylates KAP1 on serine 824 , thereby reducing its ability to bind SETDB1 hence its repressor potential ( White et al . , 2006 , 2012; Noon et al . , 2010 ) . We thus asked whether this modification could alter the consequences of KAP1 recruitment to the HCMV genome . We could immunoprecipitate the TB40-E DNA with pS824KAP1-specific antibodies in HCMV-reactivated mDC but not in latently infected HSC , with a pattern of genomic recruitment identical to that delineated with the global KAP1 antibody in both cell types ( Figure 3C , D ) . Furthermore , pS824KAP1 was similarly found associated with the replicating AD169 viral DNA in both MRC5 cells and CD34+ HSC ( Figure 3—figure supplement 1E , F ) and with the TB40-E-genome in productively infected MRC-5 fibroblasts ( Figure 3—figure supplement 1G ) . Confirming these data , immunofluorescence detected the presence of pS824KAP1 in 100% of TB40-E- or AD169-infected MRC-5 fibroblasts , but not in uninfected fibroblasts ( Figure 4A , B–D and Figure 4—figure supplement 1A ) , nor in CD34+ HSC latently infected with the TB40-E strain , which could be identified by their loss of expression of the surface protein MRP1 as recently described ( Weekes et al . , 2013 ) ( Figure 4E ) . 10 . 7554/eLife . 06068 . 012Figure 4 . mTOR inhibition prevents HCMV-induced KAP1 phosphorylation . MRC-5 ( A , B , C ) and CD34+ ( D and E ) cells , infected ( AD169 or TB40-E ) or not ( Non Infected ) with HCMV , and treated or not ( Non Treated ) with mTOR inhibitor ( Torin1 ) were examined by immunofluorescence with anti-IE and anti-MRP1 antibodies ( Alex-488 , green ) , and anti-S824 phosphoKAP1 antibodies ( Alexa 568 , red ) . DNA was stained with Dapi ( blue ) . White arrows represent latently infected cells ( MRP-1 negative cells ) . Scale Bar in white represents 10 µm for MRC-5 and 5 µm for CD34+ . All pictures are representative of slide overview from three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 01210 . 7554/eLife . 06068 . 013Figure 4—figure supplement 1 . mTOR and HCMV-associated KAP1 phosphorylation . ( A–F ) Confocal microscopy coupled to immunofluorescence was performed on MRC-5 ( A–C ) or CD34+ ( E and F ) HCMV infected ( TB40-E , AD169 ) or not ( Non Infected ) and treated or not ( Non Treated ) with mTOR ( Rapamycin , Torin1 ) or ATM ( KU55933 ) inhibitors . Staining was performed with antibodies against IE ( Alexa 488 , green ) , and pS824KAP1 ( Alexa 568 , red ) , staining DNA with Dapi ( blue ) . White scale bar , 10 µm . All pictures are representative of results obtained in three independent experiments . ( D ) KAP1 levels in MRC-5 Treated or not ( NT ) with Rapamycin ( Rapa ) Torin-1 ( Torin ) or KU55933 ( KU ) , analyzed by western-blot . Results are representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 01310 . 7554/eLife . 06068 . 014Figure 4—figure supplement 2 . mTOR and HCMV-associated KAP1 phosphorylation . ( A–C ) Confocal microscopy coupled to immunofluorescence was performed on MRC-5 ( A–C ) , HCMV infected ( TB40-E ) or not ( Non Infected ) and treated or not ( Non Treated ) with mTOR inhibitors ( Torin1 ) . Staining was performed with antibodies against IE or anti-β-Tubulin ( Alexa 488 , green ) , and pS473KAP1 ( Alexa 568 , red ) , staining DNA with Dapi ( blue ) . White scale bar , 10 µm . All pictures are representative of results obtained in three independent experiments . ( D ) In vitro phosphorylation of GST-KAP1 by recombinant mTOR , using Western blot and Ponseyu staining to detect pS824KAP1 and total KAP1 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 014 We then set to identify the kinase responsible for phosphorylating KAP1 in cells productively infected with HCMV . The ATM inhibitor KU55933 did not prevent this process , suggesting that this known KAP1 kinase did not play a primary role ( Figure 4—figure supplement 1B ) . We thus turned to the mammalian target of rapamycin ( mTOR ) , which was previously found activated in cells replicating HCMV ( Moorman and Shenk , 2010; Clippinger et al . , 2011; Clippinger and Alwine , 2012; Poglitsch et al . , 2012 ) . When TB40-E infected MRC-5 cells were treated with the mTOR inhibitors rapamycin or Torin1 ( Thoreen and Sabatini , 2009; Thoreen et al . , 2009 ) , the phosphoKAP1-specific immunofluorescence signal was suppressed ( Figure 4C and Figure 4—figure supplement 1C ) , whereas total levels of KAP1 were unaffected ( Figure 4—figure supplement 1D ) . Torin1-preventable pS824KAP1 accumulation was also documented in CD34+ HSC infected with the replicative AD169 strain ( Figure 4D and Figure 4—figure supplement 1E , F ) . In this setting , the IE protein-specific signal was reduced by the mTOR inhibitor , consistent with some repression of viral gene expression . Of note , some KAP1 phosphorylated on serine 473 was detected in MRC5 cells , but this was independent of HCMV infection and Torin1-resistant ( Figure 4—figure supplement 2A–C ) . In contrast , recombinant mTOR could phosphorylate GST-KAP1 on S824 in vitro ( Figure 4—figure supplement 2D ) . Cell fractionation analyses indicated that pS824KAP1 accumulated in the nucleus of CMV-infected cells , which also displayed increased levels of phospho-S6 ribosomal protein , a marker of mTOR activation ( Figure 5A ) . Indirect immunofluorescence and confocal microscopy further revealed mTOR-specific foci in the nucleus of TB40- or AD169-infected but not control MRC5 cells ( Figure 5B and Figure 5—figure supplement 1A ) . Finally , ChIP with an mTOR-specific antibody demonstrated that the kinase associated with the CMV genome at the same places regions as pS824KAP1 ( Figure 5C ) . Consistent with a restriction of KAP1 phosphorylation and inactivation to CMV-associated molecules , the corepressor could still mediate the transcriptional silencing of an integrated TetO-PGK-GFP cassette via a Tet repressor-KRAB fusion protein ( Wiznerowicz and Trono , 2003 ) in virus-infected cells ( Figure 5—figure supplement 1B ) . 10 . 7554/eLife . 06068 . 015Figure 5 . mTOR associates with the HCMV genome during lytic replication . ( A ) MRC-5 cells were infected 3 days ( TB40-E ) or not ( NT ) with HCMV and harvested for cell fractionation . IE1/2 , α-Tubulin , KAP1 , KAP1 pS824 , Histone 3 and phosphorylated or not S6 ribosomal protein expression/localization in Total extract , cytosolic ( soluble ) nuclear ( insoluble ) and chromatin ( chromatin ) parts were analysed by Western-Blot . ( B ) Immunofluorescence by confocal microscopy was performed on MRC-5 , HCMV infected ( TB40-E , AD169 ) or not ( Non Infected ) . Staining was performed with anti-IE ( Alex-488 , green ) , and anti-mTOR antibodies ( Alexa 568 , red ) . DNA was stained with Dapi ( blue ) . Scale Bar in white represents 5 µm . ( C ) ChIP-PCR for indicated HCMV genomic regions were performed with anti-mTOR or control-IgG on material extracted from TB40-E-infected MRC-5 . Results are presented as total input fold enrichment after HCMV Neg1 region normalization . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 01510 . 7554/eLife . 06068 . 016Figure 5—figure supplement 1 . mTOR nuclear translocation and KAP1 activity in HCMV replicating cells . ( A ) Confocal microscopy coupled to immunofluorescence was performed on MRC-5 , infected ( TB40-E , AD169 ) or not ( Non Infected ) with HCMV , using antibodies against IE- ( Alex-488 , green ) or mTOR ( Alexa 568 , red ) , staining DNA with Dapi ( blue ) . White scale bar , 10 µm . ( B ) MRC-5 cells were transduced with a lentiviral vector expressing GFP from the TetO-PGK promoter ( TetO-GFP ) with or without tTR-KRAB , infected or not with HCMV and treated or not with DOX as indicated . Results are presented as % of GFP positive cells , using non-infected and Dox treated cells as a 100% reference . ( n = 3 , p < 0 . 01 , error bars as s . d . ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 016 These results suggested that HCMV silencing might be amenable to suppression by pharmacologically induced KAP1 phosphorylation . In the absence of strictly specific mTOR activator , we turned to ATM , a kinase previously demonstrated as capable of phosphorylating KAP1 on S824 ( White et al . , 2012 ) . When HSC infected 5 days earlier with TB40-E were incubated with the ATM activator chloroquine ( Bakkenist and Kastan , 2003 ) , intracellular levels of viral RNAs and DNA increased ( Figure 6A , B ) , and infectious viral particles were released in the supernatant ( Figure 6C ) . Supporting the specificity of the observed phenomenon , Torin-1 did not block chloroquine-induced HCMV RNA , DNA and viral particle production , whereas the ATM inhibitor Ku55933 prevented this process ( Figure 6 ) . These results obtained in a first set of HSC obtained from three donors were confirmed in another group of three donors , where it was further observed that levels of viral transcripts increased in drug-treated cells only transiently after a single dose of drug but steadily if three doses were administered over 5 days ( Figure 6—figure supplement 1A ) , and that single doses of the ATM activator prevented the progressive drop in viral DNA that was observed in control cells , while repeated doses triggered an augmentation in viral DNA copies , indicative of genome replication ( Figure 6—figure supplement 1B ) . Correlating these quantitative virological data , chloroquine treatment allowed the detection by immunofluorescence of pS824KAP1-positive cells ( Figure 6—figure supplement 1C ) . These were also systematically positive for IE antigens ( Figure 6—figure supplement 1C ) , confirming that the virus harbored by these cells had exited latency . Finally , when the supernatant of TB40-E-infected HSC exposed to chloroquine was used to inoculate MRC5 fibroblasts , it induced the accumulation of viral DNA in these targets ( Figure 6—figure supplement 1D ) , demonstrating that the production of replication-competent virus had been induced by ATM-activating treatment of the latently infected HSC . Noteworthy , pS824KAP1 was detected only in HCMV IE-positive cells , which strongly suggests that KAP1 became phosphorylated in these cells through the combined action of ATM and some HCMV-encoded factor . Importantly , these effects were induced without drop in KAP1 levels or loss of surface expression of the CD34 stem cell marker ( Figure 6—figure supplement 2A , B ) . Additional evidence strongly suggested that chloroquine stimulated HCMV via KAP1 phosphorylation . First , the drug did not further boost TB40-E in MRC-5 cells , where viral replication is KAP1-insensitive ( Figure 6—figure supplement 2C , D ) . Second , it did not further stimulate HCMV gene expression and virion release in CD34+ cells depleted for KAP1 , whether or not these were complemented with a phosphorylation resistant S824A KAP1 mutant ( Figure 6—figure supplement 3A , B and Figure 6—figure supplement 4AB ) . Third , and by contrast , chloroquine did reactivate HCMV when these knockdown cells were complemented with wild-type KAP1 ( Figure 6—figure supplement 4A , B ) . 10 . 7554/eLife . 06068 . 017Figure 6 . Forcing HCMV out of latency in CD34+ cells by KAP1 pharmacological manipulation . After 5 days of infection with TB40-E , CD34+ cells were treated three times ( Day 0 , 3 and 5 ) or not ( NT ) with Chloroquine ( Chloro ) alone or in combination with the mTor inhibitor Torin-1 ( Torin ) or the ATM inhibitor KU59933 ( KU ) . ( A ) Indicated HCMV transcripts were quantified by RT-qPCR , using GAPDH and β-2 microglobulin for normalization . ( B ) HCMV DNA associated with the TB40-E-infected HSC was quantified by qPCR , normalizing with the GAPDH and albumin genes . Data are presented as average of three different experiments performed with cells from independent donors . ( C ) Supernatant from the TB40-E-infected CD34+ cells , harvested after 7 days of treatment and the viral production was quantified by classical plaque assay on MRC-5 fibroblasts . Results are presented as PFU/ml . Histogram represents an average of three different experiments performed with HSC from three independent donors ( n = 3 , p < 0 . 05 , p < 0 . 01 , error bars as s . d . ) . See Figure 6—figure supplements 5 , 6 for all individual experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 01710 . 7554/eLife . 06068 . 018Figure 6—figure supplement 1 . HCMV can be forced out of latency in CD34+ cells by pharmacological manipulation . After 5 days of infection with TB40-E , CD34+ cells were treated or not ( mock ) with chloroquine in single or multiple doses as indicated . Indicated HCMV transcripts were quantified by RT-qPCR , using GAPDH and β-2 Microglobulin for normalization ( A ) . HCMV DNA associated with the TB40-E-infected HSC was quantified by qPCR , normalizing with the GAPDH and albumin genes ( B ) . Data are presented as average of three different experiments performed with cells from independent donors . ( C ) Immuno-fluorescent staining was performed on CD34+ cells for pS824KAP1 ( pS824 ) , HCMV IE1-2 ( IE ) , or total KAP1 ( KAP1 ) , using Dapi for DNA . White scale bar , 10 µm . All pictures are representative of three independent experiments . ( D ) Supernatant from the TB40-E-infected CD34+ cells , harvested after 7 days of treatment , was used to infect MRC-5 fibroblasts , which were assessed for their viral DNA content 2 and 7 days later by qPCR . Results are presented as HCMV DNA level normalized for GAPDH and Albumin . Histogram represents an average of three different experiments performed with HSC from three independent donors ( n = 3 , p < 0 . 05 , error bars as s . d . ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 01810 . 7554/eLife . 06068 . 019Figure 6—figure supplement 2 . HCMV can be forced out of latency in CD34+ cells by pharmacological manipulation . ( A ) Immuno-fluorescent staining was performed on CD34+ cells for total KAP1 ( KAP1 ) , using Dapi for DNA . White scale bar , 10 µm . All pictures are representative of three independent experiments . ( B ) Percentage of CD34+ cells after 7 days of treatment was quantified by FACS with a PE-anti-CD34 antibody . Results are presented of fold change expression vs NT of three independent experiments . HCMV DNA ( C ) and mRNAs ( D ) were quantified by qPCR in MRC-5 cells infected for 5 days with TB40-E in the presence or absence of chloroquine . Results are presented as HCMV DNA or mRNA levels normalized for GAPDH and Albumin for DNA , GAPDH and β-2M for mRNA . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 01910 . 7554/eLife . 06068 . 020Figure 6—figure supplement 3 . HCMV can be forced out of latency in CD34+ cells by pharmacological manipulation . ( A ) RT-qPCR analysis of indicated HCMV transcripts in cord blood CD34+ cells infected with the TB40-E strain 7 days prior to being transduced or not ( NT ) with lentivectors expressing ( shKAP1 ) or not ( empty ) a small hairpin RNA against Kap1 . RT-qPCRs were performed 7 days after lentiviral transduction treated or not ( NT ) with chloroquine as described in Figure 6 . Results are presented as average of fold change expression vs NT after GAPDH and β-2M normalization ( n = 4 , p < 0 . 05 , p < 0 . 01 , error bars as s . d . ) . ( B ) HCMV production in CD34+ cells was quantified by plaque assay on MRC-5 cells . Results are presented as average of PFU/ml ( n = 3 , p < 0 . 05 , p < 0 . 01 , error bars as s . d . ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 02010 . 7554/eLife . 06068 . 021Figure 6—figure supplement 4 . HCMV can be forced out of latency in CD34+ cells by pharmacological manipulation . ( A ) RT-qPCR analysis of indicated HCMV transcripts in CD34+ cells infected with the TB40-E strain , 3 days after transduction or not ( NT ) with lentiviral vectors expressing or not ( empty ) an shRNA ( shKAP1 ) against Kap1 , together with a vector expressing an shRNA-resistant allele of the S824 KAP1 phospho-resistant mutant ( S824A ) or a wild type KAP1 ( KAP1 . opt ) lentiviral vector . After 5 days of infection with TB40-E , CD34+ cells were treated three times ( Day 0 , 3 and 5 ) or not ( NT ) with Chloroquine ( Chloro ) . Indicated HCMV transcripts were quantified by RT-qPCR , using GAPDH and β-2 microglobulin for normalization . ( B ) Supernatant from the TB40-E-infected CD34+ cells , harvested after 7 days of treatment and the viral production was quantified by classical plaque assay on MRC-5 fibroblasts . Results are presented as PFU/ml . Histogram represents an average of three different experiments performed with HSC from three independent donors ( n = 3 , p < 0 . 01 , error bars as s . d . ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 02110 . 7554/eLife . 06068 . 022Figure 6—figure supplement 5 . HCMV can be forced out of latency in CD34+ cells by pharmacological manipulation . Figure 6A source datas . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 02210 . 7554/eLife . 06068 . 023Figure 6—figure supplement 6 . HCMV can be forced out of latency in CD34+ cells by pharmacological manipulation . Figure 6B , C source datas . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 023 We also asked whether the pharmacological induction of KAP1 phosphorylation could trigger HCMV production from circulating monocytes . Monocytes were purified from the peripheral blood of seropositive individuals and exposed to three doses ( at days 0 , 3 and 5 ) of chloroquine , alone or in combination with Torin-1 or KU55933 ( Figure 7 ) . ATM activation did not trigger the differentiation of the monocytes into dendritic cells or macrophages , as verified by measuring the cell surface expression of CD1a , CD14 , CD80 and CD86 ( Figure 7—figure supplement 1 ) . However , it induced the intracellular accumulation of HCMV DNA and the release of infectious viral particles from these monocyte populations . Furthermore , Ku55933 but not Torin-1 prevented the stimulating effect of chloroquine , as noted in HSC . 10 . 7554/eLife . 06068 . 024Figure 7 . Inducing KAP1 phosphorylation releases HCMV from latency in monocytes . Monocytes from HCMV seropositive donors were purified and treated three times ( Day 0 , 3 and 5 ) or not ( NT ) with Chloroquine ( Chloro ) alone in combination with Torin-1 ( Torin ) or KU59933 . ( A ) HCMV DNA associated with the monocytes was quantified by qPCR , normalizing with the GAPDH and albumin genes . Data are presented as average of three different experiments performed with cells from independent donors . ( B ) Supernatant from the monocytes , harvested after 7 days of treatment and the viral production was quantified by classical plaque assay on MRC-5 fibroblasts . Results are presented as PFU/ml . Histogram represents an average of three different experiments performed with HSC from three independent donors ( n = 3 , p < 0 . 05 , p < 0 . 01 , error bars as s . d . ) . See Figure 7—figure supplement 2 for all individual experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 02410 . 7554/eLife . 06068 . 025Figure 7—figure supplement 1 . Monocytes do not differentiate during pharmacological reactivation of HCMV . After 7 days of treatment with chloroquine , monocytes were examined by FACS for the surface expression of CD1a , CD14 , CD80 and CD86 , analyzing results with the FlowJo software . Graphs are representative of experiments performed with cells from three different HCMV-seropositive donors . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 02510 . 7554/eLife . 06068 . 026Figure 7—figure supplement 2 . Monocytes do not differentiate during pharmacological reactivation of HCMV . Figure 7 source datas . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 026 These data demonstrate that inducing KAP1 phosphorylation results in releasing HCMV from its latent state by relieving repressive marks on the viral chromatin . Nevertheless , levels of viral gene expression achieved in this setting remain lower than observed when HCMV-harboring precursors are differentiated into macrophages or mature dendritic cells , where NF-κB has been demonstrated to drive viral transcription . We thus exposed latently infected CD34+ cells to chloroquine and TNF- α ( 5 ng/ml ) , either alone or in combination ( Figure 8 ) . With TNF alone , HCMV transcription was not induced , indicating that the virus remained latent . With chloroquine alone , significant induction of cell-associated viral transcripts ( approx . 100-fold ) and DNA ( approx . fivefold ) was measured , and viral particles were released in the supernatant ( 10 , 000 PFU/ml in the experiment depicted in Figure 8 ) . With the further addition of TNF , levels of CMV-specific RNAs increased by another 10-fold , the amounts of cell-associated viral DNA doubled , and virion production was boosted by a factor 10 . Importantly , this dual treatment did not trigger the differentiation of the HSC ( Figure 8—figure supplement 1 ) . Taken together , these results suggest a model whereby chloroquine treatment renders the HCMV chromatin permissive for transcription through KAP1 S824 phosphorylation , which then allows TNF to induce full viral gene expression via NF-kB activation . 10 . 7554/eLife . 06068 . 027Figure 8 . Combining KAP1 phosphorylation and NF-κB induction increases HCMV activation from latently infected HSC . After 5 days of infection with TB40-E , CD34+ cells were treated three times ( Day 0 , 3 and 5 ) or not ( NT ) with chloroquine ( Chloro ) in combination or not with 5 ng/ml of recombinant TNF-α ( TNF ) . ( A ) Indicated HCMV transcripts were quantified by RT-qPCR , using GAPDH and β-2 microglobulin for normalization . ( B ) HCMV DNA associated with the TB40-E-infected HSC was quantified by qPCR , normalizing with the GAPDH and albumin genes . Data are presented as average of three different experiments performed with cells from independent donors . ( C ) Supernatant from the TB40-E-infected CD34+ cells , harvested after 7 days of treatment and the viral production was quantified by classical plaque assay on MRC-5 fibroblasts . Results are presented as PFU/ml . Histogram represents an average of three different experiments performed with HSC from three independent donors ( n = 3 , p < 0 . 05 , **p < 0 . 01 , error bars as s . d . ) . See Figure 8—figure supplements 2 , 3 for all individual experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 02710 . 7554/eLife . 06068 . 028Figure 8—figure supplement 1 . Pharmacological activation of HCMV does not trigger CD34+ cells differentiation . After 7 days of treatment with chloroquine and TNF-α , cells were analyzed by FACS for CD34 surface expression . Representative of experiments performed on cells from three different donors . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 02810 . 7554/eLife . 06068 . 029Figure 8—figure supplement 2 . Pharmacological activation of HCMV does not trigger CD34+ cells differentiation . Figure 8A source datas . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 02910 . 7554/eLife . 06068 . 030Figure 8—figure supplement 3 . Pharmacological activation of HCMV does not trigger CD34+ cells differentiation . Figure 8B , C source datas . DOI: http://dx . doi . org/10 . 7554/eLife . 06068 . 030 This work sheds light on the mechanisms of HCMV persistence by revealing that , in latently infected hematopoietic stem cells , the master co-repressor KAP1 recruits HP1α and SETDB1 to the viral genome , triggering H3K9 methylation and heterochromatin formation . Our results correlate with the observation that , in latently infected CD34+ HSC or circulating monocytes isolated from seropositive individuals , chromatin at the HCMV MIEP bears H3K9me3 and HP1 ( Sinclair , 2010 ) . Our finding that KAP1 partakes in both the establishment and the maintenance of HCMV latency also corroborates the previously noted absence of more permanent silencing marks , such as DNA methylation , on latent HCMV genomes ( Hummel et al . , 2007 ) . We provide another important clue to the understanding of HCMV biology by revealing that the virus exits latency when KAP1 becomes phosphorylated on serine 824 , which blocks its ability to recruit SETDB1 hence abrogates its co-repressor potential . We further identify mTOR as responsible for this KAP1 phosphorylation switch , and capitalizing on these findings we finally demonstrate that HCMV can be forced out of latency by pharmacological activation of ATM , another KAP1 kinase , and that this effect can be potentiated by adding TNF , an inducer of NF-kB . Whether KAP1 phosphorylation relieves HCMV from latency only by dislodging SETDB1 and HP1 from the viral chromatin or through more active mechanisms remains to be determined . Interestingly , a small molecule inhibitor of LSD1/KDM1A was recently found to decrease HCMV immediate early gene transcription ( Liang et al . , 2013 ) . It could be that pS824KAP1 plays a role in attracting this H3K9 demethylase to the viral genome . There are remarkable analogies but also interesting differences between our results and recent data implicating KAP1 in the control of KSHV latency ( Chang et al . , 2009; Cai et al . , 2013 ) . In cells harboring latent KSHV genomes , KAP1 , HP1 and H3K9me3 were enriched at more than two thirds of viral promoters , and KAP1 knockdown reduced these repressive marks and stimulated virus production . Furthermore , upon induction of a lytic cycle by overexpression of the Rta viral transactivator , KAP1 dissociated from the viral genome , resulting in loss of HP1 and H3K9me3 . This correlated with phosphorylation of KAP1 on S824 , apparently mediated by the viral kinase vPK , with reciprocal decrease in the level of KAP1 sumoylation ( Chang et al . , 2009 ) . Here , while we find that a KAP1 phosphorylation switch can also force HCMV out of latency , we demonstrate that , with this virus , the phosphorylated form of the master repressor remains associated with actively transcribed viral genomes . Furthermore , we implicate the cellular kinase mTOR in KAP1 phosphorylation . It was previously noted that both the mTORC1 ( mTOR complex 1 ) and mTORC2 ( mTOR complex 2 ) arms of the mTOR pathway are activated during lytic HCMV replication ( Kudchodkar et al . , 2004 , 2006 , 2007 ) . This results in phosphorylating notably the eukaryotic initiation factor 4E ( eIF4E ) -binding protein ( 4E-BP1 ) and the p70S6 kinase ( S6K ) , which enhances viral protein translation and more generally minimizes cellular stress responses ( Clippinger et al . , 2011; Poglitsch et al . , 2012 ) . Here , we identify KAP1 as another substrate of this cascade , by observing that ( i ) the co-repressor is phosphorylated on serine 824 in cells productively infected with HCMV; ( ii ) this process is inhibited by the mTOR inhibitors rapamycin and Torin1; and ( iii ) mTOR targets the viral chromatin at KAP1-bearing genomic sites during a lytic cycle . Conversely , we could force HCMV out of transcriptional dormancy by inducing the activation of ATM , another kinase targeting KAP1 on serine 824 , in latently infected HSC and circulating monocytes . Nevertheless , our data also point to the role of viral factors in targeting this process . First , S824 phosphoKAP1 was detected only in HCMV IE antigen-positive cells . Second , KAP1 conserved its ability to repress a chromosomal target in cells where its HCMV silencing activity was abrogated by phosphorylation . This strongly suggests that mTOR is targeted to the HCMV genome , where the viral DNA-associated KAP1 molecules are then exposed to its action , a model supported by the results of our mTOR-specific chromatin immunoprecipitation . Such spatial compartmentalization would not be unprecedented , as site-specific KAP1 phosphorylation is also observed in case of DNA damage ( White et al . , 2006 ) , and as mTOR has itself been observed to be directed to sites of HCMV assembly ( Clippinger and Alwine , 2012 ) . More generally , it will be interesting to ask whether the KRAB/KAP1 pathway contributes to the latency of other herpes viruses , even though in the case of EBV the only evidence available so far is that it instead promotes viral DNA replication ( Liao et al . , 2005 ) . Latency is a dominant , cell type- and differentiation stage-specific mechanism ( Saffert et al . , 2010; Goodrum et al . , 2012 ) , as it is limited to HSC and cells of the myeloid lineage such as monocytes , and to HCMV clinical isolates ( e . g . , TB40-E ) that contain genomic regions commonly deleted in latency-incompetent laboratory strains ( e . g . , AD169 ) ( Goodrum et al . , 2007; Reeves and Sinclair , 2010; Petrucelli et al . , 2012 ) . Our findings thus suggest that , in HSC , a latency-specific gene product counters HCMV-induced mTOR-mediated KAP1 phosphorylation , for instance by either blocking the kinase or by acting as or inducing a KAP1-specific phosphatase . This warrants investigations aimed at elucidating this aspect , and at determining whether , when and , if so , downstream of what pathway mTOR becomes activated and KAP1 phosphorylated during the differentiation of HSC towards macrophages or activated DCs , the cells in which HCMV transcription becomes fully productive . KAP1 is not a DNA-binding protein , and is classically tethered to given genomic loci , for instance endogenous retroelements , by sequence-specific KRAB-ZFPs ( Friedman et al . , 1996; Wolf and Goff , 2009; Rowe and Trono , 2011; Castro-Diaz et al . , 2014 ) . By analogy , it is tempting to hypothesize that members of the KRAB-ZFP family are also involved in recognizing the HCMV DNA . In support of this model , KAP1 recruitment to the HCMV genome was lost when the KRAB-binding RBCC domain was deleted ( not illustrated ) . Since we detected close to thirty major KAP1 peaks on the HCMV DNA and failed to identify a common motif amongst all these sequences , more than one DNA-binding protein , whether or not KRAB-ZFP , is likely involved in docking the corepressor at these sites . In the case of KSHV , the latency protein LANA is responsible for tethering sumoylated KAP1 and the KAP1-bound Sin3A repressor to the Rta promoter , through its ability to interact with SUMO-2-bearing proteins . Upon hypoxia , KAP1 is de-sumoylated and together with Sin3A is released from the KSHV genome , leading to the activation of the LANA-associated HIF-1α , the inducible subunit of the hererodimeric HIF-1 ( hypoxia-induced factor 1 ) , thus activating viral gene expression ( Cai et al . , 2013 ) . By analogy , it could be that a virally encoded product contributes to targeting KAP1 to the HCMV genome . Our data have important clinical implications . First , it was recently noted that transplant recipients placed on immunosuppressive regimens that included rapamycin were less likely to suffer serious CMV-related complications than patients for whom this drug was omitted ( Andrassy et al . , 2012; Sabé et al . , 2012 ) . The present work indicates that this beneficial effect might stem at least in part from a prevention of viral reactivation . This warrants further studies to assess the positive impact of this and other mTOR inhibitors on the control of HCMV infection , whether after transplantation or in other settings . Of note , a recent study demonstrated that rapamycin did not prevent HCMV reactivation from latently infected dendritic cells in vitro ( Glover et al . , 2014 ) . However , it could be that KAP1 phosphorylation was already established in this system , and that maturation of the DCs only resulted in activating transcription factors that are strong stimulators of HCMV gene expression , such as NF-κB . Our discovery that HCMV can be forced out of latency by pharmacological manipulation suggests new avenues to eradicate infection by combining ATM and NF-κB activation with immune- or drug-based approaches aimed at killing HCMV-infected cells . The systemic administration of ATM and NF-κB activators may cause important side effects , and HCMV eradication is anyway probably unnecessary in immuno-competent individuals . However , the ex vivo treatment of HCMV-positive transplants , for instance bone marrow or HSC purified therefrom , could be readily envisioned to purge latently infected cells prior to engraftment , notably in the high-risk setting of HCMV-negative recipients . CD34+ cells from HCMV negative human cord blood were obtained from the Lausanne University Hospital ( Centre Hospitalier Universitaire Vaudois , CHUV , Switzerland ) delivery room with proper informed consent , layered on Ficoll gradient , purified with the CD34+ cell separation kit ( Miltenyi Biotec , Germany ) and freshly used for experiments without freezing to avoid differentiation . CD34+ cells were stimulated for one day in X-Vivo 15 medium ( Lonza , Switzerland ) supplied with cytokines ( 100 ng/ml of Flt-3 ligand , 100 ng/ml of SCF , 20 ng/ml of TPO , and 20 ng/ml of IL-6; Peprotech , Rocky Hill , NJ ) and 5% penicillin/streptomycin , maintained at a density of 5 . 105 cells/ml and transduced the day after stimulation with lentiviral vector at a MOI of 100 as previously described ( Barde et al . , 2013 ) , washed 24 hr later and cultured in X-Vivo 15 medium , 50 ng/ml of Flt-3 ligand , 25 ng/ml of SCF and 20 ng/ml of TPO . MRC5 fibroblasts serum and 5% penicillin/streptomycin . MRC-5 fibroblasts were transduced with a lentivector at a MOI of 10 , 3 days before HCMV infection . Monocytes were purified from HCMV seropositive donors buffy coat , obtained at the EFS ( Etablissement Français du Sang , France ) , with CD14+ magnetic selection kit ( Miltenyi Biotec ) . After 7 days of HCMV reactivation treatment , non differentiation of cells was controlled by different marker expression by FACS ( CD1a , CD14 , CD80 , CD86 ) ( Invitrogen , Carlsbad , CA ) . Cells were cultured in RPMI medium ( invitrogen ) supplemented with 10% FCS and Peni/strep . The mTOR ( Torin1 and rapamycin , Selleckchem , Houston , TX ) and ATM ( KU55933 , LuBioscience , Switzerland ) inhibitors were used at 0 . 1 and 10 µM , respectively , added 30 min before infection . The ATM activator ( chloroquine , Sigma–Aldrich , St Louis , MO ) was used at 20 µM . The recombinant TNF-α ( Peprotech ) is used at 5 ng/ml . After 10 days of TB40-E infection , CD34+ HSC were cultured for 7 days in X-VIVO-15 medium supplemented with GM-CSF ( 100 ng/ml ) , TGF-β ( 0 . 5 ng/ml ) , Flt-3 ligand ( 100 ng/ml ) , SCF ( 20 ng/ml ) and TNF-α ( 50 U/ml ) ( all from peprotech ) . This treatment leads to generation of immature dendritic cells that will be pushed to maturation with lipopolysaccharide ( LPS , 50 ng/ml ) treatment during 3 days , as described in ( Reeves et al . , 2005 ) . Cord blood was obtained after approval of the project by the local ethics committee , commission cantonale d'ethique de la recherche sur l'etre humain , as protocol 146/10 . When necessary , samples were obtained following informed consent of the participants . pLKO vectors were purchased from Sigma–Aldrich and the puromycin was replaced by eGFP ( pLKO-shKAP1 , pLKO-empty , pLKO-scramble ) . QuickChange II XL Site directed Mutagenesis Kit ( Agilent Technologies Santa Clara , CA ) was used to mutate KAP1 Serine 824 into Glutamic Acid in a pFUT plasmid containing a human Kap1 allele modified by nucleotide substitutions to resist to our shRNA . Production and titration of lentiviral vectors was performed as previously described ( Barde et al . , 2011 ) . The AD169 ( ATCC , Manassas , VA ) and TB40-E ( a gift from G Herbein , Besançon , France ) HCMV strains were used in this study . Virus stocks were generated in MRC5 or HUVEC cells , collecting particles when cytopathic effects were >90% . Supernatants were clarified of cell debris by centrifugation at 1 , 500×g for 10 min , ultracentrifuged at 100 , 000×g for 30 min at 4°C and stored at −80°C until use . Virus titers were determined by plaque assay on MRC5 cells using standard methods ( MEM2X diluted twofold with solution of 1 . 6% agarose ) . MRC-5 were infected at a MOI of 1 , while CD34+ infection were performed at a MOI of 5 , 3 days after or 7 days before transduction with lentivectors . Chromatin from 107 CD34+ was prepared and immunoprecipitated as in ( Barde et al . , 2013 ) , with KAP1- ( Abcam , UK ) , H3K9me3- ( Diagenode , Belgium ) , SETDB1- ( Abcam ) mTOR ( Abcam ) or S824 phosphoKAP1- ( Abcam ) specific antibodies . TB40-E infected CD34+ KAP1-ChIPed DNA was sent to sequencing . The 50 bp reads for KAP1 IP and Total Input were mapped to the HCMV genome ( TB40-BAC4 GenBank accession EF999921 . 1 ) using the bowtie short read aligner version 0 . 12 . 7 ( Langmead et al . , 2009 ) and allowing up to three mismatches over the whole length of the read . The signal of the total input was then subtracted from the KAP1 IP signal using a custom made PERL program ( see Source code 1 ) and regions with similar signal were merged together . The resulting regions were ranked according to their signal . The signal follows a normal distribution ( tested with Shapiro–Wilk normality test ) thus we considered as enriched regions the extreme right part of the distribution . The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus and are accessible through GEO Series accession number GSE53271 ( http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE53271 ) . SYBR green qPCR was performed to quantify enrichment at specific loci . Total Input Fold enrichment was quantified by the classical following calculation: 100 × 2^− ( CT of ChIPed DNA − CT of Total Input ) . Cellular promoters regions were used as negative ( EVX-1; GAPDH ) or positive ( ZNF-180; RP11-517P14 . 7 ) controls for PCR enrichment . 28 highly enriched peaks were found for the HCMV TB40-E strain , and two negative regions were designed for qPCR controls . MRC-5 were cultured directly on coverslips . Around 2 × 105 CD34+ were attached to slides by cytospinning with Shandon EZ Single Cytofunnel ( Thermo Fisher Scientific , Waltham , MA ) . Cells were then fixed and permeabilised with Methanol during 5 min at −20°C , and saturated with PBS 5% FCS during 2 hr at room temperature ( RT ) . Primary staining with IE- , anti-S824 phopshoKAP1- , KAP1- mTOR ( all from Abcam ) anti-KAP1 S473 phosphoKAP1 ( BioLegend , San Diego , CA ) specific antibodies was performed in PBS 5% FCS at 4°C overnight or 3 hr at RT , before addition of secondary antibodies ( Alexa-488 or -565 ) in PBS 5% FCS during 1 hr at RT . Dapi bath was performed during 10 min at RT , and slides were mounted with Fluoromount-G ( Southern Biotech , Birmingham , AL ) . Cells were sorted for GFP ( control of transduction ) and CD34 expression . RNA was extracted by Trizol and reverse transcribed with Superscript II ( both from Invitrogen ) according to the manufacturer's instructions . All qPCR were performed with SYBR green mix ( Roche , Switzerland ) . Primers used in this work were designed by Primer Express software ( Applied Biosystems ) and their sequences are available on Supplementary file 2 . Cells were washed two times with cold PBS and centrifuged . Pellets were resuspended in 500 µl of cytop Buffer ( Triton 0 . 25% , Tris HCl 10 mM , EDTA 5 mM , EGTA 0 . 5 mM and proteases inhibitor cocktail ) and incubated 3–5 min on ice . After centrifugation , supernatant was kept as soluble part ( cytoplasm ) and pellet were lysed with TNEN 250:0 . 1 buffer ( NaCl 250 mM , Tris HCl 50 mM , EDTA 5 mM , NP40 0 . 1% and proteases inhibitor cocktail ) 20–30 min on ice . After centrifugation , supernatants were kept as insoluble part ( nucleus ) and pellets were lysed in Urea 8 M , 10 min at 95°C to recover chromatin part . Total amount of protein was quantified by BCA assay . Cells lysates were subjected to SDS/PAGE on 4–12% polyacrylamide gels ( Invitrogen ) . iBlots on Nitrocellulose membrane ( Invitrogen ) were treated with KAP1- Histone 3- α-Tubulin- ( Abcam ) mTOR- S6 ribosomal protein- phospho S6 ribosomal protein ( Cell Signaling , Beverly , MA ) and βactin- ( Calbiochem , San Diego , CA ) specific antibodies , followed by polyclonal rabbit HRP-conjugated antibody , and protein bands were detected by using an ECL-plus kit ( Thermo-Scientific ) . Anti-KAP1pS824 antibody was produced in rabbits using a KLH-MBS coupled peptide with the following sequence: AcNH-CAG LSS QEL pSGG P-CONH2 from Eurogentec . GST-KAP1 and 1 U of mTOR ( 475987 Millipore , Billerica , MA ) were incubated in phosphorylation buffer ( 25 mM Hepes pH 6 . 8 , 10 mM MgCl2 , 0 . 5 mM DTT , 200 μM ATT , 1× Complete protease inhibitor cocktail from Roche , 1% BSA ) in a total volume of 15 μl for 30 min with shaking at room temperature . mTOR activity is inhibited with 100 nM of Torin1 .	Human cytomegalovirus ( HCMV ) is an extremely common virus that causes life-long infections in humans . Most individuals are exposed to HCMV during childhood , and the infection rarely causes any symptoms of disease in healthy individuals . However , in people with weaker immune systems—for example , newborn babies , people with AIDS , or individuals who have received an organ transplant—HCMV can cause life-threatening illnesses . It is difficult for the immune system to fight the infection because HCMV is able to hide in cells within the bone marrow called hematopoietic stem cells . Inside these cells , the virus can survive in a ‘dormant’ state for many years , before being reactivated and starting to multiply again . In most people , the immune system manages to control this new outbreak of HCMV , and the virus becomes dormant again , but reactivation of the virus in individuals with weakened immune systems is much more likely to cause serious illness . The results of previous studies suggest that when HCMV infects the hematopoietic stem cells , human proteins switch off the expression of many virus genes , which makes the virus inactive . The virus can be reactivated when infected stem cells change into a type of immune cell called dendritic cells , but it is not clear how this is controlled . Here , Rauwel et al . reveal that a human protein called KAP1 is responsible for switching off the virus genes in the stem cells . It does so by interacting with two other proteins to alter the structure of the DNA in these genes . However , if the stem cells are stimulated to change into dendritic cells , KAP1 becomes inactive , which allows the virus genes to be switched on . Rauwel et al . also show that it is possible to force HCMV out of its dormant state by using drugs to block the activity of KAP1 . This may aid the development of treatments that prevent the virus from causing serious illness in patients with weakened immune systems . For example , it could be used to remove dormant HCMV infections from bone marrow before it is transplanted into a new individual .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2015	Release of human cytomegalovirus from latency by a KAP1/TRIM28 phosphorylation switch
While Cre-dependent viral systems permit the manipulation of many neuron types , some cell populations cannot be targeted by a single DNA recombinase . Although the combined use of Flp and Cre recombinases can overcome this limitation , insufficient recombinase activity can reduce the efficacy of existing Cre+Flp-dependent viral systems . We developed a sensitive dual recombinase-activated viral approach: tTA-driven Recombinase-Guided Intersectional Targeting ( tTARGIT ) adeno-associated viruses ( AAVs ) . tTARGIT AAVs utilize a Flp-dependent tetracycline transactivator ( tTA ) ‘Driver’ AAV and a tetracycline response element-driven , Cre-dependent ‘Payload’ AAV to express the transgene of interest . We employed this system in Slc17a6FlpO;LeprCre mice to manipulate LepRb neurons of the ventromedial hypothalamus ( VMH; LepRbVMH neurons ) while omitting neighboring LepRb populations . We defined the circuitry of LepRbVMH neurons and roles for these cells in the control of food intake and energy expenditure . Thus , the tTARGIT system mediates robust recombinase-sensitive transgene expression , permitting the precise manipulation of previously intractable neural populations . The molecular heterogeneity of the nervous system requires a rich toolkit for precise study of distinct cell populations . Together with Cre recombinase-expressing mice , the available suite of Cre-dependent viral vectors permits the manipulation of genetically identified neural types . However , this approach neither permits the study of subpopulations of cells expressing a particular Cre allele ( Hodge et al . , 2019; Mickelsen et al . , 2019 ) nor permits the study of Cre-expressing cells within a defined Central nervous system ( CNS ) site while excluding the Cre-expressing cells in closely apposed brain regions . Restricting transgene expression to cells that express two marker genes can overcome this challenge ( Luan and White , 2007 ) . Early attempts at this approach utilized gene-specific , Cre-sensitive transgene-expressing alleles in combination with Cre alleles driven by distinct genes ( Chen et al . , 2011 ) . Another solution involves expressing two Cre fragments across different transgenes with distinct promoters , reconstituting active Cre only in cells that express both pieces ( Hirrlinger et al . , 2009 ) . These approaches generally involve substantial investments of time and resources , however , as they require generating and interbreeding multiple new gene- and experiment-specific alleles . Transgene expression can also be directed to cell types defined by two marker genes through the use of recombinase-dependent alleles ( often inserted into the Rosa26 Locus ) containing a ubiquitous promoter and tandem STOP cassettes that are each excised by distinct recombinases ( Daigle et al . , 2018; Sciolino et al . , 2016 ) . Neuroscience research often requires a higher degree of spatial specificity than afforded by the intersectional genetic models outlined above . Stereotaxic injections of recombinase-sensitive viral vectors can restrict transgene expression to a narrow anatomical region . Adeno-associated viruses ( AAVs ) generally represent the preferred viral system for Cre-dependent transgene expression , given their minimal toxicity and the speed and ease of their generation . Developing AAVs sensitive to multiple recombinases has been challenging because of the limited AAV genome size , which precludes the use of multiple recombinase-sensitive STOP cassettes ( Wu et al . , 2010 ) . The INTRSECT system overcomes this limitation by utilizing a single AAV vector that flanks the transgene coding sequence with lox and FRT sites in such a way that combinatorial expression of Cre and Flp permits expression of a functional pre-mRNA that can then be spliced to produce a mature coding sequence ( Fenno et al . , 2020; Fenno et al . , 2014 ) . The multiple inversion and splicing steps involved in this system can limit transgene expression , however , perhaps due to the relatively poor recombinase activity of Flp ( and even optimized FlpO ) . Furthermore , generating INTRSECT viruses that express new transgenes requires a relatively complex and labor-intensive design and optimization process ( Fenno et al . , 2020; Fenno et al . , 2017 ) . Seeking a dual recombinase-activated AAV system to overcome these limitations and that could be modified quickly and easily , we generated tetracycline transactivator ( tTA ) -driven Recombinase-Guided Intersectional Targeting ( tTARGIT ) AAVs composed of a Flp-dependent tetracycline transactivator ( tTA ) ‘Driver’ AAV and a tetracycline response element ( TRE ) -driven Cre-dependent ‘Payload’ AAV to express the transgene of interest . We applied tTARGIT AAVs to the study of ventromedial hypothalamic LepRb ( VMH; LepRbVMH ) neurons , which modulate metabolic adaptations to obesogenic diets ( Bingham et al . , 2008; Dhillon et al . , 2006 ) but have proven difficult to study directly due to the density and proximity of neighboring LepRb populations . Together with LeprCre and Slc16a7FlpO , tTARGIT AAVs allowed us to overcome these challenges and reveal a specific role for LepRbVMH neurons in suppressing food intake and increasing energy expenditure to promote weight loss . The density of LepRb neurons in the adjacent arcuate nucleus ( ARC ) , dorsomedial hypothalamus , and lateral hypothalamic area complicated our initial attempts to study LepRbVMH neurons using LeprCre mice and Cre-dependent vectors; viruses targeted to the VMH spread to other nearby LeprCre neurons , confounding the interpretation of our results ( data not shown ) . Because Slc17a6 , encoding the vesicular glutamate transporter 2 ( vGLUT2 ) protein , expression is largely restricted to LepRbVMH neurons and absent from most surrounding LepRb cells ( Vong et al . , 2011 ) , we generated a Slc17a6FlpO strain and crossed it with LeprCre and a novel Flp- and Cre-dependent reporter ( Rosa26RCFL-eGFP-L10a ) . We then tested the potential ability of this combination of Cre and Flp alleles to specify LepRbVMH neurons in the mediobasal hypothalamus . Although these reporter mice identified Cre- and Flp-co-expressing LepRbSlc17a6 neurons elsewhere in the brain , within the mediobasal hypothalamus this approach largely limited eGFP expression to VMH cells ( Figure 1—figure supplement 1 ) . We thus sought to use Flp- and Cre-dependent AAVs to target LepRbVMH neurons and omit manipulation of non-VMH LepRbSlc17a6 cells . We injected the INTRSECT AAV system ( Fenno et al . , 2014 ) into the VMH of Slc17a6FlpO;LeprCre mice in an attempt to express channelrhodopsin ( ChR2 ) in LepRbVMH cells . However , this approach resulted in detectable ChR2 expression in one or fewer cells per section ( Figure 1—figure supplement 2 ) . We surmised that while INTRSECT works well in systems with robust Flp and Cre activities , the poor recombinase activity of Flp and the more moderate Flp and Cre expression mediated by Slc17a6FlpO and LeprCre , respectively , might limit INTRSECT-mediated transgene expression in LepRbVMH cells . We therefore set out to develop a more sensitive AAV system to drive Cre+Flp-dependent transgene expression , using as our framework a previously described inducible gene expression system based on recombinase-dependent expression of the tetracycline transactivator ( tTA ) in combination with a TRE-driven transgene-expressing allele ( Chan et al . , 2017; He et al . , 2016 ) . We packaged this system into two viral vectors , hereafter ‘tTA-driven Recombinase-Guided Intersectional Targeting’ ( tTARGIT ) AAVs . Our tTARGIT system utilizes ‘Driver’ and ‘Payload’ AAVs . The Driver ( AAV-hSYN1-fDIO-tTA ) utilizes Flp-dependent Double-Floxed Inverted Open reading frame ( fDIO ) cassette ( Fenno et al . , 2014 ) to Flp-dependently invert tTA , permitting its expression by a human synapsin I ( hSYN1 ) promoter . This virus also contains two tetracycline operators ( TetO ) to drive a positive feedback loop and increase tTA expression ( Chan et al . , 2017 ) . The Payload AAV mediates tTA/TRE-dependent transgene expression following its Cre-mediated inversion into the sense orientation . Hence , only cells that contain both recombinases express the transgene ( Figure 1a ) . Tetracycline inhibits tTA-dependent gene expression ( Das et al . , 2016 ) , so this system mediates constitutive payload expression in target cells in the absence of tetracycline . To test the recombinase dependence of this system , we combined the Flp-dependent Driver AAV and a Payload AAV that permits the tTA/TRE-mediated Cre-dependent expression of a ChR2-TdTomato fusion protein ( ChR2-TdT; AAV-TRE-DIO-ChR2-TdT ) . We co-injected these viruses into the VMH of wild-type mice , mice that expressed either Slc17a6FlpO or LeprCre only , or Slc17a6FlpO;LeprCre mice ( Figure 1b , c ) . We detected no TdT ( DSRed-immunoreactivity [IR] ) in the VMH of wild-type or Slc17a6FlpO animals and minimal expression in LeprCre mice ( Figure 1—figure supplement 3 ) . In contrast , the Driver+Payload combination mediated robust DSRed-IR in the VMH of Slc17a6FlpO;LeprCre mice ( Figure 1c ) ; furthermore , VMH photostimulation in these mice promoted robust colocalization of DSRed- and FOS-IR , consistent with the ability of this system to activate transduced Flp+Cre-expressing cells ( Figure 1—figure supplement 4 ) . We also tested the requirement for both the Driver and Payload viruses in this system by injecting each virus alone , or both viruses together , into the VMH of Slc17a6FlpO;LeprCre animals ( Figure 1d ) . As expected , injecting either virus alone produced minimal or no detectable DSRed-IR , while co-injection of the two viruses yielded robust DSRed-IR ( Figure 1e ) . Thus , robust transgene expression by the tTARGIT system requires injection of both the Driver and Payload AAVs , as well as the presence of both Flp and Cre recombinases . The tTARGIT approach can also be modified from a Flp-ON/Cre-ON system , requiring both Flp and Cre recombinases for payload expression , to a Flp-ON/Cre-OFF system , mediating transgene expression in all Flp-expressing cells that do not contain Cre ( Figure 1—figure supplement 5; Table 1 ) . Placing the payload transgene in the forward orientation within the DIO cassette permits tTA-driven transgene expression in all Flp-expressing cells that do not contain Cre because Cre inverts the Payload transgene into the antisense orientation . We tested this system with a novel Cre-inactivated Payload virus expressing a hM3Dq designer receptor exclusively activated by designer drugs ( DREADD ) -mCherry transgene . We co-injected this Cre-OFF Payload and the Flp-dependent Driver AAV into the VMH of Slc17a6FlpO;LeprCre animals on the Cre-dependent Rosa26LSL-GFP-L10a background ( Figure 1—figure supplement 5b ) . As expected , this modified tTARGIT system drove hM3Dq-mCherry expression almost exclusively in cells that did not express the Cre-dependent GFP ( Figure 1—figure supplement 5c ) . We surmise the few GFP-IR neurons with detectable mCherry , constituting 1–7% of all mCherry cells , might result from the low Cre expression mediated by LeprCre ( Patterson et al . , 2011 ) and predict that the Flp-On , Cre-Off tTARGIT system should demonstrate complete Payload inactivation when used in conjunction with a more robustly expressing Cre allele or require longer incubation periods to fully inactivate the payload transgene . To define the projection targets of LepRbVMH neurons , we developed a payload virus ( AAV-TRE-DIO-GFP-2A-SynmRuby ) that encodes GFP plus a cotranslationally expressed synaptophysin-mRuby transgene ( Figure 2a ) . Co-injection of this tTARGIT Payload AAV and the Driver virus into the VMH of Slc17a6FlpO;LeprCre mice promoted robust VMH-restricted GFP-IR ( Figure 2b ) . mRuby detection ( DSRed-IR ) for this virus was much lower than for GFP , however ( data not shown ) ; thus , we used GFP-IR to detect projections from LepRbVMH cells . Assessing the entire CNS for the presence of GFP-IR revealed terminals in the periaqueductal gray , the arcuate , the periventricular thalamic nucleus , the periventricular hypothalamic nucleus , the bed nucleus of the stria terminalis , and the preoptic area ( Figure 2c , d ) . These are consistent with the known projections of the VMH ( Canteras et al . , 1994; Meek et al . , 2016; Zhang et al . , 2020 ) . Ablating Lepr expression from the VMH ( Nr5a1Cre-mediated ) promotes obesity associated with decreased energy expenditure in high-fat-diet-fed animals , suggesting a specific role for LepRb signaling in LepRbVMH cells in the control of energy balance via the dietary modulation of energy utilization ( Bingham et al . , 2008; Dhillon et al . , 2006 ) . To define the function of LepRbVMH cells , rather than the function of LepRb in these cells , we developed a Payload virus containing an inverted hM3Dq-mCherry transgene . We co-injected the Driver AAV and this AAV-TRE-DIO-hM3Dq-mCherry Payload virus into the VMH of Slc17a6FlpO;LeprCre animals ( LepRbVMH-Dq mice; Figure 3a ) . This approach promoted robust VMH-restricted expression of functional hM3Dq-mCherry as administration of the DREADD activator ( Pei et al . , 2008 ) , clozapine-N-oxide ( CNO ) , stimulated colocalization of FOS with DSRed-IR cells ( Figure 3b–d ) . To determine the potential modulation of energy expenditure , activity , and food intake by LepRbVMH neuron activation , we placed LepRbVMH-Dq animals in metabolic cages and administered either vehicle or CNO twice daily ( Figure 3e–j ) . Compared to vehicle administration , activating LepRbVMH neurons significantly increased 24 hr oxygen consumption ( VO2 ) and energy expenditure , both primarily due to effects during the light phase , despite decreasing ambulatory activity over 24 hr ( primarily due to effects during the dark phase ) ( Figure 3g , h ) . Additionally , the hM3Dq-mediated activation of LepRbVMH neurons also suppressed 24 hr food intake , primarily due to decreased light-phase feeding , revealing a previously unsuspected role for these cells in the suppression of feeding . CNO also decreased the respiratory exchange ratio during the light phase , consistent with the increased metabolism of fat stores due to the combination of increased energy expenditure and decreased food intake . To understand whether effects on brown adipose tissue ( BAT ) might contribute to the increased energy expenditure during LepRbVMH neuron activation , we placed temperature probes in the interscapular space of LepRbVMH-Dq animals to monitor BAT thermogenesis . Compared to controls , CNO significantly increased intrascapular temperatures in LepRbVMH-Dq animals , suggesting LepRbVMH neurons promote energy expenditure at least in part by augmenting BAT thermogenesis ( Figure 3k–l ) . As activating LepRbVMH neurons increased energy expenditure and decreased food intake , we surmised that these neurons should promote weight loss . We thus administered CNO in drinking water to LepRbVMH-Dq mice or LeprCre-only control animals ( lacking any Flp expression ) injected with the tTARGIT hM3Dq for 3 days . During this time , the body weight of LepRbVMH-Dq mice decreased by approximately 10% ( Figure 4a ) , returning to baseline following the cessation of CNO exposure . Importantly , body weight of LeprCre-only controls remained stable . While CNO treatment decreased food consumption ( largely during the second day of treatment ) and water intake ( Figure 4b–d ) , the magnitude and timing of these ingestive effects dictates that neither could account for the decreased body weight mediated by CNO , consistent with the notion that increased energy expenditure mediates the major effect of LepRbVMH cells on body weight . Hence , the use of our dual recombinase-dependent tTARGIT AAV system permitted us to determine that LepRbVMH neuron activation increased energy expenditure and decreased food intake during the inactive phase , suggesting the diurnal control of energy balance by LepRbVMH neurons . The use of sequence-specific DNA recombinases ( Cre , Flp , and others ) in conjunction with recombinase-dependent genetic alleles and viral vectors has revolutionized our ability to manipulate specific circuits and understand the central nervous system . The lack of robust viral systems to manipulate cell populations defined by the expression of multiple genes has impeded the study of more refined neural populations , however , including those identified by single cell RNA-sequencing ( Campbell et al . , 2017 ) . Our tTARGIT AAV system addresses many shortcomings of previous intersectional tools , including limitations to transgene expression and the difficulty of incorporating novel transgenes into the AAV plasmids . As the tTARGIT system is based on the use of dual AAVs , it is possible that the two viruses could compete with each other for the surface receptors required for viral internalization and/or for the cellular machinery required for transgene production , thereby reducing expression efficiency from both AAVs . Based upon our results , and since dual AAVs have been used successfully previously ( Akil et al . , 2019; Xu et al . , 2018; Yang et al . , 2016 ) and transduce the same cells in vivo ( Vardy et al . , 2015 ) , we surmise that if such competition exists with tTARGIT AAVs , it is minor . To further facilitate the study of intersectional neural populations , we developed a suite of Cre-dependent tTARGIT Payload plasmids ( Table 1 ) . While we have not yet tested the function of all of these , our experience with the transgenes that we have tested predicts similarly robust expression of the various Payload AAV transgenes . The limitations of these Payload vectors are likely to mirror those of standard viruses with Cre-dependent transgenes , including the requirement for stoichiometric transduction of/recombination in the cell type of interest to observe the effects of interfering with neuronal function . Additionally , the tTARGIT system displays some detectable payload expression , albeit at low levels and in a small number of cells , in LeprCre-only mice co-injected with Driver and Payload AAVs , as well as in Slc17a6FlpO;LeprCre mice injected with the Payload AAV alone . We surmise this results from the previously noted tTA-independent activity of the TRE ( Das et al . , 2016 ) . We expect this tTA-independent Payload transgene expression to predominately occur in Cre-expressing cells , which have inverted the transgene into the correct orientation downstream of the TRE . In our studies , leaky payload expression had little impact as the expression levels were low and failed to produce a metabolic response to CNO in drinking water . Nevertheless , the possibility of tTA-independent payload expression represents an important consideration for experimental design and requires the inclusion of appropriate controls . In addition to Cre-ON/Flp-ON tTARGIT plasmids , we also generated Cre-inactivated Payload vectors to specifically mark Cre-negative Flp-expressing cells within the injection field . Within our model , we found up to 7% of mCherry-IR cells were co-labeled with the Cre-dependent GFP reporter . We posit that increasing post-injection incubation times prior to study or using alleles with higher Cre expression than observed in LeprCre may further reduce the number of Cre-labeled cells expressing the Cre-OFF tTARGIT payload . Hence , inclusion of appropriate control will be required for studies employing tTARGIT AAVs . While the tTARGIT AAV system as we have built it is designed to constitutively express transgenes in Cre- and Flp-expressing cells without the use of tetracycline-like compounds ( usually doxycycline ) , it should be possible to decrease transgene expression from the tTARGIT AAVs by doxycycline treatment . Indeed , we built a Flp-dependent rtTA Driver virus that is predicted to require doxycycline treatment to mediate strong transgene expression . The rtTA-based system can drive low-level transcription independent of doxycycline ( Zhu et al . , 2001 ) , however , and we have not yet tested this system . The use of tTARGIT AAVs permitted us to target robust transgene expression to LepRbVMH neurons specifically , in isolation from LepRb neurons in adjacent hypothalamic nuclei . While deleting Lepr from the VMH of chow-fed mice or restoring VMH Lepr expression on an otherwise LepRb-deficient background minimally ( if at all ) alters energy balance , knockout mice fail to increase energy expenditure on high-fat chow and become more obese than controls ( Bingham et al . , 2008; Dhillon et al . , 2006; Gonçalves et al . , 2014; Senn et al . , 2019 ) . These studies thus suggest a role for leptin action on LepRbVMH cells in the control of energy expenditure , but were unable to study LepRbVMH cells more broadly . In contrast , our use of the tTARGIT system , together with Slc17a6FlpO;LeprCre mice , identified the projections of LepRbVMH cells , demonstrated their ability to acutely suppress food intake as well as promoting energy expenditure ( identifying BAT thermogenesis as a target for these cells ) , and revealed the diurnal nature of LepRbVMH neuron-mediated control of energy balance . Presumably , the finding that LepRbVMH neuron activation alters food intake and energy expenditure specifically during the light cycle suggests that these neurons may decrease in activity during the inactive/light phase , permitting us to observe the effects of artificial neuron activation during this time . In summary , we have developed a suite of dyad AAV vectors for the study of intersectional neural populations marked by the co-expression of Flp and Cre or marked by expression of Flp in the absence of Cre . This tTARGIT system yields robust dual recombinase-sensitive expression of the desired payload in vivo . With this approach , we defined the neural circuitry and functional capacity of LepRbVMH neurons . These intersectional genetic tools will facilitate the study of a broad range of dual gene-defined cell populations across the central nervous system . Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact , Martin Myers Jr ( mgmyers@med . umich . edu ) . Plasmids generated by this study have been deposited to Addgene . Mice were bred in the Unit for Laboratory Animal Medicine at the University of Michigan . These mice and the procedures performed were approved by the University of Michigan Committee on the Use and Care of Animals and in accordance with Association for the Assessment and Approval of Laboratory Animal Care and National Institutes of Health guidelines . Unless otherwise indicated , mice were provided with ad libitum access to food ( Purina Lab Diet 5001 ) and water in temperature-controlled ( 25°C ) rooms on a 12 hr light–dark cycle with daily health status checks . Rosa26 LSL-eGFP-L10a mice ( Krashes et al . , 2014 ) and LeprCre mice ( Leshan et al . , 2006 ) have been described previously . Both male and female mice were used for all studies . Sample size was determined based on previous experience . Slc17a6FlpO mice were generated using recombineering techniques as previously described ( Balthasar et al . , 2005 ) . Briefly , the FlpO transgene ( Addgene plasmid #13793 ) and a LoxP-flanked neomycin selection cassette were subcloned after an optimized internal ribosome entry sequence ( IRES ) . The IRES-FlpO-neomycin cassette was then targeted 3 bp downstream of the stop codon of Slc17a6 in a bacterial artificial chromosome . The final targeting construct containing the Slc17a6-IRES-Flpo neomycin cassette and 4 kb of flanking genomic sequence on both sides was electroporated into ES cells followed by neomycin selection . Appropriately targeted clones were identified by quantitative PCR and confirmed by southern blot analysis . Targeted clones were expanded and injected into blastocysts by the University of Michigan Transgenic Core . Chimeric offspring were then bred to confirm germline transmission of the Slc17a6-IRES-Flpo allele; the neomycin selection cassette was removed by breeding to the E2A-Cre deleter strain ( Jax stock #003724 ) . The targeting vector was developed by the Allen Brain Institute and obtained from AddGene ( plasmid #61577 ) . The neomycin resistance cassette and tdTomato sequence were removed and replaced with the eGFP:L10a coding sequences . The plasmid was then microinjected by the University of Michigan Transgenic Core into fertilized oocytes with Cas9 protein and gRNAs targeting the Rosa26 locus ( actccagtctttctagaaga ) . Tail DNA from the resulting pups was screened with PCR for the presence and proper insertion of the targeting vector . The Flp-dependent Rosa26RFL-GFP-L10a mouse was generated by germline deletion of the lox-stop-lox cassette . Mice were anesthetized with isoflurane ( 2% ) and mounted in a stereotaxic frame ( Kopf ) . Using standard surgical techniques , 150 nL of virus was injected bilaterally via a glass micropipette attached to a microinjector ( picospritzer II ) targeting the VMH ( Anterior/Posterior −1 . 3 mm; Medial/Lateral ±0 . 25 mm , Dorsal/Ventral −5 . 55 mm , relative to bregma ) . For DREADD studies , hit sites were verified by mCherry detection ( DSRed-IR ) following euthanasia . Any data from mice in which mCherry was not detected within the VMH or was detected in other hypothalamic nuclei were discarded . Data from mice with either unilateral or bilateral viral hits were included . To generate the AAV-hSyn1-TetOx2-fDIO-tTA , the tTA transgene was placed within a fDIO cassette ( Addgene plasmid #55641 , a gift from Karl Deisseroth ) . The tTA sequence was then removed from pAAV-ihSyn1-tTA ( Addgene plasmid #99120 , a gift from Viviana Gradinaru ) and replaced with the fDIO-tTA sequence . Similarly , AAV-hSyn1TetOx2-fDIO-rtTA was generated by first placing the rtTA sequence ( using rtTA sequence from Addgene plasmid #102423 , a gift from Kian Peng Koh ) in a fDIO cassette ( Addgene plasmid #55641 , a gift from Karl Deisseroth ) . The tTA sequence was then removed from pAAV-ihSyn1-tTA ( Addgene plasmid #99120 , a gift from Viviana Gradinaru ) and replaced with the fDIO-rtTA sequence . To generate the payload viruses , the GFP cassette was removed from AAV-TRE-DIO-GFP ( Addgene plasmid #65449 , a gift from Hongkui Zeng ) and replaced with ChR2-TdTomato ( Addgene plasmid #18917 , a gift from Scott Sternson ) , hM3Dq-mCherry ( Addgene plasmid #44361 , a gift from Bryan Roth ) , GFP-2A-SynmRuby ( Addgene plasmid #71760 , a gift from Liqun Luo ) , HA-Cas9 ( Addgene plasmid #61592 , a gift from Feng Zhang ) , eGFP-L10a ( provided by DPO [Allison et al . , 2015] ) , Caspase3-2A-TEVp ( Addgene plasmid #45580 , a gift from Nirao Shah ) , GCaMP6s ( Addgene plasmid #100845 , a gift from Douglas Kim ) , SwiChRca-TS-YFP ( Addgene plasmid #55631 , a gift from Karl Deisseroth ) , or TVA-mcherry+oG ( a gift from Marco Tripodi [Ciabatti et al . , 2017] ) . For control experiments presented in Figure 1 , mice were euthanized 3 weeks following viral delivery . Hit sites were verified using a marker virus ( AAV-CMV-Cas9-HA ( 18 ) ) . Upon the completion of DREADD studies , mice were injected with CNO ( 1 mg/kg ) , sacrificed 2 h post-injection , and then perfused with 10% formalin . Brains were then removed and post-fixed in 10% formalin for 24 hr , before being moved to 30% sucrose for 24 hr . Brains were then sectioned as 30 μm thick free-floating sections . Immunohistochemical and immunofluorescent staining was performed using standard procedures using anti-FOS ( 1:1000 , #2250 , Cell Signaling Technology ) , GFP ( 1:1000 , #1020 , Aves Laboratories ) , and DSRed ( 1:1000 , #632392 , Clontech ) antibodies . Images were collected on an Olympus BX51 microscope . Mice were singly housed 1 week prior to indirect calorimetry studies . Mice were placed into metabolic ( CLAMS ) cages in the University of Michigan Mouse Metabolic Phenotyping Center ( UM-MMPC ) and further equilibrated for 24 hr . Subsequently , mice were then injected twice daily ( 930 AM and 5 PM ) with vehicle for 2 days followed by an additional 2 days of twice daily ( 930 AM and 5 PM ) CNO ( 1 mg/kg ) . Data is presented as the average of the two saline days compared to the average of the two CNO days . Mice were singly housed for 1 week prior to study . Mice were given ad libitum access to standard drinking water for 48 hr . For the subsequent 72 hr , standard water was replaced by water containing CNO ( 2 . 5 mg/100 mL ) and 1% glucose ( to make the CNO palatable ) . Vehicle-treated mice received 1% glucose-containing drinking water ( lacking CNO ) on CNO treatment days . CNO-laced water was changed daily . For the final 48 hr , mice were returned to standard drinking water . Body weight , food mass , and water levels were recorded daily . Virus controls for this study were Slc17a6WT;LeprCre mice co-injected with AAV-hSYN1-fDIO-tTA and AAV-TRE-DIO-hM3Dq-mCherry . The UM-MMPC placed temperature transponders ( IPTT-300 model with corresponding DAS-7007R reader , Bio Medic Data Systems ) in the intrascapular subcutaneous tissue directly under the conjunction part of the butterfly-shaped BAT under isoflurane anesthesia . Mice were allowed to recover for 14 d before testing . One day prior to testing , ambient temperatures were increased from 22°C to 30°C . On the day of testing , mice were randomized to either CNO ( 1 mg/kg ) or saline injections , and temperatures were recorded at −10 , 10 , 20 , 30 , 45 , 60 , and 120 min relative to injection time . Following 1 week , the experiment was repeated and treatment conditions ( vehicle or CNO ) reversed . A single fiber-optic cannula ( Doric Lenses ) was implanted above the VMH ( A/P: 1 . 3 mm , M/L: 0 . 25 mm , D/V: 5 . 0 mm ) and affixed to the skull using Metabond ( Fisher ) . After 3 weeks recovery from surgery , mice were then subjected to optical stimulation using 473 nm wavelength laser using 20 mW/mm2 irradiance . Light pulses were delivered by 1 s of 20 Hz photo stimulation and 3 s resting with multiple repetitions for 1 hr . The INTERSECT pAAV-nEF Con/Fon hChR2 ( H134R ) -EYFP plasmid ( Fenno et al . , 2014 ) was procured through Addgene ( plasmid #55644 ) . All rAAV viruses were made at the University of Michigan Vector Core using ultracentrifugation through an iodixanol gradient . rAAVs were washed three times with PBS using Amicon Ultra Centrifugal Filter Units ( Millipore ) and resuspended in PBS + 0 . 001% Pluronic F68 . All viruses were packaged as AAV8 serotypes . Titers were assessed by qPCR . VirusTiter ( vg/mL ) AAV8-hSYN1-fDIO-tTA2 . 60E+13 AAV8-TRE-DIO-hM3Dq-mCherry2 . 05E+13 AAV8-TRE-DIO-ChR2-TdTomato5 . 83E+13 AAV8-TRE-DIO-GFP-2A-SynmRuby5 . 69E+13 AAV8-TRE-DIO-hM3Dq-mCherry ( Cre-OFF ) 6 . 86E+13 AAV8-nEF Con/Fon hChR2 ( H134R ) -EYFP4 . 75E+13 All data is displayed as mean ± SEM . Replicate number is included in each figure legend . Statistical analysis was performed in either Graphpad Prism eight using either t-tests or ANOVAs with Dunnet’s post hoc test or linear mixed model . p<0 . 05 was considered significant .	The brain contains hundreds of types of neurons , which differ in size , shape and behavior . But neuroscientists often wish to study individual neuronal types in isolation . They are able to do this with the aid of a toolkit made up of two parts: viral vectors and genetically modified mice . Viral vectors are viruses that have been modified so that they are no longer harmful and can instead be used to introduce genetic material into cells on demand . To create a viral vector , the virus’ own genetic material is replaced with a ‘cargo’ gene , such as the gene for a fluorescent protein . The virus is then introduced into a new host such as a mouse . Importantly , the virus only produces the protein encoded by its ‘cargo’ gene if it is inside a cell that also contains one of two specific enzymes . These enzymes are called Cre and Flp . This is where the second part of the toolkit comes in . Mice can be genetically engineered to produce either Cre or Flp exclusively in specific cell types . By introducing a viral vector into mice that produce either Cre or Flp only in one particular type of neuron , researchers can limit the activity of the cargo gene to that neuronal type . But sometimes even this approach is not selective enough . Researchers may wish to limit the activity of the cargo gene to a subpopulation of cells that produce Cre or Flp . Or they may wish to target only Cre- or Flp-producing cells in a small area of the brain , while leaving cells in neighboring areas unaffected . Sabatini et al . have now overcome this limitation by developing and testing a new set of viral vectors that are active only in neurons that produce both Cre and Flp . The vectors are called tTARGIT AAVs and allow researchers to target cells more precisely than was possible with the previous version of the toolkit . Sabatini et al . show tTARGIT AAVs in action by using them to identify a group of neurons that control how much energy mice use and how much food they eat . As well as applying the vectors to their own research on obesity , Sabatini et al . have also made them freely available for other researchers to use in their own projects .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "medicine", "tools", "and", "resources", "neuroscience" ]	2021	tTARGIT AAVs mediate the sensitive and flexible manipulation of intersectional neuronal populations in mice
Social interactions are often powerful drivers of learning . In female mice , mating creates a long-lasting sensory memory for the pheromones of the stud male that alters neuroendocrine responses to his chemosignals for many weeks . The cellular and synaptic correlates of pheromonal learning , however , remain unclear . We examined local circuit changes in the accessory olfactory bulb ( AOB ) using targeted ex vivo recordings of mating-activated neurons tagged with a fluorescent reporter . Imprinting led to striking plasticity in the intrinsic membrane excitability of projection neurons ( mitral cells , MCs ) that dramatically curtailed their responsiveness , suggesting a novel cellular substrate for pheromonal learning . Plasticity was selectively expressed in the MC ensembles activated by the stud male , consistent with formation of memories for specific individuals . Finally , MC excitability gained atypical activity-dependence whose slow dynamics strongly attenuated firing on timescales of several minutes . This unusual form of AOB plasticity may act to filter sustained or repetitive sensory signals . Chemical cues detected by the vomeronasal system convey vital social information , influencing diverse behaviors such as reproduction ( Bruce and Parrott , 1960; Kimchi et al . , 2007 ) , pair bonding ( Young and Wang , 2004 ) , parental care ( Dulac et al . , 2014; Kendrick et al . , 1992; Lévy et al . , 2004 ) , individual recognition ( Hurst , 2009 ) , and aggression ( Chamero et al . , 2007; Stowers et al . , 2002 ) . Vomeronasal pathways directly access the limbic system , consistent with their powerful role in guiding behavior , and also influence neuroendocrine centers to modify physiological and hormonal status ( Dulac and Torello , 2003; Tirindelli et al . , 2009 ) . While vomeronasal circuits often elicit stereotyped behavioral and neuroendocrine responses , they can also be highly plastic . In one striking example , female mice imprint on the pheromones of the stud male after mating , where a single salient sensory experience drives long-term changes in both behavior and the flow of sensory information to central targets ( Keverne and Brennan , 1996 ) . During the first few days after fertilization , chemosignals from unfamiliar males typically block pregnancy by altering the female’s neuroendocrine state ( Bruce and Parrott , 1960 ) . However , mating opens a plasticity window that creates a recognition memory for the stud's pheromones , so that they lose their potency and no longer disrupt embryo implantation ( Brennan and Keverne , 1997 ) . Memories are formed within hours , yet last weeks or longer ( Kaba and Keverne , 1988 ) . Sensory imprinting thus offers an opportunity to test the neural basis of a natural form of social learning in a circuit intimately coupled with intraspecies behaviors . While social experience acts on diverse neural circuits throughout the brain ( Wallace et al . , 2009; Wu et al . , 2014 ) , mating-dependent learning is strongly linked to plasticity in the accessory olfactory bulb ( AOB ) . Imprinting in females leads to local neurochemical changes ( Brennan et al . , 1995 ) , is affected by local lesions or pharmacological interventions ( Brennan and Keverne , 1997; Kaba et al . , 1994; Kaba and Keverne , 1988 ) , and can be artificially induced by manipulating AOB signaling ( Kaba et al . , 1994 ) . Pheromonal cues are encoded in AOB by the firing of mitral cells ( MCs ) , whose activity signals gender , hormonal status , and in particular , strain and/or individual identity ( Ben-Shaul et al . , 2010; Luo et al . , 2003; Tolokh et al . , 2013 ) . Well-established theories propose that learning selectively suppresses the firing of the MCs encoding the stud male so that his pheromones no longer drive neuroendocrine responses , precluding pregnancy block ( Brennan , 2004; Brennan et al . , 1990 ) . MC suppression is further proposed to depend on strengthening of local inhibitory circuits in AOB , consisting largely of granule cells ( GCs ) that supply feedback inhibition to MCs through unique dendrodendritic synapses ( Isaacson and Strowbridge , 1998; Shepherd and Greer , 1998 ) . Inhibitory plasticity is consistent with both microdialysis data ( Brennan and Binns , 2005; Brennan et al . , 1995 ) and ultrastructural changes in local interneurons ( Matsuoka et al . , 1997 , 2004 ) . Despite well-established models of learning in AOB , many key features of plasticity remain untested . To date , direct measurements of either synaptic plasticity or changes in MC output are lacking . Furthermore , while the selectivity of recognition memories for different individual or strains is thought to rely on changes in specific groups of MC , there are no data linking plasticity to functionally defined cell populations . More broadly , the nature of the neural changes that allow for adaptive changes in social behavior remain poorly understood . Here , we examined how mating affects local AOB microcircuits using targeted whole-cell recordings of identified neurons activated by the stud male in ex vivo brain slices . We found pronounced reductions in the sensitivity of AOB neurons , which unexpectedly were mediated by changes in intrinsic excitability rather than synaptic strength , suggesting a novel cellular basis for encoding sensory memories in AOB . MC firing was selectively attenuated in stud-activated neurons , suggesting a potential basis for the specificity of pheromonal learning . Changes in MC responsiveness emerged only when they were activated with repetitive patterns , suggesting that after learning the AOB may dynamically filter repetitive sensory signals from the stud male , lessening their impact on neuroendocrine status on long timescales . While inhibitory circuits are intensively studied in main olfactory bulb ( Isaacson and Strowbridge , 1998; Shepherd and Greer , 1998 ) , their role in shaping AOB output is not well characterized . We thus began by characterizing self-inhibition in AOB projection neurons , mitral cells ( MCs ) . MC self-inhibition arises from specialized dendrodendritic synapses shared with local interneurons , primarily granule cells ( GCs ) ( Figure 1A ) . We assessed self-inhibition by driving MC firing with current injection and examining the resulting synaptic feedback from interneurons . Brief , high-frequency spike trains generated only modest feedback inhibition in AOB ( Figure 1B–C ) , comparable but smaller on average than reported in main olfactory bulb ( Abraham et al . , 2010; Margrie et al . , 2001 ) . However , pronounced inhibition appeared when we drove MCs with prolonged stimuli similar in duration to chemosensory responses , which can last for tens of seconds ( Luo et al . , 2003 ) . Extended MC spike trains elicited slowly emerging but robust barrages of inhibitory postsynaptic potentials ( IPSPs ) that contributed to a strong decline in firing rate ( Figure 1D–1F; Figure 1—source data 1 ) . In contrast , the same protocol generated little detectable inhibition in MCs of the main olfactory bulb , where high firing rates were sustained throughout the train . To confirm that synaptic self-inhibition shapes MC output , we eliminated dendrodendritic feedback pathways by blocking fast synaptic transmission with NBQX , APV , and gabazine ( 5 , 25 , and 10 µM respectively ) . Pharmacologically eliminating feedback inhibition typically increased MC firing as well ( Figure 1—figure supplement 1 ) . Together , our results suggest that MC self-inhibition is substantially stronger in AOB than in main olfactory bulb , but also unusually slow to manifest , consistent with the prolonged sensory responses characteristic of this brain area . Such powerful self-inhibition by single MCs could potentially provide a basis for cell-specific control over AOB output , as previously proposed ( Brennan and Keverne , 1997 ) . 10 . 7554/eLife . 25421 . 003Figure 1 . AOB MCs express robust , slowly emerging self-inhibition . ( A ) Schematic of dendrodendritic self-inhibition pathway in MCs . ( B ) Left , dye-filled MC imaged after recording . S , soma; d , dendritic tufts that integrate sensory inputs; p , recording pipette . Right , brief , high-frequency spike trains trigger modest self-inhibition ( gray , standard ACSF; black , after blocking inhibition with 15 µM BMI; IPSP , inhibitory postsynaptic potential ) . ( C ) Pharmacologically isolated self-inhibition in AOB MCs . Colored traces show individual cells; black trace , average; mean Vinh = −1 . 4 ± 0 . 27 mV ( n = 8 cells in 5 mice ) . ( D , E ) Prolonged firing elicits robust MC self-inhibition in AOB but not MOB ( black and purple respectively ) . Boxes show expanded view of barrages of IPSPs in AOB MCs , indicated by arrowheads , which only emerge after several seconds of firing . ( F ) Self-inhibition contributes to stronger decay of MC firing rates in AOB during extended stimuli ( n = 9 and 9 cells in 5 and 5 mice for AOB and MOB respectively ) . ( G ) Initial and final firing rates during MC spike trains in MOB and AOB ( purple and gray respectively ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 00310 . 7554/eLife . 25421 . 004Figure 1—source data 1 . This spreadsheet contains the initial and final firing rates for the individual neurons shown in Figure 1G . These data can be opened with Microsoft Excel or with open-source alternatives such as OpenOffice . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 00410 . 7554/eLife . 25421 . 005Figure 1—figure supplement 1 . Robust self-inhibition regulates spiking of AOB MCs . ( A ) Eliminating self-inhibition by blocking fast synaptic transmission increased the overall firing rate of MCs . ( B ) Blocking synaptic inhibition also eliminated the barrages of IPSPs that emerged later in train and persisted after the offset of spiking . ( C ) Firing of AOB MCs is consistently enhanced after blocking inhibition ( p=0 . 06; t-test; n = 6 MCs in 6 mice ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 005 Microdialysis suggests increased bulk GABA release in AOB after mating ( Brennan and Binns , 2005 ) , consistent with enhanced inhibition , but the synaptic correlates of imprinting have not been measured directly . We next examined the effects of pheromonal learning on local inhibitory circuits . To align the timing of mating with brain slice recordings ( Figure 2A ) , we induced estrus using ovariectomy , implanted estradiol capsules , and progesterone injection ( Ström et al . , 2012 ) . We then paired females in their home cage with sexually experienced males for 4 hr to provide the mating and sensory exposure required for imprinting . Females engaged in frequent , repetitive investigation of males , particularly of facial and anogenital regions ( Figure 2—figure supplement 1A–C; mean interval , 62 . 1 ± 7 . 99 s; median , 15 . 9 s ) . Investigative behavior was elevated in mated relative to sensory-exposed females , suggesting that they experienced both heightened arousal states and increased levels of sensory input during the pairing period ( Figure 2—figure supplement 1D ) . 10 . 7554/eLife . 25421 . 006Figure 2 . Imprinting drives synaptic plasticity in both MCs and GCs . ( A ) Schematic of timeline for mating , sensory experience , and recording . ( B ) Inhibitory synaptic inputs recorded in voltage-clamped MCs from naïve , sensory-exposed , and mated mice . ( C , D ) Mating substantially increases mIPSC frequency . Left , cumulative interval distributions; mated < naïve and exposed groups , p=0 . 002 and 0 . 001 respectively . Right , mean frequency ( F = 5 . 88; Fc = 3 . 20; p=0 . 005 for mated vs . naïve; ANOVA with post-hoc Tukey test; n = 18 , 17 , and 15 cells in 5 , 5 , and 6 mice respectively ) . ( E , F ) The mean amplitude of mIPSCs was not significantly changed by imprinting ( F = 1 . 74; Fc = 3 . 20; p=0 . 19; ANOVA with post-hoc Tukey test ) , although distributions were significantly shifted towards smaller values in the mated vs . naïve and sensory-exposed groups ( p=0 . 00007 and 3 × 10−7 respectively . ( G ) Example traces showing spontaneous EPSPs in GCs from naïve , sensory-exposed and mated mice . Rasters indicate synaptic events used for analysis . ( H , I ) Mating increased mean sEPSP frequency relative to both naïve and sensory-exposed animals ( F = 6 . 64; Fc = 3 . 14; p=0 . 00037 and 0 . 038 for mated vs . exposed and naïve mice respectively; ANOVA with post hoc Tukey test; n = 17 , 19 , and 30 cells in 5 , 9 , and 12 mice ) . Interval distributions were significantly smaller for mated vs . exposed and naïve animals ( p=1×10−11 and 0 . 0008 respectively ) . ( J , K ) Mating also increased mean sEPSP amplitude in mated vs . naïve animals . Left , cumulative distribution; right , mean amplitude ( F = 3 . 56; Fc = 3 . 14; p=0 . 037 for naïve vs . mated , ANOVA with post hoc Tukey test ) . Amplitude distributions were larger for mated vs . naïve mice ( p=0 . 04 ) . NS , not significant; p<0 . 05; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 00610 . 7554/eLife . 25421 . 007Figure 2—source data 1 . This spreadsheet contains the mean frequency and amplitude data for the individual neurons used to generate the bar plots shown in Figure 2D and F ( mitral cell mIPSCs ) and 2I and 2K ( granule cell mEPSCs ) . These data can be opened with Microsoft Excel or with open-source alternatives such as OpenOffice . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 00710 . 7554/eLife . 25421 . 008Figure 2—figure supplement 1 . Mating and sensory interactions during pairing . ( A ) Video analysis of characteristic post-mating behavior between females and males . Top , mating typically occurred early during the pairing period ( red ) , followed by extensive and repetitive behavioral encounters including direct nasal contact required for vomeronasal activation ( yellow ) . Bottom , expanded view of approximately 10 min of female-male interactions . ( B ) Distribution of intervals between investigatory bouts in the mated group ( mean , 62 . 1 ± 7 . 99 s; median , 15 . 9 s; n = 554 bouts in 7 animals ) . ( C ) Investigatory behavior is elevated in mated relative to sensory-exposed groups ( mean bouts per female , 89 . 5 ± 32 . 1 vs . 38 ± 11 . 6 respectively; n = 7 animals per group ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 00810 . 7554/eLife . 25421 . 009Figure 2—figure supplement 2 . Pharmacologically isolated inhibitory synaptic currents in MCs . Top , inward currents recorded from voltage-clamped MCs using high-chloride pipette solution , measured in the presence of TTX , NBQX , and APV . Bottom , miniature currents were completely blocked by addition of 15 µM bicuculline . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 00910 . 7554/eLife . 25421 . 010Figure 2—figure supplement 3 . Synaptic effects in GCs are independent of event detection criteria . ( A ) Raw data traces with rasters showing synaptic events detected using thresholds of 0 . 25 , 0 . 45 , and 0 . 65 mV . ( B ) Mean frequency ( left ) and cumulative distribution of interevent interval ( right ) for excitatory synaptic input using a detection threshold of 0 . 45 mV . Mean frequency , 2 . 47 ± 0 . 53 , 2 . 43 ± 0 . 53 , and 4 . 69 ± 0 . 55 Hz; F = 5 . 93; Fc = 3 . 14; p=0 . 012 and 0 . 038 for mated vs exposed and naïve groups respectively . ( C ) Mean and cumulative distribution of EPSP amplitudes for the same criteria . Mean amplitudes , 1 . 42 ± 0 . 08 , 1 . 62 ± 0 . 11 , and 1 . 85 ± 0 . 08 mV for naïve , exposed , and mated groups respectively . F = 5 . 58; Fc = 3 . 14; p=0 . 0047 for mated vs . naïve groups . ( D , E ) Corresponding plots using a threshold of 0 . 65 mV . Mean frequency , 1 . 84 ± 0 . 51 , 1 . 80 ± 0 . 42 , and 3 . 38 ± 0 . 46 Hz; F = 3 . 89; Fc = 3 . 14; p=0 . 048 for mated vs naïve groups . Mean amplitude , 1 . 70 ± 0 . 08 , 1 . 86 ± 0 . 11 , and 2 . 29 ± 0 . 09 mV for naïve , exposed , and mated groups respectively . F = 10 . 52; Fc = 3 . 14; p=0 . 0021 and 0 . 0062 for mated vs . naïve and exposed groups respectively . p<0 . 05; *p<0 . 005; all p values calculated using ANOVA with post-hoc Tukey test . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 010 Immediately following mating and sensory exposure , we prepared AOB brain slices from females and examined changes in synaptic inhibition with whole-cell voltage-clamp recordings of MCs . We compared three groups: ( i ) mating plus sensory experience with a freely moving male; ( ii ) sensory-exposed controls without mating; and ( iii ) naïve mice with no prior male exposure , housed overnight in a fresh cage . We measured GABAergic input onto MCs by recording miniature inhibitory postsynaptic currents ( mIPSCs; Figure 2B ) . IPSCs were pharmacologically isolated with 5 µM NBQX , 25 µM APV , and 1 µM TTX ( Figure 2—figure supplement 2 ) . The frequency of mIPSCs was strongly increased in mated animals vs . naïve and sensory-exposed groups ( Figure 2C–D; 2 . 45 ± 0 . 37 Hz vs . 1 . 48 ± 0 . 21 Hz and 1 . 28 ± 0 . 17 Hz respectively ) . The mean amplitude of mIPSCs was similar for all three conditions , suggesting little change in the postsynaptic sensitivity of inhibitory synapses onto MCs ( Figure 2E–F; 56 . 8 ± 3 . 5 pA , 54 . 5 ± 3 . 8 pA , and 65 . 8 ± 6 . 1 pA for naïve , sensory-exposed and mated mice respectively; Figure 2—source data 1 ) . The distribution of amplitudes was shifted towards higher values , however , indicating that a subset of inhibitory synapses may be strengthened . Overall , mating experience substantially increased inhibitory input onto MCs , consistent with prior microdialysis results ( Brennan et al . , 1995 ) . Because imprinting may act on other elements of the self-inhibition pathway , we also asked whether mating alters excitatory input to granule cells , the most numerous interneurons in AOB . Using current clamp recordings from GCs , we measured the amplitude and frequency of spontaneous excitatory postsynaptic potentials ( sEPSPs; Figure 2G ) . The frequency of sEPSPs was elevated in mated relative to naïve and sensory-exposed females ( Figure 2H–I; 4 . 25 ± 0 . 71 , 3 . 57 ± 0 . 56 , and 6 . 61 ± 0 . 63 Hz for naïve , exposure and mated groups respectively ) . EPSP amplitude was also slightly enhanced in mated compared to naïve animals ( Figure 2J–K , 1 . 08 ± 0 . 08 , 1 . 18 ± 0 . 10 , and 1 . 44 ± 0 . 10 mV for naïve , exposed , and mated groups respectively ) . These differences were consistent across a wide range of EPSP detection criteria ( Figure 2—figure supplement 3 ) . Overall , learning also increased excitatory drive onto GCs , enhancing both presynaptic and postsynaptic elements of glutamatergic synapses . Together , our data suggest that imprinting upregulates both the excitatory and inhibitory components of the pathways for MC self-inhibition . Pheromonal recognition memories are specific to particular individuals or strains , implying that learning may act selectively on the particular AOB neurons activated by the stud’s chemosignals ( Keverne and Brennan , 1996 ) . To test the cellular specificity of plasticity , we identified the AOB neurons activated during mating and sensory exposure using GFP reporter lines based on the promoters for the immediate-early genes Arc and Fos ( Reijmers et al . , 2007; Wang et al . , 2006 ) . We then used fluorescence-guided recordings ( Barth , 2007 ) to evaluate cellular and synaptic changes specifically in the neural population activated during mating . We focused first on interneurons , which were robustly labeled in Arc-GFP animals . Prior to recording , we assessed GFP labeling across the GC population with 2-photon microscopy . Naïve animals showed low levels of background GFP expression ( Figure 3A ) , and 4 hr of sensory exposure to a male in the absence of mating produced little additional labeling over background ( Figure 3—figure supplement 1A ) . However , mating combined with subsequent sensory exposure drove strong GFP expression in a subset of GCs , consistent with Arc immunolabeling in AOB in response to conspecifics ( Halem et al . , 2001; Matsuoka et al . , 2002 ) . We found robust labeling in both anterior and posterior AOB , in agreement with prior reports using histochemical staining ( Brennan et al . , 1992; Halem et al . , 2001 ) . Fluorescent activity reporters therefore identify mating-activated neural populations in live AOB tissue for targeted ex vivo electrophysiological measurements . 10 . 7554/eLife . 25421 . 011Figure 3 . Synaptic plasticity is uncorrelated with activation during mating . ( A ) Arc-GFP labeling of AOB GCs activated by the stud male , visualized with live-tissue 2-photon imaging . Left , naïve control animal; right , mated female . ( B ) Fluorescence-targeted recordings of both unlabeled and labeled populations of GCs . ( C , D ) Mean amplitude and frequency of sEPSPs are similar for GFP ( - ) and GFP ( + ) GCs in mated mice ( amplitude: p=0 . 82; frequency , p=0 . 58; t-test; n = 9 and 13 cells in 10 mice , GFP ( + ) and ( - ) groups subdivided from data in Figure 3—figure supplement 1 ) . ( E , F ) GFP labeling is uncorrelated with either amplitude or frequency of spontaneous excitatory input to GCs ( regression slope not different from zero; amplitude: p=0 . 70 and 0 . 22 for sensory-exposed and mated groups respectively; frequency: p=0 . 50 and 0 . 92; linear regression t-test; n = 17 , 19 and 30 neurons in 5 , 9 , and 12 mice for naïve , exposure and mated groups ) . ( G ) Fos-GFP labeling reveals a subpopulation of mating-activated MCs ( arrowheads ) . ( H ) Targeted recordings of stud-activated MCs . ( I , J ) Mean amplitude and frequency of mIPSCs are not significantly different between GFP ( - ) and GFP ( + ) MC populations ( p=0 . 33 and 0 . 38 respectively; t-test; n = 8 and 5 cells in 5 mice; groups subdivided from data in Figure 2 ) . ( K , L ) Amplitude and frequency of mIPSCs show no correlation with Fos-GFP intensity in MCs ( regression slope not different from zero; p=0 . 64 and 0 . 97 respectively; linear regression t-test; n = 16 neurons from 5 mice ) . Dashed lines show 95% confidence intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 01110 . 7554/eLife . 25421 . 012Figure 3—source data 1 . This spreadsheet contains the mean frequency and amplitude data for the individual neurons used to generate the bar plots shown in Figures 3C , D , I and J , comparing synaptic inputs to GFP ( - ) and GFP ( + ) neurons . These data can be opened with Microsoft Excel or with open-source alternatives such as OpenOffice . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 01210 . 7554/eLife . 25421 . 013Figure 3—figure supplement 1 . Mating increases fluorescent labeling in AOB and increases inhibitory synaptic input onto MCs . ( A ) Cumulative histogram of Arc-GFP intensity . Mating , but not sensory exposure alone , increases labeling in GCs ( p<10−45 , Kolmogorov–Smirnov test; n = 14 , 9 , and 18 slices for naïve , sensory-exposed , and mated groups respectively ) . ( B ) Mating produces robust increases in Fos-GFP labeling of MCs ( p=0 . 0004 , Kolmogorov–Smirnov test; n = 8 and 7 slices from 4 and 5 mice for naïve and mated groups respectively ) . ( C ) Representative synaptic currents measured from naïve and mated females . Currents are inward due to a high chloride recording solution . ( D ) Cumulative distribution of event amplitude ( p<0 . 001; Kolmogorov–Smirnov test; n = 6 and 16 cells in 3 and 5 mice ) . ( E ) The mean amplitude of inhibitory currents is unchanged in MCs from naïve and mated ( red ) animals ( p=0 . 29; t-test ) . ( F , G ) Mating increased the frequency of inhibitory input to MCs , reflected in both the leftward shift in the distribution of event intervals and increase in mean frequency ( p=0 . 001; t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 013 Using fluorescence-guided recordings of GFP-labeled GCs ( Figure 3B ) , we asked whether the synaptic plasticity generated by mating was specific to these neurons . Unexpectedly , there was no difference between GFP ( - ) and GFP ( + ) populations of GCs for either amplitude or frequency of spontaneous excitatory input ( Figure 3C–D; amplitude , 1 . 43 ± 0 . 15 vs 1 . 48 ± 0 . 13 mV; frequency , 7 . 13 ± 0 . 98 vs . 6 . 41 ± 0 . 84 Hz for unlabeled and labeled cells respectively; Figure 3—source data 1 ) . Similarly , there was no significant relationship between the intensity of GFP expression and either amplitude or frequency of EPSPs ( Figure 3E–F ) . These data suggest that mating globally increased synaptic drive onto inhibitory GCs without apparent specificity to the neurons activated by the stud male . AOB output is relayed to behavioral and neuroendocrine centers by MCs , suggesting that memory specificity ultimately relies on changes in these neurons . Because MCs were only weakly labeled by Arc-GFP , we used an alternative Fos-GFP reporter line ( Reijmers and Mayford , 2009 ) . Fos-GFP levels were low in MCs from naïve females , but were robustly elevated in a subset of MCs after mating ( Figure 3G; Figure 3—figure supplement 1B ) . Fos-GFP also provided more extensive labeling of GCs , suggesting that it captured similar sets of activated neurons , but at a lower threshold . As in Arc-GFP animals , we found no systematic differences in MC labeling in anterior vs . posterior AOB . In mated females , approximately 28% of detected MCs were classified as GFP ( + ) ( intensity 4X greater than neuropil ) , but this is likely an overestimate due to difficulty in detecting unlabeled cells with live tissue imaging . To correlate inhibitory plasticity with activation of MCs by the stud male , we repeated our measurements of GABAergic input using fluorescence-guided recordings ( Figure 3H ) . This second dataset revealed a similar two-fold increase in mIPSC frequency in mated vs . naïve animals , with no change in amplitude ( Figure 3—figure supplement 1C–H; frequency , 0 . 86 ± 0 . 14 vs . 1 . 68 ± 0 . 17 Hz; amplitude , 58 . 02 ± 5 . 59 vs . 63 . 21 ± 5 . 53 pA for naïve vs . mated respectively ) . We evaluated the cellular specificity of synaptic changes within mated females by subdividing the second MC dataset into stud-activated GFP ( + ) neurons and a corresponding GFP ( - ) population . Similar to interneuron results , there was no significant difference in mean amplitude and frequency of mIPSCs between GFP ( - ) and GFP ( + ) MCs ( Figure 3I and J; amplitude , 65 . 98 ± 8 . 15 vs . 57 . 86 ± 3 . 15 pA for unlabeled and labeled cells respectively; frequency , 1 . 43 ± 0 . 18 vs . 1 . 83 ± 0 . 31 Hz ) . Furthermore , there was no apparent correlation between GFP levels and properties of mIPSCs ( Figure 3K and L ) . However , we cannot exclude the possibility that a larger sample may have revealed differences . Together , these results further indicate that imprinting drives synaptic plasticity in AOB inhibitory circuits . In contrast with established learning models , however , synaptic changes were widely distributed across both GC and MC populations with no apparent relationship to activation during mating . The lack of specificity in synaptic plasticity suggested that AOB output may be shaped by alternative mechanisms . One possibility is changes in intrinsic membrane excitability , which will alter the recruitment of AOB neurons by shifting the threshold for generating action potential firing . We tested for learning-induced changes in membrane excitability in AOB using graded current injections , focusing first on interneurons . The responsiveness of GCs in Arc-GFP females was enhanced after mating , so that less current was needed to initiate firing , and higher firing rates were produced by the same current steps ( Figure 4A–B ) . Increased excitability was also reflected in GC resting potentials , which were consistently depolarized in mated versus naïve animals ( Figure 4C , −72 . 1 ± 1 . 6 , –71 . 0 ± 1 . 6 , and −66 . 3 ± 1 . 2 mV for naïve , exposed , and mated groups respectively ) . Other properties , such as membrane resistance and slope of the input-output firing function , were unchanged across groups ( Figure 4D , Rinput = 469 ± 35 vs . 520 ± 34 for naïve and mated groups; p=0 . 31; ANOVA with post-hoc Tukey test ) , suggesting that increased GC responsiveness was largely determined by resting potential . Overall , mating increased the intrinsic excitability of AOB interneurons , suggesting that synaptic plasticity in inhibitory circuits is complemented by additional non-synaptic mechanisms . 10 . 7554/eLife . 25421 . 014Figure 4 . Experience alters intrinsic excitability of interneurons . ( A ) Representative responses to graded current injection for GCs from naïve , sensory-exposed and mated mice . ( B ) Current-firing plot shows a shift towards increased excitability of GCs from both mated and sensory-exposed females . ( C ) GC resting membrane potential was significantly hyperpolarized after mating ( p=0 . 008 for mated vs . naïve; ANOVA with post-hoc Tukey test; F = 5 . 18; Fc = 3 . 14; n = 17 , 19 , and 31 cells from 5 , 9 , and 12 mice for naïve , sensory-exposed , and mated groups respectively ) . ( D ) The slope of the current-firing function was similar across groups ( 0 . 28 ± 0 . 01 , 0 . 27 ± 0 . 02 , and 0 . 28 ± 0 . 01; F = 0 . 14; Fc = 3 . 15; p=0 . 87; ANOVA with post-hoc Tukey test ) . ( E ) GC resting potential was uncorrelated with intensity of Arc-GFP labeling in both sensory-exposed and mated animals ( slope not significantly different from zero; p=0 . 87 , 0 . 37 and 0 . 81 for naïve , sensory-exposed and mated groups respectively; linear regression test; n = 17 , 19 and 31 cells in 5 , 9 and 12 mice ) . ( F ) In mated females , resting potential was indistinguishable between GFP ( - ) and GFP ( + ) GCs ( −66 . 8 ± 2 . 19 vs . −65 . 4 ± 1 . 85 mV respectively; p=0 . 62 , t-test; n = 10 and 13 cells in 10 mice , subdivided from the mated group in panel E ) . NS , not significant; p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 01410 . 7554/eLife . 25421 . 015Figure 4—source data 1 . This spreadsheet contains the resting membrane potential and firing rate data for the individual neurons used to generate the bar plots shown in Figures 4C , D and F . These data can be opened with Microsoft Excel or with open-source alternatives such as OpenOffice . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 015 To test whether changes in excitability were specific to mating-activated GCs , we examined the relationship between resting potential and Arc-GFP labeling . As with synaptic measurements , GC resting potential was uncorrelated with GFP intensity ( Figure 4E ) . Furthermore , when we subdivided the GC dataset from mated animals into GFP ( - ) and GFP ( + ) populations , mean resting potential was similar for the two groups ( Figure 4F; −67 . 0 ± 2 . 2 vs . −65 . 4 ± 1 . 9 mV respectively; Figure 4—source data 1 ) . Together , the increased GC excitability after mating indicates that learning acts on intrinsic as well as synaptic properties of AOB neurons . However , intrinsic plasticity was widespread across interneurons , and lacked dependence on prior activation during mating . Because information about strain and individual identity is ultimately conveyed by MCs ( Arnson and Holy , 2013; Ben-Shaul et al . , 2010; Luo et al . , 2003 ) , the effects of learning should ultimately be reflected in their firing patterns . To further evaluate mating-dependent changes in AOB output , we compared responses to current injection in MCs from naïve and mated female Arc-GFP mice . In both groups , MCs responded to current injection with robust firing that decayed during the step ( Figure 5A ) . In contrast to GCs , however , mating had no apparent effect on MC responses to a single stimulus . We found no difference in peak firing rate , total spike count during the train , or decay of firing between naïve and mated females ( Figure 5B–D; peak firing: 15 . 0 ± 3 . 0 Hz vs . 17 . 4 ± 1 . 8 Hz; total spike count: 153 ± 28 vs . 176 ± 25 for naïve vs . mated respectively; Figure 5—source data 1 ) . Despite robust increases in inhibitory input , therefore , MC firing for a single stimulus was unchanged . 10 . 7554/eLife . 25421 . 016Figure 5 . Mating reduces the responsiveness of MCs to repetitive inputs . ( A ) MC firing to an initial current stimulus is similar for naïve and mated females . ( B ) Firing rate profile averaged across all MCs from naïve and mated animals . ( C , D ) Mating has no effect initial MC output ( peak firing rate: p=0 . 51; change in firing rate: p=0 . 55; t-test; n = 10 and 15 cells from 5 and 7 mice for naïve and mated groups respectively ) . ( E ) Example MC responses to repetitive stimulation . Firing is stable over time in naïve females , but drops dramatically over time after mating . ( F ) Average MC output across successive stimuli for naïve ( blue ) and mated females ( red ) . Firing on 10th trial is 80 ± 11% ( naïve ) vs . 36 ± 8% of 1st trial ( mated ) ; F = 10 . 01 , Fc = 4 . 30 , p=0 . 003; ANOVA with post-hoc Tukey test . Light colors show individual neurons; dark traces show mean ± SEM . ( G ) Cumulative histogram showing increased attenuation in MCs from mated animals ( p=0 . 027; Kolmogorov-Smirnov test; n = 9 and 15 neurons from 5 and 7 mice respectively ) . NS , not significant . p<0 . 05; p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 01610 . 7554/eLife . 25421 . 017Figure 5—source data 1 . This spreadsheet contains the firing rate and spike count data for mitral cells used to generate the bar plots and average data shown in Figures 5C , D and F . These data can be opened with Microsoft Excel or with open-source alternatives such as OpenOffice . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 017 Although immediate effects on MC output were not apparent , we further probed for changes over longer time periods . The hormonal changes that induce pregnancy block require prolonged AOB activity lasting several hours ( Li et al . , 1994; Rosser et al . , 1989 ) , timescales that typically encompass multiple sensory interactions ( Hull and Dominguez , 2007 ) . To approximate repeated activation of vomeronasal inputs ( see Figure 2—figure supplement 1 ) , we probed MCs with repetitive stimuli spanning several minutes ( ten current injections , 20 s in duration , repeated every 60 s ) . Surprisingly , in mated females MC firing often declined dramatically across successive trials , so that even neurons initially responding with hundreds of action potentials ceased firing entirely ( Figure 5E ) . While firing also declined in some neurons from naïve animals , MC attenuation was greatly enhanced after imprinting , so that average spike counts in mated females dropped to less than half than that of controls ( Figure 5F–G; 80 ± 11% vs . 36 ± 8% ) . Together , our data indicate that mating leads to an unusual form of plasticity in MC membrane properties , where firing gains a striking dependence on recent history of activity . This metaplasticity in intrinsic excitability offers an alternative mechanism for attenuating AOB output , dramatically suppressing MC firing to repetitive stimuli and curtailing their responsiveness on timescales of minutes . To account for individual-specific recognition memories , AOB plasticity is predicted to be expressed selectively in the MC ensemble activated by the stud . Because targeted MC recordings were precluded by weak labeling in Arc-GFP mice , we tested for selective changes in excitability by collecting a second dataset using the Fos-GFP reporter . Within mated females , we compared responses of labeled MCs with unlabeled cells that presumably represent other , non-stud chemosignals . Mating drove robust increases in Fos-GFP relative to sensory-exposed controls , generating detectable labeling in approximately 25% of MCs ( Figure 3—figure supplement 1B ) . After mating , both labeled and unlabeled MCs fired similarly to initial stimuli , consistent with responses in Arc-GFP mice ( Figure 6A–D; peak firing rate , 15 . 2 ± 3 . 2 Hz vs . 16 . 2 ± 2 . 4 Hz; total spike count , 155 ± 39 vs . 133 ± 26 for GFP ( - ) and GFP ( + ) respectively; Figure 6—source data 1 ) . We again used repetitive stimulation to probe for cell-specific plasticity in MC responsiveness . In unlabeled MCs , firing was stable across trials , or even increased slightly over time ( Figure 6E–F; spike count of 10th vs . 1st trial , 115 ± 17% , peak firing rate , 87 . 0 ± 11 . 4% ) . In contrast , the output of GFP ( + ) MCs decreased markedly over successive stimuli ( Figure 6F–H; spike count of 10th vs . 1st trial , 30 . 7 ± 12 . 3%; peak firing , 29 . 4 ± 10 . 2% ) . Results were independent of criteria for selecting GFP ( + ) and GFP ( - ) populations ( Figure 6—figure supplement 1 ) . These data indicate that reduced excitability arises specifically in the MC ensemble activated by the stud during mating and subsequent sensory experience . 10 . 7554/eLife . 25421 . 018Figure 6 . Plasticity in MC responsiveness is specific to mating-activated neurons . ( A ) Both GFP ( - ) and GFP ( + ) MCs show similar initial responses to current stimuli . ( B ) Mean firing rate profiles for GFP ( + ) and GFP ( - ) MCs in Fos-GFP females after mating ( mean ± SEM; n = 7 and 11 cells in 5 and 7 mice ) . ( C , D ) Initial MC output is similar between GFP ( - ) and GFP ( + ) groups ( firing rate , p=0 . 81; spike count , p=0 . 63; t-test ) . ( E ) Representative MC responses to repetitive stimulation in mated females . ( F ) After mating , GFP ( - ) MCs maintain consistent firing , but the output of stud-activated GFP ( + ) neurons is dramatically attenuated over time ( mean ± SEM; n = 7 and 11 cells in 7 mice ) . ( G ) Cumulative histograms indicate a shift towards greater suppression in the MCs activated during mating ( p=0 . 18; Kolmogorov-Smirnov test ) . ( H ) Mean suppression after 10 trials for GFP ( - ) and GFP ( + ) neurons ( spike count on 10th vs . 1st trial: unlabeled , 115 ± 17% , p=0 . 41; labeled , 30 . 7 ± 12 . 3%; p=0 . 0017; t-test ) . ( I ) Cumulative action potential output of GFP ( - ) and GFP ( + ) MCs , averaged across all recorded neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 01810 . 7554/eLife . 25421 . 019Figure 6—source data 1 . This spreadsheet contains the firing rate and spike count data for mitral cells used to generate the bar plots and average data shown in Figures 6C , D , F and H . These data can be opened with Microsoft Excel or with open-source alternatives such as OpenOffice . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 01910 . 7554/eLife . 25421 . 020Figure 6—figure supplement 1 . Correlated plasticity and GFP labeling are independent of selection criteria . ( A ) Arc-GFP labeling intensity for the GCs in our recording sample ( green markers ) superimposed on the distribution for all detectable GCs ( gray line , normalized exponential fit to imaging data ) . Our dataset spans the majority of intensity values . ( B , C , D ) Frequency and amplitude of sEPSPs and resting membrane potential of GCs , respectively . Results comparing GFP ( + ) and GFP ( - ) populations are consistent across various selection criteria ( upper vs . lower 1/2 , 1/3 , and 1/4 of the recorded cells ) . ( E ) Fos-GFP labeling intensity for recorded MCs as in ( A ) . Triangles indicate cells used to analyze synaptic inhibition; circles , cells used to analyze slow firing suppression . ( F , G , H ) mIPSC frequency , mIPSC amplitude , and firing rate suppression for repetitive stimuli , respectively . Results were again consistent across a range of selection criteria . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 02010 . 7554/eLife . 25421 . 021Figure 6—figure supplement 2 . Slow attenuation is absent in GFP ( + ) MCs labeled by sensory exposure alone . ( A ) Firing rates are elevated in GFP ( + ) MCs relative to unlabeled neurons . ( B , C ) Increased firing rates for initial stimuli in GFP ( + ) vs . GFP ( - ) MCs in the absence of mating . Peak rates , 30 . 5 ± 2 . 2 Hz vs . 15 . 5 ± 2 . 7 Hz respectively , mean ± SEM; p<0 . 05 , t-test; n = 5 and 7 cells in 4 mice ) . ( D ) Example MC responses to repetitive current injection for both GFP ( - ) and GFP ( + ) neurons . ( E ) In the absence of mating , both MC groups maintain stable output ( mean ± SEM; n = 5 and 7 cells in 4 mice ) . ( F ) GFP ( + ) and GFP ( - ) cells show similar levels of suppression on trial 10 ( p=0 . 77; t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 021 To ensure that MC excitability was altered by learning , rather than reflecting a pre-existing cell population in AOB , we performed parallel experiments where control females received sensory exposure to males without mating . This labeled a much smaller group of MCs , which responded with higher firing rates than unlabeled neurons in the same animals ( Figure 6—figure supplement 2A–C ) . Membrane resistance was also higher in GFP ( + ) MCs ( data not shown ) , suggesting that sensory stimulation alone preferentially recruits a group of high-excitability neurons similar to findings in neocortex ( Yassin et al . , 2010 ) . In contrast to results in mated females , however , both GFP ( - ) and GFP ( + ) populations maintained consistent firing over all trials , despite their differences in initial responsiveness ( Figure 6—figure supplement 1D–F ) . These data further indicate that dynamic changes in MC excitability result from imprinting , and emerge specifically in the population labeled during mating . We estimated the net loss of output for stud-activated MC populations by plotting cumulative spike counts for matched groups of GFP ( + ) and GFP ( - ) neurons , which showed that total spike count diverged rapidly between the two groups ( Figure 6I ) . Together , these results provide the first direct evidence of targeted , cell-specific changes in the AOB ensembles activated by conspecifics during social experience . MC firing could potentially be shaped directly by changes in intrinsic membrane properties per se , or by prolonged synaptic inhibition that outlasts the stimulus . To distinguish between these possibilities , we tested MCs after blocking fast synaptic transmission with NBQX , APV , and bicuculline ( 10 , 50 , and 10 µM respectively ) . MC suppression was intact even after eliminating local circuit interactions , indicating that it did not depend on persistent inhibition ( Figure 7—figure supplement 1 ) . We further probed the source of reduced MC firing by examining membrane properties over the course of stimulus trains . Repeated stimulation led to a progressive hyperpolarization of MC membrane potential , both in randomly selected MCs in mated Arc-GFP females ( Figure 7A ) and in GFP ( + ) MCs in mated Fos-GFP mice ( Figure 7B; Figure 7—source data 1 ) . These changes were not predicted by initial MC resting potential , which was indistinguishable between naïve and mated females in Arc-GFP mice ( Figure 7C; −58 . 0 ± 1 . 1 mV vs . −56 . 7 ± 0 . 7 mV respectively ) , and between GFP ( + ) and GFP ( - ) populations within mated Fos-GFP females ( Figure 7C; −55 . 0 ± 1 . 5 vs . −53 . 9 ± 1 . 1 mV , p=0 . 56 , t-test; n = 7 and 11 respectively ) . On average , GFP ( + ) MCs were hyperpolarized by −3 . 3 ± 0 . 5 mV vs . 0 . 2 ± 0 . 6 for GFP ( - ) neurons ( Figure 7D; p<0 . 05 , t-test ) . MC hyperpolarization was strongly correlated with loss of firing ( Figure 7E ) . Hyperpolarization was accompanied by a slight reduction in membrane resistance ( initial vs . final Rin , 486 ± 114 MΩ vs . 341 ± 27 MΩ; p=0 . 26 , t-test ) . Together , these data further indicate that reduced MC output is due to changes in intrinsic membrane properties rather than altered synaptic inhibition , imparting stud-activated neurons with a sensitivity to recent firing that progressively dampens their output . 10 . 7554/eLife . 25421 . 022Figure 7 . Loss of MC sensitivity results from progressive membrane potential hyperpolarization . ( A ) Representative MC responses to repetitive stimulation , showing initial resting potential and onset of firing for each trial . Progressive hyperpolarization was greatly enhanced in mated vs . naïve mice ( red and blue respectively ) . ( B ) Within the AOB of mated females , hyperpolarization was selectively expressed in mating-activated GFP ( + ) MC populations . ( C ) Initial resting membrane potential was similar for MCs from naïve vs . mated females ( blue and red respectively; p=0 . 76; n = 9 and 15 cells in 5 and 7 mice ) , and for labeled and unlabeled MC populations in mated animals ( gray and green; p=0 . 56; t-test; n = 7 and 11 neurons in 5 and 7 mice ) . ( D ) Mean hyperpolarization during repetitive stimulation for GFP ( + ) and GFP ( - ) MCs ( green and gray; upper and lower 1/3 of the recorded population; , p<0 . 05; t-test ) . ( E ) Progressive loss of MC responsiveness is correlated with membrane hyperpolarization . Green , GFP ( + ) ; dark gray , GFP ( - ) ; light gray , intermediate . ( F ) Membrane hyperpolarization persists for > 2 . 5 min between stimuli . Traces show Vm before and after three successive spike trains delivered 3 min apart . Red dashes show step-like hyperpolarization lasting until the next stimulus . ( G , H ) Mean hyperpolarization and normalized change in firing for MCs tested with stimuli 2 min and 3 min apart . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 02210 . 7554/eLife . 25421 . 023Figure 7—source data 1 . This spreadsheet contains the membrane potential data for mitral cells used to generate the bar plots and average data shown in Figure 7C and D . These data can be opened with Microsoft Excel or with open-source alternatives such as OpenOffice . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 02310 . 7554/eLife . 25421 . 024Figure 7—figure supplement 1 . MC hyperpolarization and attenuation of firing does not depend on persistent synaptic inhibition . ( A ) Representative MC membrane potential response after blocking fast synaptic transmission with NBQX , APV , and bicuculline ( 5 , 50 , and 10 µM respectively ) . ( B ) Decreased MC output over the course of 10 stimuli correlates strongly with membrane hyperpolarization . DOI: http://dx . doi . org/10 . 7554/eLife . 25421 . 024 The hormonal changes that drive pregnancy block occur over several hours of sensory exposure , during which animals interact intermittently at varying intervals . To further probe the time course of plasticity , we probed randomly selected MCs in mated females using stimuli spaced 2 min and 3 min apart . Each firing bout led to hyperpolarization that decayed extremely slowly , lasting until the onset of the next stimulus ( Figure 7G ) . Intervals of both 2 min and 3 min gave comparable levels of hyperpolarization and attenuated firing after 10 stimuli ( Figure 7G–H; mean ∆Vm , −5 . 64 ± 0 . 88 and −6 . 02 ± 2 . 05 mV for 2 and 3 min intervals respectively; normalized spike count , 0 . 176 ± 0 . 116 and 0 . 193 ± 0 . 089 for 2 and 3 min ) . When possible , we tested MCs again at 15 min and 35 min after the offset of stimulation , which revealed that suppression lasted for total periods of an hour or longer ( normalized spike count , 0 . 149 ± 0 . 080 and 0 . 117 ± 0 . 072 at 15 and 35 min respectively; ∆Vm , −5 . 85 ± 1 . 49 and −6 . 21 ± 1 . 97 at 15 and 35 min ) . These results indicate that MC plasticity persists on the timescales relevant to physiology and behavior in vivo . AOB inhibitory circuits had several unique properties . First , we found that MCs could strongly suppress their own firing via pronounced self-inhibition , consistent with robust GABAergic circuits in AOB ( Castro et al . , 2007; Hendrickson et al . , 2008; Shpak et al . , 2012 ) . Self-inhibition in MOB MCs , in contrast , had much weaker effects on firing . Self-inhibition in AOB also emerged surprisingly slowly , building over several seconds so that it was strongest during prolonged MC firing on the timescales of natural sensory-evoked responses ( Ben-Shaul et al . , 2010; Luo et al . , 2003 ) . The basis for differences in self-inhibition in AOB and MOB are unclear , but may be linked to the prominent role of mGluRs in AOB ( Castro et al . , 2007 ) . Our data suggest that mating acts on multiple cell types and synaptic elements in AOB inhibitory circuits . Increased frequency of excitatory input to GCs implies upregulation of release sites in MC dendrites , and enhanced amplitude suggests strengthening of postsynaptic contacts , consistent with the ultrastructural enlargements seen in the postsynaptic density ( Ichikawa , 2003; Matsuoka et al . , 2004 ) . Increased frequency of mIPSCs in MCs also suggests enhanced release of GABA from GCs , which could result either from changes in existing contacts or addition of new synapses via spine growth or recruitment of adult-born interneurons ( Mak and Weiss , 2010; Shingo et al . , 2003 ) . Overall , however , this inhibitory plasticity appeared to have little effect on MC output driven by current injection . Increased release of GABA may be counterbalanced by short-term dynamics during extended firing ( Dietz and Murthy , 2005 ) , or alternatively changes in mIPSCs may reflect top-down inputs to AOB ( Fan and Luo , 2009 ) that would contribute to spontaneous but not recurrent inhibition . Memory formation has long been proposed to rely on inhibitory plasticity , selectively targeting the stud-activated MC population ( Brennan and Keverne , 1997 ) . In contrast , we found that synaptic changes were widely distributed across both MCs and GCs with no observable dependence on activation during mating . While non-specific plasticity runs counter to existing models , our data do not exclude a role for inhibition in learning or sensory processing . GABAergic circuits may be differentially recruited in vivo , where complex natural cues activate larger MC populations ( Ben-Shaul et al . , 2010; Meeks et al . , 2010 ) . Inhibition strongly shapes MC firing in the intact brain ( Hendrickson et al . , 2008 ) , and descending pathways targeting AOB are likely to further shape sensory responses ( Fan and Luo , 2009 ) . Overall , however , the nonspecific nature of inhibitory plasticity suggests it may act in concert with other , more targeted changes in the AOB circuit . Unexpectedly , imprinting had the most striking effects on intrinsic rather than synaptic properties , suggesting an alternative cellular mechanism for storing sensory experience in AOB . MC suppression was largely absent in naïve females , but was strongly increased by mating in two separate datasets , indicating that it is was generated de novo by learning rather than reflecting pre-existing AOB populations ( Yassin et al . , 2010 ) . Experience-dependent changes in excitability often accompany synaptic modifications in both mammalian and invertebrate systems ( Daoudal and Debanne , 2003; Zhang and Linden , 2003 ) . Altered intrinsic properties support homeostatic regulation in cortical circuits , scaling cellular excitability to match long-term changes in sensory input ( Desai et al . , 1999; Turrigiano , 2011 ) . Often , learning acts to enhance excitability ( Barkai and Saar , 2001; Zhang and Linden , 2003 ) , acting to amplify responses to trained sensory inputs ( Mozzachiodi and Byrne , 2010 ) , or to select neural populations encoding the learned cue ( Yiu et al . , 2014; Zhou et al . , 2009 ) . Here , in contrast , MC excitability was strongly reduced by pheromonal learning . This sign reversal is consistent with the fact that imprinting leads to the suppression of an otherwise default neuroendocrine response to sensory input . Notably , whereas other paradigms lead to static effects on excitability that are immediately apparent on testing , changes in MCs were dynamic and only emerged after strong activation . The excitability of control and ‘imprinted’ neurons was initially indistinguishable , and responses only diverged after cells had experienced substantial firing . MC hyperpolarization accumulated after each trial and lasted for at least 30 min after the offset of stimulation , suggesting that MCs display an unusual and highly integrative form of intrinsic membrane plasticity . What are the potential advantages of intrinsic versus synaptic plasticity in AOB ? Membrane excitability offers a simple way to selectively control the output of specific MC populations , whereas inhibitory plasticity would need to be coordinated across large sets of GCs , and further targeted to the specific synapses onto stud-encoding MCs . Intrinsic excitability may be particularly well suited to mediating learning in dedicated sensory pathways coupled to stereotyped behavioral responses . Interestingly , firing was similar for labeled and unlabeled MCs after mating , and only diverged after extended activity bouts . Thus , learning does not cause simple , static changes in MC excitability per se , but instead leads to metaplastic effects that impart sensitivity to recent firing levels . In metaplasticity of synaptic strength , experience shifts thresholds for potentiation and depression via changes in NMDA subunit composition ( Abraham , 2008; Lee et al . , 2010 ) . Metaplasticity in the intrinsic excitability of MCs may rely on similar changes in composition of membrane conductances . Prolonged hyperpolarization and lowered membrane resistance are consistent with changes in potassium channels such as HCN or Ca2+-dependent K+ currents , which are present in main olfactory bulb and linked to learning in other systems ( Lin et al . , 2008; Nolan et al . , 2004 , 2003; Stackman et al . , 2002; Wang et al . , 2007 ) . MC firing is also modulated by intrinsic conductances such as CAN currents , which boost synaptic responses ( Shpak et al . , 2014 , 2012 ) and likely contributed to accelerated firing rates seen over the first few seconds of stimulation in our study . Interestingly , this current opposes the MC hyperpolarization we describe here , which appeared to dominate after more prolonged firing bouts and was most readily apparent only after mating . AOB MCs thus appear to express multiple activity-dependent conductances that dynamically modulate their firing depending on the strength , duration , and biological context of activity . The specific conductances responsible for slow MC attenuation remain to be established . Dynamic , activity-dependent changes in excitability were a unique feature of AOB MCs . How may slowly adapting sensitivity contribute to sensory processing ? One potential role is to high-pass filter vomeronasal input , preserving responsiveness at the onset of social interactions for appropriate selection of aggressive , reproductive , or parental behaviors ( Burns-Cusato et al . , 2004; Clancy et al . , 1984; Stowers et al . , 2002; Tachikawa et al . , 2013 ) , while selectively attenuating the long-lasting , repetitive AOB activity required for the neuroendocrine changes that block pregnancy ( Li et al . , 1994; Rosser et al . , 1989 ) . Similarly , slow activity dynamics may help reduce interference between memories of different individuals . Different males have similar chemical signatures ( Harvey et al . , 1989 ) , suggesting that they may also have overlapping neural representations in AOB . Eliminating the responses of stud-encoding MCs could therefore also disrupt representations of other , non-imprinted animals . Delayed changes in MC output may help minimize the impact of plasticity on overlapping sensory codes . Fos-GFP labeling was transient , limiting our measurements to a period of several hours following mating . The effects of mating on pregnancy block , however , can last for many weeks ( Brennan et al . , 1990 ) . It will be important to determine whether effects on MC excitability are maintained for similar time periods . Other , more permanent labeling strategies may allow plasticity to be tested at later time points ( Guenthner et al . , 2013; Sakurai et al . , 2016 ) . Alternatively , sensory memories could be stored initially in AOB and then transferred to other areas , as seen in other memory systems ( Preston and Eichenbaum , 2013; Ross and Eichenbaum , 2006 ) . Mating is one of several biological contexts where animals show flexibility in vomeronasal-guided behaviors . Interestingly , many of these involve the loss of an otherwise default response , similar to the effects of mating . Males shift from attack to parental behaviors towards pups ( Tachikawa et al . , 2013; Wu et al . , 2014 ) , and regulate aggression towards other males to form dominance hierarchies ( Wang et al . , 2014 ) . Behavioral responses to fear-inducing cues such as predator odors can also habituate with repeated presentation ( Takahashi et al . , 2005 ) . It is currently unclear whether the behavioral plasticity seen in other paradigms relies on similar cellular mechanisms in AOB , and it will be important for future work to test this possibility . Overall , our data reveal a novel form of cellular plasticity that emerges after mating in females , where slowly emerging , activity-dependent changes in intrinsic excitability dramatically attenuate the output of the MC ensemble activated by the stud male . It will be important for future work to test how this plasticity shapes sensory representations and neuroendocrine responses in behaving animals during social encounters . Changes in MC excitability could also contribute to flexibility in other vomeronasal-mediated behaviors , which often involve suppression of otherwise default sensory responses ( Tachikawa et al . , 2013 ) . While the AOB is a critical node in the vomeronasal pathway , MC plasticity likely acts in parallel with broader changes across the extended network of brain regions that couple chemosensory input to behavior ( Dulac et al . , 2014; Wu et al . , 2014 ) . All experiments were performed in sexually mature adult female mice 8–14 weeks of age . Reporter lines were obtained from Jackson Laboratory ( Arc-GFP , RRID:IMSR_JAX:007662; Fos-GFP , RRID:IMSR_JAX:018306 ) and bred in a C57BI/6J background . Experimental Arc-GFP animals were heterozygous , maintaining a functional Arc allele . Animals were group housed in the Boston University animal care facility on a 12 hr light/dark cycle with ad lib access to food and water . Mice were anesthetized ≥5 days prior to experiment and received bilateral ovariectomies followed by implantation of estradiol capsules ( Bakker et al . , 2003 ) . On the experimental day , estrus was induced with progesterone injection ( 16 µg/g ) , confirmed by vaginal smears and histological examination ( Caligioni , 2009 ) . At estrus onset , 3–4 hr after the beginning of the light cycle , females were paired with a sexually experienced male in their home cage for an additional 4 hr for mating and subsequent sensory exposure required for imprinting . Males typically attempted copulation within 20–30 min . Cases where males did not mount females were used as controls for sensory experience without mating . Sedated males were also used for sensory-exposure controls; activity-dependent labeling in the corresponding females was indistinguishable and these data were grouped together . Females in both mated and sensory-exposed groups were ovariectomized and progesterone-primed , while naïve females were unmanipulated . While mating success could not be evaluated in electrophysiology experiments , in a parallel group this protocol resulted in pregnancy in 9 of 11 gonadally intact females . Reproductive encounters were video recorded and scored to quantify mating and behavioral interactions . All procedures were approved by the Boston University Institutional Animal Care and Use Committee and followed guidelines set by the US National Institutes of Health . Activity-dependent labeling was visualized in each slice prior to electrophysiological recordings using a two-photon microscope ( Prairie Ultima ) with 920 nm excitation and a 20X NA 0 . 95 objective ( Olympus ) , using consistent acquisition settings for laser power and detector gain across sessions . Immediately prior to establishing recordings , we acquired additional image stacks of GFP labeling for the field of view at each recording location ( 250 µm X 250 µm ) using a 40X NA 0 . 8 objective ( Olympus ) . Intensity was quantified for all detectable neurons using a circular region of interest centered on the soma . GFP intensity was continuously distributed , presumably reflecting graded levels of prior activity . Cells were classified as unlabeled or labeled using a threshold of ≤2 or≥4 times background neuropil fluorescence respectively . For electrophysiological analysis , we performed similar analyses comparing the brightest third and dimmest 50% , 33% , and 25% of our recorded sample . Results were robust to classification threshold and comparison groups . Sagittal brain slices of AOB ( 300 μm thick ) were prepared from female mice using a vibratome ( VT1200S , Leica , Buffalo Grove IL ) . To preserve tissue health in adult animals , mice were deeply anesthetized with ketamine/xylazine and perfused transcardially with ice-cold modified artificial cerebrospinal fluid ( ACSF ) containing , in mM: 124 NaCl , 2 . 5 KCl , 1 . 25 NaH2PO4 , 25 NaHCO3 , 75 sucrose , 10 glucose , 1 . 3 ascorbic acid , 0 . 5 CaCl2 and 7 MgCl2 . Slices were maintained using ACSF containing , in mM: 124 NaCl , 3 KCl , 1 . 25 NaH2PO4 , 26 NaHCO3 , 20 sucrose , 2 CaCl2 and 1 . 5 MgCl2 , continuously oxygenated with 95/5% O2/CO2 . Slices were visualized with a two-photon microscope ( Ultima , Prairie Technologies , Middleton WI ) using a 40x water immersion objective and Dodt contrast imaging . Whole cell electrodes were pulled to tip resistances of 3–8 MΩ and contained the following internal solutions ( in mM ) : current clamp , 135 K-gluconate , 2 MgCl2 , 10 HEPES , 0 . 4 EGTA , 2 MgATP , 0 . 5 Na3GTP , 10 phosphocreatine disodium; voltage clamp , 115 CsCl , 25 TEA-Cl , 5 QX314-Cl , 0 . 2 EGTA , 4 MgATP , 0 . 3 Na3GTP and 10 phosphocreatine disodium . Alexa 594 was added to the internal solution to confirm cell identity in targeted recordings . Membrane voltage was not corrected for liquid junction potential . Electrophysiological data was collected at 29 . 5°C with a Multiclamp 700B amplifier ( Molecular Devices , Sunnyvale , CA ) and digitized at 10 kHz ( National Instruments PCI-6321 ) using custom Matlab routines ( Mathworks , Natick , MA ) . Action potential detection and analysis was performed using custom Matlab routines detecting zero-crossing membrane potentials . Changes in firing with repeated stimulation were quantified as a suppression index , calculated as the ratio of firing on the 10th vs . first trial . Synaptic currents and EPSPs were detected and analyzed in Igor Pro ( WaveMetrics , Lake Oswego , Oregon ) using Taro Tools ( https://sites . google . com/site/tarotoolsmember/ ) . Thresholds were chosen to maximize detection of synaptic events while excluding false positives due to recording noise . Thresholds were set at 10 pA for mIPSCs in MCs , and 0 . 25 mV for EPSPs in GCs . In both cases we estimate we detected at least 95–98% of events while limiting false positives to <1% , determined by visual inspection . GC results were consistent across a wide range of detection criteria . All chemicals were obtained from Sigma/Aldrich ( NBQX ) , Tocris ( BMI ) , and Alomone Labs ( TTX ) . Receptor antagonists ( APV , NBQX and Gabazine ) were applied by bath perfusion . All results reported in the text and figures represent mean ± S . E . M . Statistical significance was calculated using t-test or ANOVA as appropriate , noted in results and figure legends . Distributions of miniature and spontaneous synaptic events were analyzed with the Kolmogorov-Smirnov test . Animals were randomly assigned to naïve , sensory exposure , or mating groups after recovery from surgery . Data collection and analysis were not blind to experimental conditions .	To navigate social situations , humans and other animals need to remember who they have interacted with and how it went , and adjust their behavior in future encounters accordingly . For example , your physical actions , and even your body’s physiological responses , will be very different when you encounter the last person you kissed instead of the last person you fought with ( assuming this is not the same person ! ) . Memories of social interactions can have dramatic consequences . For instance , male mice often kill the offspring of other males . Female mice appear to have adopted a countermeasure to avoid losing a litter of pups to such aggression: they will spontaneously abort a pregnancy when exposed to chemicals called pheromones from unfamiliar males . However , when the female mouse is exposed to the pheromones of the male she mated with she maintains her pregnancy . Exactly how the memories of previous social interactions with the males affect the female’s pheromone responses is not fully understood . To investigate how the female is able to respond differently to different males , Gao et al . recorded the activity of individual neurons taken from the brain tissue of female mice who had recently mated . The recordings showed that previous social experiences produce learning-related changes in the brain of the female mouse that reduce how sensitively pheromone-detecting neurons respond to the chemical cues of the male mate . This suppresses the signals that the neurons would otherwise send to trigger an abortion in response to male pheromones . Gao et al . also used fluorescent tags to identify which neurons in the female’s brain had been activated during mating . This revealed that only those neurons that had been activated by the mate become unresponsive when the cells again encountered his pheromones . This suggests that a set of neurons in the female’s brain records the chemical ‘fingerprint’ of the mate , and can then selectively filter out that mate’s pheromone signals . Many other social interactions , such as parenting , are also strongly shaped by experience . The results presented by Gao et al . may therefore offer wider lessons for understanding how the brain targets different behaviors toward specific individuals . It will also be important to investigate how highly arousing experiences cause such powerful memories to form . This could ultimately help us to better understand – and potentially treat – conditions like post-traumatic stress disorder .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Neural mechanisms of social learning in the female mouse
For decades , the mechanism of skeletal patterning along a proximal-distal axis has been an area of intense inquiry . Here , we examine the development of the ribs , simple structures that in most terrestrial vertebrates consist of two skeletal elements—a proximal bone and a distal cartilage portion . While the ribs have been shown to arise from the somites , little is known about how the two segments are specified . During our examination of genetically modified mice , we discovered a series of progressively worsening phenotypes that could not be easily explained . Here , we combine genetic analysis of rib development with agent-based simulations to conclude that proximal-distal patterning and outgrowth could occur based on simple rules . In our model , specification occurs during somite stages due to varying Hedgehog protein levels , while later expansion refines the pattern . This framework is broadly applicable for understanding the mechanisms of skeletal patterning along a proximal-distal axis . During evolution , a number of changes in vertebrate body plan allowed terrestrial tetrapod species to thrive on land and take advantage of new habitats . For example , in contrast to the open rib cages of aquatic species , the thoracic cage became an enclosed chamber that could support the weight of the body and facilitate lung ventilation ( Janis and Keller , 2001 ) . Current tetrapod species typically have ribs that are subdivided into two segments , a proximal endochondral bony segment connected to the vertebrae , and a distal permanent cartilage segment that articulates with the sternum ( Figure 1A , B ) . The rib bones support the body wall and protect the internal organs; the costal cartilage maintains thoracic elasticity , allowing respiration while still enclosing the thoracic cage . Although clues from the fossil record are beginning to reveal when the enclosed rib cage arose during evolution ( Daeschler et al . , 2006; Pierce et al . , 2013 ) , little is known about what changes occurred during embryogenesis to extend the ribs around the body and to connect the ribs to the sternum via a costal cartilage element ( Brainerd and Brainerd , 2015 ) . Here , using genetic and computational approaches , we generate a plausible model for how two rib segments form during development . Lineage-tracing studies indicate that the sternum and ribs have different developmental origins . The sternum , like the appendicular skeleton , arises from the lateral plate mesoderm ( Cohn et al . , 1997; Bickley and Logan , 2014 ) , while the ribs and vertebrae arise from the somites ( reviewed in [Brent and Tabin , 2002] ) . Studies using chicken-quail chimera grafts have shown that the thoracic somites contribute to all portions of the ribs ( Huang et al . , 1994 ) , with a the medial somite contributing to the proximal ribs while lateral somite contributes to the distal ribs ( Olivera-Martinez et al . , 2000 ) . These results suggest that the proximal and distal progenitor populations of the rib are distinct at early somite stages rather than being intermixed . As the whole somite matures , it separates into distinct dorsal ( dermomyotome and myotome ) and ventral ( sclerotome ) compartments ( Figure 1C ) . Initially , there was some debate on the precise embryological origin of the ribs within the somite ( Kato and Aoyama , 1998; Huang et al . , 2000 ) . However , using retroviral lineage labeling which avoids the challenges of transplantation experiments , both the proximal and distal segments of the rib were shown to arise from the sclerotome compartment ( Evans , 2003 ) . It has been still unclear though , how the sclerotome becomes patterned along the proximal-distal axis . Through studies particularly of Drosophila wing/leg disc and of vertebrate limb development over the past decades , several patterning models have been conceived to explain how proximal-distal , dorsal-ventral , and anterior-posterior pattern arises ( Briscoe and Small , 2015 ) . For example , compartments could become specified based on: ( 1 ) the presence of cellular determinants , ( 2 ) the concentration of a morphogen , ( 3 ) the duration of exposure to a signaling molecule , and/or ( 4 ) the action of local relay or mutual inhibition signaling . Specification could gradually emerge over the course of organogenesis or via a biphasic process with specification occurring early in a small population of cells followed later by expansion into compartments ( recently reviewed in [Zhu and Mackem , 2017] ) . In this study , we first use genetically modified mice in which the Hedgehog ( Hh ) and apoptosis pathway is disrupted to provide clues for how two rib segments are patterned and grow . Our experiments produced unexpected results which led us to seek an explanation using Agent-Based Modeling , a simulation method based on a cell’s ability to make decisions in response to stimuli . We designed a set of simple rules that could produce a wide variety of potential phenotypes which then motivated the collection of further biological measurements . Using a refined model , we were then able to conclude that complex patterning and growth can emerge through a set of simple rules and biologically supported parameters . Furthermore , our model does not require individual cells to have necessarily received any positional information prior to Hh expression . Finally , we find that our model is essentially biphasic , with early events that define the size and fate of the progenitor populations and later events that expand the already specified population . Studying rib formation during embryogenesis , and in particular , determining how the segments of the rib cage become distinct , provides a relatively simple test-case for questions regarding skeletal patterning , while also giving clues as to the evolutionary origin of the enclosed rib cage . In addition , our studies may aid in understanding the etiology of congenital abnormalities of the rib cage ( Blanco et al . , 2011 ) . The use of agent-based modeling provides insight into how simple decision making at the cellular level could lead to the emergence of multi-component structures during skeletal development . Previous studies have demonstrated the importance of Hedgehog ( Hh ) signaling for sclerotome induction and specification . Overexpression of Sonic hedgehog ( Shh ) can produce ectopic sclerotome at the expense of dermomyotome . Furthermore , in the absence of Shh , Pax1 expression is reduced and later lost; vertebrae and the proximal rib segments then fail to form while the distal cartilaginous portions of the ribs are still present , although abnormal ( Fan and Tessier-Lavigne , 1994; Chiang et al . , 1996; Marcelle et al . , 1999 ) . Hh signaling is also well-known to be important for promoting cell proliferation , growth , and survival ( Charrier et al . , 2001; Thibert et al . , 2003 ) , and in the absence of Shh or Shh-producing cells in the floor plate and notochord , the apoptosis in the somite is greatly increased ( Teillet et al . , 1998 ) . Thus , in addition to a role for Hh signaing in somite/sclerotome induction , Hh signaling also protects somite cells from undergoing programmed cell death . To distinguish between these two roles , we decided to block cell death in Shh null animals . In previous studies , where loss of Shh results in high cell death in the developing heart and face , removing the function of Apoptotic protease-activating factor 1 ( Apaf1 ) , a central player in the mitochondrial pathway of programmed cell death , could block cell death and rescue the phenotype ( Aiyer et al . , 2005; Long et al . , 2013 ) . Perhaps similarly , removing Apaf1 on a Shh null background would inhibit somite cell death and rescue the thoracic phenotype . We therefore generated genetically modified mouse lines that lacked Shh , Apaf1 , or both . Apaf1 is required for the Cytochrome c and ATP-dependent activation of Caspase9 which leads to the subsequent activation of Caspase3 , followed by the initiation of nuclear breakdown and proteolysis ( Zou et al . , 1997 ) . Embryos carrying null alleles for Apaf1 ( Apaf1 KO ) have vastly decreased cell death in the nervous system and exhibit disruptions of the head and face skeleton likely due to a grossly overgrown CNS . In addition , interdigital death in the autopod is delayed ( Cecconi et al . , 1998; Yoshida et al . , 1998 ) . Although , the embryos develop to term , they typically die at birth and have not been examined during skeletal development . Therefore , we investigated the skeleton of Apaf1 KO embryos and found that patterning of the axial and appendicular portions is largely normal and that cartilage and bone development proceeds on schedule , indicating that Apaf1 is not required for normal skeletal development ( Figure 1D , E ) . As previously observed , Shh KO animals exhibit a failure to form the vertebral column and pronounced rib cage defects ( Figure 1F ) ( Fan and Tessier-Lavigne , 1994; Chiang et al . , 1996 ) . Absence of the proximal ribs was consistently observed amongst all Shh KO embryos stained for bone and cartilage ( E15-18 ) . In contrast , the pattern of the remaining rib segments varied ( Figure 1I ) . These cartilage portions were distally located and never mineralized suggesting that they represented the distal costal cartilages . These segments were not entirely normal as they were discontinuous , reduced in number ( ~7–8 instead of 13 ) , not properly articulated with the sternum , and positioned at abnormal angles . The clavicle and sternum were present , although the sternum was sometimes not completely fused and had missing or disorganized segments; however , it did undergo ossification on schedule . In a few cases , small condensations could be observed laterally , possibly near the chondro-costal joint ( Figure 1F ) . We then created animals double null for Shh and Apaf1 ( Shh;Apaf1 DKOs ) and analyzed the skeletal pattern . To our surprise , instead of observing a rescue of the Shh KO phenotype , Shh;Apaf1 DKO animals exhibited an even more severe skeletal phenotype . Shh;Apaf1 DKO embryos did display features of the Shh single KO ( no vertebrae and proximal ribs ) . However , in addition to these defects , the distal portion of the ribs was now missing as evidenced by the lack of alcian blue staining in the body wall ( Figure 1G ) . Rarely , a few small pieces of cartilage were present at the lateral margin , near the expected location of the chondro-costal joint ( Figure 1J ) . Apaf1 has been shown to have other roles in addition to regulating the programmed cell death pathway ( Zermati et al . , 2007; Ferraro et al . , 2011 ) . Thus , to determine if the observed effects were specific to Apaf1 , we carried out mouse crosses utilizing a Caspase3 null allele ( Casp3 KO ) . Caspase3 is an executioner caspase , that cleaves key structural proteins leading to DNA fragmentation and membrane blebbing ( Fuchs and Steller , 2011 ) . As with loss of Apaf1 , the axial skeleton of Casp3 KO animals was grossly normal ( Figure 2A , B ) . Shh;Casp3 DKO embryos exhibited a complete absence of all vertebrae , vertebral ribs and sternal ribs as was observed in the Shh;Apaf1 DKO embryos ( compare Figure 1G and Figure 2D ) . These results suggest that the observed phenotypes when Apaf1 is removed are not due to a specific non-canonical function but rather due to a general abrogation of the programmed cell death pathway . To determine if the absence of programmed cell death genes was indeed preventing cells from dying , TUNEL and/or LysoTracker assays were performed ( Fogel et al . , 2012 ) . In control embryos , evidence of normal cell death can be seen beginning as early as E9 . 0 as distinct periodic stripes of staining in the somites , extending to E10 . 5 and waning by E11 . 5 ( Figure 2E , E’ , I ) ( Sanders , 1997; Teillet et al . , 1998 ) . The absence of Apaf1 , however , results in a dramatic reduction in cell death throughout the embryo , and notably absence of cell death in the somites ( Figure 2F , F’ ) . In contrast , Shh KO embryos exhibit high cell death in the somites — with high LysoTracker-positive staining in the ventral sclerotome domain ( Figure 2G , G’ and inset , marked with brackets ) . Interestingly , the absence of Apaf1 in Shh null embryos ( Shh;Apaf1 DKOs ) results in a vast reduction in LysoTracker-positive cells ( Figure 2H , H’ ) . Like Apaf1 KO embryos , Casp3 KO embryos display a dramatic reduction in cell death compared to the normal pattern ( Figure 2I , J ) . Similarly , Shh;Casp3 DKO embryos ( Figure 2L ) , also exhibit a dramatic reduction in cell death compared to the Shh KO pattern . Thus , the absence of either Apaf1 or Casp3 results in an inhibition of programmed cell death even in Shh null animals . However , it seems counterintuitive that a reduction in cell death could lead to a more severe skeletal phenotype . Previous studies have demonstrated that Shh expression in the notochord and floor plate is essential for ventral neural tube and somite specification ( Varjosalo and Taipale , 2008 ) . However , Shh is also expressed in other developing tissues that could affect somite patterning and rib formation . For example , RNA in situ hybridization reveals that Shh is also expressed in the dorsal root ganglia , ventral neural tube , developing lungs , as well as the developing ribs themselves ( Figure 3A , B ) . To determine if Shh specifically from the notochord and floor plate is required to obtain the observed phenotypes , Shh hypomorph embryos were created utilizing a tamoxifen inducible Foxa2-CRE conditional knock-out approach ( Park et al . , 2008 ) . Using this system , the temporal influence of Shh on rib development could also be analyzed . Administration of tamoxifen at E8 . 0 did not alter Ptch1 expression , a readout of Shh signaling , in E9 . 0 somites indicating that the activity of Shh prior to cre-mediated deletion ( likely sometime between E8 . 0-E10 ) was sufficient for Ptch1 expression . As a result , these embryos had no skeletal defects in the thoracic cage; later injections also failed to generate skeletal defects . However , administration of tamoxifen at the same doses at E7 . 0 caused a discontinuity in Shh and Ptch1 expression at E8 . 0 ( see Figure 3—figure supplement 1 ) . Subsequently , embryos developed with a range of Shh KO hypomorphic phenotypes ( Figure 3 ) . Importantly , the most severe phenotypes were very similar to Shh KO animals and lacked the vertebrae and proximal ribs ( Figure 3C , F ) . The distal-most sternal ribs were present but mis-patterned , although less severely than the Shh KO animals . Furthermore , being additionally null for Casp3 resulted in failed distal rib development ( Figure 3D ) . Among the hypomorphic Shh KO animals , less severely affected animals lacked vertebrae and the proximal half of the vertebral rib , while the least affected only had abnormal vertebrae and were missing just the most proximal ends of the proximal rib ( Figure 3F–H ) . Thus , these experiments indicated that Shh signaling from the notochord and floor plate is required for normal thoracic skeletal development at stages prior to rib outgrowth and potentially even earlier . In addition , a comparison of the milder phenotypes vs . the more severe phenotypes suggests that high levels of Hedgehog signaling are required for normal proximal development and lower levels for distal development . It is well-established that somite patterning is specified by instructive signals from the surrounding tissues ( Brent and Tabin , 2002 ) with Shh from the midline being required for proper sclerotome patterning ( Fan and Tessier-Lavigne , 1994; Fan et al . , 1995; Furumoto et al . , 1999; Marcelle et al . , 1999 ) . Thus , one possibility is that the more severe skeletal phenotypes in Shh;Apaf1 DKO animals could be due to defects in somite patterning more profound than observed in Shh KO animals . We first found that compared to Shh KO embryos , Shh;Apaf1 DKO embryos were smaller ( compare panels 2G’ , H’ ) and consistently delayed at E9-E12 , ( ~3 fewer somites than average for that litter; n > 6 litters ) suggesting the Apaf1 plays a role in embryo size in the absence of Shh . Thus , we decided to compare patterning without size as a variable , by carefully stage-matching embryos by somite count and assessing for myotome and sclerotome patterning by RNA in situ hybridization . Defects in myotome development ( which could secondarily impact sclerotome ) could account for the more severe Shh;Apaf1 DKO phenotypes , however , this was not the case ( see Figure 3—figure supplement 2 ) . Or , sclerotome might never specified in Shh;Apaf1 DKO animals leading to the absence of both proximal and distal rib elements . However , while the expression domain of sclerotome markers Pax1 and FoxC2 ( Furumoto et al . , 1999; Peters et al . , 1999; Rodrigo et al . , 2003 ) are reduced in both Shh KO and Shh;Apaf1 DKO embryos , the expression profiles were similar ( Figure 3—figure supplement 2 ) . Thus , an alteration in early somite patterning does not readily explain the difference in the final skeletal phenotype between Shh KO and Shh;Apaf1 DKO animals . We next determined if it is the failure of this remaining sclerotome compartment to undergo differentiation into cartilage that distinguishes Shh KO from Shh;Apaf1 DKO animals ( schematic in Figure 4A ) . The differentiation of cartilage involves the specification of mesenchymal cells to a cartilage fate as evidenced by the upregulation of Sox9 , a master regulator of the cartilage pathway ( Akiyama et al . , 2002 ) . Cells destined to become cartilage then form aggregates which subsequently undergo compaction and condensation to form tight clusters of cells . These cells then begin to produce Aggrecan , Type II Collagen , and a specific matrix rich in acid polysaccharides detectable by alcian blue staining . Finally , the chondrogenic cells mature , slow their production in matrix , and become hypertrophic ( reviewed in [DeLise et al . , 2000] ) . We first determined if chondroprogenitors are ever specified in Shh;Apaf1 DKO embryos by examining Sox9 expression at E12 . 0 . At this stage , Sox9 expression is observed in control embryos extending from the vertebrae laterally approximately halfway around the chest ( Figure 4B ) and in Shh KO embryos in a thinner distally located reduced domain ( likely representing the precursors of the distal-most sternal ribs ) ( Figure 4C ) . Interestingly , Sox9-expressing cells are indeed present in Shh;Apaf1 DKO embryos , however , in an even thinner and shorter domain ( Figure 4D ) indicating that cartilage specification occurs in Shh;Apaf1 DKO embryos . We next determined if chondrogenic aggregates and condensations expressing Sox9 and Agc could be observed in cross-section . Aggregations and tight condensations could be seen in normal and Shh KO embryos ( Figure 4E , F;H , I ) . However , in Shh;Apaf1 DKO embryos , while some aggregation could be observed , distinct condensations were not evident ( Figure 5G , J ) . Further differentiation of cartilage involves the expression of Col2a1 , the continued expression of Sox9 , and the production of an alcian-blue-positive matrix ( Akiyama et al . , 2002 ) ( Lefebvre and Smits , 2005 ) . Col2a1 , Sox9 expression and alcian blue staining in control embryos extends from the vertebrae laterally approximately halfway around the chest ( Figure 4K , N , Q ) . Shh KO embryos exhibited staining for these markers but only in a distal portion aligned under the forearms ( Figure 4L , O , R’ , R’ ) . However , in the body wall of Shh;Apaf1 DKO embryos , no expression of Col2a1 , Sox9 , or alcian blue staining was evident ( Figure 4M , P , S , S’ ) . Thus , in summary , these assays demonstrate that in Shh KO embryos although the distal rib anlage is smaller compared to controls , differentiation proceeds normally . However in Shh;Apaf1 DKO embryos , the rib anlagen are even smaller than seen in Shh KO embryos , and while some cells turn on Sox9 and some aggregates form , they do not condense normally , and fail to differentiate . The formation of smaller aggregates in the Shh;Apaf1 DKO is at first confusing since the loss of cell death might be expected to result in increased tissue growth as has been observed in the brains of Apaf1 and Casp3 embryos ( Cecconi et al . , 1998; Yoshida et al . , 1998 ) . In contrast , while both Apaf1 KO and Casp3 KO animals have decreased cell death in the somites relative to normal , the thoracic skeleton does not appear dramatically overgrown ( Figure 1E ) . One possibility is that the normal level of cell death in the somite is not very high and so its loss does not produce overgrowth . Another possibility is that proliferation is decreased in the absence of Apaf1 , compensating for the increased survival and thus a sufficient number of cells to build the thoracic skeleton is maintained . To determine if this kind of compensation could produce the observed outcomes and also to better understand the even smaller cartilage anlage phenotype in Shh;Apaf1 DKO embryos , we decided to build an agent-based simulation using NetLogo ( Wilensky , 1999 ) to model rib outgrowth and patterning up to ~E12 . 0 . To build the simulation , we incorporated six important causal processes . These included: a Hh signal with varying concentration , variable cell death and proliferation rates , a progenitor pool that could vary in number , boundaries as would be created by surrounding tissues , and the potential effect of local cell-cell communication . We proposed that these processes predominated and therefore composed the minimal set of factors to be considered . We then represented cells as agents ( called ‘turtles’ in NetLogo ) and created an initial field of these agents randomly placed in a square to represent the location of progenitor cells within the somite . Based on our conditional ablation results ( Figure 3 ) , we assumed that cells are responsive to Hh signaling in an early time window . The field of cells was then programmed to change through time within a defined rectangular space according to simple parameters: for example , the levels , diffusion , and concentration of a Hh signal could be controlled . In addition , cell death rate and timing could be modulated along with cell proliferation and the number of agents in the initial field . A change in cell fate was simulated by changing the color of some undecided ‘cells’ to red ( for proximal ) or blue ( for distal ) at each time step with the decision determined by the amount of local Hh signal available and with an adjustable probability of conversion . To account for the reinforcement of fate by local cell-cell communication ( the ‘community effect’ [Gurdon , 1988] ) , each cell was programmed to evaluate the local distribution of red or blue cells at each time step , and to convert to the local majority color when a local super-majority of other-colored cells surrounded it . To simulate the early stages of outgrowth , the cells were programmed to move outward as they proliferated based on their degree of crowding at each time tick , with cells only moving when their local crowding was sufficiently high and then also confined in the rectangular space . When the cells hit the far right-side end of the defined space , the clock was stopped to indicate the end of outgrowth ( see Materials and methods and Supplementary file 2 for more details on the model design ) . While the agent-based simulation represents a highly simplistic scenario compared to real cells that have a particular developmental history , a complex relationship with their environment , and specific migratory properties , we were impressed to find that this small set of six processes was both necessary and sufficient to simulate a wide range of phenotypes . With the model structure established , we began to set parameters to different values based on biological insight such that different phenotypes could be generated based on a final visual outcome . It became readily apparent that in order to simulate the Shh;Apaf1 DKO , which has smaller elements than normal or even the Shh KO ( Figure 4D ) , that even in the absence of cell death , cell proliferation and/or somite size must be reduced to obtain the predicted outcome . In order to confirm this qualitative prediction and estimate more precise parameter settings , we returned to our biological samples to measure somite size and proliferation in the different genotypes . We therefore analyzed somite-matched E9 . 0 embryos when multiple somites are readily visible and the forelimb bud has begun to emerge providing an internal landmark across all genotypes . Embryos were analyzed for size and for the expression of phosphorylated histone H3 ( pHH3 ) , an indicator of cells in mitosis ( Figure 5A–D ) . To obtain an estimate of the relative size and proliferation rates for the different genotypes , we created a hierarchical Bayesian measurement model ( see Supplementary file 2 ) . In line with the observed data points ( Figure 5E , top graph ) , we discovered that any difference in size between normal and Apaf1 KO embryos was negligible and that both Shh KO and Shh;Apaf1 DKO were noticeably reduced in size ( Figure 5E , bottom graph ) . Furthermore , Apaf1 KO embryos had a reduced proliferation rate which was further reduced in the absence of Shh ( Figure 5F ) . This confirms that the loss of Apaf1 does result in a compensatory decrease in proliferation and that the effect appears to interact with being null for Shh to further reduce the proliferation rate in Shh;Apaf1 DKO embryos . Using this analysis , we then chose parameter values for proliferation and initial size for all genotype simulations equal to the median posterior sample value from the Bayesian model . Using these values ( see Table 1 and Figure 6—figure supplements 1 and 2 ) , we observed that the final pattern of the simulations visually matches the expected phenotypic outcomes ( Figure 6A–C ) . Values different from normal are shaded . Values for size and proliferation rate were chosen based on the median posterior distribution from a Bayesian model , see Supplementary file 1 . Using an initial fixed size and our newly defined baseline parameters , we found that the simulation could represent the response to a Hh concentration gradient reliably with the percent of proximal red and distal blue cells dependent on the Hh dose and gradient steepness ( Figure 6B and Figure 6—figure supplement 1 ) . In addition , when varying the initial size , with fixed morphogen parameters , smaller initial sizes are more proximalized as a larger percentage of the cells are under the influence of high Hedgehog signaling ( Figure 6—figure supplement 2 ) . Interestingly , the generation of a highly distinct border between the two elements did require that cells be programmed to sense the fate of local cells within their environment ( the ‘community effect’ ) as predicted by early studies on muscle differentiation ( Gurdon , 1988 ) ( Figure 6—figure supplement 2 ) . In addition , alterations in the rate of cell death or of proliferation , while profoundly impacting overall size , did not have a large effect on the ratios of cells that contributed to a proximal vs . distal element ( Figure 6—figure supplement 3 ) . Thus , our model displays robust reproduction of a wide range of observed phenomena ( see Video 1 for simulations for the different genotypes ) . As observed in many other contexts ( Briscoe and Small , 2015 ) , the role of Hh signaling is multi-functional and likely required for many processes during rib development including: ( 1 ) sclerotome induction , ( 2 ) maintenance of cell survival , and ( 3 ) proximal-distal specification . It remains an open question as to when proximal-distal specification occurs during early skeletal development . Two competing hypotheses , based on studies in limb development , include the Early Specification Model and the Progress Zone Model ( reviewed in [Mariani and Martin , 2003; Bénazet and Zeller , 2009] ) . In the context of the rib , we favor an early specification scenario where proximal-distal specification occurs when the cells have not yet migrated long distances away from the midline and the progenitor pools are quite small . Therefore , the transport of the Hh protein can occur over a short period of time and transport over long distances is not required to explain the results ( Phase 1 ) . Then as the embryo grows , the specified compartments would expand laterally ( Phase 2 ) ( Figure 7 ) . In recent years , this idea of early proximal-distal specification followed by expansion to establish the initial broad proximal and distal fields has been gaining traction in the field of limb development ( [Roselló-Díez et al . , 2011; Cunningham and Duester , 2015] and reviewed in Mariani [2010] ) and seems easily applicable in the case of rib patterning . Support for a role for Shh signaling at very early somite stages ( prior to E9 . 0 ) , comes from our experiments where tamoxifen administration at E7 . 0 resulted in thoracic skeletal defects similar to Shh KO embryos ( Figure 3 ) , while administration after E8 . 0 or later was not sufficient to generate a thoracic phenotype . These results also align with previous studies suggesting that Hh signaling is only initially required at presomitic mesoderm stages and that subsequently BMPs are important in axial skeletal growth to maintain a chondrogenic regulatory loop ( Zeng et al . , 2002; Stafford et al . , 2011 ) . In summary , Hh signaling plays several major roles—influencing the number of cells induced to become sclerotome , reducing the number of cells undergoing cell death , and causing rib progenitors to adopt a proximal vs . distal fate . The effect of Hh is likely early , during somite stages , and relatively short-range . Indeed , within our agent-based simulations , the length scale over which Hh concentration can influence cell fate is approximately the same as the size of the progenitor pool and most cells ‘decide’ their fate relatively early . We suggest that cells making late decisions become distal because they experience a very low Hh signal or become proximal due to influence from surrounding proximal cells . Considering the prevalence of programmed cell death during development ( Hardy et al . , 1989 ) , the relatively minor developmental defects seen when null for apoptosis genes has been surprising . One explanation has been that apoptosis genes can compensate for each other and/or that multiple genes need to be removed in order to see profound phenotypes ( Nagasaka et al . , 2010 ) . Another possibility is that cell death still occurs but by other cell death pathways ( Yuan and Kroemer , 2010 ) . However , another explanation is that the loss of these genes does block normal cell death and that the proper cell number is re-established by compensatory mechanisms that decrease cell numbers such as decreased proliferation ( Cecconi et al . , 2004; Huh et al . , 2004 ) . In the context of rib development , in the absence of Apaf1 , somite cell death is inhibited . When modeling this in our agent-based simulation , the proliferation rate must be moderately reduced to compensate for reduced cell death in order to achieve normal skeletal size . This prompted us to look more carefully at the proliferation rate and we did in fact see that it is decreased when Apaf1 is lost ( Figure 5 ) . We suggest that further perturbations ( loss of Shh ) influences the initial sclerotome cell number and this combined with a further reduction in proliferation rate leads to the more severe defects seen in the Shh;Apaf1 DKO embryos . An interesting question for future research is determining the mechanism by which loss of Apaf1 results in reduced proliferation . Two potential mechanisms are ( 1 ) the direct abrogation of the cell cycle machinery when the apoptosis pathway is blocked or ( 2 ) the indirect effect of a reduced number of dying cells that can release growth factors . Compensatory proliferation with dying cells releasing growth factors into their environment has been observed in other contexts ( Fan and Bergmann , 2008b; Jäger and Fearnhead , 2012 ) . Studies from Drosophila reveal that the signaling pathways involved in compensatory proliferation differ depending on the tissue and the developmental state of the tissue . Highly proliferating tissues have been shown to induce Tgfβ and Wnt homologues ( Pérez-Garijo et al . , 2004; Ryoo et al . , 2004 ) , while differentiating tissues activate Hh signaling ( Fan and Bergmann , 2008a ) . In the sclerotome , dying cells could be releasing growth factors that are sufficient to maintain the proliferation of the somite . In our studies , an increase in cell death in the Shh KO does correlate with a possible small increase in proliferation above normal ( Figure 5 ) . Then , in the absence of cell death ( Apaf1 KO ) , perhaps these growth factors are not released and this is a mechanism by which proliferation rate is decreased from normal . Based on our combined genetic analysis and agent-based simulations , it can be useful to think of rib development in terms of two main phases ( Figure 7 ) . During Phase 1 , which starts as the presomitic mesoderm is forming into somites ( E8 . 0-E10 . 0 ) , the dominating activities are the induction of sclerotome and the establishment of proximal/distal cell identity mediated by Hh signaling . While during Phase 2 ( E10 . 0 and onwards ) , the dominating activity is expansion as cell identity is maintained . The transition between phases likely occurs gradually as the skeletal elements increase in size . The ultimate size of a skeletal element at birth then , is determined by the number of cells left after Phase 1 , along with an expansion of that population during Phase 2 . Based on our agent-based simulations , the proliferation rate not only influences cell numbers early but , during Phase 2 can also have a profound impact on segment size because cell number increases exponentially . In the Shh;Apaf1 DKO embryos , the decrease in transient cell death likely has a minor contribution compared to the decrease in sclerotome induction ( due to loss of Hh signaling ) and the decreased proliferation rate throughout somite development ( due to loss of Apaf1 ) . Even if cartilage specification occurs , successful differentiation is still needed to achieve the final skeletal pattern . In Shh KO embryos , even though there are few cells participating , they are still able to aggregate , condense , and differentiate into matrix-producing cartilage tissue . However , in Shh;Apaf1 DKO embryos , while sclerotome is present and some cells begin to express Sox9 , the population pool is even smaller , condensation fails to occur and differentiation does not proceed ( Figure 4 ) . While cooperation of both Hh signaling and Apaf1 could be required for chondrocyte differentiation , another possibility is that chondrogenesis fails due to an insufficient number and density of cells . In vitro studies have shown that robust proliferation along with high cell number and density are very important for the production of cartilage matrix and chondrocyte differentiation ( Denker et al . , 1999; DeLise et al . , 2000; Malko et al . , 2013 ) . Thus , differentiation in vivo may also be highly dependent on these characteristics . In Shh;Apaf1 DKO embryos , a smaller initial size , along with decreased proliferation could result in a density of Sox9-expressing cells that is insufficient for differentiation to occur and thus condensation fails resulting in a more severe phenotype . The requirement for a sufficient number of cells may also account for the variation in distal rib phenotypes seen in Shh KO embryos ( Figure 1 ) . A high cell density may result in the production of growth factors at sufficient levels for differentiation . In particular , BMP and TGFβ pathways play prominent roles in differentiation ( Wang et al . , 2014 ) with the cartilage cells themselves producing the critical ligands during pre- and early condensation stages . BMP signaling is required for chondroprogenitor aggregation and condensation in vitro ( Barna and Niswander , 2007 ) and in vivo via smad1/5/8 signaling ( Pizette and Niswander , 2000; Retting et al . , 2009 ) . The application of TGFβ protein in vitro to promote differentiation has been long appreciated ( DeLise et al . , 2000 ) , while TGFβsignaling via smad2/3 has been shown to be required for the progression of chondroprogenitors toward differentiation in vivo ( Wang et al . , 2014 ) . Interestingly , smad4 , the common smad protein involved in both the BMP and TGFβ response , has been shown to be required specifically within pre-cartilage cells . Without Smad4 , Sox9-expressing cartilage cells fail to condense , fail to undergo differentiation , and any condensations that do form are not maintained . In addition , in the absence of the ability to respond to differentiation signals , they appear to adopt a connective tissue fate ( Bénazet et al . , 2012 ) . Thus , BMPs and TGFβ along with other pathways ( reviewed in ( DeLise et al . , 2000 ) are likely players in the density-dependent differentiation of chondroprogenitors . Interestingly , this autocrine mechanism for cartilage differentiation has been modeled by Newman and colleagues ( Christley et al . , 2007 ) . Using a discrete grid-based 2D multiscale computational model , they simulate cartilage condensation and differentiation considering cell number , cell behavior , and importantly TGFβ-mediated cell-cell signaling . As our model ends prior to differentiation , this model could follow nicely as a sequel . Taken together , we suggest that during rib skeletal development , cells respond to local cues in a way that can be usefully modeled through a series of simple rules . During normal development , we propose that a compensatory feedback loop between the apoptotic pathway and cell proliferation plays a critical role in achieving the proper number of skeletal progenitor cells for cartilage differentiation . The precise molecular mechanism that underlies this balance may involve growth factor signaling but this is still to be discovered . Likewise , it will be important to determine if compensatory proliferation occurs in other embryonic contexts when programmed cell death is blocked . One of the most interesting issues for future investigation is to discover the molecular mechanism by which Hh concentration specifies proximal and distal identity and importantly the timing during which this happens . In addition , the precise mechanism by which refinement of pattern occurs is unknown and future studies could determine whether local signaling from the surrounding majority cells ( community affect ) plays a role or whether some other mechanism such as reciprocal inhibition or cell re-arrangement and assortment is involved . Beyond the specifics of rib development , the use of mathematical and computational models can be extremely useful for making testable predictions . In particular , an agent-based approach allows the modeling of developing tissues while taking into consideration the behavior and fate decisions of individual cells based on local information . Furthermore , agent-based modeling is compatible with other classical differential equation-based modeling , such as reaction-diffusion and finite-element modeling which have been previously used to model limb development and other tissues ( Zhang et al . , 2013; Lau et al . , 2015 ) . In the future , these combined methods could be used to generate more precise multi-scale models ( Zhu et al . , 2010 ) which could be used not only to better understand the emergence of pattern in laboratory organisms but also to understand how changes in cell behavior during evolution could generate new and diverse forms among different species . To generate Apaf1 and Shh double null embryos , females heterozygous for Apaf1 ( RRID:MGI:3783548 ) and homozygous for a Shh ‘floxed’ conditional allele ( RRID:IMSR_JAX:004293 ) ( Apaf1-/+; Shhfl/fl ) ( Yoshida et al . , 1998; Lewis et al . , 2001 ) were crossed with males ubiquitously expressing the CRE enzyme ( RRID:IMSR_JAX:003376 ) and heterozygous for both Apaf1 and Shh ( Actb-CRE; Apaf1-/+; Shh-/+ ) leading to the production of Shh:Apaf1 DKO embryos at a 1 in 8 frequency ( 12 . 5% ) . Heterozygous embryos were used as controls . A standard cross ( Shh+/-; Casp3+/- X Shh+/-; Casp3+/- ) was established to create Shh;Casp3 DKO embryos at a 1 in 16 frequency ( 6 . 25% ) ( RRID:IMSR_EM:06370 ) ( Woo et al . , 1998 ) . Conditional Shh;Apaf1 DKO or Shh;Casp3 DKO mice were produced using a Tamoxifen-inducible Foxa2-CRE-ERT2 line ( RRID:IMSR_JAX:008464 ) ( Park et al . , 2008 ) . Tamoxifen ( Sigma ( St . Louis , MO ) ; 3 mg/mouse ) and progesterone ( Sigma; 1 . 5 mg/mouse ) were administered by intraperitoneal injection 7 . 0–10 . 0 days of gestation . Embryos were genotyped by PCR . All animal procedures were carried out in accordance with approved Animal Care and Use Protocols at the University of Southern California . An Agent-Based Model ( ABM ) was developed using the NetLogo system version 6 . 0 ( Wilensky , 1999 ) . The model operates by initially creating a field of agents representing cells ( called ‘turtles’ in NetLogo ) and then evolving this field through time according to simple rules . A varying concentration of secreted Hh is modeled by using a Gaussian type curve with peak at x = −10 and a variable width and height . The agents are initially randomly placed in a square block between x = 0 and x = sqrt ( N/1200 ) 14 and y = ± ( sqrt ( N/1200 ) 14 ) /2 . They divide with base probability of 0 . 05 per time step and die with base probability 0 . 05 per time step , and the death rate is multiplied by a variable-relative rate and a time-varying curve using a Gaussian with a variable-duration width in time . At each cell division , there is a probability to convert from yellow ( unspecified ) to red ( proximal ) or blue ( distal ) and this is based on the local concentration of Hh . Once converted ( biased in fate ) , ‘cells’ ignore the Hh concentration ( mimicking the window of competence to respond ) , but at each time point , the local concentration of red or blue cells is assessed and the cell is programmed to convert to the local majority color when a local super-majority of other-colored cells surround it . As the cells proliferate , the cells spread away from regions of high local density maintaining a density of 4–6 cells per patch . The net effect is that the cell field expands as the cells divide . The cells are prevented from moving through the boundary of the spatial field . When the cell field hits the farthest column of patches the clock is stopped , otherwise the simulation continues for the specified number of time ticks to models a slowing in growth rate as structures reach the size limit of the animal . Without this boundary effect the number of cells would grow exponentially in time and without bound . By varying the initial cell field size , the Hh concentration in space , the time duration , the coefficients that determine yellow-red and yellow-blue conversion probabilities , the cell death intensity , the proliferation rate , and the duration of the cell death in time , we can model a variety of conditions . See Supplementary file 2 for further details on the model . Code to run the simulations can be found in Source code 3 and run after downloading the free NetLogo program at: https://ccl . northwestern . edu/netlogo/ .	During animal development , the ribs grow from the back of the embryo around towards the chest . In fish , these bones simply terminate . Yet in land animals , cartilage forms at the end of the rib where it connects to the breastbone , or sternum . This encloses the chest cavity . Fogel , Lakeland et al . have now asked how the progenitor cells that develop into the ribs form these two skeletal elements – the bone element and the cartilage element – in land animals . Their approach involved genetic analysis in mice and a simple computing model . It revealed that two elements could form if the progenitor cells decide which element they will belong to based on the concentration of the diffusible protein called Hedgehog . This protein controls many aspects of animal development , and higher concentrations seem to bias the cells in a developing rib toward belonging to the bone element . Fogel , Lakeland et al . propose that this decision is locked-in early , before the rib grows outward and becomes more refined . An analysis using this simple model reproduces all the basic observations seen in the experiments with mice . The model also explains how processes like cell division and cell death control the growth of developing skeletal elements . These modeling techniques can be applied to many fields within biology , including research into the causes of birth defects , the mechanisms of tissue repair , and the evolution of skeletal diversity . An advantage to this modeling technique is that it uses only the information in each cell’s local environment to make decisions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2017	A minimally sufficient model for rib proximal-distal patterning based on genetic analysis and agent-based simulations
Bioluminescence imaging ( BLI ) is ubiquitous in scientific research for the sensitive tracking of biological processes in small animal models . However , due to the attenuation of visible light by tissue , and the limited set of near-infrared bioluminescent enzymes , BLI is largely restricted to monitoring single processes in vivo . Here we show , that by combining stabilised colour mutants of firefly luciferase ( FLuc ) with the luciferin ( LH2 ) analogue infraluciferin ( iLH2 ) , near-infrared dual BLI can be achieved in vivo . The X-ray crystal structure of FLuc with a high-energy intermediate analogue , 5’-O-[N- ( dehydroinfraluciferyl ) sulfamoyl] adenosine ( iDLSA ) provides insight into the FLuc-iLH2 reaction leading to near-infrared light emission . The spectral characterisation and unmixing validation studies reported here established that iLH2 is superior to LH2 for the spectral unmixing of bioluminescent signals in vivo; which led to this novel near-infrared dual BLI system being applied to monitor both tumour burden and CAR T cell therapy within a systemically induced mouse tumour model . Bioluminescence imaging ( BLI ) is used extensively for the sensitive , longitudinal and high-throughput monitoring of biological processes in vivo ( Xu et al . , 2016; Paley and Prescher , 2014; Mezzanotte et al . , 2017; Yao et al . , 2018; Yeh and Ai , 2019 ) . Bioluminescence light emission is produced through the catalysis of a small molecule substrate , most commonly D-luciferin ( D-LH2 ) , by a luciferase enzyme . The mutagenesis of bioluminescent enzymes has improved the sensitivity and accuracy of BLI in small animals ( Branchini et al . , 2010 ) , ( Iwano et al . , 2018 ) . However , despite its widespread use in scientific research , BLI is still largely restricted to tracking a single parameter in vivo . The ability to discretely monitor two biological parameters ( dual-BLI ) simultaneously within a single animal is highly desirable ( Xu et al . , 2016 ) with potential uses ranging from the monitoring of tumour burden alongside cellular therapy , to the visualisation of dynamic biological processes such as protein-protein interactions ( Prescher and Contag , 2010 ) . Previous approaches to dual-BLI have been disappointing . The use of multiple bioluminescent proteins which catalyse different substrates is the most frequently used method but suffers from a number of limitations . Sequential substrate administration is normally required , in addition this method commonly employs a combination of a coelenterazine and a D-LH2 utilising luciferase ( with the blue emission from the former being heavily absorbed compared to the yellow-green emission from the latter ) ( Maguire et al . , 2013 ) , ( Stacer et al . , 2013 ) . Differences in biodistribution and reaction kinetics of two substrates can make image co-registration and interpretation difficult . The development of orthogonal luciferase-luciferin pairs has solved some of these limitations but this approach still requires multiple substrate administrations ( Rathbun et al . , 2017 ) . An ideal dual-BLI approach would use two spectrally distinct bioluminescent proteins utilising a single substrate followed by spectral unmixing of the signal . However , this approach is not currently feasible using LH2 . Although luciferases can be mutated to alter the colour of their emission , a limit appears to have been reached for mutational colour modulation of firefly and related luciferases . The most red-shifted of these enzymes have maximal peak emissions between 610 and 620 nm ( Branchini et al . , 2010 ) . This is insufficient for dual BLI in vivo . Due to the differential attenuation of light by biological tissue spectral unmixing of a red-shifted luciferase paired with a standard or green-shifted enzyme is challenging , especially in deeper tissue models ( Mezzanotte et al . , 2011 ) . Shifting the emission of both enzymes into the near infrared must be achieved to allow adequate unmixing . To further red-shift peak emission we and others have turned to chemical modification of the D-luciferin ( LH2 ) substrate ( Adams and Miller , 2014 ) . We recently described the LH2 analogue infraluciferin ( iLH2 ) which has a luciferase dependent red-shifted peak emission of up to 706 nm ( Jathoul et al . , 2014 ) ( Figure 1a ) . We hypothesised that combining colour modulation of bioluminescence through mutagenesis of the FLuc protein along with red-shifting bioluminescence by chemical modification of LH2 would allow dual-BLI , an approach that has not been described previously . In this work , we explored the possible structural interactions in the enzyme that may account for the near infrared emission of iLH2 and its application to dual-BLI in vivo . First , the X-ray crystal structure of FLuc in complex with a high-energy intermediate analogue , 5’-O-[ ( N-dehydroinfraluciferyl ) -sulfamoyl] adenosine ( iDLSA ) was determined to provide insight into the FLuc-iLH2 light-emitting reaction . Next , we selected a pair of stabilised colour-shifted FLuc mutants , which emit with a 20 nm separation in peak emission wavelength with iLH2 in the near infrared . We demonstrated the ability to spectrally unmix these two biological signals in vivo at depth using iLH2 . Finally , we show a proof-of-concept of utility using this novel dual imaging technique to longitudinally monitor both tumour burden and chimeric antigen receptor ( CAR ) T cell therapy within a single animal model . To help understand the red shift in bioluminescence emission from FLuc that is observed in its reaction with iLH2 , the X-ray crystal structure of FLuc in complex with iDLSA was resolved and is shown in Figure 1b ( PDB ID: 6HPS ) . Data collection and refinement statistics ( molecular replacement ) ; and data collection , phasing and refinement statistics for mad ( semet ) structures can be found in Figure 1—figure supplement 1 . iDLSA captures FLuc in the adenylation step of the light emitting reaction ( 1H and 13C data spectra synthetic chemical compounds can be found in ) . The conformation of the iLH2 heterocyclic rings with respect to the alkene , as drawn in Figure 1a , is confirmed to be as predicted by computational studies and is the most likely conformation of the light emitting form ( Berraud-Pache and Navizet , 2016 ) . This newly crystallised FLuc structure was aligned with the reported structure of FLuc with 5’-O-[ ( N-dehydroluciferyl ) -sulfamoyl] adenosine ( DLSA ) ( PDB ID: 4G36 ) ( Sundlov et al . , 2012 ) . The structures show good alignment to each other , however there is evidence of a more open active site supported by a reduction in root-mean-squared ( RMSD ) score when aligned based on just the N-terminal domain of FLuc rather than the entire structure ( RMSD = 0 . 688 and 0 . 783 respectively ) ( Figure 1c ) . All FLuc residues in close proximity ( 4 Å ) to DLSA were also found to be within the same distance to iDLSA , with the exception of Arg437 , >4 Å away from iDLSA ( Figure 1d and e ) . We noted that despite differences in the conformation of iDLSA compared to DLSA in both 4G36 and L . cruciata 2D1S ( Nakatsu et al . , 2006 ) structures , the positions of the phenolic groups are quite similar ( ~0 . 5 Å ) . The altered position of the benzothiazole ring and the greater size of iDLSA may be the cause of a series of small active site changes that affect residues Glu311 , Arg337 , Asn338 , Gly339 , and Thr343 resulting in a total of six differences in H-bonding interactions . When specific residues implicated in the light emitting reaction ( Sundlov et al . , 2012 ) were measured between the two structures differences ranged from 0 . 7 to 1 . 6 Å; with the biggest divergence being Lys529 ( found in the C-terminal cap ) which had a 2 . 4 Å difference in the nitrogen residue found in the side chain of the amino acid ( Figure 1f ) . The resulting increase in active site polarity due to the rotation of the C-terminal cap , if maintained during the light emitting conformation , could contribute to the red-shift in light emission ( Nakatsu et al . , 2006 ) , in addition to the increased π-conjugation through the chemical structure of the emitter . This X-ray structure will help the future design of more efficient FLuc-iLH2 pairs . A range of colour-shifted , thermo- and pH stable FLuc mutants were spectrally characterised in vitro with a comparative selection of LH2 analogues proven to red-shift bioluminescence emission ( CycLuc1– Evans et al . , 2014; Aka-Lumine-HCL – Kuchimaru et al . , 2016; and iLH2– Jathoul et al . , 2014 ) . Two new luciferins NH2-NpLH2 and OH- NpLH2 have also been shown to have near infrared emissions ( Hall et al . , 2018 ) but these were reported too late to include in this study . FLuc mutants were engineered to combine mutations reported to provide superior stability ( Jathoul , 2012 ) and colour-shifting capability ( Branchini et al . , 2005 ) ( stabilising and colour shifting FLuc mutations are detailed in Materials and methods ) . The Raji B lymphoma cell line engineered to express a FLuc mutant were spectrally imaged after addition of each substrate . These cell lines were subsequently used for all in vitro and in vivo testing . Both CycLuc1 and Aka-Lumine-HCL showed a consistent red-shift in peak bioluminescence emission wavelength to ~600 nm and ~660 nm respectively for all FLuc mutants , making these substrates unsuitable for dual colour BLI ( Figure 2—figure supplement 1 ) . The data confirmed that with LH2 both FLuc_natural and FLuc_green have a peak emission of ~560 nm , whilst FLuc_red has a peak emission of ~620 nm ( Figure 2—figure supplement 1 ) ( Jathoul , 2012 ) , ( Branchini et al . , 2005 ) . When tested with iLH2 all FLuc mutants were shifted >100 nm into the near infrared but maintained their relative spectral shift [FLuc_green ~ 680 nm , FLuc_natural ~ 700 nm and FLuc_red ~ 720 nm ( Figure 2—figure supplement 1 ) ] . From this , we progressed further with two FLuc mutants , FLuc_green and FLuc_red to explore their utility for dual-BLI . The ability to spectrally unmix FLuc_green and FLuc_red ( Figure 2a ) in vitro was investigated by mixing the two FLuc_mutants expressed in the Raji B lymphoma cell line at various ratios followed by spectral imaging and unmixing with both LH2 and iLH2 ( Figure 2b ) . As would be expected from accurate spectral unmixing , the top wells were classified as containing mostly FLuc_green signal , which gradually decreased down the plate in line with the decreasing proportions of FLuc_green expressing cells , with the bottom wells being largely classified as FLuc_red signal for both LH2 and iLH2 . The percentage unmixed signal of FLuc_green and FLuc_red was plotted for each ratio of FLuc expressing cells ( Figure 2c ) . Correlation analysis was performed on this data comparing input cellular proportions with unmixed signal , giving R2 values of 0 . 9983 and 0 . 9972 for LH2 and iLH2 respectively . Even though all 18 bandpass filters equipped on the IVIS Spectrum were utilised for spectral unmixing in this in vitro testing , we appreciate that not all potential users of this novel dual bioluminescence methodology will have access to machines with such a wide selection of bandpass filters . Therefore , further analysis of our data showed that spectral unmixing could be achieved with high accuracy just using a subset of filters . For LH2 , the use of 3 bandpass filters ( 500 nm , 660 nm , 820 nm ) gave an R2 value of 0 . 9958; For iLH2 , the use of 3 bandpass filters ( 600 nm , 700 nm , 800 nm ) gave an R2 value of 0 . 9937 . The highest accuracy of spectral unmixing we could achieve using just two filters were R2 values of 0 . 9776 and 0 . 9775 for LH2 ( 500 nm and 720 nm ) and iLH2 ( 600 nm and 720 nm ) , respectively . Additionally , an experiment was carried out where FLuc_green and FLuc_red have been expressed at different levels in the same cell . Spectral bioluminescence imaging and unmixing has subsequently been performed to successfully reflect these differing expression levels with both LH2 and iLH2 ( Figure 2—figure supplement 2 ) . This spectral imaging data show that both LH2 and iLH2 can be used for dual bioluminescence reporting in vitro . To investigate the use of FLuc_green and FLuc_red with LH2 and iLH2 for in vivo dual BLI three NOD scid gamma ( NSG ) tumour models , representing increasing tissue depth ( subcutaneous , systemic and intracranial ) , were established with the Raji B lymphoma cell line expressing either FLuc_green or FLuc_red ( as described for in vitro experiments ) . All tumour models were then spectrally imaged with both LH2 and iLH2 ( Figure 3 , Figure 3—figure supplement 1 and Figure 3—figure supplement 2 ) to obtain the normalised spectra and average radiance of each FLuc mutant with both luciferins in all three in vivo models ( Figure 4 ) . The normalised bioluminescence spectra for every mouse in each model when imaged with LH2 is shown , with the total radiance for each mouse plotted to the right of the spectral plot ( Figure 4a–c ) . The data show that when imaged with LH2 both FLuc_green and FLuc_red had an average peak emission between 610–630 nm; meaning the peak emission for FLuc_red is maintained as in vitro whereas the peak emission of FLuc_green is red shifted by ~60 nm in vivo . FLuc_green also exhibited a bimodal spectral distortion , with a minor peak at ~560 nm . In contrast to LH2 , FLuc_green and FLuc_red had a ~ 20 nm separation of average peak emissions in all three animal models with iLH2 ( FLuc_green ~ 700 nm and FLuc_red ~720 nm ) ( Figure 4d–f ) . In addition to the separation of peak emission wavelengths when imaged with iLH2 in vivo , the relative intensities of FLuc_green and FLuc_red were more comparable when imaged with iLH2 than with LH2; With LH2 FLuc_red had an average radiance that was 42 ( subcutaneous ) , 4 . 12 ( systemic ) and 7 . 28 ( intracranial ) times brighter than FLuc_green ( Figure 4a–c ) . Whereas , when imaged with iLH2 the average radiance between FLuc_green and FLuc_red was 1 . 38 ( subcutaneous ) , 1 . 2 ( systemic ) and 1 . 51 ( intracranial ) times different ( Figure 4d–f ) . No statistically significant difference in relative intensities between FLuc_green and FLuc_red was found in tumour models imaged with iLH2 ( p=0 . 3414 , 0 . 4594 and 0 . 6153 for the subcutaneous , systemic and intracranial tumour models respectively , T test ) . This comparability of relative intensities between FLuc_green and FLuc_red with iLH2 means that if used as genetic reporters for dual imaging , the dynamic range of radiance values for both enzymes will be more similar , therefore giving a more accurate comparison of the processes being monitored . To validate the ability to spectrally unmix FLuc_green and FLuc_red in vivo with iLH2 a systemic Raji tumour model was established . Raji cell lines expressing the FLuc mutants were mixed in the following ratios: 90:10 , 75:25 and 50:50 for FLuc_green: FLuc_red and vice versa . After spectral BLI with both substrates , animals were sacrificed and the bone marrow was extracted for flow cytometry analysis to confirm the proportions of engrafted Raji FLuc populations ( representative examples of flow cytometry plots can be found in Figure 5—figure supplement 1 ) . Spectral unmixing was performed using Living Image ( Perkin Elmer ) by creating library spectra of FLuc_green and FLuc_red with both LH2 and iLH2 , established from the pure expressing populations obtained during in vivo spectral characterisation ( Figure 4 ) . Output images with both LH2 ( Figure 5a ) and iLH2 ( Figure 5b ) were generated for FLuc_green and FLuc_red , as well as a composite image for each substrate . The percentage signal unmixed as FLuc_green and FLuc_red with both LH2 ( Figure 5c ) and iLH2 ( Figure 5d ) was determined and correlated to the percentage population of each FLuc_mutant within the Raji cell population taken from extracted bone marrow samples and analysed using flow cytometry ( Figure 5e ) . A correlation of 0 . 99 was found with iLH2 ( R2 value , SD = 0 . 01 ) . LH2 had a correlation of 0 . 89 ( R2 value , SD = 0 . 06 ) , which was significantly different from the R2 values obtained by flow cytometry ( p<0 . 0001 , ONE-Way ANOVA with post hoc Tukey test ) . No significant difference was found between R2 values determined by flow cytometry and unmixed bioluminescence signal using iLH2 . Additionally , significant differences were found between the percentage unmixed signal using LH2 and cellular proportions determined by flow cytometry , with p values of < 0 . 0001 , 0 . 003 , 0 . 0042 and 0 . 0056 for the 90% , 75% , 50% and 25% FLuc_green conditions respectively . No significant difference was found between percentage unmixed signal using iLH2 and cellular proportions determined by flow cytometry , except for the 90% FLuc_green condition ( p=0 . 0257 ) . Therefore , iLH2 is superior to LH2 for in vivo dual reporting applications . The characterised and validated in vivo dual BLI system using FLuc mutants in combination with iLH2 was then applied to track tumour burden and CAR T cell therapy within the same animal model . A Raji B lymphoma tumour cell line expressing FLuc_green was used , and healthy human donor T cells were engineered to express CD19 CAR and FLuc_red linked via a 2A peptide . Tumour cells were first systemically engrafted , followed by administration of CAR T cells 8 days later . A control animal received tumour only . Spectral BLI using iLH2 was then performed at 3 , 4 and 6 days post CAR T cell administration on the IVIS Spectrum . Output images for the unmixed FLuc_green and FLuc_red signal , as well as the composite of the two unmixed images , for day six post CAR T cell administration were generated ( Figure 6a ) . The radiance values from the unmixed FLuc_green and FLuc_red images for both tumour control and treatment ( tumour + CAR T cells ) at 3 , 4 and 6 days post CAR T cell treatment were determined ( Figure 6b ) . As would be expected , the tumour only control showed consistently high levels of FLuc_green signal which increased overtime compared to FLuc_red , which is likely due to the close proximity of Fluc_red photons emitted by the neighbouring treatment mouse . However , the CAR T cell treatment mice initially had a higher proportion of FLuc_green signal , which was then surpassed by FLuc_red by the final imaging . This represents the homing of CAR T cells expressing FLuc_red to the FLuc_green expressing Raji B lymphoma tumour , followed by the expansion of the CAR T cells and the reduction in the tumour growth in the treatment mice . Small animal dual ( or multi-parameter ) bioluminescence is highly desirable . Currently , most dual BLI has been achieved with two luciferases each utilising a different substrate , one of which is normally the auto luminescent coelenterazine ( Maguire et al . , 2013 ) . A simpler approach would be to use a single LH2 substrate and two firefly ( or related ) luciferases which emit at different wavelengths . This in theory has advantages of higher quantum yield and a more favourable substrate . Such an approach has been attempted: for instance Mezzanotte et al tested dual BLI in vivo using LH2 with the green luciferase ( CBG99 ) with the red luciferase ( Ppy RE8 ) ( Mezzanotte et al . , 2011 ) . However , in deeper tissues the shorter wavelength component of green emitting enzyme emission is heavily attenuated by mammalian tissues leaving an ‘in vivo spectrum’ which is almost indistinguishable from that of the red luciferase . To illustrate this , and as a control for subsequent experiments , we attempted dual BLI using a pair of stabilised firefly luciferases , which are green/red shifted to 546 nm and 610 nm respectively using LH2 as a substrate ( Branchini et al . , 2007 ) . Whilst in vitro the spectra could be easily distinguished ( Figure 2 ) in vivo the spectral separation between the two FLuc enzymes was lost due to the differential attenuation of the green enzyme by biological tissue ( Figure 4 ) . An obvious solution is to red-shift both enzymes into the optic window while maintaining an adequate separation . However , although bioluminescence emission has been successfully red-shifted through mutagenesis of firefly luciferase the structure of the LH2 substrate ultimately limits this approach . Dual BLI with a single substrate in the near-infrared should be better; this has been described using BRET based reporters with an coelenterazine derived substrate ( Rumyantsev et al . , 2016 ) . To move bioluminescence into the near infrared modification of the LH2 substrate was required . Indeed , red-shifted luciferin analogues have been reported . These include CycLuc1 and Aka-Lumine , however neither permit variation in emission spectra with different Luciferase mutants ( Evans et al . , 2014 ) , ( Kuchimaru et al . , 2016; Iwano et al . , 2018 ) ( Figure 2 ) . More recently , two Naphthyl luciferins , NH2-NpLH2 and OH-NpLH2 , were reported to red shift the in vitro bioluminescence emission of CBR to 664 nm and 758 nm respectively ( emission max 614 nm with LH2 ) . Interestingly , NH2-NpLH2 was shown to have an in vitro peak emission of 730 nm with an optimised version of CBR ( CBR2 ) . The potential application of CBR and CBR2 with NH2-NpLH2 for dual-BLI in vivo was not explored; and the reported broad emission spectrum of optCBR2 in live cells is likely to make any near-infrared dual BLI with NH2-NpLH2 challenging , however this approach cannot be discounted ( Hall et al . , 2018 ) . We described previously iLH2 , which has a near-infrared emission and in contrast to other LH2 analogues maintains the 6’ hydroxyl benzothiazole group of LH2 that preserves the colour-shifts of mutant luciferases ( Jathoul et al . , 2014 ) . Analysis of the x-ray crystal structure of FLuc in complex with the iLH2 analogue ( iDLSA ) revealed that the hydrogen-bonding network , thought to be critical in stabilising the phenolate ion of the emitter , is disrupted in the FLuc-iDLSA due to the accommodation of the larger iLH2 substrate ( Branchini et al . , 2017 ) . This enables full charge delocalization of the phenolate ion , in addition to extended conjugation , resulting in a red-shifted emission as suggested by homology modelling experiments with the recently described near infrared emitting naphthyl analogue ( Hall et al . , 2018 ) . Evidence for the disruption of the network is the absence of a H-bond between Arg337 and Glu311 that is found in the DLSA structures of FLuc and L . cruciata Luc . This H-bonding interaction has been proposed to be important for stabilising an active site conformation for green light emission ( Viviani et al . , 2005 ) . In addition , the resulting increase in active site polarity due to the rotation of the C-terminal cap could also contribute to the red-shift in light emission ( Nakatsu et al . , 2006 ) . It must be noted that the crystal structure reported here has captured FLuc in the adenylation step of the reaction , therefore further computational modelling/crystal structure elucidation would be expected to provide further information on the light-emitting step of the reaction with iLH2 ( Nakatsu et al . , 2006 ) . Finally , the more open adenylation conformation of FLuc -iDLSA may affect the adenylate significantly , altering light production by decreasing the yield of the electronically excited state emitter and/or the efficiency in which the emitter produces a photon . Moreover , small differences in the binding position of the adenylate seen here , caused by positional changes of key active site residues could have a similar effect . Given that iLH2 preserves the colour modulation of mutant luciferases ( in addition to the 100 nm red shift ) , we set about to explore dual-BLI with iLH2 . We identified two stabilised FLuc enzymes with colour shifting mutations: FLuc_green ( V241I/G246A/F250S ) and Fluc_red ( S284T ) . Both exhibit a wide separation in peak emission wavelength , and were also balanced in their relative intensities . We next compared in vitro spectra with in vivo spectra from luciferase expressing cells implanted subcutaneously , systemically and intracranially ( to approximate superficial , intermediate and deep light source ) with both LH2 and iLH2 . The near-infrared emission of the FLuc mutants with iLH2 meant spectral separation was maintained for all animal tumour models . Importantly , the spectra of the two FLuc mutants remained consistent over tissue depth , meaning this dual BLI system could be applied to animal models without prior spectral characterisation . This maintenance of spectral separation , and similarity of relative intensities , between FLuc_green and FLuc_red when imaged with iLH2 meant this near-infrared bioluminescence system was found to be significantly better at spectral unmixing in vivo than using LH2 . This was demonstrated when the unmixed bioluminescent signals were correlated with actual cellular populations , determined by flow cytometry , ( R2 = 0 . 99 and 0 . 89 for iLH2 and LH2 respectively ) . One potential limitation of this system in its current state is the lower quantum yield of the FLuc-iLH2 reaction when compared to the FLuc-LH2 reaction , ~2–3 orders of magnitude dimmer . However , in this study , all mice were successfully spectrally imaged with both LH2 and iLH2 , which can be attributed to the sensitivity of the photon counting capabilities of the CCD cameras fitted in optical imagers ( Cool et al . , 2013 ) . The crystal structure reported here will be important for further optimisation of this near-infrared dual BLI system , particularly to increase the brightness of the current enzymes with iLH2 , as well as the discovery of novel luciferase colour-mutants which could be used for multi-coloured BLI in the near-infrared . Additionally , the use of this system in combination with luciferases utilising other substrates for multi-coloured BLI could also be explored , as well as its application to monitoring more complex processes in animal models ( Kleinovink et al . , 2018 ) . This work represents an important step forward in increasing the utility of BLI and opens up the window for multi-coloured BLI in the near-infrared . All manipulations were routinely carried out under an inert ( Ar or N2 ) atmosphere . All reagents were used as received unless stated . For the purposes of thin layer chromatography ( tlc ) , Merck silica-aluminium plates were used , with uv light ( 254 nm ) and potassium permanganate or anisaldehyde for visualisation . For column chromatography Merck Geduran Si 60 silica gel was used . Butyl lithium solutions were standardised with diphenyl acetic acid . Melting points are uncorrected and were recorded on a Griffin melting point machine . Infrared spectra were recorded using a Bruker Alpha ATR spectrometer . All NMR data were collected using a Bruker AMX 300 MHz or Bruker AVANCE III 600 MHz as specified . Reference values for residual solvents were taken as δ = 7 . 27 ( CDCl3 ) , 2 . 51 ( DMSO –d6 ) , 3 . 30 ( MeOD- d4 ) ppm for 1H NMR and δ = 77 . 2 ( CDCl3 ) , 39 . 5 ( DMSO –d6 ) , 49 . 0 ( MeOD- d4 ) ppm for 13C NMR . 19F NMR spectra were measured using a Bruker DX300 spectrometer , referenced to trichlorofluoromethane . Coupling constants ( J ) are given in Hz and are uncorrected . Where appropriate COSY and DEPT experiments were carried out to aid assignments . Mass spectroscopy data were collected on a Micromass LCT Premier XE ( ESI ) instrument . Elemental analysis was performed on an Exeter Analytical Inc CE-440 CHN analyser . 6- ( β-Methoxyethoxymethylether ) benzothiazole ( 1 ) , ( Muramoto et al . , 1999 ) 6- ( β-Methoxyethoxymethyl ether ) −2-formylbenzothiazole ( 2 ) , ( Anderson et al . , 2017 ) 1- ( 4-methoxycarbonylthiazole ) methyltriphenylphosphonium chloride ( Hermitage et al . , 2001; Old , 2008 ) 2’ , 3’-O-Isopropylidene-5’-O-sulfamoyladenosine ( Heacock et al . , 1996 ) were synthesised using procedures reported in the literature . 6- ( β-Methoxyethoxymethoxy ) −2- ( 2- ( 4-methoxycarbonylthiazol-2-yl ) ethenyl ) benzothiazole ( 3 ) . A suspension of aldehyde 2 ( 200 mg , 0 . 748 mmol ) and 1- ( 4-methoxycarbonylthiazole ) methyltriphenylphosphonium chloride ( 680 mg , 1 . 50 mmol ) in DMF ( 3 . 5 mL ) was cooled to 0°C and treated with K2CO3 ( 348 mg , 2 . 52 mmol ) . The resultant solution was allowed to warm to rt and stirred for 16 hr . After this time H2O ( 40 mL ) was added and the solution extracted using EtOAc ( 2 × 40 mL ) . The organics were combined and washed with H2O ( 40 mL ) , separated , dried ( MgSO4 ) , filtered and concentrated in vacuo . Purification was achieved using flash column chromatography ( 60% Et2O/Hexane ) to give three as a mixture of cis and trans isomers ( 163 mg , 54% ) . Cis: Rf = 0 . 52 ( 60 % EtOAc/Hexane ) ; 1H NMR ( 600 MHz , CDCl3 ) δ 3 . 39 ( 3H , s , OCH3 ) , 3 . 58–3 . 59 ( 2H , m , OCH2CH2O ) , 3 . 86–3 . 89 ( 2H , m , OCH2CH2O ) , 4 . 00 ( 3H , s , OCH3 ) , 5 . 36 ( 2H , s , OCH2O ) , 6 . 96 ( 1H , d , J = 12 . 8 , CHC ( N ) S ) , 7 . 25–7 . 29 ( 2H , m , ArH , CHC ( N ) S ) , 7 . 62 ( 1H , d , J = 2 . 3 , ArH ) , 8 . 06 ( 1H , d , J = 8 . 9 , ArH ) , 8 . 28 ( 1H , s , CHS ) . In CDCl3 solution the trans isomer was seen to isomerise to the cis isomer . Trans: A pure sample of trans was by separated by column chromatography to give three as a yellow solid . m . p . 112–115°C; Rf = 0 . 48 ( 60 % EtOAc/Hexane ) ; IR νmax 3101 , 2926 ( νCH ) , 2889 ( νCH ) , 2818 ( νCH ) , 1716 ( νCO ) , 1629 , 1598 , 1556 , 1495 , 1456 , 1333 , 1318 , 1281 , 1239 , 1222 , 1208 , 1162 , 1089 , 1046 , 978 cm-1; 1H NMR ( 600 MHz , CDCl3 ) δ 3 . 39 ( 3H , s , OCH3 ) , 3 . 58–3 . 59 ( 2H , m , OCH2CH2O ) , 3 . 86–3 . 88 ( 2H , m , OCH2CH2O ) , 4 . 00 ( 3H , s , OCH3 ) , 5 . 35 ( 2H , s , OCH2O ) , 7 . 22 ( 1H , dd , J = 8 . 9 , 2 . 4 , ArH ) , 7 . 59 ( 1H , d , J = 2 . 4 , ArH ) , 7 . 68 ( 2H , s , CHC ( N ) S ) , 7 . 92 ( 1H , d , J = 8 . 9 , ArH ) , 8 . 21 ( 1H , s , CHS ) ; 13C NMR ( 150 MHz , CDCl3 ) δ 52 . 8 ( CH3 ) , 59 . 2 ( CH3 ) , 68 . 0 ( CH2 ) , 71 . 7 ( CH2 ) , 94 . 0 ( CH2 ) , 107 . 5 ( CH ) , 117 . 9 ( CH ) , 124 . 1 ( CH ) , 128 . 2 ( CH ) , 128 . 7 ( CH ) , 136 . 2 ( C ) , 148 . 1 ( C ) , 148 . 9 ( C ) , 156 . 3 ( C ) , 161 . 7 ( C ) , 162 . 9 ( C ) , 165 . 6 ( C ) ; m/z ( ESI ) 407 ( 100% , M++H ) ; HRMS C18H19N2O5S2 calcd . 407 . 0735 , found 407 . 0738 . ( E ) −6- ( β-Methoxyethoxymethoxy ) −2- ( 2- ( 4-carboxythiazol-2-yl ) ethenyl ) benzothiazole ( 4 ) . A suspension of 3 ( 20 mg , 0 . 049 mmol ) in THF ( 0 . 75 mL ) and H2O ( 0 . 37 mL ) was treated with LiOH . H2O ( 5 . 0 mg , 0 . 12 mmol ) and stirred for 15 min . After this time H2O ( 10 mL ) and EtOAc ( 10 mL ) were added and the layers separated . The aqueous layer was acidified with 2 M HCl and extracted using EtOAc ( 2 × 10 mL ) , organics dried over MgSO4 , filtered and concentrated in vacuo to give 4 ( 19 mg , quant . ) as a yellow solid . m . p . 175–178°C; Rf = 0 . 20 ( 50 % EtOAc/MeOH ) ; IR νmax3101 ( νOH ) , 2917 ( νCH ) , 1680 ( νCO ) , 1598 , 1554 , 1455 , 1398 , 1320 , 1241 , 1204 , 1160 , 1101 , 1045 , 938 cm-1; 1H NMR ( 600 MHz , MeOD-d4 ) δ 3 . 33 ( 3H , s , OCH3 ) , 3 . 57–3 . 58 ( 2H , m , OCH2CH2O ) , 3 . 84–3 . 85 ( 2H , m , OCH2CH2O ) , 5 . 36 ( 2H , s , OCH2O ) , 7 . 25 ( 1H , dd , J = 8 . 9 , 2 . 4 , ArH ) , 7 . 69 ( 1H , d , J = 2 . 4 , ArH ) , 7 . 71 ( 1H , dd , J = 16 . 1 , 0 . 6 , CHC ( N ) S ) , 7 . 73 ( 1H , d , J = 16 . 1 , CHC ( N ) S ) , 7 . 80 ( 1H , d , J = 8 . 9 , ArH ) , 8 . 43 ( 1H , s , J = 7 . 5 , CHS ) ; 13C NMR ( 150 MHz , MeOD-d4 ) δ 59 . 1 ( CH3 ) , 69 . 0 ( CH2 ) , 72 . 8 ( CH2 ) , 95 . 0 ( CH2 ) , 108 . 7 ( CH ) , 119 . 0 ( CH ) , 124 . 7 ( CH ) , 128 . 5 ( CH ) , 129 . 1 ( CH ) , 130 . 1 ( CH ) , 137 . 5 ( C ) , 149 . 8 ( C ) , 150 . 0 ( C ) , 157 . 7 ( C ) , 163 . 8 ( C ) , 164 . 6 ( C ) , 166 . 7 ( C ) ; m/z ( ESI ) 393 ( 100% , M++H ) , 300 ( 9% ) ; HRMS C17H17N2O5S2 calcd . 393 . 0579 , found 393 . 0581; Anal . Calcd . for C17H16N2O5S2: C , 52 . 03; H , 4 . 11; N , 7 . 14 . Found C , 51 . 85; H , 4 . 29; N , 6 . 71% . 6- ( β-Methoxyethoxymethoxy ) −2- ( 2- ( 4-pentafluorophenoxycarbonylthiazol-2-yl ) ethenyl ) benzothiazole ( 5 ) . A solution of 4 ( 30 mg , 0 . 077 mmol ) in pyridine ( 3 . 80 mL ) was treated with EDC ( 18 mg , 0 . 096 mmol ) and pentafluorophenol ( 18 mg , 0 . 096 mmol ) and stirred at rt for 16 hr . The reaction mixture was concentrated in vacuo . Purification was achieved using flash column chromatography ( 30% Et2O/hexane ) to give five as a yellow solid ( 37 mg , 86% ) . m . p . 87–91°C; Rf = 0 . 25 ( 30 % EtOAc/Pet . Ether ) ; IR νmax 2922 , 2887 , 2835 , 1764 ( νCO ) , 1599 , 1554 , 1486 , 1470 , 1454 , 1321 , 1290 , 1260 , 1242 , 1211 , 1199 , 1180 , 1137 , 1123 , 1100 , 1060 , 1010 , 986 cm-1; 1H NMR ( 600 MHz , CDCl3 ) δ 3 . 40 ( 3H , s , OCH3 ) , 3 . 59–3 . 60 ( 2H , m , OCH2CH2O ) , 3 . 87–3 . 88 ( 2H , m , OCH2CH2O ) , 5 . 36 ( 2H , s , OCH2O ) , 7 . 23 ( 1H , dd , J = 8 . 9 , 2 . 4 , ArH ) , 7 . 60 ( 1H , d , J = 2 . 4 , ArH ) , 7 . 72 ( 1H , dd , J = 16 . 2 , 0 . 6 , CHC ( N ) S ) , 7 . 74 ( 1H , d , J = 16 . 2 , CHC ( N ) S ) , 7 . 94 ( 1H , d , J = 9 . 0 , ArH ) , 8 . 40 ( 1H , d , J = 0 . 4 , CHS ) ; 13C NMR ( 150 MHz , CDCl3 ) δ 59 . 2 ( CH3 ) , 68 . 0 ( CH2 ) , 71 . 7 ( CH2 ) , 94 . 0 ( CH2 ) , 107 . 5 ( CH ) , 117 . 9 ( CH ) , 124 . 3 ( CH ) , 124 . 9 ( C ) , 127 . 3 ( CH ) , 129 . 8 ( CH ) , 131 . 4 ( CH ) , 136 . 4 ( C ) , 137 . 3 ( C ) , 139 . 0 ( C ) , 140 . 6 ( C ) , 142 . 2 ( C ) , 144 . 7 ( C ) , 149 . 2 ( C ) , 156 . 4 ( C ) , 156 . 8 ( C ) , 162 . 5 ( C ) , 166 . 6 ( C ) ; 19F NMR ( CDCl3 , 282 ) δ −150 . 0 ( d , J = 16 . 9 , ArF ) , −157 . 14 ( app t , J = 19 . 7 , ArF ) , −161 . 9 ( dd , J = 19 . 7 , 16 . 9 , ArF ) ; m/z ( ESI ) 559 ( 100% , M++H ) ; HRMS C23H15F5N2O5S2 calcd . 559 . 0415 , found 559 . 0418 . 2’ , 3’-O-Isopropylidene-5’-O-[N- ( 6- ( β-methoxyethoxymethoxy ) -dehydroinfraluciferyl ) -sulfamoyl]adenosine ( 6 ) . A solution of 2’ , 3’-O-Isopropylidene-5’-O-sulfamoyladenosine ( 20 mg , 0 . 052 mmol ) in DMF ( 1 . 8 mL ) was treated with DBU ( 11 mg , 0 . 076 mmol ) and stirred at rt for 10 min . A solution of 5 ( 29 mg , 0 . 052 mmol ) in DMF ( 0 . 2 mL ) was then added dropwise . The reaction was stirred at rt for 16 hr . After this time pyridine ( 0 . 15 mL ) was added and the solution stirred for 4 hr . The resultant solution was concentrated in vacuo and purified using flash column chromatography ( 5% MeOH/DCM ) to give 6 ( 30 mg , 76% ) as a yellow solid . m . p . 164°C , dec . ; Rf = 0 . 32 ( 5 % MeOH/DCM ) ; IR νmax3290 ( νOH ) , 2932 ( νCH ) , 1644 ( νCO ) , 1598 , 1552 , 1505 , 1457 , 1418 , 1373 , 1291 , 1250 , 1209 , 1150 , 1103 , 1080 , 1048 , 985 cm-1; 1H NMR ( 600 MHz , DMSO-d6 ) δ 1 . 34 ( 3H , s , C ( CH3 ) 2 ) , 1 . 54 ( 3H , s , C ( CH3 ) 2 ) , 3 . 21 ( 3H , s , OCH3 ) , 3 . 46–3 . 49 ( 2H , m , OCH2CH2O ) , 3 . 74–3 . 78 ( 2H , m , OCH2CH2O ) , 4 . 09 ( 1H , dd , J = 11 . 0 , 4 . 9 , CH2OS ( O ) 2 ) , 4 . 12 ( 1H , dd , J = 11 . 0 , 4 . 9 , CH2OS ( O ) 2 ) , 4 . 42–4 . 46 ( 1H , m , OCHCH2 ) , 5 . 09 ( 1H , dd , J = 6 . 1 , 2 . 5 , CHCHO ) , 5 . 36 ( 2H , s , OCH2O ) , 5 . 41 ( 1H , dd , J = 6 . 1 , 2 . 9 , CHCHNO ) , 6 . 17 ( 1H , d , J = 2 . 9 , CHCHNO ) , 7 . 22 ( 1H , dd , J = 8 . 9 , 2 . 5 , ArH ) , 7 . 35 ( 2H , br , NH2 ) , 7 . 71 ( 1H , d , J = 16 . 1 , CHC ( N ) S ) , 7 . 79 ( 1H , d , J = 16 . 0 , CHC ( N ) S ) , 7 . 79 ( 1H , d , J = 2 . 5 , ArH ) , 7 . 95 ( 1H , d , J = 8 . 9 , ArH ) , 8 . 11 ( 1H , s , CHS ) , 8 . 12 ( 1H , s , NCH ( N ) ) , 8 . 44 ( 1H , s , NCH ( N ) ) ; 13C NMR ( 150 MHz , DMSO-d6 ) δ 25 . 2 ( CH3 ) , 27 . 1 ( CH3 ) , 58 . 1 ( CH3 ) , 67 . 3 ( CH2 ) , 67 . 6 ( CH2 ) , 71 . 0 ( CH2 ) , 81 . 6 ( CH ) , 83 . 5 ( CH ) , 83 . 9 ( CH ) , 89 . 3 ( CH ) , 93 . 3 ( CH2 ) , 108 . 0 ( CH ) , 113 . 2 ( C ) , 117 . 4 ( CH ) , 118 . 8 ( C ) , 123 . 6 ( CH ) , 125 . 2 ( CH ) , 126 . 1 ( CH ) , 128 . 5 ( CH ) , 136 . 0 ( C ) , 139 . 6 ( CH ) , 148 . 6 ( C ) , 149 . 0 ( C ) , 152 . 8 ( CH ) , 155 . 3 ( C ) , 156 . 1 ( C ) , 156 . 8 ( C ) , 162 . 8 ( C ) , 162 . 9 ( C ) , 165 . 0 ( C ) ; m/z ( ES+ ) 761 ( 100% , M++H ) ; HRMS C30H30N8O10S3 calcd . 761 . 1482 , found 761 . 1486 . 6-Hydroxy-2- ( 4-1E , 3E- ( 4-ethoxycarbonyl-4 , 5-dihydrothiazol-2-yl ) buta-2 , 4-dienyl ) benzothiazole ( iDLSA ) . A solution of 6 ( 20 mg , 0 . 026 mmol ) in TFA ( 0 . 32 mL ) was stirred at rt for 2 hr and then H2O ( 0 . 1 mL ) added and the solution concentrated in vacuo . EtOH ( 2 mL ) added and concentrated in vacuo to give the TFA salt of iDLSA ( 18 mg , 94% ) as an orange solid . m . p . 68–70°C; IR νmax3102 ( νOH ) , 1667 ( νCO ) , 1426 , 1132 cm-1; 1H NMR ( 600 MHz , DMSO-d6 ) δ 4 . 20–4 . 25 ( 2H , m , CHOH , CHOH ) , 4 . 53–4 . 59 ( 3H , m , CHOC , CH2OS ( O ) 2 ) , 5 . 96 ( 1H , d , J = 5 . 0 , CHCHNO ) , 7 . 00 ( 1H , d , J = 2 . 4 , ArH ) , 7 . 42 ( 1H , d , J = 2 . 5 , ArH ) , 7 . 72 ( 1H , d , J = 15 . 9 , CHC ( N ) S ) , 7 . 84 ( 1H , d , J = 8 . 8 , ArH ) , 7 . 95 ( 1H , d , J = 15 . 8 , CHC ( N ) S ) , 8 . 33 ( 1H , s , CHS ) , 8 . 53 ( 1H , s , NCH ( N ) ) , 8 . 59 ( 1H , s , NCH ( N ) ) . 13C NMR too weak due to poor solubility of compound . MS did not give M+ or meaningful fragment for accurate mass measurement . Copies of all 1H and 13C NMR have been deposited at https://doi . org/10 . 5061/dryad . 3j9kd51cs . Approximately 0 . 6 mg of iDLSA was suspended in 500 µL of crystallisation buffer ( 25 mM Tris-Cl containing 200 mM AmSO4 , 1 mM DTT , 1 mM EDTA ) pH 7 . 85 at 21°C . The solution was vortexed vigorously and sonicated . Most of the solid was dissolved and the concentration was determined to be ~1 mM by UV absorbance ( using an extinction coefficient of 8200 at 372 nm for this buffer and pH ) . A 20 mg/mL solution of P . pyralis luciferase ( PpyWT that contains the N-terminal peptide GPLGS- ) in the same buffer ( 500 µL ) was mixed gently with the iDLSA at room temperature and then incubated at 15°C for 20 min . The concentration of iDLSA in the protein-inhibitor mixture was determined by UV absorbance to be 620 µM , giving an inhibitor:enzyme ratio of ~3:1 , at this point the bioluminescence activity of the mixture was assayed and the enzyme was 85% inhibited . A small amount of a separate iDLSA solution ( available from solubility trials ) was added to bring the inhibitor:enzyme ratio to ~4:1 , and based on activity the enzyme was 91% inhibited . Finally , the protein-inhibitor solution was added to ~0 . 5 mg of iDLSA and mixed gently and incubated at 15°C for 15 min . Based on activity , 99% of the enzyme was inhibited and based on UV absorbance the inhibitor:enzyme ratio was 5 . 6:1 ( 950 µM:170 µM ) The protein-inhibitor solution was centrifuged and a very slight amount of inhibitor was evident . The supernatant was frozen in liquid nitrogen in ~18–50 µL aliquots and stored at −80°C . A single aliquot was thawed and the solution remained clear . The pH of this solution at 6°C should be 8 . 3 , pH 8 . 17 at 10°C , and pH 7 . 9 at 21°C . Approximately 0 . 6 mg of iDLSA was resuspended in 500 µL of buffer ( 25 mM Tris-Cl containing 200 mM ( NH4 ) 2SO4 , 1 mM DTT , 1 mM EDTA ) pH 7 . 85 at 21°C to a concentration of 1 mM as determined by UV absorbance ( extinction coefficient of 8200 at 372 nm ) . This solution was mixed with a 20 mg/mL solution of P . pyralis ( inhibitor:enzyme ratio of ~3:1 ) at room temperature and then incubated at 15°C for 20 min . The inhibitor:enzyme solution was centrifuged and the supernatant was frozen in liquid nitrogen in ~18–50 µL aliquots and stored at −80°C for future crystallisations . Crystallisations used the hanging drop vapour diffusion method . Drops containing 1–2 µl of inhibitor:enzyme solution were mixed with the same volume of well solution and equilibrated against 500 µl of well solution , incubated at 4°C , with crystals typically appearing within 48 hr . Glycerol was used as a cryoprotectant , in an optimised well solution of 150 mM ( NH4 ) 2SO4 , 50 mM HEPES pH 7 . 0 , 2% PEG 1000 . Data were collected at the Diamond Light Source on beam line IO4-1 , at wavelength 0 . 91587 Å , and 100 K . Processing and data reduction were carried out on site using CrysalisPro ( Agilent Technologies ) , and synchrotron data sets were processed and scaled by using XDS , SCALA and XIA2 programs . Molecular replacement methods were used successfully to determine the relative orientation and position of the two monomers in the asymmetric unit using the PHASER program ( McCoy et al . , 2007 ) . The starting dehydroinfraluciferin DLSA complex model was derived from the Firefly luciferase apo structure ( PDB-ID 3IEP ) with all solvent atoms and the luciferin removed . A simple rigid body refinement was sufficient to initiate refinement , with subsequent refinement and model building cycles performed using Refmac5 and Coot ( Murshudov et al . , 2011; Emsley et al . , 2010 ) . The X-ray data collection and refinement statistics have been deposited at https://doi . org/10 . 5061/dryad . 3j9kd51cs . FLuc mutants contained 11 pH and temperature stabilising mutations ( F14R/L35Q/A105V/V182K/T214C/I232K/D234G/E354R/D357Y/S420T/F465R ) ( Jathoul , 2012 ) . FLuc_green contained an additional three mutations ( V241I/G246A/F250S ) , and FLuc_red has the red-shifting mutation S284T , as well as the mutation R354I which is required to maintain the red-shift in this stabilised FLuc backbone ( Branchini et al . , 2007 ) . All FLuc mutants were codon optimised for mammalian expression and cloned into the MLV-based splicing gamma retroviral vector SFG . The Raji B lymphoma cell line used in all experiments was transduced to express a FLuc mutant , and subsequently flow-sorted for pure FLuc expressing populations using a co-expressed marker gene . For tumour cell lines FLuc . IRES was upstream of the marker gene CD34 or dNGFR as indicated . For T cells FLuc . 2A_peptide was upstream of the CAR CD19-4G7_HL-CD8STK-41BBZ . HEK-293T packaging cells were plated at a density of 200’000 cells/ml in 100 mm tissue culture dish ~24 hr prior to transfection . Transfections were performed when cells were 50–70% confluent . A bulk transfection mixture was prepared where 30 μl GeneJuice Transfection Reagent ( Merck millipore ) was added to 470 μl of plain RPMI for each supernatant to be produced . Following a 5 min incubation at room temperature , a total volume of 12 . 5 μg of DNA was added for each plate to be transfected ( for retroviral transfection: 3 . 125 μg RDF RD114 env plasmid , 4 . 6875 μg PeqPam-env gagpol plasmid , 4 . 6875 μg SFG retroviral construct ) . Following addition of plasmid DNA , the mixture was incubated for a further 15 min at room temperature prior to dropwise addition to the HEK- 293 T cell culture . Plates were gently agitated following transfection . Supernatant harvested at 48 hr was stored at 4°C , and was then combined with the 72 hr harvest prior to aliquoting and storage at −80°C . The day prior to transduction Raji B lymphoma cells ( atcc ccl-86 ) ( >90% viable ) were diluted ~1 in 10 to ensure exponential growth for transduction; also a well of non-tissue culture treated 24 well plate was coated with 8 μg/ml retronectin ( Lonza ) for every plasmid to be transduced and left at 4°C overnight . The next day retronectin was aspirated and 250 μl of each retroviral supernatant for transduction was added to a well and incubated for 30 min at room temperature . Whilst incubating , Raji cells were harvested , counted and resuspended at a concentration of 600 , 000 cells/ml . supernatants were aspirated from wells of retronectin coated plate and 500 μl of cell suspension was added to each well followed by 1 . 5 ml of the same retroviral supernatant that was previously incubated in each well . Cells were spin transduced at 1000 RCF for 40 min then returned to incubator for 48 hr before harvest and expression testing . Transduction efficiencies were assessed by flow cytometry , based on marker gene expression as indicated by antibody staining using the BD LSR FortessaX-20 . If necessary Fluorescence Activated Cell Sorting ( FACS ) was performed to obtain pure expressing populations , also based on marker gene expression as indicated by antibody staining , using the BD FACS Aria Fusion . FACS was also use to sort populations of cells expressing differing levels of FLuc_green and FLuc_red within the same cell . Concentration of antibody used was guided by manufacturer’s instructions . Anti-human CD34-PE ( clone 581 ) , anti-human CD271-APC ( clone ME20 . 4 ) and anti-mouse/human CD11B-PerCP/Cy5 . 5 ( clone M1/70 ) ( Biolegend ) . Anti-humanCD19-FITC ( clone HIB19 ) , anti-human CD20-eFluor450 ( clone 2H7 ) and Viability APC-eFluor780 ( eBioscience ) . Data were analysed using Flow Jo software ( Tree Star Inc , Oregon , USA ) . When bone marrow cells were required for flow cytometry analysis . Following animal sacrifice by CO2 narcosis and cervical dislocation , the femurs were removed and transferred to PBS pending cellular harvest . The ends of the femur were snipped off using scissors and the bone was placed in an extraction tube ( microfuge tube with holder made from 200 μl pipette tip inserted ) . Tubes were centrifuged at 1000 RCF for 60 s . Bone marrow pellet was resuspended in 50 μl Ammonium-Chloride-Potassium ( ACK ) lysing buffer ( Lonza ) and left for 60 s at room temperature before washing with PBS and passing through a 70 μm filter before pelleting . Samples were blocked 2 . 4G2 supernatant ( rat anti-mouse CD32 ) supplemented with mouse Ig FcR blocking reagent ( Miltenyl Biotec ) for 30 min at room temperature . Cells were washed with PBS and pelleted , followed by each sample being transferred to a well of a U-bottomed 96 well plate before proceeding with antibody staining . An antibody master mix containing all antibodies to be used for staining was prepared in PBS to a total volume of 100 μl per sample . Samples were left to stain at room temperature in the dark for 30 min . Samples were washed once with PBS , pelleted and transferred to FACS tubes . Beckman Coulter Flow-Checkfluorospheres were used as a stopping gate for flow cytometry analysis . Beads are supplied at 1 × 10e6 beads/ml in an aqueous solution containing preservative surfactant . To prevent toxicity to cellular samples , beads were washed once with PBS prior to addition to samples . Following centrifugation ( 400 RCF for 5 min ) , beads were resuspended in an equal volume of PBS with 10 μl of beads added to each sample . As six fluorophores were used for bone marrow analysis , compensation was performed prior to sample acquisition using OneComp eBeads ( eBioscience ) . Events were kept between 2 , 000–5 , 000 events/second , with 1000 events being recorded per sample , using flow check beads as a stopping gates ( 10% each sample ) . Flow cytometry gating first identified the lymphocyte population ( FSC-A vs SSC-A ) , exclusion of doublet cells ( SSC-A vs SSC-W ) ; antibodies detailed in Flow cytometry and Fluorescence Activated Cell Sorting were then used to gate on viable cells , exclude mouse monocyte cells ( mCD11b ) , identify the Raji tumour cell population ( CD19 and CD20 ) and finally co-expressed marker gene ( dNGFR and dCD34 ) . On day one peripheral blood mononuclear cells were isolated from a healthy donor blood using Ficoll-paque density gradient media ( GE Healthcare ) . Cells were resuspended at 2 × 106/ ml and stimulated with 1 mg/ml PHA ( Sigma ) . On day 2 cells were fed with IL-2 at a concentration of 100 U/ml ( Genscript ) . On day 3 cells were transduced as described in transduction of cell lines , with IL-2 at a final concentration of 100 U/ml . On day 6 cells were harvested and resuspended at 1 × 106/ml with 100 U/ml IL-2 and left to recover for at least 48 hr before in vivo injection . Transduction efficiency was measured using flow cytometry . For spectrographic testing of FLuc , mutants were stably transduced in the mammalian Raji B-cell lymphoma cell line . For in vitro bioluminescence assays cells were harvested , counted and 1 × 106 cells/well were resuspended in TEM buffer ( 1M Tris-acetate , 20 mM EDTA and 100 mM MgSO4 at pH 7 . 8 ) and added in triplicate to wells of a black 96-well plate ( 100 μl/well ) . If mixtures of cells were used , total cell number remained the same . For spectral testing the stage temperature of the IVIS Spectrum was set to 37°C ( automatic acquisition mode , FOV 13 . 2 , f/1 ) . iLH2 was synthesised by UCL Chemistry . Other substrates tested include D-luciferin ( Regis Technologies ) , CycLuc1 ( Merck Millipore ) and Aka-Lumine-HCL ( Wako Pure Chemical Industries ) . Substrates were dispensed into the wells using a multi-channel pipette ( at a final concentration of 300 μM ) . A 2 min delay was allowed for stabilisation of light output . Images were acquired through all 18 bandpass filters on the IVIS Spectrum ( 20 nm bandpass , 490 nm to 840 nm ) . Living image software ( Perkin Elmer ) was used for ROI analysis of spectral images and spectral unmixing analysis . Image analysis involved placing a ROI over the signal in each well . If a series of spectral images was acquired , the same ROI was placed over the well in every image for each plate . For spectral unmixing analysis , guided spectral unmixing was first used on pure expressing FLuc_green and FLuc_red populations to create a library spectra for each mutant with each substrate . The relevant library spectra was then used to perform spectral unmixing on mixed FLuc_green and FLuc_red populations ( or cellular populations expressing both enzymes ) . Data exported to Excel ( Microsoft ) and Prism ( Graphpad ) for further analysis . Spectra was normalised to peak emission for each FLuc mutant with each substrate . Due to the characterisation nature of these in vitro experiments , and the substantial amounts of precious chemicals needed to synthesise iLH2 , it was decided to repeat each in vitro experiment twice with three replicates . All animal studies were approved by the University College London Biological Services Ethical Review Committee and licensed under the UK Home Office regulations and the Guidance for the Operation of Animals ( Scientific Procedures ) Act 1986 ( Home Office , London , United Kingdom ) . All of the in vivo models used the severely immunocompromised NSG ( NOD . Cg PrkdcscidIl2rgtm1Wji/SzJ ) mouse model ( JAX mouse strain , Charles River ) . Mice were male and aged between 6–8 weeks old . Due to the characterisation , or proof of concept , nature of these experiments , and the substantial amounts of precious chemicals needed to synthesise iLH2 , it was decided to engraft 4–5 mice for every condition in each model to ensure engraftment and survival in a least three animals for each condition . Also , no specific toxicity experiments were performed , but no adverse side effects were observed with iLH2 . For engraftment of subcutaneous tumours , FLuc expressing Raji cell lines were counted and 2 × 106 cells were pelleted for each animal to be injected . Cells were washed twice in PBS before being resuspended in plain RPMI 1% HEPES ( Sigma-Aldrich ) to a concentration of 2 × 107 cells/ml and were kept on ice ready for injection . Cells were injected subcutaneously ( 100 μl bolus ) into a shaved area of the flank . Mice were left at least 5 days for tumour development before imaging . For engraftment of systemic tumours , FLuc expressing Raji cell lines were counted and 5 × 105 cells were pelleted for each animal . If mixtures of two different FLuc expressing cell lines were used , total cell number per mouse remained at 5 × 105 cells . Cells were washed twice in PBS before being resuspended in plain RPMI 1% HEPES to a concentration of 2 . 5 × 106 cells/ml and were kept on ice ready for injection . Animals were transferred to a warming chamber set at 39–42°C to facilitate peripheral vasodilation prior to intravenous ( IV ) injections . Mice were placed in a restraint and cells were injected IV ( 200 μl bolus ) into the tail vein . Mice were left at least 7 days for tumour development before imaging . For the CAR T cell model 5 × 106 CAR positive T cells were injected IV ( 200 μl bolus ) 8 days after Raji cell engraftment . For engraftment of intracranial tumours , FLuc expressing Raji cell lines were counted and 2 × 104 cells were pelleted for each animal . Cells were washed twice in PBS before being resuspended in PBS to a concentration of 1 × 104 cells/μl and were kept on ice ready for injection . Intracranial injections were performed using a stereotaxic frame fitted with a hamilton syringe . Cells were injected ( 2 μl bolus ) into the right striatum ( from bregma 2 mm right , 1 mm anterior , 4 mm down ) . Mice were left at least 7 days for tumour development before imaging . For imaging of in vivo models , LH2 and iLH2 were solubilised in sterile PBS and animals were administered with substrate ( 2 mg ( or 100 mg/kg ) of either LH2 or iLH2 in 200 μl or 400 μl bolus respectively ) via intraperitoneal ( IP ) injection . Animals were anaesthetised using 2% Isofluorine ( flow rate 1 L/min O2 ) . Spectral imaging was commenced 10 min post IP injection to allow stabilisation of light output . If the same animal was being imaged with both LH2 and iLH2 , at least 24 hr was left between imaging to allow for full clearance of substrate . In vivo bioluminescent images were acquired using IVIS Spectrum ( FOV 24 , f/1 , Medium ( 8 ) bin , automatic acquisition mode for imaging with LH2 , FOV 24 , f/1 , Medium ( 8 ) bin , 120 s acquisition , total imaging time 24 mins for imaging with iLH2 ) . These parameters are calculated to keep the binning , exposure time and f/stop within an optimal range for quantification . Up to five animals could be imaged at once and the stage was heated to 37°C . Open filter images were acquired prior to and post spectral imaging to confirm the stability of photon emission during spectral acquisition . Spectral imaging acquired images through 14 and 12 of the 20 nm bandpass filters on the IVIS Spectrum depending on substrate used ( 530–830 nm for LH2 and 590–830 nm for iLH2 ) , starting from the lowest to the highest filter . It was not necessary to acquire images through all filters as the bioluminescent emissions of FLuc mutants did not cover the full spectral range from 490 to 850 nm . Living image software was used for ROI analysis of spectral images and spectral unmixing analysis . Radiance values for bioluminescence are shown using pseudo-colour scales detailed in each image . Image analysis involved placing an ROI over the tumour signal for every animal in each model . If a series if spectral images were acquired , the same ROI was placed over tumour signal in every image for each mouse . For spectral unmixing analysis , guided spectral unmixing was first used on pure expressing FLuc_green and FLuc_red populations from spectral characterisation experiments to create a library spectra for each mutant with each substrate . The relevant library spectra was then used to perform spectral unmixing on mixed FLuc_green and FLuc_red populations . Data exported to Excel ( Microsoft ) and Prism ( Graphpad ) for further analysis . Where relevant means ± standard deviation of data given . Statistical tests used include T test and ONE-way ANOVA with post hoc Tukey’s test for multiple column comparison ( Prism , Graphpad ) . Correlation analysis performed using Microsoft excel .	Fireflies and some other insects glow to attract mates or prey . This so-called bioluminescence occurs when an enzyme called luciferase modifies the molecule luciferin , which can then emit bright yellow-green light . The gene that encodes the luciferase enzyme has been introduced into cells from mammals , including cancer cells . In the presence of luciferin , these cells begin to glow . The brightness of the bioluminescence depends on how many cancer cells are growing and dividing . The light is bright enough for the cancer cells making luciferase to be transplanted into mice so their behaviour can be examined . However , blood and other tissues absorb the yellow-green light , making it hard to see the cancer cells deep within a mouse . To circumvent this problem , researchers designed a new type of luciferin , called infraluciferin , which emits red light that shines through blood and tissues . There are now also different variants of the luciferase enzyme , which act on infraluciferin to make different shades of red light . Stowe et al . wanted to test if two different biological events happening at the same time could be observed using two shades of bioluminescent red in a single live mouse . First , a mixture of cancer cells containing two versions of luciferase were transplanted into mice . When the mice were then given infraluciferin , the two types of cancer cells could be distinguished based on the different shades of red bioluminescence . In a second experiment , Stowe et al . tracked the treatment of cancer cells with immune cells , by introducing a different version of luciferase into each of the two groups of cells . Over time , the red light produced by the immune cells grew stronger than that of the cancer cells , indicating that the number of cancer cells had decreased and that the treatment was effective . Together , this work shows that it can be simple , cheap and efficient to observe more than one cell type , or even disease , in a living system . This technique may be used by scientists to study different diseases and treatment options in mice . Importantly , it will also reduce the number of animals used to do this research .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2019	Near-infrared dual bioluminescence imaging in mouse models of cancer using infraluciferin
Human visual recognition activates a dense network of overlapping feedforward and recurrent neuronal processes , making it hard to disentangle processing in the feedforward from the feedback direction . Here , we used ultra-rapid serial visual presentation to suppress sustained activity that blurs the boundaries of processing steps , enabling us to resolve two distinct stages of processing with MEG multivariate pattern classification . The first processing stage was the rapid activation cascade of the bottom-up sweep , which terminated early as visual stimuli were presented at progressively faster rates . The second stage was the emergence of categorical information with peak latency that shifted later in time with progressively faster stimulus presentations , indexing time-consuming recurrent processing . Using MEG-fMRI fusion with representational similarity , we localized recurrent signals in early visual cortex . Together , our findings segregated an initial bottom-up sweep from subsequent feedback processing , and revealed the neural signature of increased recurrent processing demands for challenging viewing conditions . The human visual system interprets the external world through a cascade of visual processes that overlap in space and time . Visual information is transformed not only feedforward , as it propagates through ascending connections , but also from higher to lower hierarchy areas through descending feedback connections and within the same areas through lateral connections ( Ahissar et al . , 2009; Bullier , 2001; Enns and Di Lollo V , 2000; Lamme and Roelfsema , 2000; Lamme et al . , 1998 ) . This concurrent activation of a dense network of anatomical connections poses a critical obstacle to the reliable measurement of recurrent signals and their segregation from feedforward activity . As a result , our knowledge on the role of recurrent processes and how they interact with feedforward processes to solve visual recognition is still incomplete . Here we used an ultra-rapid serial visual presentation ( ultra-RSVP ) of real-world images to segregate early bottom-up from recurrent signals in the ventral pathway . We postulated that , under such rapid stimulus presentations , visual processes will degrade substantially by suppressing sustained neural signals that typically last hundreds of milliseconds . As a result , neural signals would become transient , reducing the overlap of processes in space and time and enabling us to disentangle distinct processing steps . Recent behavioral evidence exemplified the remarkable robustness of the human visual system to capture conceptual information in stimuli presented at similar rates ( Broers et al . , 2018; Evans et al . , 2011; Potter et al . , 2014 ) . Thus the underlying neural signals , while deprived , would still represent brain activity required to accomplish visual object recognition . We recorded human MEG data while participants viewed ultra-RSVP sequences with rates 17 or 34 ms per picture . Confirming our hypothesis , the rapid presentation of images segregated the activation cascade of the ventral visual pathway into two temporally dissociable processing stages , disentangling the initial bottom-up sweep from subsequent processing in high-level visual cortex . Capitalizing on this dissociation , we used multivariate pattern classification of MEG data to characterize the activation dynamics of the ventral pathway and address the following three challenges: we investigated how the evolution of the bottom-up sweep predicts the formation of high-level visual representations; we sought evidence for rapid recurrent activity that facilitates visual recognition; and we explored whether reducing visibility with higher stimulus presentation rates increases recurrent processing demands . Finally , to resolve the locus of feedforward and feedback visual signals , we tracked the spatiotemporal dynamics with a MEG-fMRI fusion approach based on representational similarity ( Cichy et al . , 2014 , 2016a; Kriegeskorte et al . , 2008 ) . To determine the time series with which individual images were discriminated by neural representations , we averaged all elements of the decoding matrix , resulting in a grand total decoding time series ( Figure 2b ) . First , we found that neural responses were resolved at the level of individual images in all 3 viewing conditions . Second , decoding accuracies decreased with faster stimulus presentation rates , reflecting the challenging nature of the RSVP task with stimuli presented for very short times . Third , peak latencies shifted earlier with faster stimulus presentation rates ( Figure 2e ) . That is , the 500 ms per picture condition reached a peak at 121 ms ( 95% confidence interval: 102-126 ms ) , preceded by the 34 ms per picture RSVP condition at 100 ms ( 94-107 ms ) , and finally the 17 ms per picture RSVP condition at 96 ms ( 93-99 ms ) ( all statistically different; P<0 . 05; two-sided sign permutation tests ) . Fourth , onset latencies shifted later with faster stimulus presentation rates ( Figure 2h ) . That is , the 500 ms per picture condition had onset at 28 ms ( 9-53 ms ) , followed by the 34 ms per picture RSVP condition at 64 ms ( 58-69 ms ) , and finally the 17 ms per picture RSVP condition at 70 ms ( 63-76 ms ) ( all statistically different; P<0 . 05; two-sided sign permutation tests ) . The decreased decoding accuracy combined with the increasingly early peak latency and increasingly late onset latency for the RSVP conditions indicate that visual activity was disrupted over the first 100 ms . Even though the highest levels of the visual processing hierarchy in humans are reached with the first 100 ms , there is little time for feedback connections from these areas to exert an effect ( Lamme and Roelfsema , 2000 ) . Thus , neural activity during the first 100 ms has been linked to the engagement of feedforward and local recurrent connections , rather than long-range feedback connections . In line with these arguments , the early peaks at 100 and 96 ms for the 34 and 17 ms per picture RSVP conditions , respectively , explicitly delineate the first sweep of visual activity , differentiating it from later neural activity that includes feedback influences from the top of the visual hierarchy . Further , if early decoding would only reflect feedforward activity , we would not expect to see onset latency differences , but we do . The fact that different stimulus durations have different onsets suggests that interactions with recurrent activity are already incorporated when the first decoding onsets emerge , arguing against the view that the early part of the decoding time course can be uniquely tied to feedforward alone ( Fahrenfort et al . , 2012; Lamme and Roelfsema , 2000; Ringach et al . , 1997 ) . Next , to investigate the generalization of our findings to any pair of images , even when they share categorical content , we evaluated whether our results hold to within-category image classification . For this , we subdivided the decoding matrix into two partitions , corresponding to within-face comparisons , and within-object comparisons . Averaging the elements of each partition separately determined the time series with which individual images were resolved within the subdivision of faces ( Figure 2b ) and objects ( Figure 2c ) . We confirmed the generalization and reliability of our findings , as our results were similar to the grand total decoding time series: individual images were discriminated by neural responses; decoding accuracies were weaker for rapid stimulus presentations; and peak and onset latencies had the same ordinal relationship as in the grand total analysis ( Figure 2fg ) . Peak and onset latencies for the grand total and within category comparisons are shown in Table 1 . In sum , decoding accuracies decreased with progressively shorter stimulus presentation times , indicating that neuronal signals encoded less stimulus information at rapid presentation rates . Onset latencies shifted late with shorter presentation times , indicating that recurrent activity exerts its influence even as the first decoding onsets emerged . Importantly , the progressively earlier peak with shorter presentation times indicated disruption of the first sweep of visual activity , thus indexing feedforward and local recurrent processing and segregating it in time from subsequent processing that includes feedback influences from high-level visual cortex . How did the disruption of the early sweep of visual activity , reported in the previous section , affect the emergence of categorical information in the RSVP conditions ? A prevalent theory posits that core object recognition is largely solved in a feedforward manner ( DiCarlo et al . , 2012; Liu et al . , 2002; Serre et al . , 2007; Thorpe et al . , 1996 ) . If this holds under rapid presentation conditions , then categorical signals would be expected to emerge with comparable dynamics regardless of stimulus presentation rates . However , opposing theories concur that feedback activity is critical for visual awareness and consciousness ( Lamme and Roelfsema , 2000; Ahissar et al . , 2009; Fahrenfort et al . , 2017 , 2012 ) . According to these theories , presenting stimuli at rapid presentation rates would ( i ) afford less time for initial stimulus evidence accumulation ( a process that in all likelihood already incorporates some local recurrent processing , as suggested by variable onset latencies reported in the previous section ) and ( ii ) lead to disruption of recurrent signals of the target stimulus due to the masking stimuli of the RSVP paradigm . These would be consistent with slowing down the speed and extent with which category information can be resolved using recurrence ( Brincat and Connor , 2006; Tang and Kreiman , 2017 ) . To differentiate between those competing theories , we computed categorical division time series . We divided the decoding matrix into partitions corresponding to within-category ( face or object ) and between-category stimulus comparisons separately for each of the three viewing conditions ( Figure 3a ) . The difference of between-category minus within-category average decoding accuracies served as a measure of clustering by category membership . We found that the categorical division time series resolved face versus object information in all three conditions ( Figure 3a ) . Consistent with the grand total decoding results , categorical neural representations were stronger in the 500 and 34 ms per picture conditions than the 17 ms per picture condition . Multidimensional scaling ( MDS ) plots ( Kruskal and Wish , 1978 ) at peak latencies for the three conditions , offering an intuitive visualization of the stimulus relationships , are shown in Figure 3c . These plots revealed strong categorical division for the 500 and 34 ms per picture conditions , followed by weaker but still distinct categorical division in the 17 ms per picture condition . The peak of the categorical division time series revealed the time points at which categorical information was most explicitly encoded in the neural representations ( DiCarlo and Cox , 2007 ) . The peak latency increased as presentation rates became progressively faster . That is , the time series for the 500 ms per picture condition peaked at 136 ms ( 130-139 ms ) , followed by the 34 ms per picture RSVP at 169 ms ( 165-177 ms ) , and the 17 ms per picture RSVP at 197 ms ( 184-218 ms ) ( all statistically different; P<0 . 05; two-sided sign permutation tests ) ( Figure 3b ) . This relationship is reverse from the peak latency of the first sweep of visual activity reported in the previous section , further stressing the existence of variable dynamics in the ventral pathway . This suggests that categorical information did not arise directly in a purely feedforward mode of processing , as this would predict comparable temporal dynamics in all conditions . Instead , it is consistent with the idea that recurrent interactions within the ventral stream facilitate the emergence of categorical information by enhancing stimulus information in challenging visual tasks ( Brincat and Connor , 2006; Hochstein and Ahissar , 2002; Rajaei et al . , 2018; Tang and Kreiman , 2017; Tapia and Beck , 2014 ) . Taken together , our results revealed variable temporal neural dynamics for viewing conditions differing in presentation time . Even though the peak latency of the first sweep of visual activity shifted earlier with higher presentation rates , as reported in the previous section , the peak latency of categorical information shifted later , stretching the time between the abrupt end of the initial visual sweep and the emergence of categorical information . This inverse relationship in peak latencies discounts a feedforward cascade as the sole explanation for categorical representations . As neuronal signals propagate along the ventral pathway , neural activity can either change rapidly at subsequent time points , or persist for extended times . Transient activity reflects processing of different stimulus properties over time in either a feedforward manner , as computations become more abstract , or a recurrent manner as neurons tune their responses . On the other hand , persistent activity could maintain results of a particular neural processing stage for later use . Our premise in introducing the ultra-RSVP task was to suppress the persistent neural activity , and in doing so better capture the transient neural dynamics that reflect distinct neural processing steps . To experimentally confirm that persistent neural activity was indeed suppressed with rapid presentation rates , we extended the SVM classification procedure with a temporal generalization approach ( Cichy et al . , 2014; Isik et al . , 2014; King and Dehaene , 2014 ) . In particular , we used a classifier trained on data at a time point t to evaluate discrimination at all other time points t’ . Intuitively , if neural representations are sustained across time , the classifier should generalize well across other time points . Temporal generalization matrices were computed by averaging decoding across all pairwise image conditions and all subjects , thus extending over time the results presented in Figure 2b . Our temporal generalization analysis confirmed that neural activity became increasingly transient at rapid presentation rates ( Figure 4 ) . While the 500 ms per picture condition had maps with broad off-diagonal significant elements characteristic of sustained representations , the RSVP conditions had narrow diagonal maps indicating transient neural patterns , with the 17 ms per picture RSVP narrower than the 34 ms per picture RSVP . The increasingly transient activity in the RSVP conditions shows that neural activity continuously transformed stimulus information in a feedforward and feedback manner , will less neural resources used to maintain information . Thus , the results confirmed our hypothesis that the ultra-RSVP task would suppress persistent neural activity . The analyses presented thus far segregated the temporal dynamics of the initial bottom-up sweep from subsequent signals incorporating recurrent activity in the ventral pathway . Furthermore , peak latencies for early and late visual signals varied inversely , consistent with feedback processing . Here we mapped visual signals on the cortex to identify where in the brain feedforward and feedback signal interact . To map the spatiotemporal dynamics of the visual processes we used a MEG-fMRI fusion method based on representational similarity ( Cichy et al . , 2014 , 2016a ) . For this , we first localized the MEG signals on the cortex and derived the time series from all source elements within two regions-of-interest ( ROIs ) : early visual cortex ( EVC ) and inferior temporal cortex ( IT ) . We selected EVC as the first region of the cortical feedforward sweep , and IT as the end point where neural patterns have been found to indicate object category ( Cichy et al . , 2014 ) . We then performed time-resolved multivariate pattern classification on the MEG data following the same procedure described earlier , only now we created pattern vectors by concatenating activation values from sources within a given ROI , instead of concatenating the whole-head sensor measurements . This procedure resulted in one MEG RDM for each ROI and time point . We compared the representational similarity between the time-resolved MEG RDMs for the two cortical regions ( EVC and IT ) and the fMRI RDMs for the same regions ( Figure a and b ) . This yielded two time series of MEG-fMRI representational similarity , one for EVC and one for IT . In all conditions , consistent with the view of visual processing as a spatiotemporal cascade ( Cichy et al . , 2014 ) , the time series peaked earlier for EVC than IT ( Figure 5c–h ) . The peak-to-peak latency between EVC and IT increased as viewing conditions became increasingly challenging with faster presentation rates: Δ=27 ms for the 500 ms per picture condition; Δ=79 ms for the 34 ms per picture RSVP; and Δ=115 ms for the 17 ms per picture RSVP ( all statistically different; two-sided bootstrap hypothesis tests; P<0 . 05 ) . This latency difference was the compounded effect of two factors . First , the EVC peak had progressively shorter latencies ( 104 vs . 87 vs . 80 ms for the 3 conditions ) , and second the IT peak had progressively longer latencies ( 131 vs . 166 vs . 195 ms for the three conditions ) . This inverse relationship between the EVC and IT peaks corroborated the findings of the previous sections , namely that a disrupted first sweep of visual activity was associated with a delayed emergence of categorical division information . It further bound the processing stages in time to the V1 and IT locations in space . Importantly , while EVC had a single peak at 104 ms and persistent representations over hundreds of milliseconds for the 500 ms per picture condition , its dynamics were transient and bimodal for the RSVP conditions . For the 34 ms per picture RSVP condition , an early peak at 87 ms was immediately followed by weak MEG-fMRI representational similarities , and then a second peak at 169 ms . For the 17 ms per picture RSVP condition , we observed similar dynamics with an early peak at 80 ms and a second peak at 202 ms , though in this case the second peak was not strictly defined because the time course did not reach significance , possibly due to compromised neural representations at such fast stimulus presentation rates . The second peak in EVC occurred at similar times as the peak in IT ( Δ=3 ms for the 34 ms per picture RSVP condition , p=0 . 06; and Δ=7 ms for the 17 ms per picture RSVP condition , p≪0 . 001; two-sided bootstrap hypothesis tests ) . This is consistent with feedback activity in EVC at the same time as IT solves visual object recognition . Table 2 summarizes latencies and 95% confidence intervals for all conditions . We note here that resolving feedback activity in EVC was possible with MEG-fMRI fusion because MEG activation patterns disentangled slow fMRI hemodynamic responses in EVC that correspond to the combined contributions of feedforward and feedback visual activity . In sum , the combination of the RSVP paradigm with MEG-fMRI representational similarity resolved bimodal dynamics for EVC . The first EVC peak offered evidence that disruption of early visual activity resulted in delayed categorical division information in IT . The second EVC peak occurred at approximately the same time as the peak in IT and is consistent with feedback activity from IT to EVC . Here we discuss the nature of visual processing and the role of recurrent dynamics for each of the two temporally distinct stages of visual processing revealed by the ultra-RSVP task . Most popular computer vision models , such as deep neural networks ( LeCun et al . , 2015 ) and HMAX ( Riesenhuber and Poggio , 1999 ) , have adopted a feedforward architecture to sequentially transform visual signals into complex representations , akin to the human ventral stream . Recent work has shown that these models not only achieve accuracy in par with human performance in many tasks ( He et al . , 2015 ) , but also share a hierarchical correspondence with neural object representations ( Cichy et al . , 2016b; Martin Cichy et al . , 2017; Yamins et al . , 2014 ) . Even though models with purely feedforward architecture can easily recognize whole objects ( Serre et al . , 2007 ) , they often mislabel objects in challenging conditions , such as incongruent object-background pairings , or ambiguous and partially occluded inputs ( Johnson and Olshausen , 2005; O'Reilly et al . , 2013 ) . Instead , models that incorporate recurrent connections are robust to partially occluded objects ( O'Reilly et al . , 2013; Rajaei et al . , 2018 ) , suggesting the importance of recurrent processing for object recognition . Unlike other studies that use stimuli that are occluded or camouflaged ( Spoerer et al . , 2017; Tang and Kreiman , 2017 ) , our RSVP task offers no obvious computation that can be embued to feedback processes when presentation times are shortened . That is , our study does not inform on the precise nature of computations needed for stimulus evidence accumulation when presentation times are extremely short . Despite this fact , our findings on the duration and sequencing of ventral stream processes can still offer insights for developing computational models with recursive architecture . First , such models should solve object categorization within the first couple hundred milliseconds , even when the feedforward and feedback pathways are compromised as in the ultra-RSVP task . Second , the timing of recurrent processes should not be predetermined and fixed , but vary depending on viewing conditions , as in the case of onset and peak latency shifts in the RSVP decoding time series . Here viewing conditions related to the speed of the RSVP task , but it is reasonable to expect other challenging conditions , such as ambiguous or partial stimuli , to exert time delays , though possibly longer ( Tang and Kreiman , 2017 ) . Third , the timing and strength of the early visual signals should inversely determine the timing of categorical representations . And fourth , feedback processes in deep models should activate early layers of the model at the same time object representations are emerging at the last layers . Despite these insights , future research is needed to understand what recurrent computations are exactly carried out by the brain to solve visual recognition under RSVP-like presentation conditions . Since its conception ( Potter and Levy , 1969 ) , the RSVP paradigm has been implemented with stimulus rates about 100 ms per item or slower . Inherent to the design , RSVP experiments have revealed the temporal limitations of human perception ( Spence , 2002 ) , attention ( Nieuwenstein and Potter , 2006 ) , and memory ( Potter , 1993 ) . Recently however , behavioral investigations have been exploring even faster presentation rates in ultra-RSVP tasks ( Broers et al . , 2018; Evans et al . , 2011; Potter et al . , 2014 ) . These experiments found that observers can detect target images at rates 13 to 20 ms per picture . Due to its effectiveness in masking stimuli by combining forward and backward masking , the ultra-RSVP paradigm could be used to address the question whether recurrent processing is necessary for recognition of objects . One view posits that a purely feedforward mode of processing is sufficient to extract meaning from complex natural scenes ( DiCarlo et al . , 2012 ) . High behavioral performance in the ultra-RSVP task has been used as an argument to support this view . Specifically , such rapid presentations of stimuli have been presumed to block recurrent activity , since low level visual representations are immediately overwritten by subsequent images and time is too short to allow multiple synaptic transmissions ( Tovée , 1994 ) . However , this interpretation has been challenged both here , with the reengagement of EVC late in the processing stream , and by the finding that the ability to detect and categorize images at such speeds depends on the efficacy of the images to mask one another ( Maguire and Howe , 2016 ) . Thus , it still remains an open question whether recurrent activity is necessary to extract conceptual meaning ( Howe , 2017 ) . Though our study did not address whether such recurrent activity can arise in more effective masking conditions that suppress visibility ( Maguire and Howe , 2016 ) , it paves the way for future studies to explore the link between stimulus visibility and recurrent neuronal processes . Such studies could vary the effectiveness of forward and backward masking to segregate the early from late visual signals , as accomplished here , and investigate under what conditions ( e . g . ambiguous or occluded input ) stimulus visibility ( King et al . , 2016; Salti et al . , 2015 ) is associated with feedback activity . As ultra-RSVP reduces visibility , future studies could also investigate whether recurrent activity is an integral component of the neural correlates of consciousness , defined as the minimum neuronal mechanisms jointly sufficient for a conscious percept ( Koch et al . , 2016 ) . Seventeen healthy subjects ( 12 female; 16 right-handed and one left-handed; age mean ± s . d . 27 . 2 ± 5 . 7 years ) with normal or corrected to normal vision participated in the RSVP experiment . They all signed an informed consent form and were compensated for their participation . The study was approved by the Institutional Review Board of the Massachusetts Institute of Technology and followed the principles of the Declaration of Helsinki . We collected MEG data using a 306-channel Elekta Triux system with a 1000 Hz sampling rate . The data was band-pass filtered with cut-off frequencies of 0 . 03 and 330 Hz . The MEG system contained 102 triple sensor elements ( 2 gradiometers and one magnetometer each ) organized on a helmet shaped array . The location of the head was measured continuously during MEG recording by activating a set of 5 head position indicator coils placed over the head . The raw MEG data was preprocessed with the Maxfilter software ( Elekta , Stockholm ) to compensate for head movements and denoise the data using spatiotemporal filters ( Taulu and Simola , 2006; Taulu et al . , 2004 ) . The Brainstorm software ( Tadel et al . , 2011 ) was then used to extract trials from −300 ms to 900 ms with respect to target onset . Every trial was baseline-corrected to remove the mean from each channel during the baseline period , defined as the time before the onset of the first mask stimulus for the RSVP task , or the target image for the 500 ms per picture condition . A 6000 fT peak-to-peak rejection threshold was set to discard bad trials , and the remaining trials were smoothed with a 20 Hz low-pass filter . Eye blink artifacts were automatically detected from frontal sensor MEG data , and then principal component analysis was used to remove these artifacts . To visualize the complex patterns of the 24 × 24 MEG RDMs , which capture the relations across the neural patterns elicited by the 24 target images , we used the first two dimensions of multidimensional scaling ( MDS ) ( Kruskal and Wish , 1978; Shepard , 1980 ) . MDS is an unsupervised method to visualize the level of similarity between different images contained in a distance matrix . Intuitively , MDS plotted the data in two dimensions where similar images were grouped together and different images far apart . fMRI data acquisition and analysis The fMRI data was collected using a 3T Trio Siemens Scanner and 32-channel head coil . The structural images were acquired in the beginning of each session using T1-weighted sequences with TR = 1900 ms , TE = 2 . 52 ms , flip angle = 9° , FOV = 256 mm2 , and 192 sagittal slices . Functional data was acquired with high spatial resolution but partial coverage of the brain covering occipital and temporal lobe using gradient-echo EPI sequence with TR = 2000 ms , TE = 31 ms , flip angle = 80° , FOV read = 192 mm , FOV phase = 100% , ascending acquisition , gap = 10% , resolution = 2 mm isotropic , and slices = 25 . The details of fMRI analysis can be found in ( Cichy et al . , 2014 ) and here we explain it briefly . SPM8 ( http://www . fil . ion . ucl . ac . uk/spm/ ) was used to analyze the fMRI data . The data was realigned , re-sliced , and co-registered with the structural images for each subject and session separately . Then a general linear model analysis was used to estimate t-value maps for each of the 24 target images . We further defined two volumetric ROIs for fMRI data analysis , V1 and IT . V1 was defined separately for each participant using an anatomical eccentricity template ( Benson et al . , 2012 ) , and corresponded to a 0–6° visual angle . IT was defined using a mask comprising bilateral fusiform and inferior temporal cortex ( Maldjian et al . , 2003 ) , keeping the most strongly 361 activated voxels from a cross-validated dataset to match the size of IT to the average size of V1 . To assess the relations between brain fMRI responses across the 24 target images , we constructed space-resolved fMRI RDMs using a correlation-based method . We conducted two types of analyses: ( 1 ) ROI-based and ( 2 ) spatially unbiased using a searchlight approach . For the ROI-based analysis , we extracted and concatenated the V1 or IT voxel t-values to form ROI-specific fMRI pattern vectors . For each pair of images , we then calculated the dissimilarity ( one minus Pearson’s rho ) between the fMRI pattern vectors , resulting in a 24 × 24 fMRI RDM indexed by the compared images . This procedure resulted in one fMRI RDM for each ROI and subject . For the searchlight-based analysis ( Kriegeskorte et al . , 2006 ) , we constructed fMRI RDMs for each voxel in the brain . In particular , for each voxel v we extracted fMRI activation values in a sphere centered at v with a radius of 4 voxels ( searchlight at v ) and arranged them into fMRI pattern vectors . For each pair of images , we then calculated the pairwise dissimilarity ( one minus Pearson’s rho ) between fMRI pattern vectors , resulting in a 24 × 24 fMRI RDM indexed by the compared images . This procedure yielded one fMRI RDM for each voxel in the brain and subject . To assess the spatiotemporal dynamics of EVC and IT , we applied a fMRI-MEG fusion approach based on representational similarity analysis ( RSA ) ( Kriegeskorte et al . , 2008; Cichy et al . , 2014 ) . The basic idea is that if two stimuli are similarly represented in MEG patterns , they should also be similarly represented in fMRI patterns , a correspondence that can be directly evaluated using the RDMs . Thus , we computed the similarity ( Spearman’s rho ) between time-resolved MEG RDMs and space-resolved fMRI RDMs . For ROI-based fMRI-MEG fusion , we used fMRI RDMs and MEG RDMs from the corresponding ROIs . In particular , for each time point we computed the similarity ( Spearman’s rho ) between the subject-averaged MEG RDM and the subject-specific fMRI RDM . This procedure yielded n = 14 time courses of MEG-fMRI representational similarity for each ROI and subject . To investigate whether maintenance of stimulus information was compromised in the RSVP relative to the 500 ms per picture condition , we extended the SVM classification procedure using a temporal generalization approach ( Cichy et al . , 2014; Isik et al . , 2014; King and Dehaene , 2014; Pantazis et al . , 2017 ) . This method involved training the SVM classifier at a given time point t , as before , but testing across all other time points . Intuitively , if representations are stable over time , the classifier should successfully discriminate signals not only at the trained time t , but also over extended periods of time that share the same neuronal representations . We repeated this temporal generalization analysis for every pair of stimuli , and the results were averaged across compared images and subjects , yielding 2-dimensional temporal generalization matrices with the x-axis denoting training time and the y-axis testing time . For statistical assessment of peak and onset latency of the time series , we performed bootstrap tests . The subject-specific time series were bootstrapped 1000 times and the empirical distribution of the peak latency of the subject-averaged time series was used to define 95% confidence intervals . A similar procedure was used to define 95% confidence intervals for onset latency . For peak-to-peak latency differences , we obtained 1000 bootstrapped samples of the difference between the two peaks , which resulted in an empirical distribution of peak-to-peak latency differences . We then used the tail of this empirical distribution to evaluate the number of bootstrap samples that crossed 0 , which allowed us to compute a p-value for the peak-to-peak latency difference . Finally , the p-values were corrected for multiple comparisons using false discovery rate at a 0 . 05 level . A similar procedure was used for onset-to-onset differences . We had one cohort of subjects for the RSVP conditions , and another for the 500 ms per picture condition . For consistency , we performed between-subject comparisons in all comparisons across the 17 , 34 , and 500 ms per picture conditions . Statistical inference relied on non-parametric statistical tests that do not make assumptions on the distributions of the data ( Maris and Oostenveld , 2007; Pantazis et al . , 2005 ) . Specifically , for the statistical assessment of classification time series , temporal generalization matrices , and MEG-fMRI representational similarities we performed permutation-based cluster-size inference . The null hypothesis was equal to 50% chance level for decoding results , and 0 for decoding differences or correlation values . In all cases we could permute the condition labels of the MEG data , which was equivalent to a sign permutation test that randomly multiplied subject responses by +1 or −1 . We used 1000 permutations , 0 . 05 cluster defining threshold and 0 . 05 cluster threshold for time series and temporal generalization maps . Data and Matlab code used for statistical analyses and producing results in main Figures 2 , 3 and 5 , are available as a Source data one file with this article .	The human brain can interpret the visual world in less than the blink of an eye . Specialized brain regions process different aspects of visual objects . These regions form a hierarchy . Areas at the base of the hierarchy process simple features such as lines and angles . They then pass this information onto areas above them , which process more complex features , such as shapes . Eventually the area at the top of the hierarchy identifies the object . But information does not only flow from the bottom of the hierarchy to the top . It also flows from top to bottom . The latter is referred to as feedback activity , but its exact role remains unclear . Mohsenzadeh et al . used two types of imaging to map brain activity in space and time in healthy volunteers performing a visual task . The volunteers had to decide whether a series of images that flashed up briefly on a screen included a face or not . The results showed that the brain adapts its visual processing strategy to suit the viewing conditions . They also revealed three key principles for how the brain recognizes visual objects . First , if early visual information is incomplete – for example , because the images appeared only briefly – higher regions of the hierarchy spend more time processing the images . Second , when visual information is incomplete , higher regions of the hierarchy send more feedback down to lower regions . This leads to delays in identifying the object . And third , lower regions in the hierarchy – known collectively as early visual cortex – process the feedback signals . This processing takes place at the same time as the higher levels identify the object . Knowing the role of feedback is critical to understanding how the visual system works . The next step is to develop computer models of visual processing . The current findings on the role of feedback should prove useful in designing such models . These might ultimately pave the way to developing treatments for visual impairments caused by damage to visual areas of the brain .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Ultra-Rapid serial visual presentation reveals dynamics of feedforward and feedback processes in the ventral visual pathway
Mycobacterium tuberculosis ( Mtb ) expresses a broad-spectrum β-lactamase ( BlaC ) that mediates resistance to one of the highly effective antibacterials , β-lactams . Nonetheless , β-lactams showed mycobactericidal activity in combination with β-lactamase inhibitor , clavulanate ( Clav ) . However , the mechanistic aspects of how Mtb responds to β-lactams such as Amoxicillin in combination with Clav ( referred as Augmentin [AG] ) are not clear . Here , we identified cytoplasmic redox potential and intracellular redox sensor , WhiB4 , as key determinants of mycobacterial resistance against AG . Using computer-based , biochemical , redox-biosensor , and genetic strategies , we uncovered a functional linkage between specific determinants of β-lactam resistance ( e . g . β-lactamase ) and redox potential in Mtb . We also describe the role of WhiB4 in coordinating the activity of β-lactamase in a redox-dependent manner to tolerate AG . Disruption of WhiB4 enhances AG tolerance , whereas overexpression potentiates AG activity against drug-resistant Mtb . Our findings suggest that AG can be exploited to diminish drug-resistance in Mtb through redox-based interventions . Mycobacterium tuberculosis ( Mtb ) displays tolerance to several clinically important antibacterials such as aminoglycosides and β-lactams ( Flores et al . , 2005a; Morris et al . , 2005 ) . Innate resistance of Mtb toward β-lactams is likely to be due to the presence of a broad-spectrum Ambler class A β-lactamase ( BlaC ) ( Flores et al . , 2005b ) . Other physiological mechanisms such as cell envelope permeability , induction of drug efflux pumps , and variations in peptidoglycan ( PG ) biosynthetic enzymes may also play a role in the β-lactam-resistance of Mtb ( Gupta et al . , 2010; Lun et al . , 2014 ) . The Ambler class A β-lactamases are mostly susceptible to inhibition by clavulanate ( Clav ) , sulbactam ( Sub ) , and tazobactam ( Taz ) ( Kurz et al . , 2013 ) . Indeed , intrinsic resistance of Mtb toward β-lactams can be overcome by combining β-lactams with Clav ( Chambers et al . , 1998; Hugonnet et al . , 2009 ) . The combined amoxicillin ( Amox ) and Clav preparation , referred to as Augmentin ( AG ) , was not only active against Mtb in vitro , but also had significant early bactericidal activity in patients with drug-resistant TB ( Chambers et al . , 1998; Cynamon and Palmer , 1983 ) . Furthermore , a combination of meropenem and Clav showed significant bactericidal activity against drug-resistant strains of Mtb ( Hugonnet et al . , 2009 ) . In view of this , there is an imminent need to investigate the mechanisms of action of β-lactams in combination with Clav against Mtb , and the potential development of resistance by the pathogen against this combination . In other bacteria , β-lactams directly interact with enzymes involved in PG synthesis . This is likely to result in killing of the pathogen through multiple mechanisms , including the induction of autolysin pathway , holin:antiholin pathway , DNA damage , and alterations in physiology ( e . g . TCA cycle and oxidative stress ) ( Tomasz , 1974; Rice et al . , 2003; Miller et al . , 2004; Kohanski et al . , 2007; Lobritz et al . , 2015 ) . The complex effects of β-lactams on both PG biosynthesis and other processes indicate that the response to β-lactams could be mediated either through direct sensing of β-lactam molecules or by their effects on bacterial physiology . In Staphylococcus aureus , a transmembrane protease ( BlaR1 ) senses β-lactam concentrations by direct binding through an extracellular domain , which activates its intra-cytoplasmic proteolytic domain resulting in cleavage of the β-lactamase repressor , BlaI , and induction of β-lactamase expression ( Gregory et al . , 1997 ) . It has been shown that Mtb expresses a homolog of BlaR1 ( encoded by Rv1845c , blaR ) , which modulates the activity of BlaC by regulating the BlaI repressor in a manner analogous to S . aureus BlaR1-BlaI couple ( Sala et al . , 2009 ) . However , BlaR orthologues in all mycobacterial species lack the extracellular sensor domain involved in binding with β-lactams ( Sala et al . , 2009 ) , indicating that mechanisms of antibiotic sensing and BlaC regulation are likely to be distinct in Mtb . Furthermore , how β-lactams influence mycobacterial physiology ( e . g . redox balance and primary metabolism ) remains unknown . Therefore , insights on how the presence of β-lactams is conveyed in Mtb to activate appropriate adaptation response are key to combating resistance and developing novel therapies . In this work , we generated a system-scale understanding of how AG affects mycobacterial physiology . Exploiting a range of technologies , we explained mechanistically that the efficacy of AG is partly dependent upon the redox physiology of Mtb . Furthermore , we have rationally described the role of a redox-responsive transcription factor , WhiB4 , in regulating the tolerance of Mtb to AG during infection . Our study demonstrates how Mtb alters its redox physiology in response to AG and identifies a major mycobacterial antioxidant , mycothiol ( MSH ) , and WhiB4 as major contributors to β-lactam tolerance . To assess the response of Mtb toward β-lactam and β-lactamase inhibitor combination ( s ) , we analyzed the transcriptome of mycobacterial cells exposed to AG . We observed that 100 µg/ml of Amox in combination with 8 µg/ml of Clav ( 10X MIC of AG ) arrested bacterial growth at 6 hr and killing was observed only after 12 hr post-exposure ( Figure 1—figure supplement 1-Inset ) . Therefore , expression changes at a pre-lethal phase ( i . e . 6-hr post -exposure ) can reveal significant insights into Mtb pathways involved in AG tolerance . A total of 481 genes were induced ( ≥2 fold; p value ≤ 0 . 05 ) and 461 were repressed ( ≥2 fold; p value ≤ 0 . 05 ) in wt Mtb upon AG-treatment ( supplementary file 1A ) . Although these results are important , the transcriptome only provides a snapshot of the mechanisms exploited by Mtb for AG tolerance . To generate a system-scale understanding , computational approaches that combine condition-specific expression data with general protein interaction data are frequently utilized to construct dynamic and stress response networks ( see Appendix 1 for detailed explanation ) . Therefore , we further generated the AG response network by combining microarray data with the protein-protein interaction ( PPI ) map of Mtb . To construct this map , we first created a comprehensive PPI of Mtb using information from experimentally validated and published interactions ( see Materials and methods ) and integrated microarray data with the PPI to generate the AG response network . In the network , a node represents a protein whose weight is based on a weighting function that captures the variation in the expression level of the corresponding gene due to drug exposure . An edge represents an interaction between two nodes , which are also weighted by a function that captures the node weights of both nodes forming an edge , as a relative importance of all edges in the network . A full description of the mathematical equations and algorithms used to generate the AG response network is beyond the scope of this study , we encourage readers to refer our original papers for detailed methodology ( Sambarey et al . , 2013; Sambaturu et al . , 2016; Padiadpu et al . , 2016 ) . Figure 1—figure supplement 1 and supplementary file 1B represent the top 1% nodes , which cover a total of 806 genes , connected through 1096 interactions to form a well-connected AG response network of Mtb . Genes belonging to diverse functional classes such as intermediary metabolism , cell wall , lipid metabolism , virulence , and information pathways are featured in the response network . In line with cell surface targeting activity of β–lactams , the cumulative node weight ( CNW ) of genes belonging to cell-wall-related processes , including PG biosynthesis , was the highest ( CNW = 30616361 . 02 ) amongst the classes affected by AG ( Figure 1A ) . Interestingly , nodes belonging to ‘intermediary metabolism and respiration’ were also significantly enriched in response to AG ( CNW = 20716788 . 92; Figure 1A ) , indicating a downstream effect of target-specific interactions of AG on fundamental metabolic processes in Mtb . Further analysis revealed that several mediators ( e . g . sigE , sigB , mprAB , and dnaK ) of cell envelope stress response ( Bretl et al . , 2014 ) function as major hub nodes and form-interconnected networks of genes important for maintaining cell wall integrity in response to AG ( Figure 1—figure supplement 2 ) . Accordingly , expression data showed induction of genes involved in PG biosynthesis , β-lactamase regulation ( blaR-blaI ) , and cell envelope homeostasis in response to AG ( Figure 1B ) . In addition , two other mechanisms involved in tolerance toward β-lactams that is outer membrane permeability ( mycolic acid biogenesis [kasA , kasB , and fabD] and omp ) and drug efflux pumps ( efpA , Rv1819c and uppP ) were also induced ( Figure 1B and supplementary file 1A ) . Altogether , Mtb responds to AG by modulating the expression of cell-envelope-associated pathways including those that are the specific targets of β-lactams . 10 . 7554/eLife . 25624 . 003Figure 1 . Network analysis identified pathways affected by AG exposure in Mtb . Wt Mtb was grown to an OD600 of 0 . 4 and treated with 100 µg/ml of Amox and 8 µg/ml of Clav ( 10X MIC of AG ) for 6 hr at 37°C . Total RNA was isolated and processed for microarray analysis as described in Materials and methods . ( A ) Cumulative node weight intensities ( CNW ) were derived by addition of the node weights of genes in a particular functional group upon exposure to AG . Node weight intensity of a gene was derived by multiplying the normalized intensity value with the corresponding fold-change ( FC ) value . Cumulative node weight intensities for different functional classes are available in Figure 1—source data 1 . ( B ) Heat map showing expression of genes ( log2fold-change , p≤0 . 05 ) that belong to cell wall processes for untreated and AG-treated Mtb from two biological samples . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 00310 . 7554/eLife . 25624 . 004Figure 1—source data 1 . Cumulative node weight intensities for different functional classes as depicted in Figure 1A . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 00410 . 7554/eLife . 25624 . 005Figure 1—figure supplement 1 . Global network of Mtb under AG stress . The transcriptome of wt Mtb treated with AG was superimposed on Mtb protein:protein interaction ( PPI ) network to extract out top 1% network consisting of 806 nodes ( genes ) . The size of the node indicates its node weight . The interactions/edges are depicted by grey arrows; the thickness of the arrows is the measure of the strength of interactions . Shapes of the nodes reflect direction of gene expression ( square: induced; arrowhead: repressed; and circle: constitutive ) . The nodes are colored according to the TubercuList functional categories – red: virulence , detoxification , and adaptation , blue: cell wall and cell processes , green: information pathways , orange: intermediary metabolism and respiration , olive green: lipid metabolism , grey: conserved hypotheticals , pink: regulatory proteins , cyan: insertion sequences and phages , and light green: PE/PPE family . Inset shows the survival curve of Mtb treated with indicated concentrations of AG . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 00510 . 7554/eLife . 25624 . 006Figure 1—figure supplement 2 . Sub-network of major hub nodes showing the top-most activities regulating response of Mtb upon AG treatment . Sub-network showing close interactions between diverse regulators of cell envelope stress ( e . g . sigB , sigE , mprA , and phoP ) and redox stress ( e . g . whiB2 , whiB3 , whiB6 , ideR , dnaK ) during AG treatment . The nodes are colored according to the functional modules they belong to and edge thickness reflects the strength of the interaction . Shapes of the nodes denote regulation of gene expression ( square: induced; arrowhead: repressed; and circle: constitutive ) . It is noteworthy that most of the functionally diverse nodes ( e . g . sigma factors , antioxidants , and redox-sensors ) converge at a common stress-responsive chaperone , DnaK , making it a major hub node coordinating AG stress response in Mtb . Aligning with our findings , studies have suggested an important role for DnaK and ClpB chaperones in promoting recovery from oxidative stress ( Fay and Glickman , 2014; Vaubourgeix et al . , 2015 ) . Functional modules based on annotations given in the TubercuList ( http://tuberculist . epfl . ch/ ) include red: virulence , detoxification , and adaptation , blue: cell wall and cell processes , green: information pathways , orange: intermediary metabolism and respiration , olive green: lipid metabolism , grey: conserved hypotheticals , pink: regulatory proteins , cyan: insertion sequences and phages , and light green: PE/PPE family . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 006 Since nodes coordinating ‘intermediary metabolism and respiration’ were the second most enriched class in response to AG , we performed a detailed examination of genes altered in this category . We found that energetically efficient respiratory complexes such as NADH dehydrogenase I ( nuo operon ) and ATP-synthase ( atpC , atpG , and atpH ) were down-regulated , whereas energetically less favored NADH dehydrogenase type II ( ndh ) , cytochrome bd oxidase ( cydAB ) , and nitrite reductase ( nirBD ) were activated in response to AG ( Figure 2 ) . The transcriptional shift toward a lesser energy state is consistent with the down-regulation of several genes associated with the TCA cycle ( sucCD , fum , mdh , and citA ) , along with an induction of glycolytic ( pfkA , pfkb , fba , and pgi ) , gluconeogenesis ( pckA ) , and glyoxylate ( icl1 ) pathways ( Figure 2 ) . Interestingly , icl1 has recently been shown to promote tolerance of Mtb toward diverse anti-TB drugs by maintaining redox homeostasis ( Nandakumar et al . , 2014 ) . These findings indicate that the maintenance of redox balance is likely to be an important cellular strategy against AG . 10 . 7554/eLife . 25624 . 007Figure 2 . AG influences multiple pathways involved in central metabolism , respiration and redox balance in Mtb . Heat maps depicting expression of genes ( log2fold-change; p≤0 . 05 ) coordinating respiration , CCM , iron-metabolism and redox balance for untreated and 6 hr of AG-treated Mtb from two biological samples . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 00710 . 7554/eLife . 25624 . 008Figure 2—figure supplement 1 . qRT-PCR analysis of Mtb exposed to different concentrations of AG for indicated time points . Fold change for each transcript was measured with respect to untreated wt Mtb by normalizing expression with the 16srRNA transcript . Error bars represent standard deviations from mean . Data are representative of at least two independent experiments done in duplicate . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 00810 . 7554/eLife . 25624 . 009Figure 2—figure supplement 2 . Comparative analysis of genes differentially regulated by AG treatment and upon depletion of mycothiol or ergothioneine buffers . Heat maps ( absolute fold change , p≤0 . 05 ) of genes differentially regulated in response to AG treatment and their status in ( A ) ergothioneine ( ΔegtA ) and ( B ) mycothiol ( ΔmshA ) mutant strains of Mtb . The expression data of MtbΔegtA and MtbΔmshA strains were obtained from a recently published study ( Saini et al . , 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 00910 . 7554/eLife . 25624 . 010Figure 2—figure supplement 3 . Overlapping regulation of genes in response to AG and oxidative stress . AG and cumene hydroperoxide ( CHP; oxidant ) response network was prepared as described in Materials and methods . The vein diagram representing nodes present in top 1% network of Mtb under AG stress and oxidative stress . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 01010 . 7554/eLife . 25624 . 011Figure 2—figure supplement 4 . Heat maps depicting gene expression profile ( log2fold-change ) of Mtb untreated or treated with 1X and 5X MIC of AG for 6 and 12 hr from at least three biological samples . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 011 The fact that there was a significant upregulation of genes involved in oxidative stress response in Mtb was indicative of the influence of AG on mycobacterial redox physiology . We found increased expression of reactive oxygen species ( ROS ) detoxifying enzymes ( ahpCD , katG , and hpx ) , antioxidant buffers ( trxB1 , trxA , trxC , and mtr ) , methionine sulfoxide reductase ( msrA ) , Fe-S cluster repair system ( Rv1461-Rv1466; suf operon ) , and intracellular redox sensors ( whiB6 , whiB2 , whiB3 , and pknG ) ( Figure 2 ) . The global regulator of oxidative stress in bacteria ( OxyR ) is non-functional in Mtb ( Deretic et al . , 1995 ) . However , we had earlier reported that a redox-sensitive DNA-binding protein ( WhiB4 ) functions as a negative regulator of OxyR-specific antioxidant genes ( e . g . ahpCD ) in Mtb ( Chawla et al . , 2012 ) . Consequently , Mtb lacking whiB4 ( MtbΔwhiB4 ) displayed higher expression of antioxidants and greater resistance toward oxidative stress ( Chawla et al . , 2012 ) . While a modest repression of WhiB4 ( ~1 . 3 fold ) in response to AG was observed in microarray experiments , qRT-PCR analysis showed a significant down-regulation ( −5 . 00 ± 0 . 27 fold; p value ≤ 0 . 001 ) as compared to unstressed Mtb ( Figure 2—figure supplement 1 ) . The breakdown of iron homeostasis is another hallmark of oxidative stress ( Imlay , 2003 ) . Accordingly , our data exhibited induction of two Fe-responsive repressors ( ideR and furB ) along with the down-regulation of genes encoding Fe-siderophore biosynthetic enzymes ( mbt operon ) and Fe-transport ( Rv1348 ) , and up-regulation of Fe-storage ( bfrB ) ( Figure 2 ) . Recently , two mycobacterial redox buffers , mycothiol ( MSH ) and ergothioniene ( EGT ) , were implicated in protection against oxidants and antibiotics ( Saini et al . , 2016 ) . We compared gene expression changes displayed by MSH and EGT mutants ( Saini et al . , 2016 ) with the AG transcriptome . Approximately 60% of genes regulated by MSH and EGT also displayed altered expression in response to AG ( Figure 2—figure supplement 2 ) , indicating overlapping roles of MSH and EGT in tolerating oxidative stress associated with AG treatment in Mtb ( Saini et al . , 2016 ) . Lastly , we performed transcriptomics of Mtb in response to a known oxidant cumene hydroperoxide ( CHP; 250 μM for 2 hr [non-toxic concentration] ) and compared with expression changes induced by AG . As shown in Figure 2—figure supplement 3 , a considerable overlap in gene expression ( ~30% ) was observed between these two conditions ( supplementary file 1C ) . More importantly , genes associated with β-lactam tolerance ( ponA2 , ispH , blaR , and kasA ) and redox-metabolism ( ahpCD , trxB1 , trxB2 , trxC , and suf ) were similarly regulated under CHP and AG challenge ( supplementary file 1C ) . It can be argued that the high concentration of AG ( 10X MIC ) can adversely affect Mtb physiology to influence primary response of AG on gene expression . To address this issue , we reassessed global changes in gene expression upon exposure to 1X and 5X MIC of AG at 6 hr and 12 hr post-treatment . A relatively small number of genes were differentially regulated by lower concentrations of AG as compared to 10X MIC ( supplementary file 1D ) . However , similar to our results using 10X MIC , we found that exposure to 1X and 5X MIC of AG increased expression of genes associated with PG biogenesis ( mur operon , ponA1 , and ponA2 ) , β-lactamase regulation ( blaI-blaR ) , cell envelope stress ( sigB , sigE , mprAB , and phoPR ) , redox metabolism ( hpx , trx system , msrA , suf operon , whiBs , and pknG ) , alternate respiration ( ndh and cydAB ) , CCM ( pfkAB , fba , icl , and aceAa ) , and efflux pumps ( efpA and uppP ) ( Figure 2—figure supplement 4 ) . Lastly , we validated our microarray data by performing qRT-PCR on a few genes highly deregulated upon treatment with 1X , 5X , and 10X MIC of AG ( Figure 2—figure supplement 1 ) . Taken together , these data indicate a major recalibration of genes regulating cell wall processes and cellular bioenergetics of Mtb in response to AG . Altered expression of genes associated with respiration and oxidative stress response indicates that AG exposure might elicit redox stress in Mtb . To investigate this , we performed a comprehensive evaluation of changes in redox physiology of Mtb upon exposure to AG . Since , NADH redox cofactor is central to metabolism and respiration , we first measured NADH/NAD+ ratio of Mtb cells exposed to 10X MIC of AG at various time points post-treatment . At pre-lethal stage ( 6 hr post-treatment ) , we did not observe any change in NADH/NAD+ ratios ( Figure 3A ) . However , a significant elevation of NADH/NAD+ ratio was detected 24 hr post-treatment , which coincides with AG-induced killing in Mtb ( Figure 3A ) . We subsequently determined accumulation of ROS by staining with an oxidant-sensitive fluorescent dye; 2' , 7'-dichlorofluorescein diacetate ( DCFDA ) in Mtb cells treated with AG ( 10X MIC ) for 3 hr and 6 hr . Early time points were considered for ROS measurements to disregard the possibility of death-mediated increase in ROS upon AG treatment . A consistent increase ( ~3-fold increase ) in DCFDA fluorescence was observed at both time points as compared to untreated control ( Figure 3B ) . Under aerobic conditions , ROS is mainly generated through univalent reduction of O2 by reduced metals , flavins , and quinones ( Imlay , 2013 ) , which mainly generates superoxide ( O2−• ) . Therefore , we determined O2−• production using a well-established and freely cell-permeable O2−• indicator , dihydroethidium ( DHE ) ( Kalyanaraman et al . , 2014 ) . It is known that DHE specifically reacts with O2−• to release fluorescent product 2-hydroxyethidium ( 2-OH-E+ ) , which can be conveniently detected by HPLC ( Kalyanaraman et al . , 2014 ) . The reaction of DHE with other oxidants produces ethidium ( E+ ) ( Tyagi et al . , 2015 ) . Due to biosafety challenges associated with a BSL3 category pathogen such as Mtb for HPLC , we measured O2−• levels inside the related but nonpathogenic Mycobacterium bovis BCG upon AG challenge . BCG cells were treated with AG ( 10X MIC ) for 3 hr and 6 hr , followed by DHE staining and HPLC . We found that BCG cells treated with AG generate peaks corresponding to O2−• ( 2-OH-E+ ) and other ROS ( E+ ) ( Figure 3C ) . The intensity of peaks was significantly higher at 6 hr post-treatment as compared to untreated control ( Figure 3C ) . As a control , we used a well-known O2−• generator ( menadione ) in our assay and similarly detected a 2-OH-E+ peak ( Figure 3C , inset ) . Thereafter , we determined whether the thiol-based antioxidant thiourea can reverse the influence of AG on viability of Mtb . Thiourea has recently been shown to protect Mtb from oxidative stress by modulating the expression of antioxidant genes ( Nandakumar et al . , 2014 ) . Mtb was co-incubated with various concentrations of thiourea and AG , and viability was measured after 10 days . Thiourea did not exert a significant effect on the survival of Mtb under normal growing conditions ( Figure 3D ) ; however , it did increase the survival of Mtb treated with 0 . 625X and 1 . 25X MIC of AG by ~10- and 5-folds , respectively ( Figure 3D ) . At higher AG concentrations ( 2 . 5X MIC ) , only 100 mM of thiourea showed a twofold protective effect ( Figure 3D ) . 10 . 7554/eLife . 25624 . 012Figure 3 . AG influences the internal redox physiology of Mtb . Wt Mtb or M . bovis BCG was grown to OD600 of 0 . 4 and treated with 10X MIC of AG ) . At indicated time points , cells were analyzed for ( A ) NADH/NAD+ estimation , ( B ) ROS measurement using oxidant-sensitive fluorescent dye; 2' , 7'-dichlorofluorescein diacetate ( DCFDA ) , and ( C ) Superoxide estimation using dihydroethidium ( DHE ) as described in Materials and methods . ( D ) Wt Mtb was grown as described earlier and exposed to specified concentrations of Amox in the presence of 8 µg/ml of Clav for 10 days in the presence or absence of thiourea and survival was measured using colony-forming unit ( CFU ) counts . Error bars represent standard deviations from the mean . p≤0 . 05 , p≤0 . 01 and **p≤0 . 001 . Data are representative of at least two independent experiments done in duplicate . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 012 The above data indicate that bactericidal consequences of AG may be dependent upon internal oxidant-antioxidant balance of Mtb . To demonstrate this unambiguously , we exploited a mycobacterial-specific non-invasive biosensor ( Mrx1-roGFP2 ) to measure the redox potential of a physiologically relevant and abundant cytoplasmic antioxidant , MSH ( Bhaskar et al . , 2014 ) . Any changes in the oxidation-reduction state of MSH can be reliably quantified by ratiometric measurement of emission at 510 nm after excitation at 405 and 488 nm ( Bhaskar et al . , 2014 ) . Mtb expressing Mrx1-roGFP2 was treated with lower ( 0 . 2X MIC ) and higher ( 10X MIC ) concentrations of AG and intramycobacterial EMSH was determined by measuring biosensor ratiometric response over time , as described previously ( Bhaskar et al . , 2014 ) . We observed a modest but consistent increase in 405/488 ratio at 6 hr and 24 hr post-treatment with 10X and 0 . 2X MIC of AG , respectively ( Figure 4A ) , indicating that antioxidant mechanisms are mostly efficient in minimizing the impact of AG-mediated ROS generation on internal EMSH of Mtb . 10 . 7554/eLife . 25624 . 013Figure 4 . AG induces oxidative shift in EMSH of Mtb in vitro and during infection . ( A ) Wt Mtb-expressing Mrx1-roGFP2 was treated with lethal ( 10X MIC ) and sub-lethal ( 0 . 2 X MIC ) concentrations of AG and ratiometric sensor response was measured at indicated time points by flow cytometry . ( B ) PMA-differentiated THP-1 cells were infected with Mtb expressing Mrx1-roGFP2 ( moi: 10 ) and treated with indicated concentrations of Amox in the presence of 8 µg/ml of Clav as described in Materials and methods . At the indicated time points , ~30 , 000 infected macrophages were analyzed by flow cytometry to quantify changes in Mtb subpopulations displaying variable EMSH as described in Materials and methods . ( C ) In parallel experiments , infected macrophages were lysed and bacillary load was measured by plating for CFU . Error bars represent standard deviations from the mean . p≤0 . 05 , p≤0 . 01 and p≤0 . 001 . Data are representative of at least two independent experiments done in duplicate . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 013 Importantly , to determine whether AG-induced oxidative stress is physiologically relevant in the context of infection , we measured dynamic changes in EMSH of Mtb inside human macrophage cell line ( THP-1 ) during infection . Infected macrophages were exposed to AG ( 1 . 25-fold to 10-fold of the in vitro MIC ) and the redox response was measured by flow cytometry . As reported earlier , Mtb cells inside macrophages displayed variable EMSH , which can be resolved into EMSH-basal ( −270 mV ) , EMSH-oxidized ( −240 mV ) , and EMSH-reduced ( −310 mV ) subpopulations ( Bhaskar et al . , 2014 ) . Treatment with AG induces significant increase in the oxidized subpopulation over time ( Figure 4B ) . In parallel , we examined whether the elevated oxidative stress correlates with the killing potential of AG during infection . Macrophages infected with Mtb were treated with 10X MIC of AG and bacillary load was monitored by enumerating colony-forming units ( CFUs ) at various time points post-infection . At 6 hr and 12 hr post-AG treatment , the effect on Mtb survival was marginal ( Figure 4C ) . However , an ~100-fold decline in CFU was observed at 24 hr and 36 hr post-AG treatment ( Figure 4C ) . More importantly , an increase in EMSH-oxidized subpopulation was observed at a time point where survival was not considerably affected ( 6 hr ) ( Figure 4B and C ) . This suggests that AG-mediated oxidative stress precedes bacterial death inside macrophages and that the intramycobacterial oxidative stress is not a consequence of AG-induced toxicity . Altogether , our data showed that AG perturbs mycobacterial redox physiology and the environment inside macrophages potentiates the mycobactericidal effect of AG . Since AG induces intramycobacterial oxidative stress , it is likely that the loss of major intracellular antioxidant , MSH , might potentiate the antimycobacterial activity of AG . To examine this , we used a MSH-negative strain ( MsmΔmshA ) ( Rawat et al . , 2002 , 2007 ) of Mycobacterium smegmatis ( Msm ) , an organism that is widely used as a surrogate for pathogenic strains of Mtb . Wt Msm and MsmΔmshA strains were exposed to various concentrations of Amox at a saturating concentration of Clav ( 8 μg/ml ) and percent growth inhibition was measured using the Alamar blue ( AB ) assay . AB is an oxidation-reduction indicator dye which changes its color from non-fluorescent blue to fluorescent pink upon reduction by actively metabolizing cells , whereas inhibition of growth by antimycobacterial compounds interferes with AB reduction and color development ( Tyagi et al . , 2015 ) . As shown in Figure 5A , at a fixed Clav concentration , MsmΔmshA exhibited ~3 and 10-fold higher inhibition at 5 μg/ml and 2 . 5 μg/ml of Amox as compared to wt Msm , respectively . At 10 μg/ml of Amox , both strains showed nearly complete inhibition ( Figure 5A ) . Next , we measured susceptibility to Clav at a fixed concentration of Amox ( 10 μg/ml ) . Higher concentrations of Clav ( 10 μg/ml ) inhibited the growth of Msm and MsmΔmshA with a comparable efficiency ( Figure 5B ) . However , while wt Msm overcomes the inhibitory effect of Amox at lower Clav concentrations , MsmΔmshA remained sensitive to Amox even at the lowest concentration of Clav ( 0 . 625 μg/ml ) ( Figure 5B ) . As shown in Figure 5B , MsmΔmshA exhibited ~7-fold greater inhibition at 0 . 625 μg/ml of Clav as compared to wt Msm . We further validated the contribution of MSH in tolerating AG by measuring the sensitivity of Msm lacking MSH-disulphide reductase ( Mtr ) activity ( MsmΔmtr ) ( Holsclaw et al . , 2011 ) and MSH-depleted ( MsmΔmshD ) strain toward Amox and AG . A twofold reduction in MIC for Amox and AG was detected in case of MsmΔmtr as compared to wt Msm , whereas MsmΔmshD remained unaffected ( Table 1 ) . Since MsmΔmshD contains only ~3% of total cellular MSH but accumulates two novel thiols ( Suc-MSH and formyl-MSH ) ( Newton et al . , 2005 ) , our data suggest that Msm can also alleviate redox stress caused by AG via Suc-MSH and/or formyl-MSH . Alongside MSH , other prominent oxidative stress defense mechanisms include the H2O2 detoxifying enzyme , catalase ( KatG ) , and NADPH-dependent thioredoxin ( TRX ) system . The extra-cytoplasmic sigma factor , SigH , is known to regulate several components of the TRX system in mycobacteria ( Raman et al . , 2001 ) . Therefore , we assessed the inhibition of Msm strains lacking KatG ( MsmΔkatG ) ( Padiadpu et al . , 2016 ) and SigH ( MsmΔsigH ) ( Fernandes et al . , 1999 ) by Amox and AG . Both strains exhibited a twofold increased susceptibility toward Amox and AG ( Table 1 ) , further confirming a link between mycobacterial redox physiology and AG efficacy . 10 . 7554/eLife . 25624 . 014Figure 5 . Mycothiol mediates tolerance to AG . Wt Msm and MsmΔmshA strains were grown to OD600 of 0 . 4 and either treated with various concentrations of ( A ) Amox and Clav ( 8 µg/ml ) or ( b ) Clav and Amox ( 10 µg/ml ) and % inhibition in growth was measured by Alamar blue ( AB ) assay as described in Materials and methods . ( C ) Wt Mtb , MtbΔmshA , mshA-comp and mshA:OE strains were exposed to 10X MIC of AG and survival was monitored by measuring CFU over time . Error bars represent standard deviations from mean . Data are representative of at least two independent experiments done in duplicate . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 01410 . 7554/eLife . 25624 . 015Table 1 . Minimum inhibitory concentrations ( MICs ) of Amox and AG for different Mycobacterium smegmatis strains . Source data file containing the images for MIC calculation is available in Table 1–source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 01510 . 7554/eLife . 25624 . 016Table 1—source data 1 . Images of Alamar blue assay plates for calculation of minimum inhibitory concentration ( MIC ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 016μg/mLStrainsAmoxAmox+clav ( AG ) wt Msm16020 + 8 MsmΔmtr8010 + 8 MsmΔmshD16020 + 8 MsmΔkatG8010 + 8 MsmΔsigH2010 + 8 To confirm that the above findings can be recapitulated in slow growing pathogenic mycobacteria ( i . e . Mtb H37Rv ) , we utilized mshA-deficient ( MtbΔmshA ) , mshA-complemented ( mshA-comp ) , and mshA-overexpressing ( mshA:OE ) strains of Mtb . The mshA:OE strain was generated by conditionally overexpressing mshA in Mtb using an anhydrotetracyline ( Atc ) -inducible system ( TetR ) ( Mehta et al . , 2016; Vilchèze et al . , 2008; Parikh et al . , 2013 ) . We performed EMSH measurements and confirmed that the overexpression of mshA shifted the ambient EMSH of Mtb from −275 ± 3 mV to −300 ± 5 mV , indicating an overall elevation in anti-oxidative potential of Mtb . The MSH-deficient strain of Mtb ( MtbΔmshA ) displayed the oxidative EMSH of >-240 mV , whereas mshA-complemented strain ( mshA-comp ) displayed a EMSH comparable to wt Mtb ( i . e . −275 ± 3 mV ) . Wt Mtb , mshA:OE , MtbΔmshA , and mshA-comp were exposed to 10X MIC of AG and growth was monitored over time by measuring CFUs . AG treatment resulted in a time-dependent decrease in the growth of Mtb strains ( Figure 5C ) . However , the decline was severe in case of MtbΔmshA as compared to wt Mtb , whereas mshA-OE showed relatively better tolerance than wt Mtb ( Figure 5C ) . Expression of mshA from its native promoter ( mshA-comp ) restored tolerance comparable to wt Mtb ( Figure 5C ) . In summary , AG exposure triggers the redox imbalance and cellular antioxidants such as MSH provide efficient tolerance toward AG . Altered expression of oxidative stress-specific genes , elevation of ROS , and perturbation of EMSH upon AG exposure suggest that intramycobacterial redox potential can serve as an internal cue to monitor the presence of β-lactams . Canonical intracellular redox sensors such as OxyR , SoxR , and FNR are either absent or rendered non-functional in Mtb ( Deretic et al . , 1995; Chawla et al . , 2012 ) . We have previously shown that Mtb features a Fe-S cluster-containing transcription factor ( WhiB4 ) , which responds to oxidative stress by regulating the expression of antioxidant genes ( Chawla et al . , 2012 ) . Since whiB4 expression is uniformly repressed by β-lactams ( e . g . meropenem and AG ) ( Lun et al . , 2014 ) and oxidative stress ( Chawla et al . , 2012 ) , WhiB4 appears to be critical in the β-lactam-induced oxidative stress response in Mtb . We assessed this connection by examining the expression of whiB4 in MSH-deficient ( MtbΔmshA ) and MSH-sufficient ( mshA-OE and mshA-comp ) strains by qRT-PCR . Expression analysis demonstrated that the whiB4 transcript was significantly repressed in MtbΔmshA ( −2 . 94 ± 0 . 22 fold ) , whereas expression is restored in mshA-OE ( −1 . 08 ± 0 . 09 fold ) and mshA-comp ( −1 . 16 ± 0 . 08 ) , in comparison to wt Mtb ( Figure 6—figure supplement 1 ) . Overall , WhiB4 regulatory function is modulated by the internal redox physiology of Mtb . Based on the above evidence , we tested the direct role of WhiB4 in β-lactam tolerance . We performed microarray analyses of MtbΔwhiB4 ( Chawla et al . , 2017 ) upon treatment with AG ( at 10X MIC of Mtb ) for 6 hr as described previously . A total of 495 genes were induced ( ≥1 . 5 fold; p value ≤ 0 . 05 ) and 423 were repressed ( ≥1 . 5 fold , p value ≤ 0 . 05 ) in the MtbΔwhiB4 as compared to wt Mtb upon AG-treatment ( supplementary file 2A ) . Our network analysis showed that diverse functional classes such as cell wall processes , virulence adaptation pathways , intermediary metabolism and respiration , and lipid metabolism were affected in MtbΔwhiB4 upon exposure to AG ( Figure 6A ) . Microarray data indicated higher expression of genes known to be involved in tolerance to β-lactams in MtbΔwhiB4 . Transcription of blaR and blaC was induced 8 . 43 ± 4 . 75 and 2 . 23 ± 0 . 19 fold , respectively , in MtbΔwhiB4 as compared to wt Mtb upon AG treatment ( Figure 6B , Figure 6—figure supplement 2 ) . Other genetic determinants of β-lactam tolerance such as PG biosynthetic genes ( murE , murF , and murG ) , penicillin-binding proteins ( Rv2864c , Rv3627c , and Rv1730c ) , and cell division and DNA transaction factors ( ftsK and fic ) ( Figure 6B ) were also up-regulated in MtbΔwhiB4 upon treatment . Furthermore , MtbΔwhiB4 showed greater expression of DNA repair genes ( SOS response ) , many of which are known to interfere with cell division and promote β-lactam tolerance in other bacterial species ( Miller et al . , 2004 ) ( Figure 6C ) . Since transcriptional data implicate WhiB4 in regulating the biosynthesis of the PG polymer , we stained the same from wt Mtb , MtbΔwhiB4 , and whiB4-OE cells using a fluorescent derivative of the PG binding antibiotic , vancomycin ( Bodipy-VAN ) and imaged the cells using confocal microscopy . To generate the whiB4-OE strain , we overexpressed WhiB4 using an inducer anhydrotetracycline ( Atc ) , in MtbΔwhiB4 as described previously ( Chawla et al . , 2012 ) . As expected , poles of wt Mtb and whiB4-OE cells were fluorescently labeled , consistent with the incorporation of nascent PG at the poles in mycobacteria ( Figure 6—figure supplement 3 ) ( Thanky et al . , 2007 ) . Interestingly , Bodipy-VAN was found to label the entire length of MtbΔwhiB4 , indicating deposition of PG along the entire body of the cylindrical cells ( Figure 6—figure supplement 3A ) . MtbΔwhiB4 cells were also marginally longer than wt Mtb ( Figure 6—figure supplement 3B ) . Further experimentations are required to understand how WhiB4 modulates PG biosynthesis and cell size . Nonetheless , our transcriptomics and imaging data are in reasonable agreement with each other and support the PG-regulatory function of WhiB4 in Mtb . 10 . 7554/eLife . 25624 . 017Figure 6 . WhiB4 regulates response to AG in Mtb . ( A ) Cumulative node weight intensities ( CNW ) of different functional classes regulated by WhiB4 upon AG treatment . ( B–G ) Heat maps depicting expression of genes ( log2fold-change , p≤0 . 05 ) coordinating cell wall processes , alternate respiration and CCM , antioxidants , DNA repair , PE and PE_PGRS and drug efflux pumps in case of Mtb and MtbΔwhiB4 treated with AG for 6 hr as described in Materials and methods . Cumulative node weight intensities for different functional classes are available in Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 01710 . 7554/eLife . 25624 . 018Figure 6—source data 1 . Cumulative node weight intensities for different functional classes as depicted in Figure 6A . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 01810 . 7554/eLife . 25624 . 019Figure 6—figure supplement 1 . qRT-PCR analysis of whiB4 expression in MtbΔmshA , mshA-comp , and mshA-OE strains . Fold change was measured with respect to untreated wt Mtb by normalizing expression with the 16srRNA transcript . Error bars represent standard deviations from mean . Data are representative of at least two independent experiments done in duplicate . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 01910 . 7554/eLife . 25624 . 020Figure 6—figure supplement 2 . qRT-PCR analysis of MtbΔwhiB4 exposed to 10X AG for 6 hr . Fold change for each transcript was measured with respect to wt Mtb exposed to 10X AG for 6 hr by normalizing expression with the 16srRNA transcript . Error bars represent standard deviations from mean . Data are representative of at least two independent experiments done in duplicate . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 02010 . 7554/eLife . 25624 . 021Figure 6—figure supplement 3 . Vancomycin-BODIPY staining of different Mtb strains . ( A ) Wt Mtb , MtbΔwhiB4 , and whiB4-OE were grown to OD600 nm of 0 . 6 and incubated with 1 µg/ml of Vancomycin-BODIPY for 16 hr and visualized by confocal microscopy ( 63X ) . The scale of images is 3 µm ( B ) Scatter plot showing the cell measurement for above mentioned strains . Each dot represents one cell ( n > 150 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 021 Other mechanisms that could link WhiB4 with drug tolerance include the heightened expression of cation transporters , ABC-transporters , PDIM lipid biogenesis ( ppsC , ppsD , ppsE , drrA , drrB , and drrC ) , and drug-efflux pumps ( Rv1258c and Rv1634 ) in MtbΔwhiB4 upon AG exposure ( Figure 6D ) . Lack of the Rv1258c pump and PDIM lipids have been reported to sensitize Mtb towards β-lactams and vancomycin ( Dinesh et al . , 2013; Soetaert et al . , 2015 ) . Several PE_PGRS genes involved in maintaining cell wall architecture and protection from oxidative stresses were up-regulated in MtbΔwhiB4 ( Figure 6E ) ( Fishbein et al . , 2015 ) . Our results also revealed that redox-metabolism is significantly altered in MtbΔwhiB4 in response to AG . For example , components of the NADH-dependent peroxidase ( ahpCD ) , peroxynitrite reductase complex ( dlaT ) , thiol-peroxidase ( tpx ) , mycothiol biosynthesis ( mshA ) , pyruvate dehydrogenase complex ( pdhA , pdhC , and aceE ) , and hemoglobin like proteins ( glbO ) were induced in AG-challenged MtbΔwhiB4 compared to wt Mtb ( Figure 6F ) . Importantly , most of these enzymatic activities are well known to confer protection against oxidative and nitrosative stress in Mtb ( Master et al . , 2002; Hu and Coates , 2009; Ung and Av-Gay , 2006; Maksymiuk et al . , 2015; Venugopal et al . , 2011; Pathania et al . , 2002 ) . Furthermore , similar to wt Mtb , primary NADH dehydrogenase complex ( nuo operon ) was down-regulated in MtbΔwhiB4 in response to AG treatment ( supplementary file 2B ) . However , compensatory increase in alternate respiratory complexes such as ndh and cydAB was notably higher in MtbΔwhiB4 than in wt Mtb , indicating that MtbΔwhiB4 is better fit to replenish reducing equivalents during drug-induced cellular stress ( Figure 6G ) . In tune with this , components of the TCA cycle and pentose phosphate pathway involved in generating cellular reductants ( NADH and NADPH ) were induced in MtbΔwhiB4 as compared to wt Mtb . We validated our microarray data by performing qRT-PCR on a few genes deregulated upon AG treatment in MtbΔwhiB4 ( Figure 6—figure supplement 2 ) . Overall , AG-exposure elicits transcriptional changes , which are indicative of a higher potential of MtbΔwhiB4 to maintain redox homeostasis upon drug exposure . We directly assessed this by examining changes in EMSH of MtbΔwhiB4 and whiB4-OE in response to AG in vitro and inside macrophages using Mrx1-roGFP2 biosensor as described earlier . Under both culture conditions , MtbΔwhiB4 robustly maintained intramycobacterial EMSH , whereas Atc-induced overexpression of whiB4 in MtbΔwhiB4 showed a significant oxidative shift ( Figure 7 ) . In sum , our results suggest that WhiB4 can mediate AG tolerance by regulating multiple mechanisms , including PG biogenesis , SOS response , and redox balance . 10 . 7554/eLife . 25624 . 022Figure 7 . WhiB4 regulates AG-induced oxidative shift in EMSH of Mtb both in vitro and during infection . ( A ) MtbΔwhiB4 and whiB4-OE expressing Mrx1-roGFP2 were treated with lethal ( 10X MIC ) and sub-lethal ( 0 . 2 X MIC ) concentrations of AG and ratiometric response was measured by flow cytometry at indicated time points . ( B ) PMA differentiated THP-1 cells were infected with MtbΔwhiB4 and whiB4-OE expressing Mrx1-roGFP2 ( MOI:10 ) and treated with indicated concentrations of Amox in the presence of Clav ( 8 µg/ml ) as described in Materials and methods . At indicated time points , ~30 , 000 infected macrophages were analyzed by flow cytometry to quantify changes in Mtb subpopulations displaying variable EMSH as described in Materials and methods . p≤0 . 05 , p≤0 . 01 and p≤0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 022 Our data indicated that WhiB4 modulates the expression of genes involved in β-lactam tolerance ( blaR and blaC ) and redox metabolism ( mshA , ahpCD , and tpx ) . Using qRT-PCR , we confirmed that the expression of blaR and blaC was 8 . 43 ± 4 . 75 and 2 . 23 ± 0 . 19 fold higher , respectively , in MtbΔwhiB4 as compared to wt Mtb upon exposure to AG ( Figure 6—figure supplement 2 ) . Next , we examined WhiB4 interaction with upstream sequences of blaR and blaC using EMSA . Earlier , we have shown that WhiB4 contains a 4Fe-4S cluster , which is extremely sensitive to degradation by atmospheric oxygen ( Chawla et al . , 2012 ) . Moreover , WhiB4 lacking the Fe-S cluster ( apo-WhiB4 ) binds DNA and represses transcription upon oxidation of its cysteine thiols and formation of disulfide-linked oligomers , while reduction of disulfides reversed WhiB4 oligomerisation , DNA binding , and repressor function ( Chawla et al . , 2012 ) . We generated thiol-reduced and -oxidized forms of apo-WhiB4 as described previously ( Chawla et al . , 2012 ) . The oxidized and reduced apo-WhiB4 fractions were incubated with 32P-labeled DNA fragments of blaC ( ~100 bp upstream ) and blaR ( ~180 bp upstream ) and blaC/blaR-promoter complex formation was visualized using EMSA . As shown in Figure 8A and B , oxidized apo-WhiB4 binds blaC/blaR-promoter DNA in a concentration-dependent manner , whereas this binding was significantly reversed in case of reduced apo-WhiB4 . Since WhiB4 bind to its own promoter ( Chawla et al . , 2012 ) , we confirmed that oxidized apo-WhiB4 binds to its promoter in concentrations comparable to that required for binding blaC and blaR upstream sequences ( Figure 8C ) . We also performed competition assays using blaC and blaR upstream sequences as positive controls , while promoter fragment of Rv0986 was utilized as a negative control . We found that 100-fold molar excess of blaC and blaR DNA fragments completely prevented apo-WhiB4 binding . However , the same concentration of an unlabeled Rv0986 promoter fragment was inefficient to out-compete apo-WhiB4 association with blaC and blaR DNA fragments ( Figure 8—figure supplement 1 ) . Next , we performed in vitro transcription assays using a highly sensitive Msm RNA polymerase holoenzyme containing stoichiometric concentrations of principal Sigma factor , SigA ( RNAP-σA ) ( Chawla et al . , 2012 ) and determined the consequence of WhiB4 on blaC transcript . As shown in Figure 8D , addition of oxidized apo-WhiB4 noticeably inhibited transcription from blaC promoter , whereas reduced apo-WhiB4 restored normal levels of blaC transcript . Lastly , we directly measured BlaC activity in the cell-free extracts derived from wt Mtb , MtbΔwhiB4 , and whiB4-OE strains using a chromogenic β-lactam nitrocefin as a substrate ( Flores et al . , 2005b ) . Cell-free extracts of MtbΔwhiB4 possessed ~70% higher and whiB4-OE showed ~30% reduced nitrocefin hydrolysis as compared to wt Mtb , respectively ( Figure 8E ) . We have earlier shown that WhiB4 predominantly exists in an oxidized apo-form upon overexpression inside mycobacteria during aerobic growth ( Chawla et al . , 2012 ) . Therefore , decreased BlaC activity upon WhiB4 overexpression is most likely a consequence of oxidized apo-WhiB4-mediated repression of blaC in vivo . To clarify the physiological relevance of redox- and whiB4-dependent transcription of blaC , we shifted the internal redox balance of whiB4-OE using a cell permeable thiol-oxidant , diamide ( 5 mM ) , or a thiol-reductant , DTT ( 5 mM ) , and measured nitrocefin hydrolysis by cell-free extracts . We have previously reported that treatment with 5 mM diamide or DTT did not adversely affect growth of Mtb ( Singh et al . , 2009 ) . However , treatment with DTT significantly reduced disulfide-linked oligomers of oxidized apo-WhiB4 to regenerate WhiB4 thiols in vivo ( Chawla et al . , 2012 ) . Pretreatment of whiB4-OE with DTT largely restored BlaC activity to MtbΔwhiB4 levels , whereas diamide did not lead to further decrease in BlaC activity ( Figure 8F ) . Effective reduction of disulfides in oxidized apo-WhiB4 by DTT may have led to loss of WhiB4 mediated DNA binding and transcriptional repression , thereby causing elevated blaC expression and activity in whiB4-OE . Taken together , these results led us to conclude that WhiB4 regulates β-lactamase expression and activity in a redox-dependent manner . 10 . 7554/eLife . 25624 . 023Figure 8 . Regulation of β-lactamase by WhiB4 in a redox-dependent manner . Oxidized ( WhiB4-SS ) and reduced ( WhiB4-SH ) forms of apo-WhiB4 were prepared . The concentrations of apo-WhiB4 used for EMSAs were 0 . 5 , 1 , 2 , and 4 μM . EMSA reactions were performed with 0 . 5 nM 32P-labelled blaC ( A ) , blaR ( B ) and whiB4 ( C ) promoter DNA fragments . C: DNA binding in the absence of WhiB4 in each panel . ( D ) Effect of WhiB4 on in vitro transcription . Single-round transcription assays show that RNAP-σA efficiently directs transcription from the blaC promoter . 100 nM of blaC promoter DNA fragment was pre-incubated with either 1 μM WhiB4-SS or WhiB4-SH and subjected to transcription by RNAP-σA as described in Materials and methods . C: blaC transcript in the absence of WhiB4 . ( E ) 100 μg of cell-free lysates derived from exponentially grown ( OD600 of 0 . 6 ) wt Mtb , MtbΔwhiB4 and whiB4-OE were used to hydrolyze nitrocefin . β-lactamase activity was measured by monitoring absorbance of hydrolyzed nitrocefin at 486 nm as described in Materials and methods . The fold change ratios clearly indicate a significantly higher or lower β-lactamase activity in MtbΔwhiB4 or whiB4-OE , respectively , as compared to wt Mtb . p-Values are shown for each comparison . ( F ) whiB4-OE strain was pre-treated with 5 mM of DTT or Diamide and β-lactamase activity in cell-free lysates was compared to MtbΔwhiB4 over time . p≤0 . 05 , p≤0 . 01 and p≤0 . 001 . Data are representative of at least two independent experiments done in duplicate . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 02310 . 7554/eLife . 25624 . 024Figure 8—figure supplement 1 . EMSA cold competition assay . ( A ) 0 . 5 μM of oxidized form of apo-WhiB4 ( WhiB4-SS ) was incubated with 0 . 5 nM 32P-labeled blaC and blaR promoter DNA fragments . Lanes 1 and 8: free probe; Lanes 2 and 9: WhiB4:blaC/blaR-promoter DNA complex . WhiB4 DNA binding was competed out using 10- ( lanes 3 and 10 ) , 25- ( lanes 4 and 11 ) , 50- ( lanes 5 and 12 ) , 100- ( lanes 6 and 13 ) , and 200- ( lanes 7 and 14 ) fold molar excess of unlabeled competitor DNA , blaC and blaR . ( B ) Competition assay of 0 . 5 μM of WhiB4-SS binding using specific ( blaC ) and non-specific ( Rv0986 ) promoter DNA fragments . Lane 1: 32P-labeled blaC DNA ( free probe ) , Lane 2: WhiB4-SS:blaC-promoter DNA complex . WhiB4 DNA binding was competed using 100-fold molar excess of unlabeled blaC DNA ( lane 3; specific ) or Rv0986 DNA ( lane 4; non-specific ) . ( C ) Competition assay of 0 . 5 μM of WhiB4-SS binding using specific ( blaR ) and non-specific ( Rv0986 ) promoter DNA fragments . Lane 1: 32P-labeled blaR DNA ( free probe ) , Lane 2: WhiB4-SS:blaR-promoter DNA complex . Competing unlabeled DNA fragments were either blaR DNA ( lane 3; specific ) or Rv0986 DNA ( lane 4; non-specific ) , both used in 100-fold molar excess concentrations . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 024 Based on the above results , we hypothesize that WhiB4-sufficient and -deficient strains would have differential susceptibility toward β-lactams . We found that MtbΔwhiB4 uniformly displayed ~4–8 fold higher MICs against β-lactams as compared to wt Mtb ( Table 2 ) . This effect was specific to β-lactams , as the loss of WhiB4 did not alter MICs for other anti-TB drugs such as INH and RIF ( Table 2 ) . More-interestingly , over-expression of WhiB4 displayed ~2–4 fold greater sensitivity toward β-lactams as compared to wt Mtb ( Table 2 ) . We predicted that if WhiB4 is controlling tolerance to β-lactams by regulating blaC expression , we would see variations in inhibitory concentrations of Clav against wt Mtb , MtbΔwhiB4 , and whiB4-OE at a fixed concentration of Amox . As expected , inhibition of MtbΔwhiB4 by 10 μg/ml of Amox requires four fold and eight fold higher Clav as compared to wt Mtb and whiB4-OE strains , respectively ( Figure 9A ) . Phenotypic data are in complete agreement with the higher and lower BlaC activity in MtbΔwhiB4 and whiB4-OE , respectively . Studies in animals and humans have demonstrated higher efficacy of β-lactams and β-lactamase inhibitor combination against MDR/XDR-TB . Our results show that WhiB4 overexpression significantly elevated the capacity of β-lactams to inhibit drug-sensitive Mtb . To investigate whether WhiB4 overexpression similarly affects growth of drug-resistant strains , we over-expressed WhiB4 in clinical strains isolated from Indian patients ( single-drug resistant [SDR; BND320] , multi-drug resistant [MDR; JAL 2261 and JAL 1934] and extensively drug-resistant [XDR; MYC 431] ) ( Bhaskar et al . , 2014; Kumar et al . , 2010 ) and determined sensitivity toward Amox ( at various concentrations ) and Clav ( 8 μg/ml ) . As expected , drug-resistant strains over-expressing WhiB4 were ~2–4 fold more sensitive to Amox and Clav combinations than controls ( Figure 9B , C , D and E ) . 10 . 7554/eLife . 25624 . 025Figure 9 . WhiB4 regulates AG tolerance in drug-sensitive and -resistant strains of Mtb . ( A ) Wt Mtb , MtbΔwhiB4 and whiB4-OE were incubated with Amox ( 10 µg/ml ) and different concentrations of Clav and % inhibition of growth was measured by AB assay as described in Materials and methods . To determine if WhiB4 modulates the sensitivity of AG in drug-resistant strains , WhiB4 was over-expressed in clinical strains ( B ) BND 320 ( C ) JAL 1934 , ( D ) JAL 2261 , and ( E ) MYC 431 . Cells were incubated with Clav ( 8 µg/ml ) and different concentrations of Amox . The percent growth inhibition was measured by AB assay as described in Materials and methods . WhiB4 modulates susceptibility to AG during acute infection in mice ( F–G ) . Inbred BALB/c mice ( n = 3 ) were given various strains of Mtb in the form of an aerosol and orally administered with Amox ( 200 mg/kg of body weight ) and Clav ( 50 mg/kg of body weight ) that is AG twice a day starting from day 3 post-infection . Bacterial burden in the lungs was assessed by checking the survival of Mtb strains using CFU analysis . Statistical significance for the pulmonic bacterial load was obtained as follows: by comparing the CFU obtained from AG-treated Wt Mtb and MtbΔwhiB4 strains: p≤0 . 01 and **p≤0 . 001 , by comparing CFU obtained from AG-treated Wt Mtb and whiB4-OE strains: + p≤0 . 05 , by comparing CFU obtained from AG-treated MYC 431 and MYC 431/whiB4-OE strains: ### p≤0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 02510 . 7554/eLife . 25624 . 026Table 2 . Minimum inhibitory concentrations ( MICs ) of cell wall targeting drugs for different Mycobacterium tuberculosis strains . Source data file for the calculation of MIC values is available in Table 2–source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 02610 . 7554/eLife . 25624 . 027Table 2—source data 1 . Percentage growth inhibition values for Mtb , MtbΔwhiB4 and whiB4-OE in presence of different drugs for calculation of minimum inhibitory concentration ( MIC ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 027Drugsµg/mlMtbMtbΔwhib4B4whiB4-OE Amoxicillin80>16040 Ampicillin5004000250 Cloxacillin400800200 Carbenicillin10244096512 Meropenem5202 . 5 Penicillin200800100 Lysozyme5020025 Vancomycin10802 . 5 Isoniazid0 . 06250 . 06250 . 03125 Rifampicin0 . 06250 . 06250 . 0625 Lastly , we asked if WhiB4 influences tolerance to AG during infection . Poor half-life of AG in mice makes it challenging to assess the efficacy of AG in vivo ( Rullas et al . , 2015 ) . However , AG induces a marginal ( ~0 . 5 log reduction ) killing of Mtb in an acute model for TB infection in mice ( Solapure et al . , 2013 ) . Therefore , we compared bacillary load of Mtb strains in the lungs of mice during acute infection ( see Materials and methods ) . Approximately 104 bacterial cells were implanted into the lungs of BALB/c mice ( Figure 9F ) and at 3 days post-infection mice were treated with AG ( 200 mg/kg body weight of Amox and 50 mg/kg body weight of Clav ) twice a day for 2 and 4 weeks . Bacterial numbers were determined in the infected lungs upon treatment . At 3 days post-infection , bacillary load was comparable between Mtb strains ( Figure 9F ) . At 2 and 4 weeks post-treatment , wt Mtb exhibited ~7-fold reduction in bacillary load than untreated mice ( Figure 9F ) . Overexpression of WhiB4 resulted in ~4- and~38-fold decline in CFU at 2 and 4 weeks , post-treatment , as compared to untreated animals ( Figure 9F ) . In contrast , MtbΔwhiB4 either displayed an increase ( ~8-fold ) or maintained a comparable bacillary load at 2 or 4 weeks post-treatment , respectively , relative to untreated mice ( Figure 9F ) . Lastly , we overexpressed WhiB4 in the MYC 431 XDR strain ( 431/whiB4-OE ) and examined AG efficacy in mice as described earlier . As expected , WhiB4 overexpression increased the sensitivity of MYC 431 toward AG treatment at 4 weeks post-infection ( Figure 9G ) . We documented that WhiB4 overexpression significantly affects the survival of Mtb in vivo , an outcome that is most likely due to WhiB4-directed repression of the antioxidant systems and β-lactamase . In conclusion , our results suggest that WhiB4 plays a central role in coordinating Mtb tolerance to AG . We revealed a redox-based mechanism underlying tolerance to a β-lactam and β-lactamase inhibitor combination , which is actively considered to treat drug-resistant Mtb infections . Importantly , these findings should be viewed in light of recent studies debating the contribution of antibiotic-induced redox perturbations in antibiotic action and tolerance ( Kohanski et al . , 2007; Foti et al . , 2012; Brynildsen et al . , 2013; Liu and Imlay , 2013; Keren et al . , 2013 ) . We have shown how the primary targets of antibiotics ( e . g . PG biogenesis and β-lactamase ) and their secondary consequences ( redox stress and metabolic perturbations ) are functionally associated with each other through a redox-sensitive transcription factor , WhiB4 , in Mtb . Considering the fact that drug-resistance in Mtb is a global burden , our results showing that WhiB4-mediated changes in redox potential of Mtb can potentiate killing of clinical drug-resistant forms of Mtb by AG are novel and unique . We identified the internal redox potential of Mtb as a crucial determinant of mycobacterial sensitivity to AG , and demonstrated the central role of WhiB4 in maintaining redox balance and regulating gene expression . Down-regulation of TCA cycle genes and up-regulation of the glyoxylate cycle in response to AG are consistent with the reports of elevated tolerance to diverse bactericidal antibiotics , including β-lactams , in bacteria with diminished fluxes through the TCA cycle ( Kohanski et al . , 2007; Nguyen et al . , 2011 ) . In concurrence with this , metabolomic profiling of Mtb in response to other anti-TB drugs elegantly showed that tolerance is accompanied with reduced TCA cycle activity and elevated fluxes through the glyoxylate shunt ( Nandakumar et al . , 2014 ) . Mtb exhibits tolerance to antibiotics during non-replicating persistence in hypoxia ( Baek et al . , 2011 ) . Under these conditions , drug tolerance was accompanied by a redirection of respiration from the energetically efficient route ( e . g . NADH dehydrogenase I ) to the less energy efficient course ( e . g . NDH , CydAB oxidase ) , and any interference with this respiratory-switch over ( e . g . CydAB mutation ) leads to resensitization of mycobacteria to antibiotics ( Rao et al . , 2008; Lu et al . , 2015 ) . This seems to be a unifying theme underlying tolerance to conventional as well as the newly discovered anti-TB drugs bedaquiline ( BDQ ) and Q203 ( Lamprecht et al . , 2016 ) . In support of this , we found that exposure of Mtb to AG elicited a transcriptional signature that indicated a shift from the energy efficient respiration to the energetically less favored pathways , as evidenced by a significant induction of ndh and cydAB transcripts and a down-regulation of nuo , cydbc1 , and atp A-H . In bacteria , including Mtb , cytochrome bd oxidase also displays catalase and/or quinol oxidase activity ( Lu et al . , 2015; Al-Attar et al . , 2016 ) , which confers protection against oxidative stress and nitrosative stress . On this basis , upregulation of cytochrome bd oxidase in response to AG is indicative of oxidative stress in Mtb . Bactericidal antibiotics , including β-lactams , have been consistently shown to produce ROS as a maladaptive consequence of primary drug-target interaction on TCA cycle and respiration ( Kohanski et al . , 2007; Lobritz et al . , 2015; Dwyer et al . , 2014 ) . While this proposal has been repeatedly questioned ( Liu and Imlay , 2013; Keren et al . , 2013 ) , it is strongly reinforced by multiple independent studies demonstrating that tolerance to antibiotics is linked to the bacterial ability to nullify antibiotic-triggered ROS toxicity ( Nguyen et al . , 2011; Wang and Zhao , 2009; Gusarov et al . , 2009; Shatalin et al . , 2011 ) . We confirmed that AG stimulates oxidative stress in Mtb in vitro and during infection . However , in contrast to other studies ( Kohanski et al . , 2007 ) , oxidative stress was not associated with a breakdown of the NADH/NAD+ homeostasis , likely reflecting efficient ETC fluxes through NDH and cytochrome bd oxidase . In Mtb , rerouting of electron fluxes through cytochrome bd oxidase increases oxygen consumption ( Lamprecht et al . , 2016 ) , which can trigger O2−• and H2O2 generation by univalent reduction of O2 by the metal , flavin , and quinone containing cofactors of the respiratory enzymes ( Imlay , 2003 , 2013 ) . Recently , it has been shown that intramycobacterial antioxidant buffer , MSH , protects Mtb from small molecule endogenous superoxide generators and ROS-generated by vitamin C ( Tyagi et al . , 2015; Vilchèze et al . , 2013 ) . Specific to AG , we found that anti-mycobactericidal activity is greatly potentiated in MSH-deficient mycobacterial strains , whereas a MSH overexpressing strain displayed tolerance . This is all consistent with the generation of ROS and MSH as key regulatory mechanisms underlying AG tolerance . Studies indicated the importance of a broader range of physiological programs such as altered metabolic state and oxidative stress as contributory factors in antibiotic resistance . However , it is not clear if specific regulatory mediators exist which can assess physiological changes to regulate both primary drug targets and secondary consequences of drug-target interactions to functionally coordinate tolerance . Mechanisms of drug tolerance are either controlled by global changes in bacterial physiology by ppGpp or toxin-antitoxin ( TA ) modules ( Harms et al . , 2016 ) . Further , regulatory systems such as SoxRS in other bacteria and WhiB7 in Mtb facilitate physiological changes required for formation of drug-tolerant persisters without specifically affecting the expression of direct targets of antibiotics ( Morris et al . , 2005; Aly et al . , 2015 ) . We , for the first time , identified WhiB4 as a transcriptional regulator of both the genetic determinants of β-lactam resistance ( e . g . β-lactamase ) and physiological changes associated with phenotypic drug tolerance in Mtb ( e . g . redox balance ) . Due to lack of an extracellular β-lactam-sensing domain in Mtb BlaR , how Mtb responds to β-lactam remains unknown . While several possibilities including the involvement of serine/threonine protein kinases ( PknA/PknB ) containing β-lactam interacting PASTA domains are suggested to regulate BlaR-BlaI activity ( Sala et al . , 2009; Mir et al . , 2011 ) , our findings implicate internal redox balance and WhiB4 in responding to β-lactams . We detected that oxidized apo-WhiB4 binds and represses the expression of BlaR and BlaC , whereas reduction reversed this effect . Loss of WhiB4 derepresses BlaR and stimulates the expression and activity of BlaC , possibly via BlaR-mediated cleavage of the repressor of blaC ( i . e . BlaI ) . In addition to blaC , BlaI also binds to the promoters of genes encoding cytochrome bd oxidase and ATP synthase ( Sala et al . , 2009 ) , both of which showed higher expression in MtbΔwhiB4 . Altogether it indicates that regulatory function of BlaI is dependent upon the ability of WhiB4 to coordinate blaR expression in response to redox changes associated with β-lactam exposure . Our findings indicate a possible regulatory loop between the electron transport chain and β-lactam-induced oxidative stress where WhiB4/BlaI/BlaR may act as an important link between them ( Figure 10 ) . The biogenesis of PG is an energy requiring process and the cell wall damage caused by β-lactam antibiotics can perturb membrane function thereby affecting respiration , ATP generation , and redox balance . All these events can cause metabolic paralysis leading to inhibition of PG biogenesis and death . Supporting this notion , a recent study on the mechanisms of β-lactam toxicity showed that β-lactams cause metabolic instability due to activation of a futile cycle of PG biogenesis and degradation ( Cho et al . , 2014 ) . Therefore , tolerance to β-lactams would require active cooperation between mechanisms to maintain metabolic function , redox balance and β-lactamase activity , which are partly regulated by WhiB4 in Mtb . Under unstressed conditions , uncontrolled expression of genes such as blaC and cydAB is prevented by WhiB4-mediated DNA binding and repression of the blaR-blaI locus . This is possible since the WhiB4 Fe-S cluster is uniquely sensitive to oxygen and a fraction of WhiB4 exists in the apo-oxidized form inside aerobically growing Mtb ( Chawla et al . , 2012 ) . Since oxidized apo-WhiB4 is known to repress its own expression ( Chawla et al . , 2012 ) , Mtb can down-regulate the expression of whiB4 by elevating the levels of oxidized apo-WhiB4 in response to oxidative stress caused by β-lactams . The down-regulation of WhiB4 can reduce its negative influence on gene expression , necessary to adjust the expression of blaI , blaR , and blaC as well as genes involved in maintaining respiration and redox balance to neutralize β-lactam toxicity ( Figure 10 ) . Our data confirmed this by demonstrating consistent repression of whiB4 expression by AG treatment , oxidative stress , and upon MSH loss ( in MtbΔmshA ) . The down-regulation of whiB4 in MtbΔmshA is most likely a compensatory strategy to tolerate AG in the absence of MSH , albeit unsuccessfully , indicating that both WhiB4 and MSH are together required to tolerate β-lactam antibiotics in Mtb . Furthermore , induction of mshA in MtbΔwhiB4 in response to AG indicates a redox-regulatory loop between WhiB4 and MSH to tolerate oxidative and antibiotic stress in Mtb . Lastly , considering that WhiB4 might affect gene expression by altering nucleoid architecture ( Chawla et al . , 2012 ) , the exact details of how WhiB4 regulates global gene expression is a part of an ongoing study . Data from this genome-scale DNA binding study ( ChIP-seq ) indicate that WhiB4 binds largely in a non-specific fashion to the Mtb chromosome with a particular preference to GC-rich regions including an intergenic region of blaI-blaR ( >70% GC-rich ) ( manuscript in preparation ) . 10 . 7554/eLife . 25624 . 028Figure 10 . Model showing redox basis of AG tolerance in Mtb . Cell wall damage caused by AG can perturb the membrane integrity thereby affecting respiratory chain , redox balance , and ATP generation . All of this results in metabolic instability and AG-induced killing . To tolerate AG , Mtb redirects respiration from the energetically efficient route ( e . g . NDH1 , CyBC1 ) to the energetically poor one ( e . g . NDH2 , CyBD ) , and carbon metabolism from the TCA cycle to glyoxylate , glycolysis and gluconeogenesis . Rerouting of electron flux through CyBD can trigger generation of ROS ( O2−• and H2O2 ) by univalent reduction of O2 via metal- , flavin- , and quinone-containing respiratory enzymes . The intramycobacterial redox buffer , MSH , detoxifies ROS to protect Mtb from AG . The oxidative shift in EMSH of Mtb in response to AG serves as a cue to calibrate the expression of β-lactamase , PG enzymes , carbon metabolism , antioxidants , and alternate respiration via WhiB4 . Under native conditions , O2-induced loss of WhiB4 Fe-S cluster generates oxidized apo-WhiB4 , which binds and represses the expression of blaR and blaC . Reduction of oxidized apo-WhiB4 disulfides reversed this effect . Down-regulation of whiB4 in response to AG derepresses blaR and stimulates expression of blaC directly and/or indirectly via BlaR-mediated cleavage of the blaC repressor ( i . e . BlaI ) to induce AG tolerance . Accumulation of oxidized apo-WhiB4 upon overexpression led to hyper-repression of BlaC activity and oxidative shift in EMSH to potentiate mycobactericidal activity of AG . Since genes associated with alternate-respiration ( e . g . CyBD ) and energy metabolism ( e . g . ATP synthase ) are also regulated by BlaI , our results suggest cross-talk between WhiB4 and BlaI pathways resulting in AG tolerance of Mtb . Altogether , WhiB4 couples the changes in the redox physiology of Mtb triggered by AG to the expression of genes involved in antibiotic tolerance and redox homeostasis . MA: Mycolic acid , CM: Cytoplasmic membrane , NDH1: NADH-dehydrogenase I ( nuo operon ) , NDH2: NADH dehydrogenase 2 ( ndh ) , CyBD: Cytochrome BD oxidase , CyBC1: Cytochrome BC1-aa3 oxidase , F0F1 ATP syn: ATP Synthase , PBP: Penicillin-binding proteins and SDH: Succinate Dehydrogenase . Bold or dashed arrows indicate increased or decreased electron flow through respiratory complexes , respectively , based on gene expression data . DOI: http://dx . doi . org/10 . 7554/eLife . 25624 . 028 In summary , our study discovered a new redox-based mechanism of AG tolerance in Mtb . In particular , WhiB4 functions as an important regulatory protein that integrates internal redox changes triggered by β-lactams to fine-tune the expression of both genetic and phenotypic determinants of antibiotic tolerance in Mtb . Based on this work , we predict that compounds/drugs targeting bacterial systems that remediate oxidative damage ( e . g . 4-butyl-4-hydroxy-1- ( 4-hydroxyphenyl ) −2-phenylpyrazolidine-3 , 5-dione ) ( Gold et al . , 2012 ) , elevate endogenous ROS ( e . g . clofazimine/vitamin C ) ( Bhaskar et al . , 2014; Vilchèze et al . , 2013 ) , inhibit respiration ( e . g . Q2O3 ) ( Lamprecht et al . , 2016 ) , and block ATP homeostasis ( e . g . bedaquiline ) ( Lamprecht et al . , 2016 ) could be be effective companions to potentiate the action of β-lactam and β-lactamase combinations in Mtb . Details of mycobacterial strains and reagents used in this study are given in Supplementary file 3A and Supplementary file 3B . The mycobacterial strains were grown aerobically in 7H9 broth or 7H11 agar supplemented with 0 . 2% glycerol , Middlebrook Oleic acid Albumin Dextrose-Catalase ( OADC ) or 1X Albumin Dextrose Saline ( ADS ) enrichment and 0 . 1% Tween 80 ( broth ) . E . coli cultures were grown in LB medium . Antibiotics were added as described earlier ( Chawla et al . , 2012 ) . For WhiB4 overexpression , whiB4-OE strain was grown aerobically to an OD600 of 0 . 3 , followed by induction with 200 ng/ml anhydrotetracycline ( Atc ) at 37°C for 18 hr . The human monocytic cell line THP-1 ( RRID:CVCL_0006 ) was differentiated using 10–15 ng/ml phorbol 12-myristate 13-acetate ( PMA ) and cultivated for infection experiments as described previously ( Padiadpu et al . , 2016 ) . THP-1 ( ATCC TIB-202 ) cells authenticated by STR analysis by ATCC were treated with 25 µg/ml of Plasmocin for 3 weeks and tested negative for mycoplasma contamination by DE-MyoX Mycoplasma PCR Detection Kit . Sensitivity to various drugs was determined using the microplate alamar blue assay ( AB ) . AB assay was performed in 96-well flat bottom plates . Mtb or Msm strains were cultured in 7H9-ADS medium and grown till exponential phase ( OD600 of 0 . 6 ) . Approximately 1 × 105 bacteria were taken per well in a total volume of 200 µl of 7H9-ADS medium . Wells containing no Mtb were used for autofluorescence control . Additional controls consisted of wells containing cells and medium only . Plates were incubated for 5 days ( Mtb ) or 16 hr ( Msm ) at 37°C , 30 µl ( 0 . 02% wt/vol stock solution ) Alamar blue was added . Plates were re-incubated for color transformation ( blue to pink ) . Fluorescence intensity was measured in a SpectraMax M3 plate reader ( Molecular Device ) in top-reading mode with excitation at 530 nm and emission at 590 nm . Percentage inhibition was calculated based on the relative ﬂuorescence units and the minimum concentration that resulted in at least 90% inhibition was identiﬁed as MIC . Mycobacterium bovis BCG was cultured in 5 mL of Middlebrook 7H9 medium with 10% albumin-dextrose-saline ( ADS ) supplement at 37°C and grown till OD600 of 0 . 4 . The cultured bacteria were centrifuged to aspirate out the medium and re-suspended with fresh 7H9 medium . This bacterial solution was incubated with AG for 3 hr and 6 hr time points and 100 μM DHE was added for 1 hr in dark . The suspension was centrifuged to aspirate out any excess compounds and DHE in the medium . The collected bacterial pellet was re-suspended with acetonitrile and the cells were lysed using a probe sonicator for 3 min on ice . The cell lysate was then removed by centrifugation and the supernatant was separated and injected in Agilent high-performance liquid chromatograph ( HPLC ) attached with a fluorescence detector ( excitation at 356 nm; emission at 590 nm ) for analysis . Zorbax SB C-18 reversed-phase column ( 250 × 4 . 6 mm , 5 μm ) was used and water: acetonitrile ( 0 . 1% trifluoroacetic acid ) was applied as mobile phase while flow rate was maintained at 0 . 5 ml/min . The HPLC method used was as described previously ( Kalyanaraman et al . , 2014 ) . NADH/NAD+ ratios upon AG treatment were determined by NAD+/NADH Quantification Kit ( Sigma-Aldrich , USA ) . Mtb cells were cultured to OD600 of 0 . 4 and treated with Amox-Clav combination for various time points ( 6 , 12 , and 24 hr ) . Ten milliliter of culture was harvested and washed with 1X PBS and NADH/NAD+ ratio was determined according to the manufacturer’s instructions . ROS generation upon AG treatment was assessed using a peroxide detection agent , 5- ( and 6 ) -chloromethyl-2′ , 7′-dichlorodihydrofluorescein diacetate , acetyl ester ( CM-H2DCFDA; Invitrogen USA , ThermoFisher Scientific ) . The reagent is converted to a fluorescent product by cellular peroxides/ROS as determined by flow cytometry . Mtb cells were cultured to mid-logarithmic phase ( OD600 of 0 . 4 ) , and AG treatment was given for 3 hr and 6 hr . At each time point , 500 μL of culture was aliquoted and incubated with 20 μM of CM-H2DCFDA in dark ( 30 min ) at 37°C . Cells were washed with 1X PBS and analyzed by FACS Verse flow cytometer ( BD Biosciences , San Jose , CA ) . CM-H2DCFDA fluorescence was determined ( excitation at 488 nm and emission at 530 nm ) by measuring 10 , 000 events/sample . β-lactamase activity in Mtb strains was determined using a spectrophotometer by hydrolysis of nitrocefin , a chromogenic cephalosporin substrate that contains a β-lactam ring . Bacterial cultures were grown to an OD600 of 0 . 6–0 . 8 , and cells were harvested and lysed using bead beater ( FastPrep Instrument , MP Bio ) . The cell-free lysate was clarified by centrifugation and 100 μg of lysate was incubated with 100 µM Nitrocefin . Hydrolysis of nitrocefin was monitored at 486 nm using a SpectraMax M3 plate reader ( Molecular Devices ) at regular intervals . Fold activity was calculated based on changes in absorbance at 486 nm over time . Normalization was performed by Bradford estimation of total protein in the cell-free lysates . For microarray analyses , wt Mtb and MtbΔwhiB4 strains were cultured to an OD600 of 0 . 4 and exposed to 1X ( 10 µg/ml of Amox and 8 µg/ml of Clav ) , 5X ( 50 µg/ml of Amox and 8 µg/ml of Clav ) and 10X ( 100 µg/ml of Amox and 8 µg/ml of Clav ) MIC of AG for 6 hr or 12 hr . For CHP stress , wt Mtb grown similarly was treated with 250 µM of CHP for 2 hr and samples was processed for microarrays . Total RNA was isolated from samples ( taken in replicates ) , processed and hybridized to Mtb Whole Genome Gene Expression Profiling microarray- G2509F ( AMADID: G2509F_034585 , Agilent Technologies PLC ) and data were analyzed as described ( Mehta et al . , 2016 ) . DNA microarrays were provided by the University of Delhi , South Campus , MicroArray Centre ( UDSC-MAC ) . RNA amplification , cDNA labeling , microarray hybridization , scanning , and data analysis were performed at the UDSC-MAC as described ( Mehta et al . , 2016 ) . Slides were scanned on a microarray scanner ( Agilent Technologies ) and analyzed using GeneSpring software . Results were analyzed in MeV and considered significant at p≤0 . 05 . The normalized data from the microarray gene expression experiment have been submitted to the NCBI Gene Expression Omnibus and can be queried via Gene Expression Omnibus series accession number GSE93091 ( AG exposure ) and GSE73877 ( CHP exposure ) . Global PPI network was generated using the dataset described in the previous studies ( Szklarczyk et al . , 2015; Balázsi et al . , 2008; Cui et al . , 2009; Wang et al . , 2010; Zeng et al . , 2012; Ghosh et al . , 2013; Turkarslan et al . , 2015 ) . The RRIDs of two of these networks are SCR_005223 ( STRING ) and SCR_003167 ( Database of interacting proteins based on homology ) . After constructing the global PPI network of Mtb , we then extracted those interactions that are specific for genes present in our transcriptome data . Our microarray-specific network consists of 34035 edges and 4016 nodes . The expression data was used for assigning weights to nodes and edge in the PPI network to make it condition-specific . The formalism of node and edge weight calculation is given below . Node weight: We calculated node weight ( NW ) values for each node in the network by multiplying the normalized intensity values with the corresponding fold-change ( FC ) values . These values were uniformly scaled by multiplying with 104 . NWi=FCixNormalizedsignalintensity where i denotes the node in the network . Edge weight: In order to calculate the edge weight values , we first calculated Edge-betweenness ( EB ) using NetworkX , a python package ( https://networkx . github . io/ ) . The other link for the codes is; https://github . com/networkx/networkx , where the users can directly pull the codes for usage . The GitHub link for Zen library used for computing shortest paths in the network is; https://github . com/networkdynamics/zenlib/tree/master/src/zen . These values were scaled by multiplying with 106 . The node weight values were used to calculate the edge weight ( EW ) values as follows: -EW=EBxNWixNWj where i and j denotes nodes present in an edge . Edgecost=1/EW The main focus of the study was to identify the key players involved in regulating the variations in different conditions . We carried out shortest path analysis on the condition-specific networks and selected the paths that are most perturbed in these conditions . We implemented shortest path algorithm to obtain the results . The edge cost values were used as an input for calculating all vs . all shortest paths in each condition using Zen ( http://www . networkdynamics . org/static/zen/html/api/algorithms/shortest_path . html ) . More than 9 , 000 , 000 paths were obtained for each condition . In order to analyze the more significant paths , we ordered the paths on the basis of their path scores . Path score is the summation of the edge cost that constitutes a path . Based on the formula considered for calculating edge cost , lower path score indicates that the nodes in the path have higher expression . So , instead of analyzing 9 , 000 , 000 paths , we considered subnetworks , which comprise of top 1% of the network . These networks were visualized using Cytoscape 3 ( Shannon et al . , 2003 ) . Our response networks competently explain the perturbations in the system upon exposure to different situations such as AG treatment and/or disruption of whiB4 . The networks were further co-related to graph-theory-based methods and differentially regulated paths were recognized in each condition to construct sub-network for each condition ( supplementary file 3C ) . Total RNA was isolated as described previously ( Chawla et al . , 2012 ) and cDNA was synthesized ( after DNase treatment ) from 500 ng isolated RNA . Random oligonucleotide primers were used with iScript Select cDNA Synthesis Kit for cDNA synthesis . Gene-specific primers ( supplementary file 3D ) were selected for RT-PCR ( CFX96 RT-PCR system , Bio-Rad ) and iQ SYBR Green Supermix was used for gene expression analysis . In order to obtain meticulous expression levels , PCR expression was normalized and CFX Manager software ( Bio-Rad ) was utilized for data analysis . Gene expression was normalized to Mtb 16S rRNA expression . The histidine-tagged WhiB4 purification and generation of reduced or oxidized apo-WhiB4 was done as described previously ( Chawla et al . , 2012 ) . For EMSA assays , the promoter fragments of whiB4 , blaC , and blaR ( ~100–180 bp upstream of translational start codon ) were PCR amplified from the Mtb genome and the 5’ end was labeled using [γ-32P]-ATP labeled oligonucleotides by using T4 polynucleotide kinase ( MBI Fermentas ) as per the manufacturer’s instructions ( supplementary file 3D ) . Binding reactions were performed in 1X TBE buffer ( 100 mM Tris , 90 mM boric acid and 1 mM EDTA; pH 8 . 33 ) for 30 min and 5% polyacrylamide gel was used to resolve protein-DNA complexes . For competition with unlabeled DNA , fragments of blaC , blaR , and Rv0986 ( ~100–180 bp upstream of translational start codon ) were PCR amplified from the Mtb genome and used in various amounts to outcompete binding of oxidized apo-WhiB4 to 32P-labeled DNA fragments . Gels were exposed to auto radiographic film and visualized via phosphoimaging ( GE ) . 50 nM of DNA fragment containing the blaC promoter and apo-WhiB4 ( oxidized or reduced ) were incubated in transcription buffer; 50 mM Tris HCl , ( pH 8 . 0 ) , 10 mM magnesium acetate , 100 μM EDTA , 100 μM DTT , 50 mM KCl , 50 μg/ml BSA , and 5% glycerol ) for 30 min at room temperature . Single-round transcription assay was initiated by the addition of Msm RNAP-σA holo enzyme ( 100 nM ) , 100 µM NTPs and 1 µCi α-32P-UTP and incubated at 37°C for 20 min . Reactions were terminated with 2X stop dye ( 95% formamide , 0 . 025% ( w/v ) bromophenol blue , 0 . 025% ( w/v ) xylene cyanol , 5 mM EDTA and 0 . 025% SDS and 8 M urea ) and heated at 95°C for 5 min followed by snap chilling in ice for 2 min . The transcripts were resolved by loading samples on to 6% urea-PAGE . Mycobacterial strains expressing Mrx1-roGFP2 were grown in 7H9 medium to OD600 of 0 . 4 and exposed to various concentrations of AG . For measurements , cells were treated with 10 mM N-Ethylmaleimide ( NEM ) for 5 min at room temperature ( RT ) followed by fixation with 4% paraformaldehyde ( PFA ) for 15 min at RT . After washing three times with 1X PBS , bacilli were analyzed using BD FACS Verse Flow cytometer ( BD Biosciences ) . The biosensor response was measured by analyzing the ratio at a fixed emission ( 510/10 nm ) after excitation at 405 and 488 nm as described ( Bhaskar et al . , 2014 ) . Data were analyzed using the FACSuite software . For measuring intramycobacterial EMSH during infection , THP-1 cells were treated with 15 ng/ml phorbol 12-myristate 13-acetate ( PMA ) for 20 hr to differentiate them into macrophages . Differentiated cells were then allowed to rest for 2 days , to ensure a resting phenotype before infection . PMA-differentiated THP-1 cells were infected with Mtb strains expressing Mrx1-roGFP2 at a multiplicity of infection ( MOI ) of 10 for 4 hr at 37°C . After 4 hr of infection , cells were washed with pre-warmed RPMI and amikacin treatment ( 0 . 2 mg/ml for 2 hr ) was given to remove extracellular bacteria . Subsequently cells were washed and resuspended in fresh RPMI media containing various concentrations of Amox ( 12 . 5 , 25 , 50 , and 100 µg/ml of Amox ) and Clav ( 8 µg/ml ) for 6 , 12 , 24 , and 36 hr . At the indicated time points , infected macrophages were treated with NEM/PFA , washed with 1X PBS , and analyzed by flow cytometry as described previously ( Padiadpu et al . , 2016 ) . Mtb strains were grown aerobically to OD600 of 0 . 4 , followed by treatment with various concentrations of AG . At defined time-points post-exposure , cells were serially diluted and plated on OADC-7H11 agar medium for enumerating CFUs . To determine the effect of AG during infection , ~20 , 000 THP-1 cells ( PMA differentiated ) were infected with wt Mtb in a 96-well plate ( MOI:10 ) as described earlier ( Padiadpu et al . , 2016 ) . Briefly , THP-1 monocytes were treated with 15 ng/ml of PMA for 20 hr to differentiate them into macrophages . Differentiated cells were then allowed to rest for 2 days and infected with Mtb H37Rv expressing Mrx1-roGFP2 at a MOI of 10 for 4 hr at 37°C . After 4 hr of infection , cells were washed with pre-warmed RPMI and amikacin treatment ( 0 . 2 mg/ml for 2 hr ) was given to remove extracellular bacteria . Subsequently , cells were washed , fresh RPMI media was added and infected macrophages were exposed to AG ( 100 µg/ml of Amox and 8 µg/ml of Clav ) . At various time points , macrophages were lysed using 0 . 06% SDS-7H9 medium and released bacteria were serially diluted and plated on OADC-7H11 agar medium for CFU determination . The pattern of nascent PG synthesis was observed by fluorescent staining as described ( Thanky et al . , 2007 ) . Mtb strains were grown to exponential phase ( OD600 0 . 6 ) in 7H9 medium . 1 ml of culture was incubated with 1 µg/ml Vancomycin-BODIPY ( BODIPY FL Vancomycin ) for 16 hr under standard growth conditions . Cells were pelleted to remove excess stain and fixed with PFA . After washing with 1X PBS , culture aliquots ( 20 µl ) were spread on slides and allowed to air dry . The bacterial cells were visualized for BODIPY FL Vancomycin fluorescence ( excitation at 560 nm and emission at 590 nm ) in a Leica TCS Sp5 confocal microscope under a 63X oil immersion objective . Staining pattern of more than 150 cells was observed for each strain and cell length was measured using Image J software . For the acute model of infection , BALB/c mice were infected by aerosol with 10 , 000 bacilli per mouse with the Mtb H37Rv , Mtb∆whiB4 , whiB4-OE , MYC 431 , and MYC 431/whiB4-OE strains as described previously ( Solapure et al . , 2013 ) . For assured over-expression of WhiB4 , doxycycline ( 1 mg/ml in 5% sucrose solution ) was supplied in drinking water . Dosages of Amox and Clav were maintained at 200 mg/kg of body weight and as 50 mg/kg of body weight , respectively , and the drugs were administered orally twice a day . At specific time points , mice were sacrificed and their lungs were removed and processed for investigation of bacillary load . CFUs were determined by plating appropriate serial dilutions on 7H11 ( supplemented with OADC ) plates . Colonies were observed and counted after 3–4 weeks of incubation at 37°C . Statistical analyses were performed using the GraphPad Prism software ( RRID: SCR_002798 ) . The statistical significance of the differences between experimental groups ( and controls where appropriate ) was determined by two-tailed , unpaired Student’s t test . Differences with a p value of ≤ 0 . 05 were considered significant . This study was carried out in strict accordance with the guidelines provided by the Committee for the Purpose of Control and Supervision on Experiments on Animals ( CPCSEA ) , Government of India . The protocol was approved by the Committee on the Ethics of Animal Experiments of the International Centre for Genetic Engineering and Biotechnology ( ICGEB ) , New Delhi , India ( Approval number: ICGEB/AH/2011/2/IMM-26 ) . All efforts were made to minimize the suffering .	A bacterium called Mycobacterium tuberculosis causes tuberculosis in humans . Multiple antibiotics are available to treat this infection , yet around one million people still die from tuberculosis each year . One of the reasons that the number of deaths is so high is because many M . tuberculosis cells have become resistant to these drugs . Therefore , new drug treatments are urgently needed to tackle the disease . When cells are under stress – for example , when a bacterial cell is exposed to an antibiotic – they can increase the production of chemicals known as reactive oxygen species . These chemicals are vital to many processes in cells , but if their levels get too high they can kill cells by damaging DNA and other molecules . To prevent this damage , bacterial cells produce molecules , such as mycothiol , to neutralize the excess reactive oxygen species . A therapy called Augmentin is used to fight many different types of bacterial infection . It combines an antibiotic known as amoxicillin with another drug that blocks the activity of a bacterial enzyme responsible for breaking down amoxicillin-like drugs . Augmentin can also kill M . tuberculosis cells , but it was not clear exactly how it works , or how the bacteria might be able to develop resistance to this treatment . Here , Mishra et al . combined a computational technique known as network analysis with experiments to study the affect of Augmentin on M . tuberculosis . The experiments reveal that M . tuberculosis cells can develop resistance to Augmentin by increasing the production of an enzyme that breaks down the antibiotic and by neutralizing reactive oxygen species with help of mycothiol . Augmentin treatment can decrease the production of a protein called WhiB4 in the bacteria . This protein is involved in detecting when cells are stressed and regulates the levels of both mycothiol and the enzyme that breaks down the antibiotic . Increasing the production of this protein made the bacterial cells more susceptible to Augmentin treatment by decreasing the levels of active mycothiol and reducing the production of the enzyme that breaks down the antibiotic drug . These findings suggest that Augmentin could be more effective against drug-resistant tuberculosis and other bacterial infections if it is combined with a drug that can alter the levels of reactive oxygen species inside bacterial cells . The next step is to search for new molecules that may be able to perform such a role .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2017	Efficacy of β-lactam/β-lactamase inhibitor combination is linked to WhiB4-mediated changes in redox physiology of Mycobacterium tuberculosis
Epithelioid hemangioendothelioma ( EHE ) is a vascular sarcoma that metastasizes early in its clinical course and lacks an effective medical therapy . The TAZ-CAMTA1 and YAP-TFE3 fusion proteins are chimeric transcription factors and initiating oncogenic drivers of EHE . A combined proteomic/genetic screen in human cell lines identified YEATS2 and ZZZ3 , components of the Ada2a-containing histone acetyltransferase ( ATAC ) complex , as key interactors of both fusion proteins despite the dissimilarity of the C terminal fusion partners CAMTA1 and TFE3 . Integrative next-generation sequencing approaches in human and murine cell lines showed that the fusion proteins drive a unique transcriptome by simultaneously hyperactivating a TEAD-based transcriptional program and modulating the chromatin environment via interaction with the ATAC complex . Interaction of the ATAC complex with both fusion proteins indicates that it is a key oncogenic driver and unifying enzymatic therapeutic target for this sarcoma . This study presents an approach to mechanistically dissect how chimeric transcription factors drive the formation of human cancers . Epithelioid hemangioendothelioma ( EHE ) is a vascular sarcoma that arises over a wide age range and can occur essentially anywhere anatomically . When it occurs in the lung or the liver , it frequently presents with metastasis early in its clinical course ( Goldblum and Weiss , 2014 ) . Two mutually exclusive gene fusions define this sarcoma which lacks an effective medical therapy . The WWTR1-CAMTA1 gene fusion is encoded by a t ( 1;3 ) ( p36;q25 ) chromosomal translocation ( Errani et al . , 2011 , Tanas et al . , 2011; Tanas et al . , 2016 ) resulting in a TAZ-CAMTA1 fusion protein present in 85–90% of all EHE tumors ( Errani et al . , 2011; Tanas et al . , 2011 ) . The other ~10–15% of EHE contain a t ( X;11 ) ( p11;q22 ) chromosomal translocation encoding a YAP1-TFE3 gene fusion ( Antonescu et al . , 2013 ) . Homologous to TAZ-CAMTA1 , YAP-TFE3 fuses the N-terminus of YAP in frame to the C-terminus of TFE3 ( transcription factor E3 ) ( Antonescu et al . , 2013 ) . The Hippo pathway is a serine/threonine kinase cascade involved in regulating tissue growth and organ size and is composed of the STE20-like protein kinases 1 and 2 ( MST1/2; homologs of Drosophila Hpo ) ( Harvey et al . , 2003; Wu et al . , 2003; Udan et al . , 2003; Pantalacci et al . , 2003; Dong et al . , 2007; Zhao et al . , 2007 ) and the large tumor suppressor 1 and 2 ( LATS1/2; homologs of Drosophila Wts ) ( Justice et al . , 1995; Xu et al . , 1995; Tapon et al . , 2002 ) . In mammalian cells , the nuclear effectors of the Hippo pathway are TAZ ( gene is WWTR1 ) and YAP , two paralogous transcriptional coactivators ( Dong et al . , 2007; Zhao et al . , 2007; Camargo et al . , 2007 ) . Hippo signaling limits TAZ and YAP activity by phosphorylating various serine residues , promoting their accumulation in the cytoplasm and ubiquitin-mediated degradation ( Dong et al . , 2007; Zhao et al . , 2007 ) . In the nucleus , TAZ and YAP bind to the TEA domain ( TEAD ) family of transcription factors to drive their transcriptional program ( Zhao et al . , 2008; Zhang et al . , 2009 ) . TAZ and YAP have emerged as central oncoproteins ( Harvey et al . , 2013 ) in many different cancers including breast ( Chan et al . , 2008; Zhao et al . , 2012 ) , colorectal ( Wang et al . , 2013 ) , liver ( Zender et al . , 2006 ) , lung ( Xie et al . , 2012 ) , pancreas ( Zhang et al . , 2014 ) , thyroid ( de Cristofaro et al . , 2011 ) , and sarcomas ( Fullenkamp et al . , 2016; Merritt et al . , 2018; Isfort et al . , 2019; Eisinger-Mathason et al . , 2015; Crose et al . , 2014 ) . Despite this observation , point mutations within the Hippo signaling pathway are relatively rare in most cancers ( Harvey et al . , 2013 ) . This is especially true of TAZ and YAP , which instead appear to be more commonly mutated by virtue of gene fusions , including WWTR1 ( TAZ ) -CAMTA1 and YAP1-TFE3 . TAZ-CAMTA1 transforms cells by driving a predominantly TAZ-related transcriptional program that is mediated by binding to TEAD4 ( Tanas et al . , 2016 ) . Although homologous regions of YAP and TAZ are found in the YAP-TFE3 and TAZ-CAMTA1 fusion proteins , detailed functional studies of YAP-TFE3 have not been performed . Furthermore , the mechanism by which these hybrid transcription factors regulate transcription and subsequent cell transformation has not been defined . Herein we identify the Ada2a-containing histone acetyltransferase ( ATAC ) complex as a key epigenetic modifier of the TAZ-CAMTA1 and YAP-TFE3 oncogenic transcriptional programs . The ATAC complex is one of two metazoan complexes that incorporates the GCN5 or closely related PCAF histone acetyltransferase subunit , and primarily acetylates lysine 9 of histone 3 ( H3K9 ) and to a lesser degree lysine 14 ( H3K14 ) ( Nagy et al . , 2010 ) . There is emerging evidence that the ATAC complex is an oncogenic driver in cancer . YEATS2 , one of the subunits of the ATAC complex , has been shown to be functionally important to the pathogenesis of non-small cell lung cancer ( Mi et al . , 2017 ) . Amplification of YEATS2 has been identified in ovarian cancer , head and neck cancer , esophageal cancer , and uterine cancer ( Mi et al . , 2017 ) , as well as both well-differentiated and dedifferentiated liposarcoma ( Beird et al . , 2018 ) . Herein , we show that the ATAC complex is a key epigenetic mediator of the oncogenic properties of both the TAZ-CAMTA1 and YAP-TFE3 fusion proteins and thereby a potential therapeutic target for essentially all cases of EHE . The WWTR1 ( TAZ ) -CAMTA1 gene fusion fuses the end of exon 2 or exon 3 of WWTR1 to a breakpoint with exon 9 of CAMTA1 ( 2 , 3 ) . This results in the N terminus of TAZ being fused in frame to the C terminus of CAMTA1 . The N terminus of TAZ contains the TEAD binding domain and 14-3-3 protein binding site ( Kanai et al . , 2000 ) . The WW domain may be absent or present , depending on the type of gene fusion . The C terminus of CAMTA1 contributes the transactivating domain , the TIG domain , ankyrin domain , IQ domain , and nuclear localization signal ( Figure 1A; Tanas et al . , 2016; Finkler et al . , 2007; Long et al . , 2014 ) . Similarly , YAP1-TFE3 fuses exon one from YAP1 in frame to exon 4 of TFE3 resulting in the N terminus of YAP fused in frame to the C terminus of TFE3 . The N terminus of YAP contains the TEAD-binding domain ( Gaffney et al . , 2012; Figure 1A ) . The C terminus of TFE3 contains the transactivating domain , basic helix loop helix domain , and leucine zipper ( Antonescu et al . , 2013; Kauffman et al . , 2014 ) . TFE3 is a member of the MiTF family of basic helix-loop-helix containing transcription factors , consisting of four structurally similar genes ( TFE3 , TFE3B , TFEC , and MITF ) , that are involved in regulating tissue-specific functions of cell differentiation ( Kauffman et al . , 2014; Steingrimsson et al . , 2002 ) . TFE3 is ubiquitously expressed and activates transcription by binding to the µE3 motif within the immunoglobulin heavy-chain enhancer ( Beckmann et al . , 1990; Henthorn et al . , 1991 ) . TFE3 has recently been shown to promote cell survival in starvation conditions by regulating lysosomal biogenesis and homeostasis while also having a role in clearance of cellular debris ( Martina et al . , 2014; Slade and Pulinilkunnil , 2017 ) . TFE3 is also involved in gene fusions in renal cell carcinoma and alveolar soft part sarcoma ( Kauffman et al . , 2014; Argani et al . , 2003; Klatte et al . , 2012 ) . In order to dissect the function of the YAP-TFE3 fusion protein , we expressed it in NIH 3T3 cells , which have been utilized to study various components of the Hippo pathway ( Zhao et al . , 2007; Zhao et al . , 2008; Zhang et al . , 2009 ) . We also expressed the fusion protein in SW872 cells , a human liposarcoma cell line , chosen because it has an attenuated ability to grow in an anchorage independent manner in either soft agar or poly-HEMA ( poly 2-hydroxyethyl methacrylate; data not shown ) and because there are no EHE cell lines available . Both TAZ-CAMTA1 and YAP-TFE3 were able to promote colony formation in soft agar in SW872 cells ( Figure 1—figure supplement 2E ) . To investigate YAP-TFE3’s ability to transform cells , full-length YAP , full-length TFE3 , the truncated portions of YAP or TFE3 present in the fusion protein , as well as YAP-TFE3 were expressed in NIH 3T3 cells ( Figure 1—figure supplement 1 ) which were then grown in soft agar ( Figure 1B ) and poly-HEMA ( Figure 1C ) . YAP-TFE3 promoted anchorage independent growth while full length YAP , full-length TFE3 , or the truncated portions of YAP or TFE3 present in the fusion protein did not . Expressing the same constructs in SW872 cells ( Figure 1—figure supplement 1 ) and evaluating proliferation on poly-HEMA yielded similar results ( Figure 1D ) . The above findings indicate that YAP-TFE3 is a neomorphic protein combining properties of both YAP and TFE3 to transform cells . YAP and TAZ dynamically shuttle between the cytoplasm and nucleus and this is regulated by Hippo pathway-dependent and -independent mechanisms ( Manning et al . , 2020 ) . Furthermore , the abundance of both YAP and TAZ is limited by ubiquitin-mediated degradation which is controlled by the Hippo pathway ( Liu et al . , 2010; Huang et al . , 2012 ) . The relative abundance of YAP and TAZ in the nucleus and cytoplasm is modulated by multiple features including cell confluence and mechanotransduction cues ( Manning et al . , 2020; Zanconato et al . , 2016 ) . Like YAP , we found YAP-TFE3 to be localized within the nucleus when cells were plated under sparse conditions . Under confluent conditions , YAP was enriched in the cytoplasm and its abundance was reduced , while YAP-TFE3 remained predominantly nuclear ( Figure 1E–F; Figure 1—figure supplement 2A–B ) . Nuclear and cytoplasmic fractionation confirmed that YAP was no longer localized in the nucleus and instead accumulated in the cytoplasm where it was subsequently degraded . In contrast , YAP-TFE3 remained in the nucleus upon cell confluence ( Figure 1—figure supplement 2C ) . Similarly , YAP localization shifted from the nucleus to the cytoplasm upon cell detachment , while YAP-TFE3 remained constitutively nuclear ( Figure 1—figure supplement 2D ) . Taken together , the above data suggest that YAP-TFE3 is no longer subject to Hippo pathway-mediated suppression , likely because the key residues targeted by LATS1/2 ( S127 and S381 ) ( Zhao et al . , 2007 ) are not present in YAP-TFE3 . To study the function of YAP/TAZ fusion proteins in vivo , NIH 3T3 cells or SW872 cells transduced with empty vector , TAZ-CAMTA1 , or YAP-TFE3 were injected into subcutaneous tissue of Prkdcscid: Il2rgtm1Wjl/SzJ ( NSG ) mice . Mice with NIH 3T3-derived tumors expressing YAP-TFE3 or TAZ-CAMTA1 demonstrated a shorter overall survival compared to empty vector ( p<0 . 0001 for both YAP-TFE3 and TAZ-CAMTA1 ) ( Figure 1G ) . Tumors expressing YAP-TFE3 and TAZ-CAMTA1 formed palpable masses ~ 7 days post-injection ( Figure 1H ) indicating time to tumor initiation was similar for NIH 3T3 cells expressing either fusion protein . NIH 3T3 cells containing empty vector formed tumors , as has been described ( Greig et al . , 1985 ) , but tumor initiation was delayed by approximately 10 days . The slopes of the growth curves for the two fusion proteins were similar to empty vector , and this was confirmed by Ki-67 labeling ( Figure 1—figure supplement 3A and B ) . Although this is seemingly discrepant with in vitro data showing that YAP-TFE3 drives proliferation ( Figure 1C and D ) , the in vitro proliferation being measured was in the context of anchorage independent growth , which may play a more important role in tumor initiation . Mice were sacrificed when tumors reached 2 cm in the greatest dimension ( according to animal use protocol ) . NIH 3T3 cell-derived tumors expressing YAP-TFE3 or TAZ-CAMTA1 exhibited a more rounded ( epithelioid ) cytomorphology and a greater degree of cellular pleomorphism as compared to control tumors ( Figure 1—figure supplement 3A ) . Necropsy and pathological evaluation of the lungs demonstrated the presence of microscopic metastases in the YAP-TFE3 ( p=0 . 0299 ) but not the TAZ-CAMTA1 ( not significant ) cohorts ( Figure 1I ) . In the mildly tumorigenic SW872 cell line ( Stratford et al . , 2012; Zhang et al . , 2013a; Li et al . , 2014 ) , expression of YAP-TFE3 and TAZ-CAMTA1 decreased latency to tumor formation compared to empty vector in vivo ( Figure 1J ) , similar to NIH 3T3 cells . Cytological changes in SW872 cell-derived tumors were less pronounced than in NIH 3T3 cells ( Figure 1—figure supplement 3B ) . The above data show that both fusion proteins play a role in tumor initiation and can drive tumorigenesis in vivo . TAZ and YAP are transcriptional co-activators that lack DNA binding domains and form a physical complex with DNA binding transcription factors in order to drive transcription , the best characterized of which are TEAD1-4 . TAZ and YAP interact with TEAD1-4 via the TEAD binding domain in the YAP and TAZ amino termini ( Zhao et al . , 2008; Zhang et al . , 2009 ) , while the WW domains of YAP and TAZ mediate interactions with other transcription factors including TBX5 ( Murakami et al . , 2005 ) , Runx2 ( Hong et al . , 2005 ) , PPAR-γ ( Hong et al . , 2005 ) , and SMADs ( Varelas et al . , 2008 ) . We previously showed that TEAD4 is required for the oncogenic function of TAZ-CAMTA1 ( Tanas et al . , 2016 ) . To determine if the WW domain is essential for cell transformation , alanines were substituted for the two essential tryptophans in the WW domain of TAZ-CAMTA1 ( TAZ-CAMTA1ΔWW ) . While the S51A mutant ( disrupts TEAD binding ) of TAZ-CAMTA1 significantly reduced colony formation in soft agar/growth on poly-HEMA ( Tanas et al . , 2016 ) , the TAZ-CAMTA1ΔWW mutant did not ( Figure 1—figure supplement 2F and G ) . This confirms that TAZ-CAMTA1 function is dependent on the TEAD binding domain but not the WW domain . YAP-TFE3 ( Figure 1A ) contains the TEAD binding domain , but not the WW domain , suggesting that the TEAD binding domain mediates the YAP-TFE3 transcriptional program . To test the hypothesis that TEAD binding is necessary to activate the oncogenic YAP-TFE3 transcriptional program , Tead1 was knocked down in NIH-3T3 cells expressing YAP-TFE3 ( Figure 2A; Figure 2—source data 1 ) . Knock-down of Tead1 reduced anchorage independent growth on poly-HEMA ( Figure 2B ) and in soft agar ( Figure 2C ) with two different short hairpin RNAs ( shRNAs ) . A YAP-TFE3 S94A mutant was created to disrupt its interaction with TEAD ( homologous to S94 in full-length YAP and required for TEAD binding Zhao et al . , 2008 , which was confirmed by co-immunoprecipitation ( Figure 2D ) ) . A prominent band at a lower molecular weight than full-length YAP-TFE3 S94A was identified , likely representing an unstable degradation product . The S94A mutation completely abrogated YAP-TFE3-driven anchorage-independent growth on poly-HEMA ( Figure 2E ) and in soft agar ( Figure 2F ) . By quantitative RT-PCR , the YAP-TFE3 S94A mutant showed a sevenfold decrease in expression of Ccn2 ( Ctgf ) , a YAP target gene , compared to the YAP-TFE3 control ( Figure 2G ) . Chromatin immunoprecipitation-quantitative PCR for the CCN2 promoter ( which contains the TEAD-binding sequence ) showed a sixfold decrease in CCN2 promoter binding by YAP-TFE3 S94A as compared to YAP-TFE3 control ( Figure 2H ) . Using the 8XGTIIC luciferase reporter containing eight TEAD consensus binding sequences ( Dupont et al . , 2011 ) , YAP-TFE3 S94A showed decreased luciferase activity compared to the YAP-TFE3 control ( Figure 2I and Figure 1—figure supplement 1E ) . Collectively , these data suggest that YAP-TFE3 , like TAZ-CAMTA1 , requires TEAD transcription factors to drive its oncogenic transcriptional program ( Figure 2J ) . From this working model , we hypothesized that since TEAD binding is essential for YAP-TFE3 and TAZ-CAMTA1-driven transformation , the YAP-TFE3 and TAZ-CAMTA1 transcriptomes would essentially recapitulate the YAP and TAZ transcriptional programs , respectively . To test this hypothesis , RNA-Seq was performed on NIH 3T3 cells and SW872 cells stably expressing TAZ4SA ( hyperactivated form of TAZ due to alanines being substituted for four serines that can be phosphorylated by the LATS1/2 kinases [Lei et al . , 2008] ) , TAZ-CAMTA1 , CAMTA1 , YAP5SA ( hyperactivated form of YAP due to alanines being substituted for five serines that can be phosphorylated by the LATS1/2 kinases [Zhao et al . , 2007] ) , YAP-TFE3 , and TFE3 , as well as cells transduced with empty vector control ( Figure 1—figure supplement 1; Figure 3—source data 1 and 2 ) . Differentially expressed genes ( FDR < 0 . 05 ) showed considerable overlap between the TAZ4SA and TAZ-CAMTA1 transcriptomes; 70% of genes differentially expressed in the TAZ-CAMTA1 transcriptome were also present in the TAZ4SA transcriptome with a –log10[hypergeometric density] ( HGD ) of 105 . 6 ( p=1 . 92×10−106 ) . By comparison , a much smaller percentage of overlap ( 9% ) was identified between the TAZ-CAMTA1 and CAMTA1 gene sets with an HGD of 34 . 1 ( p=4 . 71×10−35 ) . A significant percentage ( 29% ) of differentially expressed genes in the TAZ-CAMTA1 transcriptional program were not induced by either TAZ4SA or CAMTA1 ( Figure 3A; Figure 3—source data 1 ) in NIH 3T3 cells . Similarly , 31% of the TAZ-CAMTA1 gene set in SW872 cells were unique and not found in the TAZ or CAMTA1 transcriptional programs ( Figure 3B; Figure 3—source data 2 ) . As in NIH 3T3 cells , the amount of overlap between TAZ-CAMTA1 and TAZ4SA in SW872 cells was greater ( 67% of TAZ-CAMTA1 genes; HGD of 72 . 6 ) than that between TAZ-CAMTA1 and CAMTA1 ( 9% of TAZ-CAMTA1 genes; HGD of 27 . 6 ) . The YAP-TFE3 transcriptional program demonstrated similar findings . In NIH 3T3 cells , 78% of genes differentially expressed in the YAP-TFE3 transcriptome were present in the YAP5SA transcriptional program . Comparatively , only 10% of differentially expressed genes in the YAP-TFE3 gene set were also present in the TFE3 transcriptome . In NIH 3T3 cells 20% of the differentially expressed genes were unique to the YAP-TFE3 transcriptome and not found in the YAP5SA or TFE3 transcriptional programs ( Figure 3C; Figure 3—source data 1 ) . Similarly , in SW872 cells 47% of the YAP-TFE3 transcriptional program was found only in the YAP-TFE3 transcriptome and not in YAP5SA or TFE3 controls ( Figure 3D; Figure 3—source data 2 ) . Mirroring findings in NIH 3T3 cells , the amount of overlap between YAP-TFE3 and YAP5SA ( 53% of YAP-TFE3 genes; HGD of 220 . 1 ) was greater than that between YAP-TFE3 and TFE3 ( 0 . 2% of YAP-TFE3 genes; HGD of 1 . 5 ) . A principal component analysis ( PCA ) revealed that the TAZ-CAMTA1 and YAP-TFE3 transcriptomes in NIH 3T3 cells grouped more closely together than the transcriptomes of their cognate full length proteins ( Figure 3E ) . In SW872 cells , the TAZ-CAMTA1 and YAP-TFE3 transcriptomes diverged from one another , but were still distinct from the TAZ4SA and YAP5SA transcriptomes ( Figure 3—figure supplement 1A ) , providing further support that the fusion proteins drive a transcriptional program that incorporates some elements of the TAZ/YAP transcriptomes but is also significantly altered from the baseline TAZ and YAP transcriptional program . This was corroborated by qRT-PCR experiments , which showed that TAZ-CAMTA1 and YAP-TFE3 promote upregulation of the well-defined TAZ/YAP transcriptional targets CCN2 ( CTGF ) /Ccn2 and CCN1 ( CYR61 ) /Ccn1 , but also genes unique to the TAZ-CAMTA1 and YAP-TFE3 transcriptomes such as Fras1 ( an extracellular matrix protein ) , Erbb3 ( Erb-B2 Receptor Tyrosine Kinase 3 ) ( Figure 3F ) , and FBLN5 ( an extracellular matrix protein ) ( Figure 3G ) . iPathwayGuide analysis performed showed that the top cancer-related signatures activated by TAZ-CAMTA1 in SW872 cells ( Figure 3H ) are the PI3K-Akt signaling pathway , Hippo signaling pathway ( Figure 3—figure supplement 1B ) , focal adhesion , proteoglycans , and extracellular matrix ( ECM ) -receptor interaction ( mirroring qRT-PCR results in Figure 3F and G ) . The top cancer-related signatures activated by YAP-TFE3 in SW872 cells ( Figure 3I ) also included the PI3K-Akt signaling pathway , as well as the MAPK and JAK-STAT signaling pathways . A microRNA and apoptosis signature were also identified . To determine if the modified transcriptional programs of the fusion proteins were driven by altered DNA binding , we performed chromatin immunoprecipitation ( ChIP ) sequencing studies on SW872 cells using Flag-tagged TAZ4SA , TAZ-CAMTA1 , CAMTA1 , YAP5SA , YAP-TFE3 , and TFE3 constructs . Annotation of ChIP peaks with respect to their nearest gene transcriptional start site ( TSS ) demonstrated that CAMTA1 and TFE3 ChIP peaks were localized in closer proximity to the TSS ( 0–1 kb ) . In contrast , TAZ4SA , TAZ-CAMTA1 , YAP5SA , and YAP-TFE3 predominantly occupied distal intergenic sequences ( 3 kb or greater ) consistent with active enhancer sequences ( Figure 4A–E; Figure 4—figure supplement 1A–D ) as confirmed by the presence of overlapping H3K27ac ChIP peaks ( Zhang et al . , 2013b; Creyghton et al . , 2010; Calo and Wysocka , 2013 ) ( 69% overlap in TAZ-CAMTA1 distal intergenic sequences; 68% overlap in YAP-TFE3 distal intergenic sequences ) ( Figure 4D and E ) . Although the fusion proteins most frequently bound to distal intergenic sequences/active enhancers , they also bound to sequences closer to the TSS ( e . g . promoters regions ≤ 1 kb from the TSS ) ( Figure 4D and E ) , which are also enriched for H3K27ac marks ( Creyghton et al . , 2010 ) . Comparison of TAZ-CAMTA1 and H3K27ac genomic occupancy around the TSS of MAFK and HOXA1 ( Figure 4—figure supplement 1E , F ) , two differentially expressed genes in TAZ-CAMTA1-expressing cells ( Figure 3—source data 2 ) , showed that H3K27ac peaks closely overlapped and surrounded TAZ-CAMTA1 bound chromatin , validating the above approach . Thus TAZ-CAMTA1 and YAP-TFE3 maintain the genomic occupancy characteristics of increased preference for distal intergenic/enhancer sequences as demonstrated by TAZ and YAP ( Figure 4—figure supplement 1A , C; Zanconato et al . , 2015; Galli et al . , 2015; Stein et al . , 2015 ) , but also occupy sequences proximal to the TSS . ChIP peaks were annotated to their nearest gene features ( as described in Materials and methods ) . Mirroring the RNA-Seq data , 56% of TAZ-CAMTA1 genes containing ChIP peaks were present in the TAZ-CAMTA1 data set , but not the TAZ4SA or CAMTA1 data sets ( Figure 4F ) . Similarly , 46% of YAP-TFE3 genes containing ChIP peaks were unique to the YAP-TFE3 data set and not present in the YAP5SA or TFE3 controls ( Figure 4G ) . Overall , the chromatin binding profile of TAZ-CAMTA1 more closely resembled TAZ4SA ( HGD of 52 . 9 ) than CAMTA1 ( HGD of 43 . 8 ) . Likewise , YAP-TFE3 bound chromatin was more like that of YAP5SA ( HGD of 166 . 8 ) than TFE3 ( HGD of 11 . 8 ) . Unbiased motif enrichment analysis ( Figure 4—source data 1 ) showed that approximately the same proportion of TAZ4SA and TAZ-CAMTA1 peaks contained the TEAD binding sequence ( 58% vs . 61% , respectively ) ( Figure 4F ) . Interestingly , there was a decrease in the frequency of FOS/JUN binding sites in TAZ-CAMTA1 ( 31% ) as compared to TAZ4SA ( 50% ) , which appeared to be offset by an increase in Early growth response gene-2 ( EGR2 ) binding motifs ( 23% vs . 13% for TAZ4SA ) . EGR2 is a transcription factor important to neural development ( Warner et al . , 1998 ) as is CAMTA1 ( Long et al . , 2014 ) . Motif enrichment analysis showed a decrease in the fraction of YAP-TFE3-bound peaks containing the TEAD-binding sequence ( 38% ) as compared to YAP5SA-bound peaks ( 51% ) ( Figure 4G ) . Microphthalmia-associated transcription factor ( MiTF ) is a member of the TFE/MiTF transcription factor family known to complex with TFE3 ( Steingrimsson et al . , 2002 ) . Interestingly , the MiTF binding motif was enriched in YAP-TFE3 ( 58% ) compared to YAP5SA ( 16% ) ( Figure 4G ) , suggesting that , like TAZ-CAMTA1 , TFE3 confers DNA binding properties to the YAP-TFE3 protein in addition to events mediated by YAP’s interaction with TEAD transcription factors . Consistent with this , we found that YAP-TFE3 indeed co-immunoprecipitates with TFE3 , and that disruption of the bHLH domain in the TFE3 portion of YAP-TFE3 that is responsible for dimerization altered anchorage-independent growth ( Figure 4—figure supplement 2A ) . Overall , the motif enrichment analysis explains , at least in part , our observation that the chromatin binding profiles of the fusion proteins overlap significantly , but also differ from , full-length TAZ and YAP . Genes bound by the fusion proteins or controls were additionally validated by ChIP-qPCR ( Figure 4H ) . To determine if genes bound by the fusion proteins were differentially expressed , a plot was generated that arrayed differentially expressed genes in terms of increasing statistical significance along the Y axis , and genes in terms of increasing ChIP peak ‘score’ ( i . e . statistical significance ) along the X axis for TAZ-CAMTA1 , YAP-TFE3 , and controls ( Figure 4—figure supplement 2B–F; Figure 4—source data 2 ) . This analysis revealed well known YAP/TAZ target genes such as CCN1 ( CYR61 ) , CCN2 ( CTGF ) , AMOTL2 , AJUBA , LATS2 , and TEAD1 , and genes that have not been identified as YAP/TAZ targets , such as MAFK . Thirteen percent of differentially expressed genes for TAZ-CAMTA1 were directly bound by the transcription factor ( Figure 4—figure supplement 3A ) . Similarly , 15% of differentially expressed genes for YAP-TFE3 were directly bound by the fusion protein ( Figure 4—figure supplement 3B ) . Kyoto encyclopedia of genes and genomes ( KEGG ) analysis of TAZ-CAMTA1 bound genes that are also differentially expressed in SW872 TAZ-CAMTA1 cells showed enrichment for RHO GTPase-dependent pathways , EPHA-dependent pathways , RUNX2-dependent pathways , as well as other pathways known to be regulated by TAZ/YAP ( Hong et al . , 2005; Dupont et al . , 2011; Edwards et al . , 2017; Figure 4—figure supplement 3C ) . Previous studies indicated that TAZ and YAP form physical complexes with multiple different with chromatin modifying proteins ( Moya and Halder , 2014 ) . To test the hypothesis that the fusion proteins modify chromatin structure , Assay for Transposase-Accessible Chromatin ( ATAC ) -sequencing studies were performed on SW872 cells expressing the above constructs . Heat map and histogram analysis of ATAC peaks showed that TAZ-CAMTA1 and CAMTA1 preferentially opened chromatin in areas of the genome more distal to the TSS than TAZ4SA ( Figure 5A , Figure 5—figure supplement 1A , B ) . Similarly , although less pronounced , YAP-TFE3 preferentially opened chromatin architecture more distal to the TSS than YAP5SA ( Figure 5B and Figure 5—figure supplement 1A , B ) . These results are in keeping with the genomic occupancy profiles of the fusion proteins defined by ChIP-Seq ( Figure 4C–E ) which were predominantly bound to enhancer sequences . Analysis of differentially accessible chromatin annotated to the nearest gene feature ( as described in Materials and methods; note that multiple ATAC peaks could be assigned to the same gene ) ( Figure 5C , D , and Figure 5—figure supplement 1C , D ) showed that 15 , 779 genes with annotated ATAC peaks were shared between TAZ-CAMTA1 and all of the controls while 16 , 809 genes were shared between YAP-TFE3 and all the controls , representing the baseline levels of transposase accessible chromatin . We initially hypothesized that the accessible chromatin landscape for the fusion proteins would overwhelmingly resemble that of full length TAZ and YAP . However , hypergeometric analysis showed that the accessible chromatin profile of TAZ-CAMTA1 most closely resembled that of CAMTA1 ( Figure 5C ) , while the -log10 ( hypergeometric density ) HGD between YAP-TFE3 and YAP5SA ( 106 . 0 ) was roughly equivalent to that between YAP-TFE3 and TFE3 ( 98 . 0 ) ( Figure 5D ) . This analysis ( Figure 5C , D ) also showed that both TAZ-CAMTA1 and YAP-TFE3 promoted chromatin accessibility for unique genes not found in the control data sets , making them available for either transcriptional activation or repression . More transposase accessible , non-baseline genes were present with expression of TAZ-CAMTA1 ( 3147 genes ) and YAP-TFE3 ( 2499 genes ) than TAZ4SA ( 656 genes ) and YAP5SA ( 1629 genes ) , respectively . 'Diffbind' differential binding affinity analysis of consensus-derived peak sets was carried out with respect to empty vector controls . Dimensionality reduction with PCA showed that TAZ-CAMTA1 and YAP-TFE3 samples clustered together with CAMTA1 and TFE3 , suggesting a greater degree of similarity in ATAC peak affinity patterns , while TAZ4SA or YAP5SA were distinct and separate ( Figure 5E ) . These results are largely consistent with the hypergeometric analysis mentioned above . The homology of the resultant chromatin landscape for the two fusion proteins , despite the structural dissimilarity of the C termini of CAMTA1 and TFE3 , independently supports a convergent function for the two fusion proteins . Integrating annotation results from all three NGS modalities ( RNA- , ChIP- , and ATAC-seq ) showed that the majority of genes bound by the fusion proteins , as well as those which were differentially expressed , were within euchromatin regions of the genome ( Figure 5F , G ) . To further evaluate whether the fusion protein DNA-binding sites were nested within transposase accessible chromatin , we evaluated the combined ChIP-Seq and ATAC-Seq data for MAFK and HOXA1 , two genes that were bound by one or both of the fusion proteins and simultaneously differentially expressed ( Figure 4—figure supplement 2C , F ) . Evaluating the combined ChIP/ATAC signal tracks for MAFK and HOXA1 in TAZ-CAMTA1 and YAP-TFE3 expressing cells ( Figure 5—figure supplement 1E , F ) suggested that the fusion proteins combine the DNA-binding activity of the TEAD domains of TAZ/YAP with the chromatin remodeling properties of CAMTA1 and TFE3 for these genes . To address the hypothesis that CAMTA1 and TFE3 confer altered chromatin remodeling properties to TAZ-CAMTA1 and YAP-TFE3 , we utilized proximity-dependent biotinylation ( BioID ) and tandem mass spectrometry . TAZ-CAMTA1 , YAP-TFE3 , and full-length controls were fused on their N terminus to BirA* , an abortive biotin ligase , in a tetracycline-inducible vector , and stably transfected in Flp-In T-Rex HEK 293 cells . Proteins in proximity to the BirA-tagged bait proteins were biotinylated , purified , and analyzed by tandem mass spectrometry and SAINT statistical analysis ( Figure 6A ) . Prey proteins detected with at least 10 spectral counts in at least one biological replicate were represented in dot plot format as a composite of the relative abundance and spectral count , where the size of the dots represents the relative abundance and darkness represents the spectral count ( the false discovery rate is represented as the color coding at the periphery of the dot ) . Sixty-eight prey proteins were identified that established high-confidence proximity interactions with the fusion proteins or the full-length controls ( Figure 6—figure supplement 1A–E ) . The YAP-TFE3 proximity interactome was composed predominantly of histone modifying enzymes ( 35% ) , transcriptional regulators ( 26% ) , proteins involved in chromatin remodeling ( 18% ) , and transcription factor/co-activators ( 15% ) ( Figure 6B ) . Similarly , the TAZ-CAMTA1 proximity interactome was composed predominantly of histone modifying enzymes ( 27% ) , transcriptional regulators ( 18% ) , transcription factor/co-activators ( 15% ) , and chromatin remodeling ( 13% ) . A smaller percentage of the TAZ-CAMTA1 interactome was involved in mediating cell-cell junctions ( 5% ) and SUMO processing ( 4% ) . To identify proteins to prioritize for further study , we subtracted proteins that showed a decreased proximity interaction with the fusion proteins relative to full-length TAZ and YAP; many of which ( e . g . MPDZ , PATJ ) are part of the Crumbs complex , and the decreased interaction can be explained in part by loss/mitigation of the WW domain . This left 55 and 34 prey proteins in the TAZ-CAMTA1 and YAP-TFE3 interactomes , respectively . Known components of the Hippo pathway were subtracted from further analysis , leaving 49 proteins in the TAZ-CAMTA1 data set , and 31 proteins in the YAP-TFE3 data set . We then retained only prey proteins that demonstrated an increased interaction with the fusion proteins relative to full length TAZ and YAP , leaving 47 proteins in the TAZ-CAMTA1 data set and 28 remaining proteins in the YAP-TFE3 data set . The two data sets were intersected to demonstrate 27 shared proteins ( Figure 6C–E ) . As shown in the table listing the function of the shared proteins ( Figure 6D ) and a chart giving an overview of their composition ( Figure 6E ) , the shared prey proteins were enriched for epigenetic/transcriptional regulators . Proteins with weak interactions with both fusion proteins ( generally these were in the lowest category of spectral count/relative abundance and FDR > 5% ) were eliminated , leaving 18 proteins ( Figure 6F ) . The observation that so many proximity interactions were shared between TAZ-CAMTA1 and YAP-TFE3 , despite having very different C-termini , indicated these shared prey proteins may be important to the oncogenic mechanism of TAZ-CAMTA1 and YAP-TFE3 . To further investigate these 18 prey proteins , an RNAi screen was performed on SW872 cells expressing TAZ-CAMTA1 using five short hairpin constructs ( shRNA ) for each gene/protein ( Figure 7—source data 1 ) . These lines were subsequently evaluated by poly-HEMA assay ( Figure 7A ) utilizing the observation that TAZ-CAMTA1 promoted anchorage independent growth in SW872 cells on poly-HEMA ( Figure 7B ) and in soft agar ( Figure 1—figure supplement 2E ) . The number of shRNAs that demonstrated a deficit in anchorage independent growth as compared to shEV and shNT were recorded to rank order the genes/proteins in the interactome for further study ( Figure 7C ) . Genes that demonstrated five shRNAs promoting a deficit in anchorage independent growth included YEATS2 ( Figure 7C , D ) , ZZZ3 ( Figure 7C , E ) , and EP400 ( Figure 7C , F ) . Other genes/prey proteins with more than two shRNAs showing a phenotype included NCOA2 ( Figure 7G ) and TCF20 ( Figure 7H ) . Remaining prey proteins/genes with at least one shRNA showing a phenotype are also shown ( Figure 7C; Figure 7—figure supplement 1A–F ) . YEATS2 knock-down and a reduction in anchorage independent growth was validated in SW872 cells expressing YAP-TFE3 using the two shRNAs with the best knock-down ( Figure 7—figure supplement 1G–I ) . Significantly , YEATS2 and ZZZ3 , the two top ranking hits in the RNAi screen , are both members of the Ada2a-containing histone acetyltransferase complex ( ATAC ) . YEATS2 is the scaffolding protein for other subunits of the ATAC complex and binds to H3K27ac via its YEATS domain ( Mi et al . , 2017 ) . The ZZ domain of ZZZ3 binds the histone H3 tail and in combination with YEATS2 anchors the ATAC complex to acetylated H3 ( Mi et al . , 2018 ) . These proteins bind to other proteins in the complex including Ada2a as well as the GCN5 catalytic subunit ( Mi et al . , 2017; Mi et al . , 2018; Grant et al . , 1997 ) . GCN5 or the closely related PCAF subunit ( Nagy and Tora , 2007 ) are responsible for acetylating H3K9 , which subsequently opens the surrounding chromatin structure promoting transcription ( Mi et al . , 2017 ) . In addition , KAT14 , a third subunit of the ATAC complex , was identified by BioID mass spectrometry to interact with TAZ-CAMTA1 and YAP-TFE3 ( Figure 6C ) , further emphasizing the functional significance of this complex and an additional impetus to prioritize YEATS2 and ZZZ3 for further study . Knock-down of YEATS2 in SW872 TAZ-CAMTA1 cells was confirmed by quantitative RT-PCR ( Figure 8—figure supplement 1A ) and resulted in loss of anchorage-independent growth by soft agar assay ( Figure 8—figure supplement 1B ) . Similarly , knock-down of ZZZ3 in SW872 TAZ-CAMTA1 cells was confirmed by western blot ( Figure 8—figure supplement 1C ) and the loss of anchorage-independent phenotype was replicated by soft agar assay ( Figure 8—figure supplement 1D ) . We then performed co-immunoprecipitation studies to validate the BioID experiments . Of the three subunits of the ATAC complex identified by BioID mass spectrometry , KAT14 showed the strongest interaction with TAZ-CAMTA1 and YAP-TFE3 by co-immunoprecipitation ( Figure 8—figure supplement 1E–G ) . To investigate the role of the ATAC complex in mediating the transcriptional programs of TAZ-CAMTA1 and YAP-TFE3 , we performed RNA-Seq on SW872 cells expressing TAZ-CAMTA1 and YAP-TFE3 with stable knock-down of YEATS2 or ZZZ3 . Principal component analysis showed that knock-down of YEATS2 and ZZZ3 altered both the TAZ-CAMTA1 and YAP-TFE3 transcriptomes , compared to shRNA control ( Figure 8A ) . Knock-down of YEATS2 and ZZZ3 had a greater effect on the final transcriptional program of YAP-TFE3 than TAZ-CAMTA1 , likely reflecting the smaller number of epigenetic modifiers interacting with YAP-TFE3 as compared to TAZ-CAMTA1 . Aligning the most down-regulated genes for TAZ-CAMTA1 shYEATS2 , TAZ-CAMTA1 shZZZ3 , YAP-TFE3 shYEATS2 , and YAP-TFE3 shZZZ3 revealed an enrichment of extracellular matrix proteins ( FBLN5 , SPP1 , GPC4 , and LAMA1 ) ( Figure 8B and Figure 8—source data 1 ) and validated by qRT-PCR ( Figure 8C ) . iPathwayGuide analysis showed that MAPK signaling , a pathway significantly upregulated in SW872 YAP-TFE3 cells ( Figure 3I ) was significantly down-regulated with YEATS2 knock-down ( Figure 8D and Figure 8—figure supplement 2A ) . Further , PI3K-Akt signaling , activated in SW872 cells expressing TAZ-CAMTA1 or YAP-TFE3 ( Figure 3H , I and Figure 8—figure supplement 2B ) , was also decreased in SW872 cells expressing YAP-TFE3 and either YEATS2 or ZZZ3 shRNA ( Figure 8D and Figure 8—figure supplement 2C–D ) . Since the ATAC complex is a HAT complex thought to increase the amount of transcriptionally available chromatin , we intersected genes differentially expressed with YEATS2 or ZZZ3 knockdown with open chromatin unique to SW872 YAP-TFE3 expressing cells . We found that 71% of differentially expressed genes ( DEG ) with YEATS2 knock-down and 74% of DEG with ZZZ3 knock-down were within euchromatin unique to YAP-TFE3 expressing cells ( Figure 8E , F ) . These findings support a working model that the ATAC complex interacts with the C termini of TAZ-CAMTA1 and YAP-TFE3 via the KAT14 subunit at predominantly H3K27ac-positive active enhancer regions , and subsequently acetylates H3K9 to modulate a TEAD-based transcriptional program ( Figure 8G ) . Towards validating these findings in clinical samples , we noted that ECM proteins were consistently represented in previous analysis of the TAZ-CAMTA1 and YAP-TFE3 transcriptomes includes those altered by knock-down of components of the ATAC complex ( Figure 3H and Figure 8B ) . To validate expression of various ECM proteins upregulated in TAZ-CAMTA1 and YAP-TFE3 expressing cells ( Figure 3—source data 2 ) , we performed immunohistochemistry for COL1A1 ( Collagen type I alpha one chain ) , CTGF ( Connective tissue growth factor ) , VTN ( vitronectin ) , and FBLN5 ( Fibulin-5 ) on EHE clinical samples and compared expression with two histological mimics of EHE , epithelioid angiosarcoma ( E-AS ) and epithelioid hemangioma ( EH ) . While CTGF , FBLN5 , and COL1A1 were expressed in EHE ( Figure 8—figure supplement 2E and F ) , they were also expressed in E-AS and EH . In contrast , VTN was expressed in all EHE samples , but was essentially absent in epithelioid angiosarcoma and epithelioid hemangioma ( Figure 8H ) . To determine if expression of YEATS2 and ZZZ3 could be prognostically important in sarcomas other than EHE , we utilized RNA-Seq expression data from sarcoma clinical samples from The Cancer Genome Atlas ( TCGA ) database . Via Kaplan-Meier analysis , we showed that increased expression of YEATS2 ( Figure 8I ) or ZZZ3 ( Figure 8J ) predicted a poorer prognosis across different histological types of sarcoma , suggesting the ATAC complex may be a key oncogenic driver in multiple sarcomas in addition to EHE . This is consistent with recent findings identifying amplification of YEATS2 in sarcomas such as well differentiated liposarcoma and dedifferentiated liposarcoma ( Beird et al . , 2018 ) . Our current studies show that rather than simply recapitulating the TAZ and YAP transcriptional programs , TAZ-CAMTA1 and YAP-TFE3 drive a transcriptome that overlaps with , but is significantly different from that of each full-length protein . Further , the degree to which the fusion protein transcriptomes do overlap with the TAZ/YAP transcriptional programs is due to shared TEAD transcription factor binding . This is borne out via evaluation of the chromatin binding profiles of the two fusion proteins which showed similar levels of enrichment for TEAD-binding sites compared to full-length TAZ and YAP . Our findings are in keeping with previous functional studies of TAZ-CAMTA1 showing that it requires binding to TEAD4 in order to drive its oncogenic transcriptional program and transform cells ( Tanas et al . , 2016 ) . Although the transcriptional programs of TAZ-CAMTA1 and YAP-TFE3 significantly overlap with the TAZ/YAP transcriptomes , they also differ in important ways . In both cell lines examined here , approximately 20–40% of differentially expressed genes were unique to the TAZ-CAMTA1- and YAP-TFE3-induced transcriptomes , indicating that the fusion proteins do not only regulate transcription via TEADs . Accordingly , unbiased motif enrichment analysis of TAZ-CAMTA1 revealed an enrichment for the EGR2 binding motif while YAP-TFE3 showed an enrichment for the MITF-binding motif . This indicates that the genomic occupancy of TAZ-CAMTA1 and YAP-TFE3 is mediated by the C termini of CAMTA1 and TFE3 in addition to interactions with the TEAD transcription factors . Bioinformatic approaches predicting the chromatin binding of transcription factors that contain multiple DNA binding sites have been developed ( Vandel et al . , 2019 ) , which in combination with structure-function studies should yield additional insight into how these chimeric transcription factors interface with the genome . We found that TAZ-CAMTA1 and YAP-TFE3 increased the amount of transcriptionally active chromatin when compared to TAZ and YAP overexpression . Although genes embedded in transcriptionally accessible portions of the genome showed a degree of overlap between the fusion proteins and either TAZ or YAP , hypergeometric analysis showed that the euchromatin landscape induced by TAZ-CAMTA1 more closely resembled that of CAMTA1 , while that of YAP-TFE3 showed an almost equivalent level of similarity to TFE3 as it did to YAP . This observation is reflected in the PCA analysis , which showed that the distribution of transposase accessible chromatin for the fusion proteins was more closely related to each other , CAMTA1 , and TFE3 than to TAZ/YAP . This indicates that CAMTA1 and TFE3 confer chromatin remodeling properties to the fusion proteins , while the N termini of TAZ and YAP essentially contribute the TEAD binding domain . This is consistent with previous studies that have demonstrated that TAZ/YAP’s interactions with epigenetic modifiers are key to their function . For example , in Drosophila , Yorkie ( YAP orthologue ) can interact with Ncoa6 , a subunit of the Trithorax-related histone H3 lysine four methyltransferase complex to drive tissue growth . Importantly , conjugation of Ncoa6 to Scalloped ( TEAD orthologue ) alone was sufficient to functionally activate the Yorkie transcriptional program ( Oh et al . , 2014; Qing et al . , 2014 ) . In mammalian/cancer cells , YAP/TAZ-driven transcription is facilitated by binding to BRD4 , a member of the bromodomain and extraterminal ( BET ) protein family of epigenetic modifiers that mediates interaction between enhancer sequences and associated promoters ( Zanconato et al . , 2018 ) . Taken together , the above studies indicate that additional fine tuning of the TEAD-based transcriptional program is essential to TAZ/YAP function in both physiological and cancer-related contexts . Although the overlap in euchromatin profile between the fusion proteins and CAMTA1/TFE3 can be partially explained by DNA-binding elements conferred by the C termini of CAMTA1 and TFE3 , it also raises questions about how the TAZ-CAMTA1:ATAC and YAP-TFE3:ATAC complexes form . It remains to be determined how the fusion proteins dynamically recruit epigenetic modifiers to contact points in the genome where TEAD and non-TEAD binding sites ( e . g . MITF or EGR2 ) are in close proximity . Although we found that TAZ-CAMTA1 and YAP-TFE3 interact with numerous epigenetic complexes , the combined proteomic/RNAi screen demonstrated that the ATAC complex is the most functionally relevant for cell transformation . Ada2a-containing acetyltransferase ( ATAC ) and Spt-Ada-Gcn5-acetyltransferase ( SAGA ) are two closely related metazoan histone acetyltransferase complexes that both contain GCN5 or the closely related PCAF histone acetyltransferase catalytic subunits . While the SAGA complex is predominantly localized to promoter sequences , the ATAC complex is found at both promoter and enhancer sequences ( Krebs et al . , 2011 ) . This is consistent with our observation that TAZ-CAMTA1 and YAP-TFE3 , which are predominantly localized at enhancer sequences , interact primarily with the ATAC complex rather than the SAGA complex . Although the ATAC complex can regulate both housekeeping and tissue specific functions , its binding to enhancer sequences is cell-type specific ( Krebs et al . , 2011 ) . Furthermore , knock-down of subunits of the ATAC complex in cells expressing TAZ-CAMTA1 or YAP-TFE3 affected only a limited set of active genes rather than globally altering gene expression , another key piece of data demonstrating that the ATAC complex modulates a specific transcriptional program . The implication of these data is that the ATAC complex is recruited to a particular set of target genes and interacts with a specific set of transcription factors rather than ubiquitously interacting with many different transcription factors across the genome . This is corroborated by our BioID mass spectrometry data which showed that the ATAC complex preferentially binds to both fusion proteins as compared to full-length TAZ and YAP . Importantly , our study also indicates that the ATAC complex represents a potential therapeutic target for EHE , a sarcoma currently lacking an effective medical therapy , regardless of which fusion protein it harbors . In addition , RNA expression profiling data show that expression of YEATS2 and ZZZ3 have prognostic significance in sarcoma clinical samples across various histological types , suggesting that the ATAC complex might represent a key oncogenic driver in other sarcomas . Future studies elucidating therapeutic strategies targeting the ATAC complex are warranted . Despite the observation that CAMTA1 and TFE3 are from entirely different protein families , several lines of evidence reveal that the function of TAZ-CAMTA1 and YAP-TFE3 are remarkably similar . As mentioned above , the two fusion proteins are similar in terms of the genes they induce , their chromatin-binding profiles , and how they alter the chromatin landscape . This raises the question of whether other TAZ/YAP gene fusions show similar convergence of function . Instead of point mutations , gene fusions are the most common genetic alterations of TAZ ( WWTR1 ) and YAP in cancer . Structurally similar WWTR1 ( TAZ ) /YAP1 gene fusions have also been identified in nasopharyngeal carcinoma ( YAP1-MAMLD2 ) ( Valouev et al . , 2014 ) , lung cancer ( YAP1-BIRC2; WWTR1-SLC9A9 ) ( Dhanasekaran et al . , 2014 ) , cervical squamous cell carcinoma and endocervical adenocarcinoma ( YAP1-SS18 ) ( Hu et al . , 2018; Szulzewsky et al . , 2020 ) , poromas/porocarcinomas ( YAP1-MAML2 and YAP1-NUTM1 ) ( Sekine et al . , 2019 ) , ependymomas ( YAP1-MAMLD1 and YAP1-FAM118B ) ( Szulzewsky et al . , 2020; Pajtler et al . , 2015; Pajtler et al . , 2019 ) , ossifying fibromyxoid tumor ( KDM2A-WWTR1 ) ( Kao et al . , 2017 ) , NF2-wildtype meningiomas ( YAP1-MAML2 , YAP1-PYGO1 , YAP1-LMO1 ) ( Sievers et al . , 2020 ) , retiform and composite hemangioendotheliomas ( YAP1-MAML2 ) ( Antonescu et al . , 2020 ) , and sclerosing epithelioid fibrosarcoma ( YAP1-KMT2A ) ( Puls et al . , 2020; Kao et al . , 2020 ) . Here , we have outlined an approach integrating NGS modalities with combined proteomic/genetic screens to gain insight into the transcriptional regulatory mechanisms of the TAZ-CAMTA1 and YAP-TFE3 fusion proteins . At its most essential , the mechanism of the two fusion proteins is the same in that a TEAD-based transcriptional program is modulated by interaction with an epigenetic modifier , the ATAC complex . Additional studies are needed to determine if YAP/TAZ/TEAD fusion proteins in other cancers also drive tumorigenesis via the ATAC complex or do so by recruiting different chromatin-modifying complexes . Double Flag TAZ-CAMTA1 and Double Flag CAMTA1 were cloned into pBabeNeo as previously described . Double Flag TAZ4SA was derived utilizing previously constructed TAZ S89A ( Tanas et al . , 2016 ) . Additional S66A , S117A , and S311A mutations were introduced by site directed mutagenesis using the QuikChange II Site-directed mutagenesis kit ( Agilent #200521 ) and the following primers: Double Flag YAP , YAP-TFE3 and TFE3 constructs were cloned into pBabeNeo as previously described ( Tanas et al . , 2016 ) . The pPGS-3HA-TEAD1 construct was obtained courtesy of Dr . Kunliang Guan . Triple Myc-tagged KAT14 plasmid ( Vector ID: VB18 p18-119pmf; Vector name: pRP[Exp]-Puro-CMV>BamHI/3xMyc-KAT14/EcorRI ) was synthesized by Vector builder , Shenandoah , TX . NIH 3T3 mouse fibroblasts and SW872 human liposarcoma cells were obtained from American Type Culture Collection ( ATCC , Manassas , VA , USA ) . NIH 3T3 cells were cultured in DMEM media containing 10% bovine calf serum ( Invitrogen-Life Technologies ) , 1 mM sodium pyruvate , and 50 µg/mL penicillin/streptomycin . SW872 cells were cultured in DMEM media ( according to ATCC recommendations ) containing 10% fetal bovine serum ( Invitrogen-Life Technologies ) , 1 mM sodium pyruvate , and 50 µg/mL penicillin/streptomycin . All cells were cultured at 37°C and 5% CO2 . Retroviral transfection with pBabeNeo constructs was performed by transfecting PhoenixA retroviral packaging cells with 10 µg of plasmid DNA . Transfection was done using Lipofectamine Reagent and Plus Reagent ( Invitrogen-Life Technologies ) according to manufacturer’s instructions . Supernatant was collected at 48 and 72 hr after transfection , filtered with 0 . 45 µm filters and supplemented with 8 µg/mL polybrene ( EMD Millipore , Burlington , MA , USA ) . Serial transductions ( 48 and 72 hr supernatants ) were applied to either SW872 or NIH 3T3 cells for 8 hr each . Pooled stable lines were generated by selecting with G418 for 2 weeks . PLKO . 1-puro constructs for knock-down of Tead1 were obtained from Sigma-Aldrich ( MISSION shRNAs ) . Constructs were transfected into LentiX ( HEK293T ) cells , along with pcMVΔ8 . 12 and pVSVG packaging plasmids , using Lipofectamine Reagent and Plus Reagent ( Invitrogen ) according to manufacturer’s instructions . Supernatants were collected at 48 hr , filtered using 0 . 45 µm filters , and polyethylene glycol ( PEG ) 8000 was added to a final concentration of 12% , and stored at 4°C overnight . The following day , 48 hr supernatants were centrifuged and viral pellets were resuspended in 0 . 45 µm filtered 72 hr media and supplemented with 8 µg/mL ( EMD Millipore , Burlington , MA , USA ) added to either SW872 or NIH 3T3 cells overnight . Pooled stable lines were generated by selection in puromycin . Anti-FLAG antibody ( mouse monoclonal clone M2 ( catalog # F3165; RRID:AB_259529 ) utilized for immunofluorescence and western blot ( 1:1000 ) was obtained from Sigma-Aldrich ( St . Louis , MO , USA ) . Anti-β-actin antibody ( AC-15; catalog #A544; RRID:AB_4767441 ) utilized for western blot ( 1:10 , 000 ) was obtained from Sigma-Aldrich ( St . Louis , MO , USA ) . Alexa 568-conjugated secondary antibody ( catalog# A11031; RRID:AB_144696 ) was obtained from Invitrogen-Life Technologies ( Grand Island , NY , USA ) . Anti-alpha tubulin antibody ( clone DM1A , catalogue #T9026; RRID:AB_477593 ) was obtained from Sigma-Aldrich ( St . Louis , MO , USA ) . Anti-H3 antibody ( clone 1G1 , catalogue# sc-517576; RRID:AB_2848194 ) was obtained from Santa Cruz Biotechnology ( Dallas , TX , USA ) . Anti-TEAD1 antibody ( clone EPR3967 Errani et al . , 2011 , catalogue #AB133533; RRID:AB_2737294 ) was obtained from Abcam ( Cambridge , MA , USA ) . Anti-HA antibody ( clone F-7 , sc-7392; RRID:AB_627809 ) was obtained from Santa Cruz Biotechnology ( Dallas TX , USA ) . Anti-ZZZ3 antibody ( catalogue# ab118800; RRID:AB_10901749 ) was obtained from Abcam ( Cambridge , MA , USA ) . Anti-Myc antibody ( catalogue # 2272S; RRID:AB_10692100 ) was obtained from Cell Signaling Technologies ( Danvers , MA , USA ) . Anti-TFE3 antibody ( catalogue# HPA023881; RRID:AB_1857931 ) was obtained from Sigma-Aldrich . Horseradish peroxidase-conjugated secondary antibodies used for western blots were obtained from Santa Cruz Biotechnology ( catalog# sc-2055; RRID:AB_631738 , sc-2054; RRID:AB_631748 , or sc-2033; RRID:AB_631729 ) and used at 1:5000 or 1:10 , 000 ) . Antibodies used for immunohistochemistry include anti-Ki-67 from Abcam ( ab#16667 , clone SP6 ) . Anti-vitronectin antibody ( mouse anti-human monoclonal antibody , catalogue# LS-B2118 ) , anti-COL1A1 antibody ( clone 3G3 , catalogue# LS-B5932 ) , anti-CTGF antibody ( rabbit polyclonal , catalogue# LS-B3284 ) , and anti-fibulin five antibody ( rabbit polyclonal , catalogue# LS-B14518 ) obtained from LifeSpan BioSciences ( Seattle , WA ) . Dilution for the above antibodies were as follows: anti-Ki-67 ( 1:100 ) , anti-vitronectin ( 1:400 ) , anti-COL1A1 ( 1:200 ) , anti-CTGF ( 1:200 ) , and anti-fibulin 5 ( 1:200 ) . For all of those markers , the antigen retrieval was Citrate Buffer pH 6 . 0 , 110°C for 15 min , in Biocare Decloaker ( Biocare Medical , Concord , CA , USA ) . Secondary antibodies were anti-mouse for VTN and COL1A1 , anti-rabbit for Ki-67 , FBLN and CTGF , from Dako Envision HRP System ( Dako North America , Inc , Carpentaria , CA , USA ) . Cell pellets were lysed in radioimmunoprecipitation assay ( RIPA ) buffer , containing cOmplete Protease Inhibitor Cocktail ( EDTA-free ) ( Roche ) and PhosSTOP Phosphatase Inhibitor Cocktail ( Roche ) according to the manufacturer’s instructions . Total protein concentration was measured using Pierce BCA Protein Assay Kit ( ThermoFisher Scientific , Waltham , MA , USA ) . Between 50 and 100 µg of total protein was loaded onto a gradient ( 4–15% ) polyacrylamide gel ( BioRad , Hercules , CA , USA ) . Proteins were then transferred to a polyvinylidene difluoride ( PVDF ) membrane and probed with antibodies described above . Each experiment was repeated at least twice . One mg of total cellular protein ( whole cell lysates ) were cleared by incubating with 2 . 0 µg of the control IgG and 20 µL vortexed Protein G PLUS-Agarose beads ( Santa Cruz , sc-2002 ) for 30 min with rocking at 4°C . Supernatant transferred and incubated with 10 µg primary antibody or IgG control and incubated at 4°C overnight with rocking . 40 µL of vortexed Protein G PLUS agarose beads added and incubated at 4°C on a rocker platform for 4 hr . After centrifugation and removal of the supernatant , the cell pellet was re-suspended and washed three times with 1 mL cold 1X PBS followed by an additional wash with 1 mL 1X RIPA buffer . The pellet/beads were then resuspended in 2X Laemmli buffer , boiled , then loaded onto polyacrylamide gel . Nuclear and cytoplasmic fractionation was performed with the Nuclear Extract Kit ( Active Motif , 40010 ) . Cells collected from 15 cm cell culture plates were centrifuged and resuspended in 500 µL 1X Hypotonic buffer and incubated for 5 min on ice . Twenty-five µL of detergent was added to resuspended cells which were subsequently vortexed for 10 s . After centrifugation , the supernatant cytoplasmic fraction was removed and stored . Remaining nuclear pellet was resuspended in 50 µL complete lysis buffer followed by vortexing . After centrifugation , the supernatant nuclear fraction was removed and stored . Cells were plated in chambered slides ( 2 . 5 × 104 for sparse conditions or 5 × 104 for confluent conditions ) and allowed to grow until desired timepoint . Cells were fixed with 4% paraformaldehyde ( in 1X PBS ) for 15 min and then washed with 1X PBS . Cells were permeabilized and blocked with 0 . 3% Triton X-100% and 3% fetal bovine serum for 30 min . Primary antibodies , as described above , were diluted in 3% fetal bovine serum and incubated on the cells overnight at 4°C in a humidity chamber . The following morning , the primary antibody was removed , cells were washed with 1X PBS , and then incubated with Alexa Fluor 568-conjugated secondary antibody ( described above ) for 45 min to 1 hr at room temperature . Immunofluorescence was visualized using a Leica DFC3000G microscope/camera and Leica Application Suite Advanced Fluorescence imaging software or an Olympus BX-61 microscope ( Tokyo , Japan ) with cellSens imaging software . Each experiment was repeated at least twice . The base layer of 0 . 5% agarose was plated into 6-well plates ( 2 mLs/well ) and allowed to solidify . For SW872 cells , 5 × 103 cells/2 mL in 0 . 35% agarose was added to form the top layer . For NIH 3T3 cells , 1 × 104 cells/2 mL in 0 . 35% agarose was added to form the top layer . Plates were left in the hood and allowed to solidify at room temperature for 3 hr . Colonies were allowed to grow for 2–3 weeks at 37°C and 5% CO2 before imaging . Each experiment was repeated at least twice . Poly ( 2-hydroxyethyl methacrylate ) , or poly-HEMA , solution was made at 20 mg/mL in 95% ethanol . 96-well tissue culture plates were coated with 130 µL of poly-HEMA solution and UV sterilized overnight . Cells were plated ( 2 × 103 cells/100 µL for SW872 or 3 × 103 cells/100 µL for NIH 3T3 ) into multiple wells and incubated at 37°C and 5% CO2 . Proliferation was assessed every other day by adding 10 µL of Cell Counting Kit-8 reagent ( Dojindo Molecular Technologies , Inc , Rockville , MD , USA ) . The plates were then incubated for 1 hr at 37°C and 5% CO2 and then absorbance at 450 nm was measured using a Synergy H1 Hybrid Multi-Mode Microplate Reader ( Biotek , Winooski , VT , USA ) . NOD scid gamma ( NSG ) mice were obtained from The Jackson Laboratory ( Bar Harbor , ME , USA ) and carry the strain NOD . Cg-Prkdcscid Il2rgtm1Wjl/SzJ ( RRID:IMSR_JAX:005557 ) . All animal work were approved by the University of Iowa Institutional Animal Use and Care Committee . SW872 cells containing either empty vector , YAP-TFE3 , or TAZ-CAMTA1 and NIH 3T3 cells containing empty vector , YAP-TFE3 , or TAZ-CAMTA1 ( 1 × 107 cells/500 µL PBS ) were injected into the flank of NSG mice . Mice were 6 months of age at the time of NIH 3T3 injection , and 2–3 months of age at the time of SW872 injection . Xenografts were repeated twice , using 5–10 mice per group . SW872 and NIH 3T3 cells were scraped and collected using TRIzol reagent ( Ambion-Life Technologies ) . Total RNA was isolated using the PureLink RNA mini kit ( Invitrogen-ThermoFisher Scientific ) . On-column DNase ( Invitrogen ) treatment was performed according to manufacturer’s instructions . Purified RNA was quantified using a NanoDrop ( ThermoFisher ) and 1 µg was converted to cDNA using SuperScript III First-Strand Synthesis System ( Invitrogen ) according to manufacturer’s instructions . PCR amplification was performed in technical triplicates on the Applied Biosystems QuantStudio 3 Real-Time PCR System ( Applied Biosystems-ThermoFisher ) . TaqMan Universal PCR Master Mix ( Applied Biosystems-ThermoFisher ) and PrimeTime standard qPCR primer/probe sets from Integrated DNA Technologies ( Iowa City , IA , USA ) were utilized . The qPCR cycling conditions were as follows: 95 °C10:00 ( 95 °C0:15 , 60 °C1:00 ) 40 . Relative quantitation was performed utilizing the delta-delta CT method and POLR2A/Polr2a ( RNA Polymerase II ) CT values as the reference control . Each experiment was repeated at least twice . The following primers and probes were used: HEK293 cells ( 4 × 105 ) stably containing empty vector , YT , YAP , YT S94A , YAP 5SA , TC , TAZ , TC S51A , or TAZ 4SA were transfected with the pGL3b-8xGTIIC firefly luciferase reporter plasmid ( containing 8 TEAD1-4 binding sites; GTIIC also known as MCAT and Hippo response element ) ( Dupont et al . , 2011 ) and the renilla luciferase reporter plasmid ( pRL-TK Renilla , ( Promega , Madison , WI , USA ) ) in a well from a six-well plate ( biological triplicates ) . Two days after transfection , the cells were collected and lysed , and extracts were assayed in technical triplicate for firefly and renilla luciferase activity using the Dual Luciferase Reporter Assay ( Promega ) and a Synergy H1 Hybrid Multi-Mode Microplate Reader ( Biotek , Winooski , VT , USA ) . Each experiment was repeated at least twice . NIH 3T3 and SW872 cells were stably transfected , in triplicate , expressing the following proteins empty vector ( EV ) , YAP 5SA , TFE3¸YAP-TFE3 ( YT ) , TAZ 4SA , CAMTA1 , and TAZ-CAMTA1 ( TC ) . All cell lines , except those containing CAMTA1 , were grown to confluence for 48 hr before collection . Cell lines containing CAMTA1 were collected at sub-confluence . Total RNA was extracted using Trizol Reagent ( Ambion-Life Technologies ) . Total RNA was isolated using the PureLink RNA mini kit ( Invitrogen-ThermoFisher Scientific ) . On-column DNase ( Invitrogen ) treatment was performed according to manufacturer’s instructions . Transcription profiling using RNA-Seq was performed by the University of Iowa Genomics Division using manufacturer recommended protocols . Initially , 500 ng of DNase I-treated total RNA was used to enrich for polyA containing transcripts using oligo ( dT ) primers bound to beads . The enriched RNA pool was then fragmented , converted to cDNA and ligated to sequencing adaptors containing indexes using the Illumina TruSeq stranded mRNA sample preparation kit ( Cat . #RS-122–2101 , Illumina , Inc , San Diego , CA ) . The molar concentrations of the indexed libraries were measured using the 2100 Agilent Bioanalyzer ( Agilent Technologies , Santa Clara , CA ) and combined equally into pools for sequencing . The concentration of the pools was measured using the Illumina Library Quantification Kit ( KAPA Biosystems , Wilmington , MA ) and sequenced on the Illumina HiSeq 4000 genome sequencer using 75 bp paired-end SBS chemistry . Barcoded samples were pooled and sequenced using an Illumina HiSeq 4000 in the Iowa Institute of Human Genetics ( IIHG ) Genomics Core Facility . Average reads per sample for SW872 samples was 93 M ( ±16 M ) . Reads were demultiplexed and converted to FASTQ format , then processed with ‘bcbio-nextgen’ ( https://github . com/chapmanb/bcbio-nextgen; RRID:SCR_002630 version 1 . 0 . 8 ) . SW872 reads were aligned against the ‘hg38’ reference genome ( genome FASTA and annotations derived from Ensembl release 78 ( ftp://ftp . ensembl . org/pub/release78/fasta/homo_sapiens/dna/Homo_sapiens . GRCh38 . dna . toplevel . fa . gz ) ) . Reads from NIH 3T3 cells were aligned against ‘mm10’ ( genome FASTA and annotations derived from ftp://ftp . ensembl . org/pub/release-97/fasta/mus_musculus/ ) reference genome using ‘Hisat2’ ( Kim et al . , 2015 ) . Reads were also pseudo-aligned to the transcriptome using Salmon ( Patro et al . , 2017 ) . Transcript-level abundances were converted to gene-level counts with ‘tximport’ ( Soneson et al . , 2015 ) . QC was performed with ‘qualimap’ and ‘samtools’ ( Li et al . , 2009 ) . Differential gene expression analysis was performed with DESeq2 ( Love et al . , 2014 ) . Gene lists were analyzed using Advaita’s ‘iPathwayGuide’ ( https://www . advaitabio . com/ipathwayguide; Advaita Bioinformatics , Ann Arbor , MI ) . SW872 cells stably expressing Flag-tagged EV , YAP 5SA , TFE3 , YT , TAZ 4SA , CAMTA1 , and TC were grown to 90% confluence in 15 cm tissue culture plates ( in biological triplicate ) . Chromatin immunoprecipitation for H3K27ac performed on SW872 cells expressing the above constructs was performed in biological duplicates . Cells were cross-linked and prepared using the SimpleChIP Enzymatic Chromatin IP Kit ( Cell Signaling , #9003 , Danvers , MA ) . Protocol was performed according to manufacturer’s instructions . Lysates were sonicated for three sets of 20 s pulses in ice bath , using a VirTis Virsonic 60 Sonicator at setting 6 , with a 30 s ice bath incubation between sonication pulses or Qsonic Q800R3 3 sets of 45 s pulse , 30 s rest , 50% amplitude . Chromatin digestion and concentration was verified to be within the desired range using the 2100 Agilent Bioanalyzer ( Agilent Technologies , Santa Clara , CA ) . The SimpleChIP ChIP-seq DNA Library Prep Kit for Illumina ( Cell Signaling , #56795 ) was used to prepare the DNA library according of manufacturer’s instructions . Immunoprecipitation was performed with the anti-DYKDDDDK Tag Rabbit monoclonal antibody ( Cell Signaling #14793 ) , anti-acetyl H3K27 ( Cell Signaling , clone D5E4 , #81735 ) or provided isotype rabbit IgG control ( Cell Signaling ) . The molar concentrations of the indexed libraries were measured using the 2100 Agilent Bioanalyzer ( Agilent Technologies , Santa Clara , CA ) and combined equally into pools for sequencing using an Illumina HiSeq 4000 in the Iowa Institute of Human Genetics ( IIHG ) Genomics Core Facility . ChIP-seq analysis was carried out using ‘bcbio-nextgen’ in ‘analysis: chip-seq’ mode . Reads were pooled across replicates to obtain improved read depth , then trimmed for low quality and adapter read-through with ‘Atropos’ ( version 1 . 1 . 16 ) ( Didion et al . , 2017 ) with adapter sequences ( ‘AGATCGGAAGAG’ , ’CTCTTCCGATCT’ ) and options set as ‘--quality-cutoff=5 --minimum-length=25’ . Trimmed reads were aligned to the ‘GRCh37/hg19’ reference assembly with ‘BWA-MEM’ ( Li and Durbin , 2009 ) using default parameters . Multi-mapping reads were removed before peak calling , and ‘greylist’ regions were detected and removed ( https://github . com/roryk/chipseq-greylist; RRID:SCR_002630 version 1 . 0 . 8 ) . Peaks were called using ‘macs2’ ( Zhang et al . , 2008 ) with command-line options ‘-f BAMPE’ , ‘-g 2 . 7e9’ , ‘-B’ , and ‘-q 0 . 10 . ’ ( FDR = 0 . 1 ) Narrow peak files were imported into the R and annotated with ‘ChIPseeker’ ( Yu et al . , 2015 ) and ‘clusterProfiler’ packages ( Yu et al . , 2012 ) . Peak ranges for motif analysis were written out as FASTA using ‘Biostrings’ ( https://bioconductor . org/packages/release/bioc/html/Biostrings . html ) and ‘BSgenome’ packages ( https://bioconductor . org/packages/release/bioc/html/BSgenome . html ) . Motif enrichment analysis was carried out with MEME suite ( http://meme-suite . org ) using the ‘AME’ and ‘FIMO’ modules and HOCOMOCO ( v11 ) database . PCR amplification was performed in technical triplicates on the Applied Biosystems QuantStudio 3 Real-Time PCR System ( Applied Biosystems-ThermoFisher ) . The PCR reaction contained SimpleChip Universal master mix ( catalogue # 88989 ) Cell Signaling Technologies ( Danvers MA , USA ) , 5 µM of primers and 2 µL of ChIP DNA or 2% input DNA . Cycling conditions as follows , 3 min at 95°C for initial denaturation followed by 40 cycles of 15 s at 95°C then 1 min at 60°C . Percent input was calculated . The following primers were used: A total of 50 , 000 cells were trypsinized and pelleted . Cells were washed once with 50 µL of cold 1x PBS buffer . After pelleting , cell pellet was resuspended in 50 µL cold lysis buffer ( 10 mM Tris-HCl , pH 7 . 4 , 10 mM NaCl , 3 mM MgCl2 , 0 . 1% IGEPAL CA-630 ) . The suspension was centrifuged again and supernatant discarded leaving a nuclear pellet . A transposition reaction was prepared containing 25 µL Tagment DNA buffer ( #15027866 ) , 2 . 5 µL TDE1 Nextera Tn5 Transposase ( #15027865 ) , and 22 . 5 µL Nuclease Free H2O . The nuclear pellet was resuspended in the transposition reaction mixture . The transposition reaction was incubated at 37°C for 30 min . After transposition , fragmented DNA purified using a Qiagen miniElute PCR purification kit ( catalogue #28004 ) per the manufacturer’s instructions . Transposed DNA fragments were PCR amplified using the Nextera XT Index Kit ( #15055293 ) per the manufacturer’s instruction to generate indexed libraries . Analysis of ATAC-seq data was carried out as follows . Reads were trimmed using ‘NGmerge’ in ‘adapter removal’ mode ( Gaspar , 2018 ) . Trimmed reads were then aligned as described above [command line settings: “-c 250 t 1 v 3’] and sorted with ‘samtools’ ( Li et al . , 2009 ) . Peaks were called with ‘genrich’ ( https://github . com/jsh58/Genrich; RRID:SCR_002630 version 1 . 0 . 8 ) in ‘atacseq’ mode [with flags set as ‘-E blacklists/wgEncodeHg19ConsensusSignalArtifactRegions . bed -v -j -r -e MT , GL000191 . 1 , GL000192 . 1 , GL000193 . 1 , GL000195 . 1 , GL000199 . 1 , GL000205 . 1 , GL000206 . 1 , GL000208 . 1 , GL000212 . 1 , GL000214 . 1 , GL000216 . 1 , GL000217 . 1 , GL000219 . 1 , GL000220 . 1 , GL000222 . 1 , GL000223 . 1 , GL000224 . 1 , GL000225 . 1 , GL000226 . 1 , GL000228 . 1 , GL000235 . 1 , GL000243 . 1’ . ] Genrich removes PCR duplicates , excludes named regions , and blacklist regions , and properly accounts for multi-mapping reads . Peaks were called individually on each replicate . Peak files were imported along with BAM files into DiffBind ( https://bioconductor . riken . jp/packages/3 . 0/bioc/html/DiffBind . html ) ( Ross-Innes et al . , 2012 ) and a consensus peakset was determined . A binding affinity matrix containing ( normalized ) read counts for each sample at each consensus site was computed . ‘DESeq2’ ( Love et al . , 2014 ) was used to test contrasts for differential chromatin accessibility . Finally , PCA analysis was performed to show how samples cluster at differentially accessible chromatin . Full-length YAP , TFE3 , TAZ , and CAMTA1 , along with YAP-TFE3 and TAZ-CAMTA1 were cloned into pcDNA5 FRT/TO FLAG‐BirAR118G vectors , containing an N-terminal E . coli biotin ligase with an R118G mutation , as described in Roux et al . , 2012 and Lambert et al . , 2015 . These bait proteins of interest were then stably expressed in Flp-In T-REx HEK293 cells as previously described ( Couzens et al . , 2013 ) . Expression was then induced using 1 µg/mL tetracycline for 24 hr . The cells were also treated with 50 µM biotin at the time of induction . After 24 hr , cells harvested and processed ( Lambert et al . , 2015 ) . Two biological replicates were made for each cell line . Affinity purification and proximity biotinylation coupled to mass spectrometry were performed as described in Lambert et al . , 2015 . SAINT ( significance analysis of interactome ) analysis ( Choi et al . , 2011 ) was performed on the mass spectrometry data , using 10 controls compressed to 5 . Only proteins with iProphet protein probability ≥ 0 . 95 were used . Recovery of bait peptides was monitored to ensure that expression levels of the bait proteins were similar ( Figure 6—figure supplement 1D , E ) . Results are expressed in dotplot format . Columns show bait proteins , while the rows list the names of the identified prey proteins . Each prey protein is represented as a dot , with color signifying average spectral count , the darkness indicating average spectral count between the two biological replicates ( the darker the dot , the higher the average spectral count ) , and the size represents the relative abundance . The ring around the dot indicates the false discovery rate ( the darker the ring , the higher the confidence ) . The data was filtered so that each prey had a minimum of 10 spectral counts in at least one of the biological replicates . Results were also filtered to exclude components of trypsin , biotin , and streptavidin . Six epithelioid hemangioendotheliomas , six epithelioid hemangiomas , and five epithelioid angiosarcomas were obtained from the files of the Department of Pathology , University of Iowa Hospitals and Clinics . Internal Review Board approval was obtained ( IRB# 201609806 ) . For soft agar assays , statistical significance was evaluated using an unpaired two-tailed t-test . For poly-HEMA proliferation assays , statistical significance was evaluated using fold change increase in proliferation at day 10 with an unpaired two-tailed t-test . For quantitative RT-PCR , standard deviation was calculated from fold change values for each triplicate . For Kaplan-Meier curves significance was determined by Log-rank ( Mantel-Cox ) test . Graphs were made using the average of the technical replicates . Each experiment was repeated at least twice . For NGS studies , biological triplicates were generated by separately transducing NIH 3T3 and SW872 cells with the various contructs followed by stable selection . Hypergeometric testing was performed using the phyper ( ) function in the stats R package ( v3 . 6 . 3 ) set to assess enrichment and the lower tail set to false . Hypergeometric density was calculated using the related dhyper function and converted using the negative log10 of the output . For gene expression data , the population was set to the total number of recovered genes with the mean of five counts across all samples . Both the ChIP-seq and ATAC-seq , the population was set as the total number of genes annotated across all conditions . Error bars were used to define one standard deviation . For all panels , *p<0 . 0001 , p<0 . 001 , p<0 . 01 , p<0 . 05 .	The proliferation of human cells is tightly regulated to ensure that enough cells are made to build and repair organs and tissues , while at the same time stopping cells from dividing uncontrollably and damaging the body . To get the right balance , cells rely on physical and chemical cues from their environment that trigger the biochemical signals that regulate two proteins called TAZ and YAP . These proteins control gene activity by regulating the rate at which genes are copied to produce proteins . If this process becomes dysregulated , cells can grow uncontrollably , causing cancer . In cancer cells , it is common to find TAZ and YAP fused to other proteins . In epithelioid hemangioendothelioma , a rare cancer that grows in the blood vessels , cancerous growth can be driven by a version of TAZ fused to the protein CAMTA1 , or a version of YAP fused to the protein TFE3 . While the role of TAZ and YAP in promoting gene activity is known , it is unclear how CAMTA1 and TFE3 contribute to cell growth becoming dysregulated . Merritt , Garcia et al . studied sarcoma cell lines to show that these two fusion proteins , TAZ-CAMTA1 and YAP-TFE3 , change the pattern of gene activity seen in the cells compared to TAZ or YAP alone . An analysis of molecules that interact with the two fusion proteins identified a complex called ATAC as the cause of these changes . This complex adds chemical markers to DNA-packaging proteins , which control which genes are available for activation . The fusion proteins combine the ability of TAZ and YAP to control gene activity and the ability of CAMTA1 and TFE3 to make DNA more accessible , allowing the fusion proteins to drive uncontrolled cancerous growth . Similar TAZ and YAP fusion proteins have been found in other cancers , which can activate genes and potentially alter DNA packaging . Targeting drug development efforts at the proteins that complex with TAZ and YAP fusion proteins may lead to new therapies .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "cancer", "biology" ]	2021	TAZ-CAMTA1 and YAP-TFE3 alter the TAZ/YAP transcriptome by recruiting the ATAC histone acetyltransferase complex
Local and cross-border importation remain major challenges to malaria elimination and are difficult to measure using traditional surveillance data . To address this challenge , we systematically collected parasite genetic data and travel history from thousands of malaria cases across northeastern Namibia and estimated human mobility from mobile phone data . We observed strong fine-scale spatial structure in local parasite populations , providing positive evidence that the majority of cases were due to local transmission . This result was largely consistent with estimates from mobile phone and travel history data . However , genetic data identified more detailed and extensive evidence of parasite connectivity over hundreds of kilometers than the other data , within Namibia and across the Angolan and Zambian borders . Our results provide a framework for incorporating genetic data into malaria surveillance and provide evidence that both strengthening of local interventions and regional coordination are likely necessary to eliminate malaria in this region of Southern Africa . Renewed efforts against malaria have resulted in substantial gains in malaria control , with active plans to eliminate malaria from 35 countries ( Newby et al . , 2016 ) . Malaria elimination requires that national and regional strategies consider the impact of local and cross-border importation on local transmission ( Cotter et al . , 2013; Marshall et al . , 2016a; Wangdi et al . , 2015; WHO , 2017 ) . This is particularly important for eliminating countries that share porous borders with areas of higher transmission , where importation can play a major role in sustaining or reestablishing local transmission ( Sturrock et al . , 2015 ) . Identifying within-country and cross-border blocks of high parasite connectivity and coordinating elimination strategies accordingly will likely be required for national and regional success . Coordinating the optimal interventions to deploy when and where depends on understanding the impact of imported malaria infections on local transmission . If transmission is self-sustained locally , local control measures such as vector control will be necessary . If importation strongly connects the local parasite population to an external one , then interventions aimed at reducing malaria in these sources of importation or otherwise reducing vulnerability to importation may additionally be needed or even take precedence ( Cotter et al . , 2013 ) . Currently , the extent of importation is estimated primarily by taking recent travel histories of malaria cases ( Sturrock et al . , 2015 ) and , less commonly , from more general estimates of human mobility . However , routine collection of travel data is not universal , even in areas nearing elimination , and requires a robust surveillance system . When these data are collected , they are often incomplete ( e . g . only the most recent travel is reported ) , or are otherwise inaccurate ( e . g . due to disincentives such as reduced access to free healthcare when a patient reports foreign nationality ) ( Marshall et al . , 2016b; Pindolia et al . , 2012 ) . Even with an accurate travel history , it can be difficult to tell with confidence whether malaria parasites were acquired locally or during travel . Beyond these factors , obtaining travel history only from those presenting with symptomatic malaria does not address the contribution of asymptomatic carriers , which may be the population primarily responsible for importation . Thus , travel data alone are often unable to accurately capture the relative contribution of malaria importation to local transmission . Since travel data are often limited , approaches based on movement of the overall human population using anonymized mobile phone data have been developed to create a generalizable and scalable framework for estimating movement of malaria parasites . However , these methods rely heavily on modeling assumptions , assume that the movement patterns of mobile phone owners and the at risk population are similar , and have not been used to measure international travel ( Pindolia et al . , 2012; Ruktanonchai et al . , 2016; Tatem , 2014; Tatem et al . , 2014; Wesolowski et al . , 2012; Zhao et al . , 2016 ) . Since travel history and other data on human movement are limited in their ability to provide reliable information on malaria parasite connectivity , it may be valuable for control programs to additionally collect data on parasite genetics ( Wesolowski et al . , 2018 ) . In principle , data on the genetics of malaria parasites have the potential to provide the most direct measure of parasite connectivity and to identify relevant sources and sinks of parasite movement ( Auburn and Barry , 2017; Escalante et al . , 2015 ) . However , there have been few efforts to systematically collect and genotype malaria infected individuals at sufficient spatial and temporal scale or density to be useful in this regard . In addition , it has been difficult to detect relevant spatial signals in parasite genetic data using existing population genetic methods , particularly in areas such as sub-Saharan Africa that have high levels of population diversity and polyclonal infections ( Anderson et al . , 2000; Mobegi et al . , 2012 ) . Ideally , multiple complementary sources of human and parasite data would be compared and integrated to better understand the movement of malaria parasites and contribution to transmission at various spatial scales . As part of the Elimination 8 ( E8 ) initiative , a regional effort to eliminate malaria from Southern Africa , Namibia has been successful in decreasing malarial morbidity and mortality ( Elimination 8 , 2015 ) . However , this success has recently stalled , with the number of reported cases increasing in the last few years ( Nghipumbwa et al . , 2018; WHO , 2017 ) . To achieve the national malaria elimination target of 2020 , it will be critical to reassess the elimination strategy in northern Namibia , where nearly all cases in the country are reported . Of particular concern is determining the contribution and spatial scale of local transmission to malaria within northern Namibia , which should guide the geographic coverage and relative timing of local interventions , and the contribution of importation from neighboring Angola and Zambia , which should guide cross-border strategies . To address these concerns , we systematically collected parasite genetic data and human mobility data – travel history from confirmed malaria cases and national mobile phone call data records – from the region in 2015–16 . From these data , we aimed to determine the importance of local transmission and imported malaria , and to compare estimates of parasite connectivity at various spatial scales obtained from the different data sources . Our results demonstrate strong evidence for local transmission in northern Namibia , provide insight into patterns of parasite connectivity within Namibia and across national borders , and demonstrate the feasibility of efficiently generating actionable information for malaria control by augmenting traditional surveillance data with a direct evaluation of the parasite population . Within-host diversity and population level genetic diversity were assessed by health district ( n = 4 ) and health facility catchment ( n = 29 ) . Across health districts , infections from Rundu and Andara had greater within-host diversity than those from Nyangana and Zambezi districts as demonstrated by higher multiplicity of infection ( MOI ) and lower within-host fixation index ( Figure 2—figure supplement 1 ) , consistent with the higher malaria incidence and higher proportion of imported malaria cases in these districts . Overall , the genetic diversity of the parasite population was high throughout the study area ( median HE = 0 . 79 [IQR: 0 . 60–0 . 85] ) , though lower in Zambezi than the other three health districts ( Figure 2—figure supplement 1 ) . When stratified by health facilities , the patterns of within host and population diversity showed variability within districts , providing supporting evidence of fine-scale heterogeneity of malaria transmission in the study area ( Figure 2—figure supplement 2 ) . For example , infections detected at Rundu district hospital had the highest within-host diversity , which may be due to the larger proportion of patients who traveled to or resided in Angola ( reported by 13% of patients ) . Infections from Rundu also had the highest population diversity , which may be attributable to the large catchment area of this facility ( 48% came from beyond the study region ) . Cases were then classified as local or imported based on recent travel history and the location of residence . Infections from individuals with a history consistent with importation had higher within-host diversity than those without , despite having similar population-level diversity ( Figure 2A–C ) . These data suggest a lower rate of superinfection and thus less local transmission in northeastern Namibia compared to the international source populations . Existing model and distance ( multidimensional scaling and phylogenetic tree ) based methods did not identify any spatial clustering between health facilities or health districts ( Figure 2—figure supplement 3 ) . Population measures of genetic differentiation ( GST and Jost’s D ) also showed no relationship with geographic distance between health facilities ( Figure 2—figure supplement 4 ) . However , a novel analysis evaluating the distribution of genetic relatedness between all infections , including polyclonal infections , revealed a strong spatial signal . For this analysis , we identified highly related pairwise connections ( identity by state , IBS ) between health facilities , using comparisons between Namibia and non-neighboring African countries as a null distribution . The distribution of the pairwise genetic relatedness within the study area diverged from the null distribution at ≥0 . 5 , and these highly related infection pairs were responsible for the majority of the spatially informative genetic signal ( Figure 2D and E ) . We found a decay in genetic relatedness with increasing geographic distance within each of the two regions of Namibia ( Figure 2F and G , p<0 . 0001 , Mantel test ) , suggesting that there was sufficient sustained local transmission occurring in both study areas to create a strong spatial gradient in parasite populations . To evaluate the local connectivity of parasite populations , we computed the pairwise genetic connectivity between infections sampled from different health facilities . In Zambezi , 60% ( 9/15 ) of the pairwise connections were highly related . However , we observed overall lower connectivity and fewer highly related pairwise connections 39% ( 99/253 ) in the Kavango East region ( Figure 3A , Figure 3—source data 1 ) . The degree of parasite connectivity between health facility catchments was heterogeneous , with some very well-connected health facilities ( i . e . highly related connections to most other facilities ) and others relatively unconnected . For example , two health facilities in Zambezi and four in Kavango East were connected to most other health facilities in each region ( connectivity score = 0 . 65–0 . 91 , Figure 3B , Figure 3—source data 1 ) . In Zambezi , the two most connected clinics also had the highest incidence of malaria ( Mumbengegwi et al . , 2018 ) and were in close proximity to the Angolan border and Kavango East region than the other clinics . In contrast , Rundu district hospital , the largest health facility in the study area , was only connected to a few other health facilities ( connectivity score = 0 . 05 ) , consistent with the high genetic diversity and large catchment of this hospital , extending well beyond the study area . Overall , 17% of the pairwise genetic connectivity measures between health facilities in Kavango East and Zambezi regions were highly related , providing evidence of mixing between these parasite populations . Health facilities that were the most genetically connected within a region were also the most connected between regions ( Figure 3C ) , suggesting that specific localities may represent priority targets for interventions to efficiently reduce within and between region transmission . We also sought to estimate parasite connectivity using human mobility data . Estimates of time at risk for infection with malaria parasites were quantified from travel surveys and population estimates of human movement within Namibia derived from mobile phone calling data . Few individuals reported at least one night spent away from their residence location ( mean 1% ) , with a higher percentage ( 17% ) of mobile phone subscribers’ overall time spent outside of their primary residence tower ( Figure 1C ) . To estimate parasite mixing patterns from both data sets , we calculated the proportion of time spent at all destinations scaled by the relative malaria incidence to create a proportion of time at risk measure for each individual that was then aggregated to quantify mixing between locations . In both travel survey and mobile phone data , individuals spent the majority of their time at their location of residence , and mixing patterns inferred from both data sources found that individuals spent similar amounts of time at risk for importing malaria to Kavango East and Zambezi regions ( Figure 4—figure supplement 1 ) . To compare the various data sets at equivalent spatial scales , we aggregated the genetic data to mobile phone catchments ( n = 14 ) and travel survey destination locations ( n = 8 ) . In the mobility data alone , both mobile phone catchments ( n = 14 ) and travel survey destination locations ( n = 8 ) were strongly connected to their neighbors , with few travelers between Kavango East and Zambezi ( Figure 4 ) . Clusters identified using modularity maximization of these networks also highlight a strong spatial signature , where contiguous locations formed clusters . Using the same procedure and spatial areas , clusters were identified for the aggregated parasite genetic data . Clusters identified from mobile phone data shared some similarity to those identified from the genetic data ( grouping similarity measure: Rand Index = 0 . 76 , Figure 4 and Figure 4—figure supplement 2 , Figure 4—source data 1 ) . However , genetic data identified a substantial amount of parasite connectivity between locations that was not detected by mobile phone data . There was little agreement between clusters identified from the travel survey data and genetic data ( grouping similarity measure: Rand Index = 0 . 46 ) , albeit limited by a smaller number of geographic units due to the coarser spatial scale of the travel data . The precision of results obtained from travel data was also limited by the relatively small number of individuals who reported travel within the study area . Overall , these results suggest that both travel survey and mobile phone data have limitations in capturing the structure of parasite connectivity within northeastern Namibia detected using genetic data . To evaluate cross-border connectivity , geographic regions were aggregated to nine locations: four health districts in Namibia , three locations in Angola , and two provinces in Zambia . Mobile phone data were limited to Namibia , not allowing for evaluation of cross-border connectivity . Data from the travel survey identified some sources of cross-border importation into Namibia , with the most prominent connections being from Rundu to southern Angola ( 2 . 8% of cases reporting travel to this area ) and Zambezi to Western Zambia ( 1 . 8% of cases , Figure 5A ) . However , these data were only able to identify symptomatic cases with a direct history of travel and would not identify any cases which failed to report relevant history or those cases which may have originated from importation via asymptomatic carriers or transmissible individuals otherwise not detected by the routine surveillance . Therefore , we augmented travel history with genetic data to estimate the underlying connectivity of the parasite populations from the same nine locations . To evaluate connectivity using parasite genetics , we analyzed genetic data collected from Namibia ( this study ) as well as additional data from Angola and Zambia . Infections with malaria parasites from Namibia and northern Angola were not closely related ( mean proportion of highly related infections = 0 . 00004 [Range = 0–0 . 00015] ) . However , parasites between health districts of Namibia; between Namibia and southern Angola; and between Namibia and Western and Southern provinces of Zambia were more closely related ( Figure 5B ) . Overall , infections from Namibia were on average 142 times more likely to be genetically related to those from southern Angola and 191 times to Zambia than to those from northern Angola , indicating substantial parasite mixing within the geographically connected Namibia-Angola-Zambia regional block . In contrast , this finding suggests limited parasite connectivity between northern and southern Angola , though limited , non-contemporaneous sampling within Angola makes it difficult to make more detailed conclusions about transmission in this country . Within Namibia , infections from Andara and Nyangana were 3 and 4 times more likely to be genetically related to Zambezi than infections sampled from nearby Rundu , respectively ( Figure 5B ) . When evaluating specific connections between the nine locations , we estimated the direction of parasite flow in addition to the degree of connectivity between areas by weighting the pairwise proportion of highly related infections by malaria incidence to account for the differential risk of infections in different areas . Results from this analysis provided substantially more information on regional connectivity than using travel history data alone . The four studied health districts within Namibia were connected to each other but had stronger connections to nearby cross-border locations than to each other ( Figure 5C and D , Figure 5—source data 1 ) . The most important source populations were Calai and Dirico ( border towns in southern Angola ) , followed by Western and Southern provinces of Zambia . Although there was evidence of importation into both Kavango East and Zambezi regions from other countries , based on these data the Zambezi region was estimated to receive high rates of importation from a larger number of sources ( i . e . , it was a dominant sink population ) . It is clear that achieving elimination of malaria will require strategic coordination of local and regional interventions guided by accurate intelligence on parasite movement; what has not been clear is how to best obtain this information . In this study , we demonstrated that augmentation of traditional malaria surveillance with parasite genetic data added substantially to the understanding of transmission epidemiology in a critical region of Southern Africa straddling the border between Namibia , Angola , and Zambia . First , parasite genetics provided positive evidence that the majority of malaria cases observed in northeastern Namibia were due to local transmission , evidenced by the strong fine-scale spatial structure in the genetic data . It would be difficult to explain such consistent spatial clustering of highly related parasites and the observed decay with distance if local transmission did not predominate . This is a key piece of programmatically relevant information that would have been difficult to confirm with negative evidence , that is merely based on a lack of history consistent with importation , especially given potential disincentives for individuals to report living outside of Namibia . Second , the addition of genetic data to travel histories provided more detailed and extensive evidence of parasite connectivity over hundreds of kilometers , both within Namibia and across borders from Angola and Zambia . Conclusions regarding the origins and relative magnitude of malaria importation from genetic data were distinct from those obtained from travel history alone , which were likely limited by sparsity and bias , and from mobility from the mobile phone data , which were unable to inform cross-border movement in this study and appeared to underestimate the importance of long-distance connections within Namibia . Although cross-border movement is possible to obtain from mobile phone calling data , for example if the handset ID was used instead as an anonymized ID , data available for this study were limited to national travel patterns . Malaria programs will require targeted interventions at sub-national scales to effectively achieve and sustain elimination ( WHO , 2017 ) . The success of such programs , for example targeted vector control , focal screening and treatment , and mass drug administration will largely be dependent on tailoring interventions to drivers of ongoing transmission ( WHO , 2014 ) . Using a novel analytic framework , we found that despite a signal of predominantly local transmission ( within tens of kilometers ) , parasite populations in Namibia remain highly connected at longer scales within and between the two administrative regions ( over hundreds of kilometers ) . In this context , restricting interventions to a relatively small area , such as a region , may result in improved malaria control but is unlikely to achieve elimination unless any malaria transmission from imported parasites is completely prevented . Indeed , limiting elimination efforts to national boundaries may be doomed to fail for the same reasons , for example it may be necessary for Namibia to coordinate efforts with Angola and Zambia to eliminate transmission within its own boundaries ( Khadka et al . , 2018 ) . Consistent with this hypothesis , we found that parasite populations within Namibia were in many cases more closely connected to those across the border than to other parasites from the same country . At a more nuanced level , variations in the degree of genetic connectivity between areas we observed could be used to optimize the order in which interventions may be most efficiently implemented . For example , all else being equal , targeting interventions to locations with greater genetic connectivity than those with lower connectivity may be more effective in fragmenting the parasite population and reducing the influx of malaria from pockets of transmission . This finding is consistent with the previous observation that malaria cases were clustered in northeastern Namibia ( Smith et al . , 2017; Tatem et al . , 2014 ) . In addition , enhancing malaria surveillance and access to care , for example through additional clinics or border posts , may be effective in reducing the extent of cross-border importation if deployed in areas with high measured importation rates . We estimated malaria parasite connectivity from mobile phone data , travel history , and parasite genetics , allowing us to compare estimates based on human population movement to more direct estimates of the connectivity of parasite populations . In principle , anonymized mobile phone data can provide a continuous and inexpensive source of human mobility data . Within Namibia , connectivity estimates derived from vast amounts of mobile phone data roughly mirrored those derived from genetics , though they were predominated by small scale movements and did not capture the extent of longer distance connections revealed by the genetic data . This difference could be due to differential patterns of movement of people with access to mobile phone and those who potentially transmit malaria , limitations in the accuracy of malaria incidence estimates used as inputs in the model , or the temporal difference between the two data sets ( Ruktanonchai et al . , 2016 ) . An important current limitation of call data records analyzed was the inability to provide any information on movement beyond national boundaries . In contrast to mobile phone data , travel data collected from symptomatic cases were sparser , estimated less travel , and were collected at a coarser spatial scale , limiting agreement with the genetic data within Namibia . However , travel data were able to provide low-resolution information and demonstrated evidence of cross-border importation by infected individuals , albeit possibly biased by patient omission on international travel or residence . Estimates derived from genetics are likely to be more comprehensive , as even with perfect accuracy travel history only captures movements of the interviewed patient while genetics can record evidence of movement through multiple generations of transmission . Importantly , information on travel allowed us to greatly extend the utility of genetic data , providing a means of ‘sampling’ parasites from beyond the study site in those with a definitive history of international travel . Travel history remains a critical part of routine surveillance , and when collected reliably and ideally at finer spatial scale than done here will likely provide important information on its own ( Smith et al . , 2017; Tejedor-Garavito et al . , 2017 ) and in conjunction with genetic data . The genetic data used in this study were generated via traditional methods – a panel of 26 microsatellites – but were well-suited for the intended application and captured strong spatial signal over local and regional scales . Particular strengths of these data were the ability to capture information from polyclonal infections ( 77% of the study population ) given the multiallelic nature of the loci , and to obtain robust results from easily collected field samples ( dried blood spots and used rapid diagnostic tests ) . Targeted deep sequencing of short , multiallelic haplotypes may provide similarly rich data from polyclonal infections , allowing greater flexibility in the number and location of loci , facilitating easier comparisons across data sets , and taking advantage of continual advances and cost savings in sequencing technology ( Aydemir et al . , 2018; Lerch et al . , 2017 ) . Generating P . falciparum whole genome sequence data from Southern Africa would facilitate rational selection of the most informative sequence targets for local and regional parasite movement . Advances in analytical methods would likewise improve the quality of information obtained from parasite genetics ( Wesolowski et al . , 2018 ) . Our methods for computing genetic relatedness , like the genetic data themselves , were relatively simple but provided useful information on relative connectivity between geographic areas . In this study site , classical methods for measuring or visualizing genetic differentiation ( e . g . , Gst , STRUCTURE , or phylogenetic trees ) had limited utility due to the marginal differences in allele frequencies between geographically proximal locations in this relatively compact study site and the inability to utilize all information from polyclonal infections . Currently , there are few established analytical approaches to quantify fine-scale genetic connectivity between locations with predominantly polyclonal infections . Estimation of pairwise genetic relatedness between all infections in this study allowed the incorporation of data from all parasites detected in infections and extraction of useful signals of recent transmission created by recombination and cotransmission of multiple parasites . However , more sophisticated bioinformatics , statistical , and modeling tools that are designed to take advantage of genetic data from polyclonal infections and map these onto quantitative , calibrated estimates of migration rates would transform the utility of genetic data for understanding operationally relevant transmission patterns . The availability of such tools would provide a strong rationale for coordinated collection of regional data on parasite genetics , allowing for more systematic evaluation of malaria transmission and generalized utility . The incorporation of parasite genetics added an important dimension to the understanding of local and cross-border malaria transmission epidemiology and connectivity in this area of the Elimination 8 region of Africa . Our results , showing strong connectivity between malaria parasite populations over hundreds of kilometers within Namibia and across national borders , calls for strengthening the simultaneous coordination of efforts between the Elimination eight countries . Furthermore , our data demonstrate the feasibility and added value of systematically integrating genetic data into national and regional surveillance efforts , particularly when the goal is elimination and movement of malaria parasites may threaten this goal or influence interventions . A combination of human mobility and parasite genetic data is proposed to mitigate limitations of each individual data source in isolation and to provide the most robust intelligence to guide local and regional strategy . Ethical approval for the study was obtained from the Institutional Review Boards of the University of Namibia and the University of California , San Francisco ( Identification numbers 15–17422 and 14–14576 ) . Informed consent was obtained from all participants or the parents of all children participated in the Zambezi study . For the Kavango study , IRB approval was obtained but no informed consent was collected as all samples ( used RDTs ) and de-identified data were collected during routine surveillance . We enrolled 4643 symptomatic Plasmodium falciparum cases from the outpatient clinics of 29 health facilities in two regions of northeastern Namibia: Kavango East and Zambezi . Diagnosis of all cases was confirmed by rapid diagnostic test ( RDT ) . In Kavango East , 3871 symptomatic cases from 23 health facilities were enrolled and used RDTs were collected from March to June 2016 . In the Zambezi region , 772 symptomatic cases from six health facilities were enrolled between February 2015 and June 2016 and dried blood spots ( DBS ) were collected at the time of diagnosis . In both locations , additional patient information such as age , residence , local and international travel history were collected . The health facilities in Kavango East and Zambezi were located within 204 km and 87 km of each other , respectively . During the time of RDT or DBS collection , study participants were asked about their location of residence as well as any overnight travel to non-residence locations . In the Kavango travel survey , individuals were asked if there was any travel to a select number of locations including neighboring towns , other districts/provinces in Namibia and Angola , and other neighboring countries . In the Zambezi travel survey , individuals were able to provide information about travel to any location . These free-response questions were geocoded to the village and regional levels and included both national and international destinations . Individuals were also asked to provide information on the duration of the trip ( in days ) . When analyzing travel survey data , each individual’s time over the prior 30 days was allocated based on their location of residence and the reported time spent away from their residence . To compare with the genetic data , health facility catchments were aggregated to the corresponding travel survey location based on the location of the catchment centroid . Mobile phone call data records were obtained from October 2010 to September 2011 . In total , 1 . 19 million unique individual subscribers were recorded at 197 mobile phone towers in Namibia ( Ruktanonchai et al . , 2016 ) . Of these , 14 towers with a total of 98 , 104 subscribers were located within the study area . Travel patterns between mobile phone tower catchment areas were calculated using previously developed methods ( Ruktanonchai et al . , 2016; Tatem , 2014; Wesolowski et al . , 2012 ) . Briefly , individuals were assigned a primary mobile phone tower based on the most frequently used tower at night . Trips to other mobile phone tower catchments were inferred if their primary daily location was recorded at another tower and was not limited to only night time use . All other time was assumed to be spent at their primary mobile phone tower . Individuals were aggregated to a single primary tower location , and mobility per mobile phone tower catchment was calculated as a distribution of time spent at each one of the other mobile phone tower catchments , including the time spent at the primary tower location . Mobile phone tower catchments and health facility catchments , although covering the same geographic area , did not correspond to a one-to-one match . When comparing mobility data from the mobile phone tower catchments with the genetic data , health facilities were aggregated to tower catchments based on the location of the catchment centroid . DBS and used RDTs were stored with desiccant at −20°C until transportation and processing . DNA was extracted from 6 mm punches of DBS and strips of used RDTs using the Saponin-Chelex method ( Plowe et al . , 1995 ) . For RDTs , the cassettes were opened using a thin metal spatula and DNA was extracted from the nitrocellulose strip in accordance with the worldwide antimalarial resistance network guidelines ( Molecular Module , 2011 ) , with the exception that DNA extraction was performed in deep 96-well plates . For all samples extracted from DBS , parasite density was quantified using var-ATS ultra-sensitive qPCR ( Hofmann et al . , 2015 ) and samples with more than 10 parasites/µL of blood were genotyped . Given the large number of RDT samples collected , a subset was selected for extraction and genotyping as follows . If less than 100 RDTs were collected from a given clinic , all were genotyped . If more than 100 RDTs were collected from a given clinic , any cases with travel history and 100 cases without travel history were genotyped . In addition , all samples were genotyped from one hospital ( Nyangana Hospital ) to validate subsampling . For samples extracted from RDTs , parasite density was quantified on a subset . When positive , parasite density was almost always above the genotyping threshold ( n = 320 , median = 13612 parasites/µL of blood ) . A total of 2990 samples were genotyped using 26 microsatellite markers as described previously ( Liu et al . , under preparation ) . Briefly , two-rounds of PCR protocol were used to amplify the 26 microsatellite loci . The multiplex primary PCR was performed in 4 groups using two different PCR conditions . 1 µL of the amplified product was then used as a template for the individual PCR for each marker . PCR products were then diluted and sized by denaturing capillary electrophoresis on an ABI 3730XL analyzer with GeneScan 400HD ROX size standard ( Thermo Fisher Scientific ) . The resulting electropherograms were analyzed using microSPAT software ( Murphy , 2018 ) to automate identification of true alleles and differentiate real peaks from artifacts . A total of 2585 samples with data in at least 15 or more loci were included in these analyses ( S1 Data ) . Additional data from comparably genotyped microsatellite datasets from northern Angola ( from Cabinda , Bengo , Uige and Zaire provinces ) collected between January and December 2014 ( n = 137 , Liu et al . , under preparation ) and southern Zambia ( from Choma district ) collected between January 2015 and April 2016 ( n = 96 , Pringle et al . , 2018 ) were also analyzed . Genotyping data from all samples were combined and processed with similar software settings to avoid variability in allele calling . The within-host diversity of infections was determined using multiplicity of infection ( MOI ) and the FWS metric . MOI was determined as the second highest number of alleles detected at any of the 26 loci , allowing for the possibility of false positive allele calls . The FWS metric is a measure of the within-host diversity of an individual infection relative to the population level genetic diversity . A high FWS indicates low within-host diversity relative to the population ( e . g . low risk of inbreeding ) . FWS was calculated as described previously ( Roh et al . , 2019; Auburn et al . , 2012 ) . Briefly , FWS was calculated for each infection using the formula , FWS=1-HwHs where Hw= heterozygosity of the individual and Hs= heterozygosity of the local parasite population . Within host heterozygosity was estimated based on the number of alleles detected at each locus . Mean FWS was calculated for each individual by taking the mean across all loci . Population level genetic diversity was estimated using expected heterozygosity ( HE ) and calculated using the formula , HE = nn-11-∑pi2 , where n is the number of genotyped samples and pi is the frequency of the ith allele in the population . Within-host and population level genetic diversity were then compared by health districts , health facilities and between local and imported cases . Imported cases include residents of Angola and Zambia and those individuals with a reported travel history to Angola and Zambia in the last 30 days . Methods for computing genetic relatedness between infections , incorporating data from all alleles detected at a loci , are lacking due to the difficulty of accurate reconstruction of haplotypes from polyclonal infections . Most existing methods either rely only on haplotypes constructed in monoclonal infections or ‘reconstructing’ haplotypes from only the dominant alleles in polyclonal infections ( i . e . not utilizing all the alleles detected in a polyclonal infection ) . In this study , we computed allele sharing between pairs of infections , allowing us to utilize all the detected alleles at a loci in polyclonal infections . For all successfully genotyped samples , pairwise genetic relatedness between infections was calculated using a modified identity by state ( IBS ) metric ( Jacquard et al . , 1974; Pringle et al . , 2018 ) . Briefly , IBS was computed based on the number of shared alleles between pairs of infections , in both mono- and poly-clonal infections . The overall pairwise IBS was calculated as:IBS=1n∑i=1nSiXiYiwhere n is the number of genotyped loci , Si is the total number of shared alleles at locus i between samples X and Y; Xi is the number of alleles in sample X at locus i and Yi is the number of alleles in sample Y at locus i . Within the Namibia dataset , a total of 3 , 365 , 700 pairs of infections from 29 health facilities were analyzed . To investigate geographic clustering , individuals were aggregated to health districts . Population structure was inferred to determine whether haplotypes ( estimated from dominant alleles ) clustered into distinct genetic populations ( K ) using the software MavericK ( Verity and Nichols , 2016 ) . Clustering was further evaluated by a neighbor-joining phylogenetic tree computed using the ‘ape’ package ( Paradis et al . , 2004 ) and PCA analysis using the pairwise genetic distances ( 1-IBS ) determined above . Jost’s D ( Jost , 2008 ) and GST ( Nei and Chesser , 1983 ) were used to estimate genetic differentiation between pairwise comparisons of clinics . Briefly , Jost’s D and GST were calculated using the formulas: D=HT-HS1-HSnn-1 and GST=HT-HS/HT , respectively , where HT and HS are the overall and the sub-population heterozygosity , respectively and n is the number of sampled populations . The values of Jost’s D and GST range from 0 ( no genetic differentiation between populations ) to 1 ( complete differentiation between populations ) . To investigate connectivity at different spatial scales , highly related infection pairs were identified . In order to determine pairs of infections which were more related than expected by chance , we used genotyping data with a similar MOI distribution from countries in West , Central and East Africa ( n = 432 , data from Liu et al . , under preparation ) . These countries are not geographically connected to the study area , thus there is likely limited direct parasite connectivity . The distribution of pairwise genetic relatedness between these and Namibia samples was estimated and used as the expected distribution of relatedness in the absence of a direct transmission link and/or a recent importation event ( i . e . , a null distribution ) . For each pair of locations and the null distribution , IBS values were binned into 20 bins . For each bin , the difference in the proportion of observed and expected pairs under the null distribution was computed . The last bin at which the observed proportion was greater than the null distribution , starting from 1 to 0 , was used as a cut-off to determine the proportion of highly related infections . To investigate spatial connectivity between locations , the proportions of highly related infections above the cut-off were compared . The median of the cut-off was an IBS of 0 . 55 . The overall proportion of highly related infections was calculated as the sum of the proportions of observed pairs above the estimated cut-off minus the proportion in the null distribution . The statistical significance of connectivity was determined by bootstrapping over the IBS values 1000 times and correcting for multiple comparisons using a Bonferroni correction , generating a 95% confidence interval for each pair of locations . In total , there were eight travel survey destinations and 14 mobile phone towers that overlapped with the health facilities of the study area . We scaled the time spent , estimated from either data set , based on the ratio of incidence in the destination versus the corresponding health facility ( Figure 4—source data 1 ) . When multiple health facilities fell within a single travel survey destination or mobile phone catchment , the average incidence was used . These data were used to estimate the proportion of time at risk and possible source locations of importations for each health facility ( measure of parasite mixing ) . Clusters were determined using a hierarchical modularity maximization algorithm ( Newman , 2006 ) from either the incidence scaled travel between travel survey destinations , mobile phone tower catchments , or the proportion of highly related samples from the genetic data . We clustered the genetic data from Kavango and Zambezi separately in order to identify sub-regional structure in Kavango . For both the travel or parasite mixing data , Kavango and Zambezi were able to cluster together . We then compared the cluster agreement estimated from mixing calculated from the proportion of time spent at risk ( mobile phone data , travel survey ) or the genetic data using a Rand Index ( Rand , 1971 ) which is a measure of similarity between two data clusters . For a set of n locations ( L ) and two clusters ( X , Y ) of L , the Rand index is calculated as:R=a+bn2 Where a is the number of pairs of elements in L that are in the same subset in X and Y and b is the number of pairs of elements in L that are in different subsets in X and Y . To estimate cross-border importation , all genotyped infections with a residence in northeastern Namibia and with no reported international travel were aggregated to the respective district . Individuals who reported international travel or with an international residence were assigned to the destination of the reported travel or the residence location . The majority of these individuals reported either a residence in or travel to locations in the nearest bordering province of southern Angola ( n = 219 genotyped cases ) and Western Province of Zambia ( n = 9 , genotyped cases ) . In addition , previously genotyped infections from different provinces of northern Angola ( n = 137 ) and Southern Province of Zambia ( n = 96 ) were included in the analyses . Cross-border analyses did not include any mobile phone data since available data were limited to cell towers within Namibia . Pairwise proportions of highly related infections were compared between four health districts in Namibia ( Rundu , Nyangana , Andara and Zambezi ) ; four locations in Angola ( Northern Angola , Calai and Dirico municipalities , and elsewhere in southern Angola ) and two locations in Zambia ( Western and Southern Provinces ) . The relative importation estimate between pair of locations was calculated as:Importationestimatesfromtravelhistorydata:IAB=TABiAiBandIBA=TBAiBiAImportationestimatesfromparasitegeneticdata:IAB=GiAiBandIBA=GiBiA Where IABandIBA are importation estimates based on genetic data from location A to location B and vice versa; TABandTBA is the proportion of time at risk for those individuals who reported travel from location A to B and vice versa; G is the proportion of highly related infections between location A and B and iAandiB represent malaria incidence from local health system data at locations A and B , respectively .	The number of malaria cases has dropped in some Southern Africa countries , but others still remain seriously affected . When people travel within and between countries , they can bring the parasites that cause the disease to different areas . This can fuel local transmission or even lead to outbreaks in a malaria-free area . When new malaria patients are diagnosed , they are often asked to report their recent travel history , so that the origin of their infection can be tracked . In theory , this would help to spot regions where the disease is imported from , and design targeted interventions . However , it is difficult to know exactly where the parasites come from based on self-disclosed travel history . At best , this history can provide information about that persons infection but nothing further in the past; at worst this history can be completely incorrect . Parasite DNA , on the other hand , has the potential to bring with it an indelible record of the past . To address the problem of determining where malaria infections came from , Tessema , Wesolowski et al . focused on Northern Namibia , a region where malaria persists despite being practically absent from the rest of the country . Patients movements were assessed using mobile phone call records as well as self-reported travel history In addition , samples a single drop of blood were taken so that the genetic information of the parasites could be examined . Combining genetic data with travel history and phone records , Tessema , Wesolowski et al . found that , in Northern Namibia , most people had gotten infected by malaria locally . However , the genetic analyses also revealed that certain infections came from places across the Angolan and Zambian borders , information that was much more difficult to obtain using self-report or mobile phone data . A new , separate study by Chang et al . also supports these results , showing that , in Bangladesh , combining genetic data with travel history and mobile phone records helps to track how malaria spreads . Overall , the work by Tessema , Wesolowski et al . indicate that , in Northern Namibia , it will be necessary to strengthen local interventions to eliminate malaria . However , different countries in the region may also need to coordinate to decrease malaria nearby and reduce the number of cases coming into the country . While genetic data can help to monitor how new malaria cases are imported , this knowledge will be most valuable if it is routinely collected across countries . New tools will also be required to translate genetic data into information that can easily be used for control and elimination programs .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "epidemiology", "and", "global", "health" ]	2019	Using parasite genetic and human mobility data to infer local and cross-border malaria connectivity in Southern Africa
From humans to vinegar flies , exposure to diets rich in sugar and fat lowers taste sensation , changes food choices , and promotes feeding . However , how these peripheral alterations influence eating is unknown . Here we used the genetically tractable organism D . melanogaster to define the neural mechanisms through which this occurs . We characterized a population of protocerebral anterior medial dopaminergic neurons ( PAM DANs ) that innervates the β’2 compartment of the mushroom body and responds to sweet taste . In animals fed a high sugar diet , the response of PAM-β’2 to sweet stimuli was reduced and delayed , and sensitive to the strength of the signal transmission out of the sensory neurons . We found that PAM-β’2 DANs activity controls feeding rate and satiation: closed-loop optogenetic activation of β’2 DANs restored normal eating in animals fed high sucrose . These data argue that diet-dependent alterations in taste weaken satiation by impairing the central processing of sensory signals . Consumption of diets high in sugar and fat decreases the perception of taste stimuli , influencing food preference and promoting food intake ( Bartoshuk et al . , 2006; Sartor et al . , 2011; Ahart et al . , 2019; May et al . , 2019; Weiss et al . , 2019; Kaufman et al . , 2018 ) . Recent studies have examined the effects of these diets on the sensitivity of the peripheral taste system and the intensity of taste experience ( May et al . , 2019; Maliphol et al . , 2013; Kaufman et al . , 2018; Weiss et al . , 2019 ) , but how exactly taste deficits increase feeding behavior is not known . Orosensory signals determine the palatability or ‘liking’ for foods ( Berridge and Kringelbach , 2015 ) , but they also promote meal termination via a process called ‘sensory-enhanced ( or mediated ) satiety’ ( Chambers et al . , 2015 ) . Indeed , foods that provide longer and more intense sensory exposure are more satiating , reducing hunger and subsequent test-meal intake in humans ( Yeomans , 2017; Bolhuis et al . , 2011; Ramaekers et al . , 2014; Viskaal-van Dongen et al . , 2011; Cecil et al . , 1998; Forde et al . , 2013 ) . Specifically , sensory signals are thought to function early in the satiety cascade ( Blundell et al . , 1987 ) by promoting satiation and bringing the on-going eating episode to an end ( Blundell et al . , 2010; Bellisle and Blundell , 2013 ) . This is in contrast to nutrient-derived signals , which develop more slowly and consolidate satiety by inhibiting further eating after the end of a meal ( Blundell et al . , 2010; Bellisle and Blundell , 2013 ) . We reasoned that if orosensory attributes like taste intensity are important to curtail a feeding event , then diet-dependent changes in taste sensation could promote feeding by impairing sensory-enhanced satiation . Here we investigated the relationship between diet composition – specifically high dietary sugar – the central processing of sweet taste signals , and satiation by exploiting the simple taste system and the conserved neurochemistry of the fruit fly D . melanogaster . Like humans and rodents , D . melanogaster flies exposed to palatable diets rich in sugar or fat overconsume , gain weight , and become at-risk for obesity and develop phenotypes associated with metabolic syndrome ( Musselman and Kühnlein , 2018 ) . We recently showed that , in addition to promoting feeding by increasing meal size , consumption of high dietary sugar decreased the electrophysiological and calcium responses of the Gr64f+ sweet sensing neurons to sweet stimuli , independently of weight gain ( May et al . , 2019 ) . These physiological changes in the Gr64f+ cells reduced the fruit flies’ taste sensitivity and response intensity . Opto- and neurogenetics manipulations to correct the responses of the Gr64f+ neurons to sugar prevented animals exposed to high dietary sugar from overfeeding and restored normal meal size ( May et al . , 2019 ) . Thus , the diet-dependent dulling in sweet taste causes higher feeding in flies , but how does this happen ? How do alterations in the peripheral sensory neurons modulate a behavior as complex as feeding ? To better understand how this occurs , we decided to examine the effects of high dietary sugar and taste changes in the central processing of sweet stimuli by dopaminergic neurons ( DANs ) . Indeed , while the neural pathways that bring sensory information from the periphery to higher order brain regions are unique across organisms , dopaminergic circuits process sweet taste information in humans , rodents , and fruit flies . Interestingly , the reinforcing effects of sugar taste and nutrient properties are relayed via distinct dopaminergic pathways in these organisms ( Yamagata et al . , 2015; Huetteroth et al . , 2015; Tellez et al . , 2016; Thanarajah et al . , 2019 ) . In flies , DANs in the Protocerebral Anterior Medial ( PAM ) cluster respond to the sweet sensory properties to signal sugar reward ( Burke et al . , 2012; Liu et al . , 2012 ) , reinforce short term appetitive memories ( Yamagata et al . , 2015; Huetteroth et al . , 2015 ) , and promote foraging and intake ( Tsao et al . , 2018; Musso et al . , 2019 ) . We hypothesized that diet-dependent impairments in the peripheral responses to sugar could influence the way sweet taste information is transduced through PAM-DANs to affect feeding behavior and obesity risk . Here we show that in flies fed a high sugar diet the presynaptic responses of a specific subset of PAM DANs to sweet taste are decreased and delayed . These changes are specific to sweet stimuli and mediated by high dietary sugar . Further , we report that the reduction in the central processing of sweet taste information increases the duration and size of meals: closed-loop optogenetic stimulation of a specific set of PAM DANs corrected meal size , duration , and feeding rate . Together , our results argue that diet-dependent alterations in the central processing of sweet sensory responses delay meal termination by impairing the process of sensory-enhanced satiation . We previously showed that the calcium responses of the sweet sensory neurons to sucrose were decreased in animals fed high dietary sugar ( May et al . , 2019; Vaziri et al . , 2020 ) . To ask if the transmission of the sweet taste signal out of these neurons was also lower , we expressed the genetically encoded vesicular release sensor synaptobrevin-pHluorin ( syb-pHluorin ) ( Poskanzer et al . , 2003 ) in the sweet taste neurons using the Gustatory Receptor 64f ( Gr64f ) GAL4 driver and measured the in vivo fluorescence from the presynaptic terminals in the Sub Esophageal Zone ( SEZ ) in response to stimulation of the proboscis with 30% sucrose ( Figure 1A ) . We found that the syb-pHluorin fluorescent changes upon sugar presentation were markedly decreased when flies were fed a high sugar diet ( SD , 30% sucrose ) for 7 days , compared to age-matched flies fed a control diet ( CD , ~8% sucrose ) ( Figure 1 ) . These data suggest that both the responses of the sweet sensing Gr64f+ neurons to sugar and the transmission of the sweet taste signal are decreased by exposure to the SD . Since the involvement of DANs in feeding behavior and in central processing of sensory information is a homologous feature across organisms ( Yamagata et al . , 2015; Huetteroth et al . , 2015; Tellez et al . , 2016; Thanarajah et al . , 2019 ) , we decided to center on this DAN circuit as a possible link between diet-dependent changes in sweet responses , higher feeding , and weight gain . In flies , DANs in the Protocerebral Anterior Medial ( PAM ) cluster that are labeled by the R48B04-GAL4 transgene and innervate the β’2 and γ4 compartments of the Mushroom Body ( MB ) , respond to sweet sensory properties ( Huetteroth et al . , 2015; Yamagata et al . , 2015 ) ; neurons of this population also centrally reinforce water taste ( Lin et al . , 2014 ) . Here we focused on the β’2 compartment because of its role in processing of the taste properties alone , compared to γ4 which is modulated by both taste and additional factors , such as internal state ( Lin et al . , 2014; Yamagata et al . , 2015 ) . In addition to labeling ~60 DANs in each PAM cluster , R48B04 is expressed in other neurons , including the Pars Intercelebralis . To avoid or minimize potential confounding effects of its expression in other compartments , we used FlyLight to visually identify GAL4 lines that label subsets of PAM-β’2 , but have limited expression in other compartments ( Aso and Rubin , 2016 ) . We selected the split-GAL4 line MB301B which had been implicated in foraging and feeding ( Tsao et al . , 2018; Musso et al . , 2019 ) and which labels ~12 TH+ PAM-β2β'2a and only shows a few , sparse projections in the ventral nerve cord and SEZ ( Figure 2A and Figure 2—figure supplement 1A ) . We used the presynaptically targeted GCaMP6s::Bruchpilot::mCherry ( Kiragasi et al . , 2017 ) to record the response of MB301B neurons to stimulation of the labellum with 30% sucrose . We observed an increase in signal in the β’2compartment ( rose ) , showing that these PAM-β’2 neurons process sweet sensory information ( Figure 2B , grey lines; Figure 2—figure supplement 1B ) . Next we measured the responses of MB301B neurons to sucrose taste in flies fed a SD for 7 days and we found a nearly 50% decrease ( Figure 2B , rose lines; Figure 2—figure supplement 1B ) . Furthermore , when we looked at both the average and individual traces , we saw a ~ 600 millisecond delay in the peak responses to the sucrose stimulus delivery to the labellum ( Figure 2C ) . A decrease in calcium responses also occurred when the proboscis was stimulated with a lower concentration of sucrose ( 5% ) , but we did not find a delayed response , suggesting that the changes in the timing of the processing may be unique to higher sugar concentrations ( Figure 2—figure supplement 1C and D ) . No sugar taste responses were recorded in β2 ( green ) , consistent with the idea that it is not involved in taste processing ( Figure 2D; Yamagata et al . , 2015 ) . Thus , the central processing of sweet stimuli in PAM-β’2 MB301B neurons is both decreased and delayed by exposure to a high sugar diet . The reduction and delay in central responses to sugar taste in PAM-β’2 DANs on a SD could be due either to the lower transmission of the sensory signal out of the peripheral sweet taste neurons ( Figure 1 ) or to the metabolic side effects of the high nutrient diet . To differentiate between these possibilities , we took multiple approaches . In addition to sweet stimuli , PAM-β’2 neurons also respond to water ( Lin et al . , 2014 ) ; we reasoned that if high dietary sugar unspecifically changed the activity of the PAM-β’2 , we would expect flies on the SD to also exhibit impaired central responses to water . However , the magnitude and timing of the β’2 response to water stimulation of the labellum were unchanged between flies on a CD or SD ( Figure 3A , B , and Figure 3—figure supplement 1A; water stimulation was delivered in the same flies as in Figure 2 ) . Thus , the decrease in PAM-β’2 responses in flies fed a SD is specific to the sweet sensory stimulus . This argues that the overall ability of these DANs to respond to stimuli is not generally affected , and the reduction observed on a SD could occur because of the diet-dependent changes in the sweet taste neurons in the periphery ( May et al . , 2019 and Figure 1 ) . To further probe this question , we fed flies a high fat diet ( FD ) , which has the same caloric content of the high sugar diet ( SD ) and promotes fat accumulation , but does not decrease the responses of the Gr64f+ sensory neurons to sugar stimuli ( May et al . , 2019 ) . If changes in PAM-β’2 responses to sugar taste occur because of the metabolic side-effects of high nutrient density ( i . e , fat accumulation ) – rather than via changes in the sweet sensory neurons’ output – we would expect a FD to also induce PAM-β’2 dysfunction . However , a FD diet had no effect on the PAM-β’2 responses to sucrose or water stimulation of the labellum in MB301B > GCaMP6s::Bruchpilot::mCherry flies ( Figure 3C , D and Figure 3—figure supplement 1B , C ) . Together , these two lines of evidence argue that the dysfunction in the processing of sweet taste stimuli in the PAM-β’2 neurons of flies on a SD is linked to alterations in the peripheral sensory processing of sugar taste caused by high dietary sugar . To test this hypothesis more directly , we examined the effect of correcting sweet taste sensation on the responses of the PAM-β’2 MB301B neurons to sugar . To rescue the sweet taste deficits caused by a high sugar diet we fed flies an inhibitor of the metabolic-signalling enzyme O-GlcNAc-Transferase ( OGT ) , which we previously found to be responsible for decreasing sweet taste on a SD ( May et al . , 2019 ) . In accordance with our previous findings on OGT ( May et al . , 2019 ) , supplementing the flies’ diet with 75 µM of OSMI-1 ( OGT-small molecule inhibitor 1 ) resulted in no changes in PER between a CD and SD ( Figure 3—figure supplement 1D ) . In these flies , the calcium responses of PAM-β’2 neurons to sucrose stimulation of the labellum were identical in SD+OSMI and CD+OSMI flies . Although we cannot exclude the possibility that the OGT inhibitor also acted elsewhere outside the sensory neurons , our data support the idea that deficits in the peripheral responses drive impairments in the central processing of sweetness ( Figure 3E , F ) . Together , these orthogonal lines of evidence suggest that the impairments in the central processing of sweet sensory information by DANs are mediated by deficits in peripheral sweet taste responses . We previously showed that a diet-dependent dulling of sweet taste drives higher feeding behavior and weight gain by increasing the size and duration of meals ( May et al . , 2019 ) . Since sweet taste deficits underlie the changes in PAM-β’2 activity , we reasoned that impairments in the central processing of orosensory signals may also play a role in promoting higher feeding in animals fed a high sugar diet . Specifically , if PAM-β’2 neurons were critical for integrating sweet taste information into feeding decisions , then normalizing their activity may also prevent increased eating and weight gain when flies are exposed to a SD . To test this possibility we expressed the light-activated cation channel ReaChR ( Inagaki et al . , 2014 ) in the MB301B neurons , and used the optoFLIC , a feeding frequency assay ( Ro et al . , 2014 ) modified for closed-loop optogenetic stimulation ( May et al . , 2019 ) , to normalize the change in activity of PAM-β’2 neurons only when the flies were interacting with the food starting at day 3 . MB301B > ReaChR flies that did not receive retinal supplementation ( ATR , all-trans-retinal is required to form a functional light-sensitive opsin ) exhibited the characteristic increase in feeding behavior on 20% sucrose ( Figure 4A , rose line ) ; however , MB301B > ReaChR +ATR animals , which were activated by light , had stable feeding for 10 days ( Figure 4A , peach line ) . Control animals on 20% sucrose had more feeding interactions per meal and longer meal duration with more days on the SD ( Figure 4B and C , rose lines ) , consistent with our previous data ( May et al . , 2019 ) . In particular , we found that a SD induced a lengthening of the peak-to-end of the meal by ~4 hr , suggesting that the satiation process is delayed in these animals ( Figure 4D , rose line ) . However , feeding-paired stimulation of PAM-β’2 neurons stabilized the size and duration of the meal , as well as the time to satiation , over the entire duration of the experiment ( Figure 4B , C and D , peach lines ) . Interestingly , stimulation of the Gr64f+ sweet taste neurons also corrected these two aspects of meal structure ( May et al . , 2019 ) . Importantly , flies in which these PAM-β’2 DANs were activated still developed taste deficits on a SD ( Figure 4—figure supplement 1A ) , arguing against the possibility that PAM-β’2 DANs stimulation prevents increased feeding by rescuing the taste changes in the Gr64f+ neurons . Instead , our data suggest that PAM-β’2 DANs modulate meal structure and feeding behavior by integrating the sensory signal from the periphery . Interestingly , identical stimulation of PAM-β’2 DANs in flies fed 5% sucrose resulted in higher feeding ( Figure 4—figure supplement 1B ) , as previously showed with both sucrose food and water ( Musso et al . , 2019 ) ; this indicates the the context of animal’s diet and the basal activity state of PAM-β’2 DANs are important to control eating . In accordance with the stable feeding patterns recorded on the optoFLIC in animals fed a SD , we found that activation of PAM-β’2 DANs also prevented diet-induced obesity in animals fed high dietary sugar ( Figure 4—figure supplement 1C ) . Interestingly , PAM-β’2 DANs labeled by MB301B seem to play a unique role in this process . Activation of different subpopulations of PAM-β’2 with eight distinct GAL4 transgenes ( Aso and Rubin , 2016; Aso et al . , 2014b ) ( MB056B , MB109B , MB042B , MB032B , MB312B , MB196B , MB316B , some of these also express in γ4 ) failed to rescue diet-induced obesity ( Figure 4—figure supplements 1D and 2A ) . Comparison of the anatomy of these lines showed that MB301B projects more anteriorly and ventrally than these other lines ( Figure 4—figure supplement 2A ) . However , the ventral expression mostly regards the β2 compartment , which it is shared with only one of the other lines ( MB032B ) ; meanwhile the MB301B β’2 expression , while more anterior than other lines , does overlap with some ( MB032B , MB196B , MB042B ) . ( Figure 4—figure supplement 2A ) . Further , flies with activation of nutrient-responsive PAM DANs ( Yamagata et al . , 2015; Huetteroth et al . , 2015 ) , which express in β2 , still accumulated fat as controls when fed high dietary sugar , suggesting that effects of MB301B neuron activation come from the sweet-responsive β’2 compartment ( Figure 4—figure supplement 1E ) . Since the FLIC records feeding interactions every 200 milliseconds ( Ro et al . , 2014 ) , we used this information to look at how feeding rate changed during a meal , as this has been linked to the process of satiation . To do this , we first calculated the number of feeding events per meal , where a feeding event is defined as a succession of consecutive feeding interactions above an established signal threshold , ( see Materials and methods , and Ro et al . , 2014 ) . We next divided the number of events per meal by the duration of each meal per day to obtain a feeding rate and to control for the fact that meals last longer on a SD . We found that both the feeding events per meal and the feeding rate increased with chronic exposures to high dietary sugar ( Figure 5A and B ) . However , optogenetic stimulation of PAM-β’2 prevented these increases and maintained a stable number of events and a constant feeding rate per meal over the duration of the experiment . We next examined whether the feeding rate changed during the course of the meal , by calculating it before and after the peak of meal feeding ( Figure 5C , diagram ) . The feeding rate past the peak of the meal increased with time in animals fed 20% sucrose , but stayed the same in flies with activation of PAM-β’2 neurons ( Figure 5C ) . Interestingly , the pre-peak eating also increased gradually with exposure to high dietary sugar ( Figure 5D ) . Together , these data suggest that diet-dependent impairments in PAM-β’2 neurons promote overfeeding by impairing satiation , and specifically by affecting the feeding rate during a meal . Since , PAM-β’2 neurons process sensory experiences from the periphery , our experiments argue that this phenomenon is connected to sensory-enhanced satiation . Together we propose that the central processing of sensory experiences during a meal by PAM-β’2 DANs , controls feeding rate and sensory-enhanced satiation . This process is altered by high dietary sugar , leading to an attenuated satiation process and higher feeding ( Figure 5E ) . In this study we found that diet-dependent changes in sensory perception promote feeding and weight gain by impairing the central dopaminergic processing of sweet taste information . When animals consume a high sugar diet , the responses to sweet taste of a distinct population of PAM DANs innervating the β’2 compartment of the MB are decreased and delayed . These alterations in dopaminergic processing increase the eating rate and extend the duration of meals , leading to attenuated satiation , higher feeding , and weight gain ( Figure 5E ) . Interestingly , we observed a reduction in PAM DAN responses only when flies ate diets that resulted in sweet taste deficits; consumption of an equal calorically-rich lard diet that did not impact taste had no effect on the PAM DANs responses . Similarly , animals fed high dietary sugar exhibited differences in PAM-β’2 responses to sweet , but not water taste stimuli , reinforcing the idea that PAM DAN alterations occur because of lower signal transmission from the sensory neurons ( Figure 5E ) . Indeed , correcting sweet taste deficits by feeding fruit flies an inhibitor of the enzyme O-GlcNAc Transferase – which we previously found to be required for taste impairments– prevented impairments in PAM-β’2 responses , although we cannot exclude that this could also be due to its effect beyond the sensory neurons ( but not in the MB301B neurons , Figure 3—figure supplement 2A ) . Here , we propose a model where diet-dependent changes in taste intensity and sensitivity reduce the central processing of sensory stimuli to cause weaker and attenuated satiation . Interestingly , our anatomical analysis of several PAM-β’2 DANs lines showed that the neurons labeled by MB301B minimally overlap with other split-GAL4 lines purported to express in β’2 ( A degree of overlap is not unexpected and does not necessarily imply that the neurons accessed by these driver lines perform identical functions . ) MB301B is largely distinct from these other lines both ventrally and anteriorly . Ventral expression of MB301B enters the β2 compartment , which has been implicated in nutrient reward; however , this compartment is not responsive to sweetness and its activation still resulted in diet-induced obesity . Thus , MB301B expression in anterior β’2 could represent an unique MB compartment that is part of a circuit dedicated to sweet taste processing and feeding behavior . A weakness of the current study is that we were unable to follow the transmission of the taste signal from the primary sensory neurons through the different circuits that eventually communicate with PAM . Studies that will identify taste projection neurons genetically will allow us to further probe this point in the future . Further , while the split-GAL4 line used in this study expresses in PAM-β2β’2 neurons in the central brain , it also labels a few projections in the ventral nerve cord/SEZ , which could also contribute to some of the effects measured here . Interestingly , two studies previously found that neurogenetic or closed-loop optogenetic activation of MB301B PAM-β2β’2 neurons resulted in an increase in foraging behavior ( Tsao et al . , 2018 ) and higher feeding to both water and sucrose substrates ( Musso et al . , 2019 ) , respectively . We also measured an increase in feeding behavior with closed-loop optogenetic activation of MB301B neurons in flies fed a control diet ( 5% sucrose ) , confirming these observations . However , applying the same optogenetic protocol when animals consumed a high sugar diet resulted in lower eating and a protection from diet-induced obesity . Since these neurons have lower activity in flies on a high sugar diet , we propose that the optogenetic stimulation in this context functions as a normalization of the activity , rather than activation in the absence of the stimulus . This suggests that variations in relative PAM DANs activity , rather than their absolute output , may modulate feeding behavior in flies exposed to high dietary sugar . It will also be interesting to ask how the activity of these neurons relates to other aspects of feeding behavior , such as the acceptance of low quality or bitter foods . Studies in rodents and humans have delineated the importance of sensory signals to modulate satiation and terminate meals . This process , termed sensory-enhanced satiation ( Chambers et al . , 2015 ) , plays an early role in the satiety cascade before post-oral nutrient-derived signals consolidate satiety ( Bellisle and Blundell , 2013; Blundell et al . , 1987 ) . Data show that higher sensory intensity and oral exposure promote stronger satiation ( Bolhuis et al . , 2011; Ramaekers et al . , 2014 ) . For example , high sensory characteristics , such as saltiness and sweetness , enhanced the satiating effect of both low and high energy test drinks ( Yeomans and Chambers , 2011; Yeomans et al . , 2014 ) , decreased consumption of pasta sauce ( Yeomans , 1998; Yeomans , 1996 ) , yoghurt ( Vickers et al . , 2001 ) and tea ( Vickers and Holton , 1998 ) . However , the neural basis for this phenomenon is unknown . Here we characterized the circuit-based mechanisms of sensory-enhanced satiation by exploiting the simplicity of the fruit fly system . We show that sensory-enhanced satiation involves the central dopaminergic processing of peripheral sweet taste stimuli by a dedicated group of PAM-β’2 neurons . Given the role of PAM DANs transmission in reinforcing appetitive memories ( Burke et al . , 2012; Liu et al . , 2012 ) , this discovery is significant because it suggests that satiation may involve a learning or rewarding component and that diet composition may direct food intake by influencing this aspect . Indeed , sensory cues function as a predictor of nutrient density and set expectations for how filling different types of foods should be ( Chambers et al . , 2015; McCrickerd and Forde , 2016; Yeomans , 2017 ) . This information could be used to modulate the feeding rate during the meal and initiate the process of meal termination without relying uniquely on nutrient-derived cues , which arrive later ( Bellisle and Blundell , 2013; Blundell et al . , 1987 ) . The idea that sensory cues could set cognitive expectations about the fullness of future meals is also in line with the known roles of DA in promoting the formation of appetitive memories . In fruit flies , PAM DANs promote the formation of short-term associative memories based on taste and long-term associative memories based on nutrient density by modulating plasticity of the postsynaptic Mushroom Body Output Neurons ( MBONs ) ( Cohn et al . , 2015; Owald et al . , 2015 ) . MBONs are , in turn , connected to pre-motor areas like the Central Complex ( Aso et al . , 2014a ) – the fly genetic and functional analog of the basal ganglia ( Strausfeld and Hirth , 2013 ) – providing an anatomical route to modulate aspects of feeding such as proboscis extension ( Chia and Scott , 2019 ) , the analogue of licking or chewing rate . Interestingly , some MBONs receive input from both the taste ( β’2 ) and nutrient ( γ5 ) compartments , raising the possibility that sensory and nutrient memories may be integrated in the same cells to regulate different aspects of the satiety cascade ( satiation vs . satiety ) . In flies , the mode and timing of DA delivery onto the MBONs is critical to establish the strength and valence of the associations ( Handler et al . , 2019 ) . The delay and decrease we measured in animals on a high sugar diet could impair MBON synaptic plasticity and the formation of new appetitive memories ( Cohn et al . , 2015; Owald et al . , 2015 ) . If this is the case , we would expect that flies on this diet may be insensitive to new learning , use old food memories to predict the filling effects of the meal , and thus overshoot their food intake . This is consistent with the idea elegantly espoused by Kroemer and Small , 2016 who explain the decrease in DA transmission with diet or obesity in a reinforcement learning framework . A different possibility , however , is that alterations in PAM DAN processing are not related to reinforcement learning per se , but instead to a decrease in overall reward receipt . In this light , sensory signals would cue reward not learning , and the pleasure experienced during eating would promote satiation and curb food intake . The idea that decreases in the sensitivity of the reward system increases food intake has been described as the ‘reward deficit’ theory of obesity ( Wang et al . , 2002 ) , which also draws a parallel between the effects of drugs of abuse and that of sugar on the brain . Our results are consistent with both reinforcement learning and reward deficit scenarios , as well as with other integrated theories of obesity ( Stice and Yokum , 2016 ) ; future experiments examining the role of circuits downstream of PAMs , and especially the involvement of MBONs , will differentiate between these possibilities . In addition to contributing to the current body of evidence connecting diet with DA alterations in mammals ( DiFeliceantonio and Small , 2019; Kroemer and Small , 2016; Geiger et al . , 2009; Friend et al . , 2017; van de Giessen et al . , 2013 ) , our results also show that at least some of these alterations are due to diet , and not obesity . In particular , we speculate that some of the changes in DA transmission observed with diet exposure in rodents and humans may be due to impairments in sensory processing , since humans and rodents also process the taste and nutritive properties of sugar separately ( Tellez et al . , 2016; Thanarajah et al . , 2019 ) . It will be particularly interesting to test whether mimicking the effects of a high sugar diet on these DANs using optogenetics will promote feeding behavior . In conclusion , our experiments demonstrate that by reducing peripheral taste sensation , a high sugar diet impairs the central DA processing of sensory signals and weakens satiation . These studies forge a causal link between sugar – a key component of processed foods – taste sensation , and weakened satiation , consistent with the fact that humans consume more calories when their diets consist of processed foods ( Hall et al . , 2019 ) . Given the importance of sensory changes in initiating this cascade of circuit dysfunction , understanding how diet composition mechanistically affects taste is imperative to understand how the food environment directs feeding behavior and metabolic disease . All flies were maintained at 25°C in a humidity-controlled incubator with a 12:12 hr light/dark cycle . For all experiments , males were collected under CO2 anesthesia , 2–4 days following eclosion , and housed in groups of 20–30 within culture vials . The GAL4/UAS system was used for cell-type specific expression of transgenes . Stocks used are listed in the Key Resources Table . As control we used w1118Canton-S flies ( gift from Anne Simon , University of Western Ontario ) , which were obtained by backcrossing a w1118 strain ( Benzer lab , Caltech ) to Canton-S ( Benzer lab , Caltech ) for 10 generations . Flies were transferred to each diet 2–4 days after eclosion in groups of 30 animals per vial and fed on experimental diets ( SD or FD ) for 7 days with age-matched controls on CD . The composition and caloric amount of each diet was as below: Adult age-matched male flies , following 7 days of CD or SD , were fasted on a wet Kimwipe for 18–24 hr before prepping for in vivo confocal laser imaging . As previously described ( May et al . , 2019; LeDue et al . , 2015 ) , the preparation consisted of a fly affixed to a 3D-printed slide with melted wax around the head and on the dorsal part of the thorax . Distal tarsal segments were removed to prevent interference of the proboscis stimulus , and the proboscis was wax-fixed fully extended with the labellum functional and clear of wax so that proboscis contraction and extension could not perturb the brain’s position . A glass coverslip was placed such that artificial hemolymph ( 108 mM NaCl , 8 . 2 mM MgCl2 , 4 mM NaHCO3 , 1 mM NaH2PO4 , 2 mM CaCl2 , 5 mM KCl , 5 mM HEPES ) placed over the head did not touch the proboscis . Data were acquired with a FV1200 Olympus confocal microscope , a 20x water immersion objective , and a rate of 0 . 254 s per frame . Stimuli consisted of a brief touch of a small Kimwipe soaked in milliQ water or 30% sucrose solution to the labellum . Responses to both sucrose and water were measured in the same fly . For calcium imaging experiments , n counts each ROI , of which there are two per fly . OptoFLIC was run as previously described ( May et al . , 2019 ) . Briefly , adult flies 3–5 days past eclosion were placed on ATR food and kept in the dark for 3 days until starting the optoFLIC . optoFLIC experiments were run in an incubator with consistent 25°C and 30–40% humidity , on a dark/dark light cycle to prevent ambient-light activation of the ReaChR . Following two days recording of feeding activity on the FLIC food without LED activation , a protocol for closed-loop feeding-triggered LED activation was begun . The LED activation protocols were as follows: For experiments with MB301B > ReaChR , 200 ms of red ( ~627 nm ) light pulsing at frequency 60 Hz and with a pulse width of 4 ms was triggered by every food interaction signal over 10 . n = 1 is a single animal . Immunofluorescence protocol was performed as described in Dus et al . , 2015 . Briefly , brains were dissected in 1xPBS from male MB301B > RFP flies 3–5 days post-eclosion , then fixed in 4% paraformaldehyde in 1xPBS for 20 min , blocked in blocking buffer ( 10% normal goat serum , 2% Triton X-100 in 1xPBS ) , and incubated overnight at RT in anti-TH ( rabbit polyclonal Ab from Novus Bio ) 1:250 in dilution buffer ( 1% normal goat serum , 0 . 25 Triton X-100 in 1xPBS ) . Secondary antibody was goat anti-rabbit Alexa Fluor 488 diluted 1:500 in dilution buffer , and brains were washed then incubated with secondary antibody overnight at RT . Brains were mounted in FocusClear between two coverslips and imaged within 24 hr . Following the protocol in Tennessen et al . , 2014 , we assayed total TAG levels normalized to total protein in whole male flies . To assay , flies were CO2-anesthetized and flash frozen . Pairs of flies were homogenized in lysis buffer ( 140 mM NaCl , 50 mM Tris-HCl pH 7 . 4 , 0 . 1% Triton-X ) containing protease inhibitor ( Thermo Scientific ) . Separation by centrifugation produced a supernatant containing total protein and TAGs . Protein reagent ( Thermo Scientific Pierce BCA Protein assay ) was added to the supernatant and the standards and incubated for 30 min at 37°C , then tested for absorbance at 562 nm on a Tecan Plate Reader Infinite 200 . TAG reagent ( Stanbio Triglycerides LiquiColor Test ) was added to supernatant and standards , incubated for 5 min at 37°C , then tested for absorbance at 500 nm . n = 1 is two flies per homogenate . Flies were fasted for 24 hr in a vial with a Kimwipe dampened with 2 mL of milliQ-filtered deionized ( milliQ DI ) water and tested for the proboscis extension response ( PER ) ( Shiraiwa and Carlson , 2007 ) . Water and all tastants were tested manually via a solution-soaked Kimwipe . Sucrose solutions were dissolved in milliQ water and presented in descending order by concentration . Groups of 10–15 flies were tested simultaneously . n = 1 equals a single animal . For each fly , ∆F/F0 was calculated from a baseline of 10 frames recorded just prior to the stimulus ( sucrose or water ) . Area under the curve ( AUC ) was calculated by summing the ∆F/F0 values from the initiation of the response to its end . Peak ∆F/F0 is the single maximum acquired within a response , and latency to peak was calculated by determining the time between the stimulus delivery and the peak response . OptoFLIC analysis of daily food interactions , meal size , and meal duration was performed as previously described ( May et al . , 2019 ) . R code used can be found on Github ( https://github . com/chrismayumich/May_et_al_optoFLIC; copy archived at https://github . com/elifesciences-publications/May_et_al_optoFLIC; May , 2020 ) . Briefly , food interactions were determined by calculating a moving baseline on the raw data and selecting signals which surpassed threshold above baseline . These signals were then summed in 30 min bins . From the binned data , daily food interactions and the start and end of meals were calculated . The evening meal was used for all meal-based calculations to control for variability in meal shape . Meal size and duration were derived using meal start and end . Post-peak feeding duration was quantified as [ ( time of meal end ) - ( time of meal peak ) ] . An event is defined as a string of consecutive food interactions . R code used to extract event information can also be found on the Github link above . To calculate events per meal , the number of events between the meal start and meal end per meal were summed for each fly . Feeding rate was quantified as [ ( events per meal ) / ( meal duration ) ] per meal per fly . Pre- and post-peak feeding rates were quantified , using the time of the meal peak determined by food interactions , also used to calculate post-peak feeding duration , as [ ( number of events pre- or post-peak ) / ( pre- or post-peak feeding duration ) ] . Pre-peak feeding duration was quantified as [ ( time of meal peak ) - ( time of meal start ) ] . Overlays created in Virtual Fly Brain ( Milyaev et al . , 2012 ) .	Obesity is a major health problem affecting over 650 million adults worldwide . It is typically caused by overeating high-energy foods , which often contain a lot of sugar . Consuming sugary foods triggers the production of a reward signal called dopamine in the brains of insects and mammals , which reinforces sugar-consuming behavior . The brain balances this with a process called ‘sensory-enhanced satiety’ , which makes foods that provide a stronger sensation of sweetness better at reducing hunger and further eating . High-energy food was scarce for most of human evolution , but over the past century sugar has become readily available in our diet leading to an increase in obesity . Last year , a study in fruit flies reported that a sugary diet reduces the sensitivity to sweet flavors , which leads to overeating and weight gain . It appears that this sensitivity is linked to the effectiveness of sensory-enhanced satiety . However , the mechanism linking diets high in sugar and overeating is still poorly understood . One hypothesis is that fruit flies estimate the energy content of food based on the degree of dopamine released in response to the sugar . May et al . compared the responses of neurons in fruit flies fed a normal diet to those in flies fed a diet high in sugar . As expected , both groups activated the neurons involved in the dopamine reward response when they tasted sugar . However , when the flies were on a sugar-heavy diet , these neurons were less active . This was because the neurons responsible for tasting sweetness were activated less in flies fed a high-sugar diet , leading to a lowered response by the neurons that produce dopamine . The flies in these experiments were genetically engineered so that the dopamine-producing neurons could be artificially activated in response to light , a technique called optogenetics . When May et al . applied this technique to the flies on a sugar-heavy diet , they were able to stop these flies from overeating . These findings provide further evidence to support the idea that a sugary diet reduces the brain’s sensitivity to overeating . Given the significant healthcare cost of obesity to society , this improved understanding could help public health initiatives focusing on manufacturing food that is lower in sugar .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2020	Dietary sugar inhibits satiation by decreasing the central processing of sweet taste
Microglial dysfunction is a key pathological feature of Alzheimer's disease ( AD ) , but little is known about proteome-wide changes in microglia during the course of AD and their functional consequences . Here , we performed an in-depth and time-resolved proteomic characterization of microglia in two mouse models of amyloid β ( Aβ ) pathology , the overexpression APPPS1 and the knock-in APP-NL-G-F ( APP-KI ) model . We identified a large panel of Microglial Aβ Response Proteins ( MARPs ) that reflect heterogeneity of microglial alterations during early , middle and advanced stages of Aβ deposition and occur earlier in the APPPS1 mice . Strikingly , the kinetic differences in proteomic profiles correlated with the presence of fibrillar Aβ , rather than dystrophic neurites , suggesting that fibrillar Aβ may trigger the AD-associated microglial phenotype and the observed functional decline . The identified microglial proteomic fingerprints of AD provide a valuable resource for functional studies of novel molecular targets and potential biomarkers for monitoring AD progression or therapeutic efficacy . Microglia play fundamental roles in a variety of neurodegenerative diseases , including AD ( McQuade and Blurton-Jones , 2019 ) . Changes in brain immunity , together with extracellular Aβ deposition and neurofibrillary tangles , are major pathological culprits in AD ( Gjoneska et al . , 2015; Guillot-Sestier and Town , 2013; Holtzman et al . , 2011; Shi and Holtzman , 2018 ) . The importance of microglia in AD pathogenesis is well illustrated by the increasing number of identified AD risk genes which are expressed in microglia and have functions in brain immunity ( Cuyvers and Sleegers , 2016; Guerreiro et al . , 2013; Jansen et al . , 2019; Jonsson et al . , 2013; Karch and Goate , 2015; Lambert et al . , 2009; Naj et al . , 2011; Sims et al . , 2017 ) . For example , the triggering receptor expressed on myeloid cells 2 ( TREM2 ) and apolipoprotein E ( APOE ) are major genetic risk factors for sporadic AD that are expressed by plaque-associated microglia and involved in Αβ clearance ( Bradshaw et al . , 2013; Castellano et al . , 2011; Kleinberger et al . , 2014; Parhizkar et al . , 2019; Reddy et al . , 2009; Wang et al . , 2015 ) . It has also been shown that microglial phagocytosis decays over the course of AD ( Hickman et al . , 2008; Koellhoffer et al . , 2017; Orre et al . , 2014a; Solito and Sastre , 2012; Zuroff et al . , 2017 ) . Along these lines , Aβ clearance was found reduced in sporadic AD and it is assumed to be a key factor in the pathogenesis ( Mawuenyega et al . , 2010; Saido , 1998; Wildsmith et al . , 2013 ) . Importantly , Aβ clearance defects in AD microglia are reversible ( Daria et al . , 2017; Krabbe et al . , 2013 ) and enhancing microglial phagocytic function has been explored as a therapeutic approach since substantial reduction of Aβ burden in mice appears to correlate with cognitive benefits ( Bacskai et al . , 2001; Bard et al . , 2000; Bohrmann et al . , 2012; Janus et al . , 2000; Lathuilière et al . , 2016; Morgan et al . , 2000; Nicoll et al . , 2006; Nicoll et al . , 2003; Schenk et al . , 1999; Schlepckow et al . , 2020; Sevigny et al . , 2016; Wilcock et al . , 2004 ) . However , when and how microglia change along AD progression is still not clear . Thus , understanding molecular alterations of microglia at different stages of AD is crucial and a pre-requisite for developing safe and efficacious therapy . Transcriptional expression profiles of microglia were previously revealed under physiological , neurodegenerative or neuroinflammatory conditions ( Butovsky et al . , 2014; Galatro et al . , 2017; Gosselin et al . , 2017; Götzl et al . , 2019; Grabert et al . , 2016; Holtman et al . , 2015; Kamphuis et al . , 2016; Krasemann et al . , 2017; Mazaheri et al . , 2017; Orre et al . , 2014a; Orre et al . , 2014b; Wang et al . , 2015; Yin et al . , 2017 ) . Transcriptional signatures were also recently reported at single-cell resolution , demonstrating regional and functional heterogeneity of brain myeloid cells ( Hammond et al . , 2019; Jordão et al . , 2019; Keren-Shaul et al . , 2017; Mathys et al . , 2017; Sala Frigerio et al . , 2019; Zhou et al . , 2020 ) . In neurodegenerative mouse models , two major profiles have been proposed along the spectrum of microglial alterations . One is the homeostatic microglial signature that occurs under physiological conditions and is characterized by the expression of several genes , including P2ry12 , Tmem119 and Cx3cr1 . The other key signatures , referred to as disease-associated microglia ( DAM ) , microglial neurodegenerative phenotype ( MGnD ) or activated response microglia ( ARM ) are observed under neurodegenerative conditions ( Keren-Shaul et al . , 2017; Krasemann et al . , 2017; Sala Frigerio et al . , 2019 ) and characterized by increased expression of Apoe , Trem2 , Cd68 , Clec7a and Itgax ( Cd11c ) , among others . These changes were quantified using RNA transcripts , but transcript levels do not necessarily reflect protein levels which ultimately control cell function ( Böttcher et al . , 2019; Mrdjen et al . , 2018; Sharma et al . , 2015 ) . Importantly , a recent study postulated that transcriptomic profiles of microglia from another AD mouse model ( 5xFAD ) do not correlate well with proteomic changes ( Rangaraju et al . , 2018 ) , suggesting the existence of additional translational or post-translational regulation mechanisms in AD microglia . Additionally , little is known about Aβ-associated changes in the microglial proteome in a time-resolved manner , or which proteome alterations underscore microglial dysfunction . Accordingly , we analyzed the microglial proteome at distinct stages of Aβ pathology in two commonly used mouse models of amyloidosis; the APPPS1 ( Radde et al . , 2006 ) , and the APP-KI mice ( Saito et al . , 2014 ) . In contrast to the APPPS1 mouse model that overexpresses mutated human amyloid precursor protein ( APP ) and presenilin-1 ( PS1 ) , the APP-KI model bears endogenous levels of APP with a humanized Aβ sequence containing three AD mutations ( NL-G-F ) , and has no alterations of PS1 ( Radde et al . , 2006; Saito et al . , 2014 ) . Our study determines the proteome of microglia from APPPS1 and APP-KI mice in a time resolved manner , starting from pre-deposition to early , middle and advanced stages of amyloid deposition and reveals a panel of Microglial Aβ Response Proteins ( MARPs ) that progressively change throughout Aβ accumulation . Although both mouse models display severe microglial alterations at late stages of Aβ pathology , the occurrence of MARP signatures differs and appears earlier in the APPPS1 mice . Strikingly , the kinetic differences in proteomic profiles correlated with the presence of fibrillar Aβ , rather than dystrophic neurites , suggesting that fibrillar Aβ aggregates may trigger the AD-associated microglial phenotype and corresponding functional decline . The time-resolved microglial profiles may serve as benchmark proteomic signatures for investigating novel microglial targets or monitoring the efficacy of future pre-clinical studies aiming at microglial repair . Amyloid plaque deposits appear at similar ages ( between 6–8 weeks ) in APPPS1 and APP-KI mouse models ( Radde et al . , 2006; Saito et al . , 2014 ) . To reveal the dynamics of microglial proteomic alterations across different amyloid stages , we analyzed microglia from 1 , 3 , 6 and 12 month old APPPS1 and APP-KI mice and their corresponding age-matched wild-type ( WT ) mice ( Figure 1—figure supplement 1A ) . To facilitate proteomic analysis , we first optimized the microglial isolation procedure . CD11b positive microglia were isolated from mouse cerebrum using MACS technology . The purity of the CD11b-enriched fraction was controlled by fluorescence activated cell sorting ( FACS ) , revealing that 97% of isolated cells were CD11b positive ( Figure 1—source data 1A ) . Of note , only 0 . 49% of CD11b positive cells were detected in the CD11b-depleted fraction ( Figure 1—source data 1B ) , demonstrating high isolation efficiency . Isolated microglia were lysed and then measured by LC-MS/MS using label-free quantification ( LFQ ) of proteins . Next , we optimized the data acquisition method for microglial proteome analysis . Recently , it was shown that Data Independent Acquisition ( DIA ) for LFQ of proteins identifies and quantifies consistently more peptides and proteins across multiple samples , compared to Data Dependent Acquisition ( DDA ) ( Bruderer et al . , 2015 ) . Thus , we first evaluated the performance of DDA vs . DIA using microglial lysates from WT and APPPS1 mice . DDA identified 53912 peptides on average compared to 74281 peptides identified by DIA , representing a 37 . 8% increase in detection by DIA method ( Supplementary file 1 ) . Overall , the main advantage of DIA was the improved consistency of protein quantifications among the replicates and the identification of proteins with lower abundance , leading to a 29% increase of relatively quantified proteins from 4412 with DDA to 5699 with DIA ( Figure 1—figure supplement 1B and C , Supplementary file 1 ) . Therefore , we chose the DIA acquisition method to also generate the proteome dataset of APP-KI microglia . We detected a consistent relative quantification of proteins with an overlap of 93 . 5% ( 5500 proteins ) between the two investigated mouse models ( Figure 1—figure supplement 1D ) , supporting our selection of DIA as a robust method for microglial proteomic analysis . For the comparative analysis of proteomic changes , we defined a threshold of a log2 fold change larger than 0 . 5 or smaller than −0 . 5 with a p-value less than 0 . 05 , and significance after False Discovery Rate ( FDR ) correction . No data imputation was performed . According to Aβ burden in both mouse models , we refer to one month of age as a pre-deposition stage , and to 3 , 6 and 12 months of age as early , middle and advanced stages of amyloid pathology , respectively ( Figure 1—figure supplement 2 ) . At the pre-deposition stage ( 1 month ) , microglial proteomes of APPPS1 and APP-KI mice did not show significant alterations compared to WT ( Figure 1A and B ) , demonstrating that microglia are not affected prior to development of Aβ pathology . At 3 months of age , microglia in APPPS1 mice already displayed a significant up-regulation of 332 proteins and down-regulation of 678 proteins , compared to WT microglia ( Figure 1C , Supplementary file 2A ) . In contrast , APP-KI microglia were hardly affected at 3 months of age ( Figure 1D , Supplementary file 2B ) , which is particularly surprising because both mouse models show comparable amyloid burden at this stage ( Figure 1—figure supplement 2 ) . At 6 months of age , microglia in APPPS1 mice displayed 309 up-regulated and 261 down-regulated proteins , compared to WT microglia ( Figure 1E , Supplementary file 2A ) . In contrast to 3 months of age ( Figure 1D ) , APP-KI mice displayed a substantial alteration of their microglial proteome at 6 months of age , illustrated by 140 up-regulated and 151 down-regulated proteins ( Figure 1F , Supplementary file 2B ) . Still , microglial alterations in 6 month old APP-KI mice were less pronounced compared to the proteome of APPPS1 mice ( Figure 1E and F ) . Noteworthy , by 12 months of age , APPPS1 microglia revealed a significant up-regulation of 776 proteins and down-regulation of 633 proteins , while APP-KI microglia displayed 704 up-regulated and 666 down-regulated proteins ( Figure 1G and H , Supplementary file 2A and B ) . Overall , our data show that amyloid plaque accumulation triggers microglial progression towards an AD-associated phenotype in both mouse models , but that response dynamics are different in APPPS1 and APP-KI microglia . Next , we determined protein alterations that first appear in early , middle or advanced stages of Aβ deposition and remain altered through all analyzed stages , thus following amyloid accumulation . To this end , we selected the APPPS1 mouse model as a reference since it displays earlier changes and therefore provides a better time resolution of protein alterations to amyloid response , compared to the APP-KI model ( Figure 2A ) . Only proteins with a consistent quantification in all samples of an age group were used for relative quantification . Furthermore , in order to determine robust and model-independent Aβ-triggered microglial alterations , we only selected MARPs that were altered with a significantly changed abundance in both mouse models ( even if in APP-KI microglia changes appear later ) . This analysis identified 90 early , 176 middle , and 435 advanced MARPs ( Figure 2—source data 1 , Figure 2—figure supplement 1A ) . The most strongly regulated MARPs with early , middle and advanced response are displayed in corresponding heatmaps ( Figure 2B–D ) . Early MARPs included several of the previously identified transcriptional DAM markers ( Keren-Shaul et al . , 2017 ) such as ITGAX ( CD11c ) , APOE , CLEC7a , LGALS3 ( Galectin-3 ) and CD68 , which were found with an increased abundance ( Figure 2B ) . Moreover , proteins involved in antigen presentation such as CD74 , H2-D1 , TAP2 , TAPBP and H2-K1 were revealed as up-regulated early MARPs . In addition , we discovered prominent changes in interferon signaling represented by the up-regulation of early MARPs , including MNDA , OAS1A , IFIT3 , ISG15 , GVIN1 , STAT1 and 2 ( Figure 2B ) . Even though early MARPs were mainly up-regulated , we also identified early MARPs with a decreased abundance , including KRAS , a protein involved in cell proliferation and the endocytosis regulator EHD2 among others ( Figure 2B ) . A gene ontology ( GO ) cluster enrichment analysis of early MARPs revealed that up-regulated proteins were enriched for immune and viral response , interferon beta and cytokine response , antigen processing and presentation as well as biotic and lipid response ( Figure 3A , Figure 3—figure supplement 1A and D ) . Thus , these processes represent first molecular alterations which progressively follow Aβ plaque pathology . The middle MARPs included the up-regulated proteins FABP3 , FABP5 , CD63 , TREM2 , MIF and GUSB ( Figure 2C ) , demonstrating a progressive conversion of the microglial proteome towards a disease state that accompanies Aβ accumulation . Importantly , middle MARPs also reveal down-regulation of the proposed homeostatic markers such as CX3CR1 , TMEM119 and P2RY12 ( Figure 2C ) . Among the down-regulated middle MARPs , we identified additional chemotaxis and cell migration related proteins like SYK , FER , CX3CL1 , and BIN2 ( Figure 2C , Figure 2—source data 1 ) , underscoring a loss of key microglial functions throughout AD progression . Advanced MARPs represent proteins that were only altered upon extensive amyloid pathology . This group included up-regulation of proteins involved in calcium ion binding such as NCAN , MYO5A , HPCAL4 , TTYH1 and GCA and down-regulation of proteins that play a role in the endocytosis/lysosomal system such as TFEB , TFE3 and BIN1 ( Figure 2D , Figure 2—source data 1 ) . In addition , different G protein-coupled receptor signaling proteins , including GNG2 , GNG5 and GNG10 , also displayed a decreased abundance ( Figure 2D ) . Furthermore , we observed a high correlation of MARP signatures between the two models at 12 months ( Figure 2—figure supplement 1B ) . A GO cluster enrichment analysis of middle and advanced MARPs identified down-regulation of biological processes including cell motility , migration and chemotaxis , as well as cell development and proliferation ( Figure 3B and C , Figure 3—figure supplement 1B , C , E and F ) . Conversely , we found an up-regulation of protein glycosylation and carbohydrate metabolism ( Figure 3C , Figure 3—figure supplement 1E and F ) . Additionally , alterations in ion transport processes involving ion homeostasis and pH regulation were also detected ( Figure 3C ) . These findings indicate that after an initial inflammatory response , several cellular processes related to chemotaxis and phagocytosis are progressively dysregulated upon increased Aβ deposition . Importantly , our proteomic analysis also detected alterations in proteins related to different genetic risk factors of AD ( Karch and Goate , 2015 ) , including significantly increased levels of APOE , TREM2 , and INPP5D , and decreased levels of PLCG2 , ABI3 , and BIN1 in both mouse models ( Figure 2—figure supplement 2A and B ) . In addition , we compared MARP signatures with the previously published single cell transcriptome study of 5xFAD mice ( Keren-Shaul et al . , 2017 ) to visualize the overlap , as well as differences , between proteomic and transcriptomic microglial profiles ( Figure 2B–D ) . When comparing the overlapping proteome of 12 months old APPPS1 and APP-KI mice with the transcriptome of 5xFAD mice ( Keren-Shaul et al . , 2017 ) , we were able to quantify 3348 common proteins/transcripts , whereas 2152 and 2841 gene products were only quantified on protein and transcript level , respectively ( Figure 4A ) . Comparison of our proteomic signatures with the microglial transcripts ( Keren-Shaul et al . , 2017 ) revealed an overlap of 227 unidirectionally regulated , whereas 263 and 849 gene products were only regulated on protein and transcript level , respectively ( Figure 4B ) . While transcriptomics demonstrates a similar regulation of a large number of early MARPs , we found less overlap for middle and advanced MARPs ( Figure 2B–D ) . We also identified proteins with an inverse regulation compared to transcriptomic signatures such as the early MARP RPL38 , middle MARPs MCM3 and GFPT1 or advanced MARPs CDC88A , GALNT2 , EIF4B and CHMP6 ( Figure 2B–D , Figure 2—source data 1 ) . Furthermore , the advanced MARP HEXB showed a consistent up-regulation in our proteomic analysis ( Figure 2—source data 1 ) , despite being previously anticipated as a homeostatic gene . Overall , our study presents a robust and reliable method to track microglial proteome and provides a resource that maps changes in brain immunity during different phases of Aβ accumulation . Next , we validated proteomic changes by western blot analysis using isolated microglia from 12 month old APPPS1 and APP-KI mice . This analysis confirmed the pronounced increase of the early MARPs APOE and CD68 , the middle MARPs TREM2 and FABP5 , as well as reduced levels of the advanced MARP CSF1R ( Figure 4C ) in both transgenic mouse models compared to WT mice . Furthermore , proteomic changes were also validated by immunohistochemistry in order to visualize spatial distribution of altered microglial proteins in APPPS1 and APP-KI mice . Immunohistological analysis of 3 month old APPPS1 mice already revealed increased immunoreactivity of selected MARPs such as CLEC7a ( Figure 5 ) , TREM2 ( Figure 5—figure supplement 1 ) and APOE ( Figure 5—figure supplement 2 ) that mark initial stages of microglial activation in AD . This increase was detected in IBA1 positive microglia that were clustering around amyloid plaques , but not in microglia further away from plaques and was – in agreement with our proteomic data – less pronounced in 3 month old APP-KI mice . Accordingly , at 12 months , both APPPS1 and APP-KI mice showed a similar increase in the levels of selected MARP CLEC7a ( Figure 6 ) and decreased levels of TMEM119 ( Figure 7 ) compared to the WT mice , once again in microglia surrounding amyloid plaques . Taken together , we validated selected microglial proteomic alterations from our dataset by applying biochemical and immunohistochemical methods . In addition , we confirmed the kinetic differences in AD-associated proteomic signatures of APPPS1 and APP-KI microglia . Our data suggest that interaction between microglia and Aβ is likely triggering the proteomic changes as they could be observed in plaque-associated microglial population . The magnitude of proteomic microglial changes was found to correlate with Aβ plaque accumulation throughout disease progression . However , the appearance of MARP signatures differed between the models and occurred earlier in the APPPS1 mice ( Figure 1C and D , Figure 2A ) despite the comparable plaque load observed in both mouse models ( Figure 1—figure supplement 2 ) . Thus , it appears possible that the nature of amyloid plaques is different between the APPPS1 and APP-KI mice . To examine this , we analyzed amyloid plaques in 3 , 6 and 12 month old APPPS1 and APP-KI mice by immunohistochemistry . We used the anti-Aβ antibody NAB228 ( Abner et al . , 2018 ) to detect Aβ plaques , and Thiazine Red ( ThR ) to visualize fibrillar amyloid ( Daria et al . , 2017; Figure 8A ) . In agreement with amyloid plaque pathology reported in this model ( Radde et al . , 2006 ) , APPPS1 mice contained fibrillar amyloid plaque cores already at 3 months of age . In contrast , fibrillar Aβ was barely detectable in APP-KI mice at 3 months of age ( Figure 8A ) . The amount of fibrillar Aβ in APP-KI mice increased at 6 and 12 months , but overall remained lower compared to the APPPS1 mice ( Figure 8A ) . The reduced levels of fibrillar Aβ in APP-KI mice were also confirmed by biochemical analysis in which fibrillar Aβ was specifically detected via immunoblot of the insoluble brain fraction ( Figure 8B ) . Despite the reduced levels of fibrillar Aβ , Aβ coverage was increased in 3 month old APP-KI compared to the APPPS1 mice ( Figure 8C ) . To obtain further information on the conformational state of amyloid plaque cores in both mouse models , we performed spectral analysis using two luminescent conjugated oligothiophenes ( LCOs ) as reported previously ( Rasmussen et al . , 2017; Figure 8—figure supplement 1A–D ) . The quadro-formyl thiophene acetic acid ( qFTAA ) LCO binds to dense core amyloid fibrils and the hepta-formyl thiophene acetic acid ( hFTAA ) LCO seems to recognize both amyloid fibrils and less dense pre-fibrillar amyloid aggregates ( Klingstedt et al . , 2011; Nyström et al . , 2013 ) . In contrast to APP-KI , a prominent qFTAA signal was detected in APPPS1 mice at 3 months of age , revealing dense core fibrillar Aβ ( Figure 8—figure supplement 1A and C ) . As expected , we could detect the hFTAA signal in both models at 3 months of age , visualizing the less dense , pre-fibrillar amyloid aggregates . To compare the levels of dense fibrillar versus pre-fibrillar Aβ , we quantified the ratio between peak intensities of qFTAA ( emission at 502 ) and hFTAA ( emission at 588 ) in both models at 3 and 12 months ( Figure 8—figure supplement 1D ) . This analysis revealed significantly reduced levels of fibrillar Aβ in the APP-KI mice at the age of 3 months and only a trend towards reduction at the age of 12 months ( Figure 8—figure supplement 1A–D ) . Thus , we demonstrate prominent differences in Aβ plaque fibrillization between APPPS1 and APP-KI mice at the age of 3 months . To determine what triggers microglial reactivity in AD , we first quantified microglial recruitment to Aβ plaques in both mouse models . This analysis was done at the early pathological stage ( 3 months ) , where we identified prominent differences in the proteome regulation ( Figure 1C and D , Figure 2A ) as well as in the amount of fibrillar Aβ ( Figure 8A and B , Figure 8—figure supplement 1A–C ) between the two mouse models . Immunohistochemical analysis revealed IBA1 positive , amoeboid microglia recruited to large , ThR positive , fibrillar Aβ aggregates in APPPS1 mice ( Figure 9A ) . Of note , we observed intracellular fibrillar Aβ in APPPS1 microglia in close contact to the plaque core as previously reported ( Bolmont et al . , 2008 ) . Despite the significantly smaller fibrillar Aβ aggregates in APP-KI mice , we could observe IBA1 positive microglia polarized towards the fibrillar Aβ , rather than to the surrounding non-fibrillar Aβ positive material ( Figure 9A ) . Quantification analysis revealed increased clustering of IBA1 positive microglia around Aβ plaques in APPPS1 compared to the APP-KI mice ( Figure 9B ) , despite their overall larger Aβ plaque size ( Figure 9C ) , supporting that fibrillar Aβ conformation , rather than plaque size , are responsible for microglial recruitment . Likewise , we observed increased CD68 immunoreactivity around Aβ plaques in the APPPS1 compared to the APP-KI mice ( Figure 9D and E ) . However , CD68 signal per individual microglial cell in the plaque vicinity was similar in both models ( Figure 9F ) , suggesting that differences in AD-associated microglial proteins are due to the number of recruited microglia rather than differences in their individual CD68 protein levels . One of the Aβ modifications that can be readily detected in AD brains and favours fibrillar conformation is pyroglutamate-modified Aβ ( pE3-Aβ ) ( Dammers et al . , 2015a; Dammers et al . , 2015b ) . To test whether pE3-Aβ may be triggering microglial reactivity , we assessed pE3-Aβ levels in 3 and 12 month old APPPS1 and APP-KI mice by immunohistochemistry , using a previously validated antibody ( Hartlage-Rübsamen et al . , 2018; Figure 9—figure supplement 1A–D ) . As we could only detect pE3-Aβ reactivity in 12 month old animals , but not at 3 months ( Figure 9—figure supplement 1A and B ) , it is less likely that this modification is responsible for microglial recruitment and differences in the proteome between the models . Similarly as observed for fibrillar Aβ , levels of pE3-Aβ were reduced in the APP-KI compared to the APPPS1 mice ( Figure 9—figure supplement 1C and D ) . Besides Aβ , microglial recruitment has also been associated with neuritic damage ( dystrophic neurites ) ( Hemonnot et al . , 2019 ) . Accordingly , we analyzed dystrophic neurite pathology in 3 month old APPPS1 and APP-KI mice , using an antibody against APP that accumulates in these structures ( Cummings et al . , 1992; Sadleir et al . , 2016 ) . As previously reported ( Radde et al . , 2006 ) , amyloid plaques in the APPPS1 mice were surrounded by prominent dystrophic neurites ( Figure 9G ) . Interestingly , despite the reduced load of fibrillar Aβ , we readily detected dystrophic neurites in the APP-KI mice ( Figure 9G ) . Moreover , our quantification analysis revealed an increased dystrophic neurite area in the APP-KI compared to the APPPS1 mice ( Figure 9H ) and reduced number of microglial cells recruited to this area ( Figure 9I ) . Therefore , the differences in early microglial recruitment to APPPS1 plaques and the consecutive proteomic changes are less likely to be triggered by dystrophic neurites . Altogether , we hypothesize that microglial recruitment may be triggered by the fibrillar Aβ content of amyloid plaques , which drives the acquisition of MARP signatures . The differences observed in the dynamics of microglial response to amyloid in the APPPS1 and the APP-KI mice prompted us to examine the association between microglial phagocytic function and the appearance of MARP signatures . To this end , we assessed the phagocytic capacity of microglia from 3 and 6 month old APPPS1 and APP-KI mice compared to the corresponding age-matched WT microglia using the E . coli-pHrodo uptake assay ( Götzl et al . , 2019; Kleinberger et al . , 2014 ) . We already detected phagocytic impairments in 3 month old APPPS1 microglia , which was reflected by a prominent decrease in the amount of intracellular E . coli particles ( Figure 9J , Figure 9—source data 1A ) and a reduced number of CD11b positive cells that were capable of E . coli uptake ( Figure 9K , Figure 9—source data 1B ) . Notably , APPPS1 phagocytic impairment did not change further in 6 month old microglia , suggesting that microglial functional deficits , as measured by the E . coli uptake assay , were fully established already at 3 months of age and characterized by early MARPs . In contrast , APP-KI microglia remained functional at 3 months , but at 6 months displayed similar impairments as seen in APPPS1 microglia ( Figure 9J and K ) . Overall , we observed different kinetics of microglial dysfunction among mouse models which correlate with the appearance of MARPs and , in turn , with the presence of fibrillar Aβ . This study presents an in-depth and time-resolved proteome of microglia isolated across different stages of Aβ accumulation in the APPPS1 and APP-KI mouse models , resulting in the identification of early , middle , and advanced MARPs . We propose that the structure of amyloid plaques ( fibrillar versus non-fibrillar ) triggers the molecular alterations of microglia . Key microglial signatures encompass proteins with a central function in microglial biology and AD pathogenesis . Moreover , our functional analysis shows that early MARP signatures already reflect microglial phagocytic dysfunction . To achieve robust and reproducible relative quantification of microglial proteins from single mice , we improved the yield of acutely isolated microglia to an average of 2 × 106 cells per mouse brain , compared to recently published protocols ( Flowers et al . , 2017; Rangaraju et al . , 2018 ) . Next , by establishing the more sensitive DIA method for protein quantification , we improved the number of consistently identified proteins by 29 . 3% and obtained an average of 5699 ( APPPS1 ) and 5698 ( APP-KI ) relatively quantified proteins . Notably , our analysis enhanced the detection of low abundance proteins and did not require data imputation . Although we included both male and female mice for the analysis of microglial proteome , our study was not powered to detect sex-specific differences that have been reported in microglia ( Sala Frigerio et al . , 2019 ) . The advancement to previous studies ( Rangaraju et al . , 2018; Sharma et al . , 2015 ) is also exemplified by quantification of membrane proteins , including well known microglial homeostatic markers TMEM119 or P2RY12 . We also measured alterations in proteins that were postulated to be only altered at the transcriptional level in AD microglia ( Rangaraju et al . , 2018 ) , including up-regulation of middle MARPs FABP3 , FABP5 , PLP2 and MIF . In summary , our study achieved a major improvement in quantitative proteomic analysis of rodent microglia ( Flowers et al . , 2017; Rangaraju et al . , 2018; Thygesen et al . , 2018 ) . This methodological advance enabled us to map microglial changes across diverse stages of Aβ pathology in two widely explored pre-clinical models of amyloidosis . Generated proteomic profiles characterize microglia under diseased conditions and can be used as a resource to track changes upon microglial therapeutic modification , such as Aβ immunotherapy . Such studies would facilitate discovery of clinically relevant molecular alterations that are necessary for microglial functional repair , monitoring disease progression and therapeutic efficacy . The TREM2/APOE axis plays a key role in the regulation of the microglial transcriptional program and guides the homeostatic/DAM signature switch ( Jay et al . , 2017; Keren-Shaul et al . , 2017; Krasemann et al . , 2017 ) . Our time-resolved proteomic analysis observed major rearrangements of the microglial proteomic landscape in both APPPS1 and APP-KI mice and revealed a partial overlap between MARPs and transcriptional profiles of DAM and homeostatic microglia ( Keren-Shaul et al . , 2017 ) , but also identified additional microglial marker proteins throughout different stages of Aβ deposition . Early MARPs include proteins of the interferon response , which is consistent with the recently identified interferon-responsive microglial sub-population in AD mice ( Sala Frigerio et al . , 2019 ) . Numerous up-regulated early MARPs , including CD74 , CTSD , CTSH , CTSZ , HEXA , GLB1 , CD68 , NPC2 and CLN3 , reflect alterations in endo-lysosomal homeostasis as an early pathological insult in AD microglia ( Van Acker et al . , 2019 ) . Additionally , factors of the fatty acid and cholesterol metabolism are altered throughout all pathological phases . Up-regulated are the early ( APOE , ACACA , and SOAT1 ) middle ( FABP3 , FABP5 , NCEH1 , APOD , AACS , ACOX3 , HACD2 ) and advanced MARPs ( ACOT11 , ACSBG1 , ECHS1 , ELOVL1 , and FASN ) and down-regulated are several middle and advanced MARPs ( NAAA , FAM213B , HPGD , HPGDS , and PRKAB1 ) , linking microglial lipid dyshomeostasis and AD pathology . An inflammatory response in AD is suggested by the significant up-regulation of early MARPs LGALS3 and its binding protein ( LGALS3BP ) . Recent findings indicated that the LGALS3/TREM2 signalling pathway , that acts as an inflammatory regulator of amyloid plaque formation , may also be of relevance for AD pathology in humans ( Boza-Serrano et al . , 2019 ) . Further evidence that some of the presented proteomic alterations of rodent microglia may be relevant for human disease is given by the detection of up-regulated early/middle microglial MARPs , including CD68 , TREM2 and ITGAX in microglia surrounding amyloid plaques in postmortem AD brains ( Hopperton et al . , 2018 ) . However , a recent study suggested little overlap between DAM and homeostatic profiles of rodent microglia and human AD patients , implicating a limitation of mouse models of AD ( Zhou et al . , 2020 ) . Of note , microglial signatures have mostly been studied in amyloidosis mouse models and amyloid triggered microglial alterations cannot be directly compared with human microglia from advanced AD stages that are exposed to both amyloid and tau pathology . As microglia emerge as a promising therapeutic target in AD , additional MARP signatures should be validated in human tissue . In particular , early MARPs that are strongly increased in both AD mouse models may serve as a resource to identify novel AD biomarkers and more specific microglial positron emission tomography ( PET ) tracers that are urgently needed to monitor microglial reactivity in vivo ( Edison et al . , 2018; Hemonnot et al . , 2019 ) . Middle and late MARPs reveal a decrease of microglial homeostatic functions affecting chemotaxis , cell migration and phagocytosis ( e . g . , CX3CR1 , SYK , P2RY12 , BIN2 , TFEB and TFE3 ) and thus mark AD progression . It is still being discussed which is the main trigger for microglial recruitment to amyloid plaques and their molecular switch from a homeostatic to a neurodegenerative phenotype ( Hemonnot et al . , 2019; Jung et al . , 2015; Krasemann et al . , 2017 ) . Our study proposes that microglial recruitment to Aβ deposits and their corresponding disease-associated proteomic alterations may be triggered by fibrillar Aβ . This response could be mediated by dense core fibrillar amyloid and/or smaller fibrillar oligomers that have been proposed as neurotoxic species ( Haass and Selkoe , 2007 ) . Of note , microglial recruitment did not correlate with neuritic dystrophies as we detected prominent neuritic dystrophies in the APP-KI mice that bear less fibrillar Aβ . Similar plaque morphology with less fibrillar Aβ , as observed in APP-KI mice , has also been reported in AD mice deficient for TREM2 or APOE that also have less microglial cells recruited to amyloid plaques and display prominent neuritic dystrophies ( Parhizkar et al . , 2019; Sala Frigerio et al . , 2019; Ulrich et al . , 2018; Wang et al . , 2015; Yuan et al . , 2016 ) . APOE may have a dual role and control the transcriptional/translational response of microglia to amyloid as well as amyloid plaque compactness that directs microglial recruitment and thus creates a regulatory feedback-loop . These findings are strengthened by the relevance of ApoE and Trem2 as genetic risk factors of AD ( Karch and Goate , 2015 ) . Fibrillar Aβ as the trigger for microglial recruitment is also supported by the human pathology where neuritic plaques in AD brains were found surrounded by microglia . In contrast , microglial clustering was not detected at diffuse plaques lacking fibrillar Aβ cores ( D'Andrea et al . , 2004 ) . Although DAM signatures , that include upregulation of phagocytosis and lysosomal genes , have been suggested as a protective response ( Keren-Shaul et al . , 2017 ) , there is still a lack of direct experimental evidence linking DAM profiles to improved microglial phagocytic function . In addition , it has been proposed that MGnD microglia may represent a dysfunctional phenotype ( Krasemann et al . , 2017 ) . Importantly , our study demonstrates a functional link between proteomic changes and reduced phagocytosis by AD microglia . This is in agreement with Aβ−dependent early phagocytic dysfunction of APPPS1 microglia reported previously ( Krabbe et al . , 2013 ) . Our study shows that APPPS1 microglia start acquiring early MARPs at the age of 3 months , which is already accompanied by reduced phagocytic function . In contrast , less altered proteomic signatures of 3 month old APP-KI microglia correlated with preserved phagocytic function . Pronounced MARP signatures that appeared later in APP-KI microglia ( 6 months ) were subsequently in accordance with phagocytic impairments . Therefore , differences in plaque fibrillization in both mouse models did not only affect microglial recruitment and activation , but also the phagocytic function of microglia . This functional link should be examined further using physiological substrates of microglia such as Aβ , myelin or synaptosomes ( McQuade and Blurton-Jones , 2019 ) . Reduced phagocytosis of AD microglia might be related to observed proteomic alterations in lysosomal proteins or cell receptors . TREM2 , which we found increased in both mouse models , plays an important role in phagocytosis as mutations of TREM2 related to AD and FTLD impair phagocytic activity of microglia ( Kleinberger et al . , 2014 ) . However , up-regulation of the TREM2/APOE axis involves up-regulation of many phagocytic or lysosomal proteins ( e . g . , cathepsins or CD68 ) that are part of MARPs and altered in APPPS1 and APP-KI microglia . Similarly , also transcriptional analysis revealed a downregulation of homeostatic and upregulation of DAM or MGnD program within microglial cells in the vicinity of Aβ plaques ( Keren-Shaul et al . , 2017; Krasemann et al . , 2017 ) . The increase in lysosomal or phagocytic gene signatures may reflect a compensatory mechanism initiated as a response of microglia to Aβ accumulation in order to enhance phagocytic function . Eventually , this limited microglial response fails to translate into improved Aβ clearance capability . Phagocytosis might also be altered through differential regulation of toll like receptors ( TLR ) . Among the TLRs , TLR2 , an Aβ binding receptor ( Liu et al . , 2012; McDonald et al . , 2016 ) , showed the strongest increase with age while TLR9 was significantly reduced in APPPS1 and APP-KI mice . Along these lines , TLR2 deficiency reduced the inflammatory response of microglia to Aβ42 , but increased Aβ phagocytosis in cultured microglia ( Liu et al . , 2012 ) while TLR9 is associated with improved Aβ clearance ( Scholtzova et al . , 2009 ) . Thus , differential regulation of TLRs might contribute to the reduced phagocytic activity of aged APPPS1 microglia ( Daria et al . , 2017 ) . Additionally , many purinergic receptors ( e . g . , P2RX7 , P2RY12 or P2RY13 ) , which are important regulators of chemotaxis , phagocytosis , membrane polarization , and inflammatory signaling and thus emerged as possible microglial targets in AD ( Calovi et al . , 2019; Hemonnot et al . , 2019 ) , were found down-regulated in both AD mouse models . P2RY12 is regarded as a marker for ramified non-inflammatory microglia ( Mildner et al . , 2017 ) that is reduced in response to Aβ plaques and therefore represents a homeostatic microglial marker ( Keren-Shaul et al . , 2017; Krasemann et al . , 2017 ) . In contrast , P2RX4 , a purinergic receptor that is likely to be involved in shifting microglia towards a pro-inflammatory phenotype ( Calovi et al . , 2019 ) or myelin phagocytosis ( Zabala et al . , 2018 ) had an increased abundance in both AD models . Taken together , our data emphasize alterations of purinergic receptor signaling in AD microglia that may regulate a morphological change towards amoeboid microglia with reduced motility and increased pro-inflammatory activity . Our study confirms that both mouse models are valuable tools for studying Aβ−induced pathological changes of microglia that are remarkably comparable at advanced stages of amyloidosis . However , the observed differences in the dynamics of early , middle and late MARPs in APPPS1 and APP-KI mice should be considered for the design of pre-clinical studies of microglial repair and will require different time windows for microglial modulation . In conclusion , we tracked pathological alterations of microglia in two AD mouse models using a proteomic approach . Our work demonstrates that microglial alterations are triggered as a response to Aβ deposition as pre-deposition stages do not reveal proteomic alterations . The conversion to MARPs is supported by changes in TREM2-APOE regulation mechanism . AD microglia display pronounced interferon stimulation , increased antigen presentation , alterations in cell surface receptors , lipid homeostasis and metabolism . These proteomic changes in microglia seem to occur as a response to fibrillar Aβ and are reflected in amoeboid microglial morphology and impaired phagocytic capacity . Finally , our proteomic dataset serves a valuable research resource providing information on microglial alterations over different stages of Aβ deposition that can be used to monitor therapeutic efficacy of microglial repair strategies . Male and female mice of the hemizygous APPPS1 mouse line overexpressing human APPKM670/671NL and PS1L166P under the control of the Thy1 promoter ( Radde et al . , 2006 ) , homozygous AppNL-G-F mouse line ( Saito et al . , 2014 ) and the C57BL/6J ( WT ) line were used in this study . Mice were group housed under specific pathogen-free conditions . Mice had access to water and standard mouse chow ( Ssniff Ms-H , Ssniff Spezialdiäten GmbH , Soest , Germany ) ad libitum and were kept under a 12/12 hr light-dark cycle in IVC System Typ II L-cages ( 528 cm2 ) equipped with solid floors and a layer of bedding . All animal experiments were performed in compliance with the German animal welfare law and have been approved by the government of Upper Bavaria . Primary microglia were isolated from mouse brains ( cerebrum ) using MACS technology ( Miltenyi Biotec ) according to manufacturer's instructions and as previously described ( Daria et al . , 2017 ) . Briefly , olfactory bulb , brain stem and cerebellum were removed and the remaining cerebrum was freed from meninges and dissociated by enzymatic digestion using a Neural Tissue Dissociation Kit P ( Miltenyi Biotec ) . Subsequently , mechanical dissociation was performed by using three fire-polished glass Pasteur pipettes of decreasing diameter . CD11b positive microglia were magnetically labelled using CD11b MicroBeads , loaded onto a MACS LS Column ( Miltenyi Biotec ) and subjected to magnetic separation , resulting in CD11b-enriched ( microglia-enriched ) and CD11b-depleted ( microglia-depleted ) fractions . Obtained microglia-enriched pellets were either washed twice with HBSS ( Gibco ) supplemented with 7 mM HEPES , frozen in liquid nitrogen and stored at −80°C for biochemical or mass spectrometry analysis , or resuspended in microglial culturing media and used for phagocytosis assay as described below . Microglia were isolated from three transgenic animals and their three corresponding WT controls for each age and genotype . We included both male and female mice into proteomic analysis and their distribution is outlined in Supplementary file 3 . Microglia-enriched pellets from individual animals were analysed separately and lysed in 200 µL of STET lysis buffer ( 50 mM Tris , 150 mM NaCl , 2 mM EDTA , 1% Triton , pH 7 . 5 ) at 4°C with intermediate vortexing . The samples were centrifuged for 5 min at 16000 x g at 4°C to remove cell debris and undissolved material . The supernatant was transferred to a LoBind tube ( Eppendorf ) and the protein concentration estimated using the Pierce 660 nm protein assay ( ThermoFisher Scientific ) . A protein amount of 15 µg was subjected to tryptic protein digestion applying the the filter aided sample preparation protocol ( FASP ) ( Wiśniewski et al . , 2009 ) using Vivacon spin filters with a 30 kDa cut-off ( Sartorius ) . Briefly , proteins were reduced with 20 mM dithiothreitol and free cystein residues were alkylated with 50 mM iodoacetamide ( Sigma Aldrich ) . After the urea washing steps , proteins were digested with 0 . 3 µg LysC ( Promega ) for 16 hr at 37°C followed by a second digestion step with 0 . 15 µg trypsin ( Promega ) for 4 hr at 37°C . The peptides were eluted into collection tubes and acidified with formic acid ( Sigma Aldrich ) . Afterwards , proteolytic peptides were desalted by stop and go extraction ( STAGE ) with self-packed C18 tips ( Empore C18 SPE , 3M ) ( Rappsilber et al . , 2003 ) . After vacuum centrifugation , peptides were dissolved in 20 µL 0 . 1% formic acid ( Biosolve ) and indexed retention time peptides were added ( iRT Kit , Biognosys ) . For label-free quantification ( LFQ ) of proteins , peptides were analyzed on an Easy nLC 1000 or 1200 nanoHPLC ( Thermo Scientific ) which was coupled online via a Nanospray Flex Ion Source ( Thermo Sientific ) equipped with a PRSO-V1 column oven ( Sonation ) to a Q-Exactive HF mass spectrometer ( Thermo Scientific ) . An amount of 1 . 3 µg of peptides was separated on in-house packed C18 columns ( 30 cm x 75 µm ID , ReproSil-Pur 120 C18-AQ , 1 . 9 µm , Dr . Maisch GmbH ) using a binary gradient of water ( A ) and acetonitrile ( B ) supplemented with 0 . 1% formic acid ( 0 min . , 2% B; 3:30 min . , 5% B; 137:30 min . , 25% B; 168:30 min . , 35% B; 182:30 min . , 60% B ) at 50°C column temperature . For DDA , full MS scans were acquired at a resolution of 120000 ( m/z range: 300–1400; automatic gain control ( AGC ) target: 3E+6 ) . The 15 most intense peptide ions per full MS scan were selected for peptide fragmentation ( resolution: 15000; isolation width: 1 . 6 m/z; AGC target: 1E+5; normalized collision energy ( NCE ) : 26% ) . A dynamic exclusion of 120 s was used for peptide fragmentation . For DIA , one scan cycle included a full MS scan ( m/z range: 300–1400; resolution: 120000; AGC target: 5E+6 ions ) and 25 MS/MS scans covering a range of 300–1400 m/z with consecutive m/z windows ( resolution: 30000; AGC target: 3E+six ions; Supplementary file 4 ) . The maximum ion trapping time was set to ‘auto’ . A stepped normalized collision energy of 26 ± 2 . 6% was used for fragmentation . Microglia from APPPS1 mice were analyzed using DDA and DIA for method establishement . Microglia from APPPS1 and APP-KI mice were compared using DIA as it outperformed DDA . For data acquired with DDA , the data was analyzed with the software Maxquant ( maxquant . org , Max-Planck Institute Munich ) version 1 . 6 . 1 . 0 ( Cox et al . , 2014 ) . The MS data was searched against a reviewed canonical fasta database of Mus musculus from UniProt ( download: November the 1st 2017 , 16843 entries ) supplemented with the sequence of human APP with the Swedish mutant and the iRT peptides . Trypsin was defined as a protease . Two missed cleavages were allowed for the database search . The option first search was used to recalibrate the peptide masses within a window of 20 ppm . For the main search peptide and peptide fragment mass tolerances were set to 4 . 5 and 20 ppm , respectively . Carbamidomethylation of cysteine was defined as static modification . Acetylation of the protein N-term as well as oxidation of methionine was set as variable modification . The FDR for both peptides and proteins was set to 1% . The ‘match between runs’ option was enabled with a matching window of 1 . 5 min . LFQ of proteins required at least one ratio count of unique peptides . Only unique peptides were used for quantification . Normalization of LFQ intensities was performed separately for the age groups because LC-MS/MS data was acquired in different batches . A spectral library was generated in Spectronaut ( version 12 . 0 . 20491 . 11 , Biognosys ) ( Bruderer et al . , 2015 ) using the search results of Maxquant of the APPPS1 dataset . The library includes 122542 precursor ions from 91349 peptides , which represent 6223 protein groups . The DIA datasets of both mouse models were analyzed with this spectral library ( version 12 . 0 . 20491 . 14 . 21367 ) with standard settings . Briefly , the FDR of protein and peptide identifications was set to 1% . LFQ of proteins was performed on peptide fragment ions and required at least one quantified peptide per protein . Protein quantification was performed on maximum three peptides per protein group . The data of APPPS1 microglia was organized in age-dependent fractions to enable separate normalization of the data . All LC-MS/MS runs of the APP-KI dataset were normalized against each other because all samples were analyzed in randomized order in one batch . The protein LFQ reports of Maxquant and Spectronaut were further processed in Perseus ( Tyanova et al . , 2016 ) . The protein LFQ intensities were log2 transformed and log2 fold changes were calculated between transgenic and WT samples separately for the different age groups and mouse models . Only proteins with a consistent quantification in all samples of an age group were considered for statistical testing . A two-sided Student’s t-test was applied to evaluate the significance of proteins with changed abundance ( from log2 fold change of transgenic versus WT microglia per age group ) . Additionally , a permutation based FDR estimation ( threshold: FDR = 5% , s0 = 0 . 1 ) was used to perform multiple hypothesis correction ( Tusher et al . , 2001 ) . A log2 fold change larger than 0 . 5 , or smaller than −0 . 5 , a p-value less than 0 . 05 and significant regulation after FDR filtering were defined as regulation thresholds criteria . The same thresholds were used for the comparison with transcriptomics data . Gene ontology enrichment analysis was performed with the web-tool DAVID ( version 6 . 8 ) ( Huang et al . , 2009a; Huang et al . , 2009b ) using GO_FAT terms . Up- and down-regulated early , middle and advanced MARPs were clustered separately for biological process , cellular component , and molecular function with all 5500 proteins , consistently quantified in APPPS1 and APP-KI microglia , as a customized background . A medium classification stringency was applied . An enrichment score of 1 . 3 was defined as threshold for cluster enrichment . RIPA lysates were prepared from brain hemispheres , centrifuged at 100000 x g ( 60 min at 4°C ) and the remaining pellet was homogenized in 0 . 5 mL 70% formic acid . The formic acid fraction was neutralized with 20 × 1 M Tris-HCl buffer at pH 9 . 5 and used for Aβ analysis . For Aβ detection , proteins were separated on Tris-Tricine gels ( 10–20% , Thermo Fisher Scientific ) , transferred to nitrocellulose membranes ( 0 . 1 µm , GE Healthcare ) which were boiled for 5 min in PBS and subsequently incubated with the blocking solution containing 0 . 2% I-Block ( Thermo Fisher Scientific ) and 0 . 1% Tween 20 ( Merck ) in PBS for 1 hr , followed by overnight incubation with rabbit polyclonal 3552 antibody ( 1:2000 , Yamasaki et al . , 2006 ) . Antibody detection was performed using the corresponding anti-HRP conjugated secondary antibody ( Santa Cruz ) and chemiluminescence detection reagent ECL ( Thermo Fisher Scientific ) . Microglia-enriched pellets were resuspended in 100 µL of STET lysis buffer ( composition as described above for mass spectrometry , supplemented with protease and phosphatase inhibitors ) , kept on ice for 20 min and then sonicated for 4 cycles of 30 s . Cell lysates were then centrifuged at 9600 x g ( 5 min . at 4°C ) and pellets discarded . Protein concentration was quantified using Bradford assay ( Biorad ) according to manufacturer instructions . 10 µg of two independent microglial lysates per genotype were loaded on a bis-tris acrylamide gel ( 8% or 12% ) and subsequently blotted onto either a PVDF or nitrocellulose membrane ( Millipore ) using the following antibodies: TREM2 ( 1:10 , clone 5F4 , Xiang et al . , 2016 ) ; APOE ( 1:1000 , AB947 Millipore ) ; CD68 ( 1:1000 , MCA1957GA , AbDserotec ) ; CSF1R ( 1:1000 , 3152 , Cell Signaling ) and FABP5 ( 1:400 , AF1476 , R and DSystems ) . Blots were developed using horseradish peroxidase-conjugated secondary antibodies ( Promega ) and the ECL chemiluminescence system ( Amersham ) or SuperSignal West Pico PLUS ( Thermo Scientific ) . An antibody against GAPDH ( 1:2000 , ab8245 , Abcam ) was used as loading control . We analyzed 3 and 12 month old APPPS1 and APP-KI mice of both sexes , as outlined in Supplementary file 5 . Mice were anesthetized i . p . with a mixture of ketamine ( 400 mg/kg ) and xylazine ( 27 mg/kg ) and transcardially perfused with cold 0 . 1M PBS for 5 min followed by 4% paraformaldehyde ( PFA ) in 0 . 1 M PBS for 15 min . Brains were isolated and postfixed for 20 min in 4% PFA in 0 . 1 M PBS and transferred to 30% sucrose in 0 . 1 M PBS for cryopreservation . Brains were embedded in optimal cutting temperature compound ( Tissue-Tek O . C . T . , Sakura ) , frozen on dry ice and kept at −80°C until sectioning . 30 µm coronal brain sections were cut using a cryostat ( CryoSTAR NX70 , Thermo Scientific ) and placed in 0 . 1 M PBS until staining . Alternatively , sections were kept in anti-freezing solution ( 30% glycerol , 30% ethylenglycol , 10% 0 . 25 M PO4 buffer , pH 7 . 2–7 . 4% and 30% dH2O ) at −20°C and briefly washed in 0 . 1M PBS before staining . Briefly , free-floating sections were permeabilized with 0 . 5% Triton-PBS ( PBS-T ) for 30 min , blocked either in 5% normal goat Serum or 5% donkey serum in PBS-T for 1 hr and incubated overnight at 4°C in blocking solution with the following primary antibodies: IBA1 ( 1:500 , 019–19741 , Wako ) , IBA1 ( 1:500 , ab5076 , Abcam ) , NAB228 ( 1:2000 , sc-32277 , Santa Cruz ) , CD68 ( 1:500 , MCA1957GA , Bio-Rad ) , TREM2 ( 1:50 , AF1729 , R and DSystems ) , APP-Y188 ( 1:2000 , ab32136 , Abcam ) , CLEC7a ( 1:50 , mabg-mdect , Invivogen ) , TMEM119 ( 1:200 , ab209064 , Abcam ) , APOE-biotinylated ( HJ6 . 3 , 1:100 , [Kim et al . , 2012] ) , 3552 ( 1:5000 , [Yamasaki et al . , 2006] ) and pE3-Aβ ( J8 , 1:500 , [Hartlage-Rübsamen et al . , 2018] ) . pE3-Aβ immunostaining required heat antigen retrieval ( 25 min at 95°C ) in citrate buffer ( 10 mM , pH 6 . 0 ) prior blocking . After primary antibody incubation , brain sections were washed 3 times with PBS-T and incubated with appropriate fluorophore-conjugated or streptavidine-fluorophore conjugated ( for APOE biotinylated antibody ) secondary antibodies ( 1:500 , Life Technologies ) together with nuclear stain Hoechst 33342 ( 1:2000 , H3570 , Thermo Fisher Scientific ) , for 2 hr at room temperature ( RT ) . Fibrillar dense core plaques were stained with ThR ( Sigma Aldrich , 2 µM solution in PBS ) for 20 min in the dark at RT ( after secondary antibody staining ) . Sections were subsequently washed 3 times with PBS-T , mounted onto glass slides ( Thermo Scientific ) , dried in the dark for at least 30 min , mounted using Gel Aqua Mount media ( Sigma Aldrich ) and analyzed by confocal microscopy . 3 month old APPPS1 and APP-KI mice were analyzed for dystrophic neurites , plaque size , microglial recruitment and CD68 and total Aβ coverage . pE3-Aβ coverage was analyzed in 12 month old APPPS1 and APP-KI mice . All quantification analysis were performed in a blinded manner and included at least three mice per genotype . Mice sex and numbers of biological and technical replicates for all immunohistological experiments are summarized in Supplementary file 5 . For the analysis of microglial recruitment , plaque and dystrophic neurite size and CD68 coverage , 30 z-stack images ( 10–12 µm thick ) of randomly selected plaques from neocortical regions were acquired per experiment using a confocal microscope ( 63X water objective with 2x digital zoom , 600 Hz , Leica TCS SP5 II ) . Microscopy acquisition settings were kept constant within the same experiment . Maximal intensity projection pictures from every z-stack were created using ImageJ software and for every image , a defined region of interest ( ROI ) was manually drawn around every plaque ( including microglia recruited -in contact- to the plaque ) . APP ( Y188 antibody ) and CD68 coverage area were quantified using the ‘Threshold’ and ‘Analyze Particles’ ( inclusion size of 1-Infinity ) functions from ImageJ software ( NIH ) within the ROI . The area covered by CD68 was normalized to the total Aβ plaque area ( NAB228 antibody ) or was divided by the number of microglia ( IBA1 positive cells ) recruited to the plaque within the ROI . The absolute values of area covered by neuritic dystrophies or Aβ plaques are represented in square micrometers ( µm² ) . Microglial recruitment to plaques was quantified by counting the number of microglia ( IBA1 positive cells ) around amyloid plaques through the z-stack images within the defined ROI using the cell counter function of ImageJ software . Number of microglial cells at amyloid plaques was normalized to the area covered by Aβ ( NΑΒ228 antibody ) or by dystrophic neurites ( APP , Y188 antibody ) and expressed as number of microglial cells per µm² of Aβ plaque . For the quantification of total Aβ and pE3-Aβ coverage at 3 and 12 months of age , respectively , 18 images were systematically taken from comparable neocortical regions using a confocal microscope ( 20X dry objective , 600 Hz , Leica TCS SP5 II ) . Quantification of Aβ and pE3-Aβ areas was performed using a self-programmed macro with ImageJ software outlined in Supplementary file 6A and B , respectively . Representative images from microglial recruitment analysis ( IBA1 positive cells and CD68 coverage ) were taken using the confocal microscope ( 63X water objective with 2x digital zoom , 400 Hz , Leica TCS SP5 II ) . Representative images of microglia polarized towards amyloid cores and of microglial recruitment and dystrophic neurite size were taken using a 63X confocal water objective with 3x digital zoom . For immunohistological validation of the proteome and analysis of amyloid pathology , representative images from comparable neocortical regions were taken by confocal microscopy using the same settings for all three different genotypes ( WT , APPPS1 and APP-KI ) . Low magnification pictures were taken with 20X dry confocal objective with 2x digital zoom and higher magnification ones with 63X confocal water objective with 3x digital zoom ( 400 Hz , Leica TCS SP5 II ) . Images of Aβ pathology ( NΑΒ228 antibody ) were taken with a tile scan system covering comparable cortico-hippocampal regions ( 10X confocal dry objective , 400 Hz , Leica TCS SP5 II ) . Representative images of Aβ composition and microglia ( NAB228 , ThR and IBA1 ) and pE3-Aβ were taken with a confocal 20X dry objective ( 400 Hz , Leica TCS SP5 II ) . 3 and 12 month old APPPS1 and APP-KI mice were analysed using LCOs . Information including mice sex and biological and technical replicates is outlined in Supplementary file 5 . LCO analysis was performed as previously described ( Rasmussen et al . , 2017 ) . Briefly free-floating brain sections from PFA-perfused mice ( as described in Immunohistochemistry ) were incubated with two LCOs , qFTAA and hFTAA ( 2 . 4 µM qFTAA and 0 . 77 µM hFTAA in PBS ) , for 30 min at RT in the dark and subsequently washed 3 times with PBS-T . Sections were mounted , air-dried and mounted with Dako mounting medium . Randomly chosen images from neocortex ( 4–10 images per mouse ) were acquired with a confocal 40X oil objective on a Zeiss LSM 510 META confocal microscope equipped with an argon 458 nm laser for excitation and a spectral detector . Emission spectra were acquired from 470 to 695 nm and values of each plaque were normalized to their respective maxima . For each mouse , 20 plaques were randomly chosen from the images , normalized values were averaged for each mouse and the mean was taken for each mouse line ( spectrum ) . The ratio of the intensity of emitted light at the blue-shifted peak ( 502 nm from qFTAA ) and red-shifted peak ( 588 nm from hFTAA ) was used as a parameter for spectral distinction of different Aβ conformations in plaque cores . Microglial phagocytosis was performed similarly as previously described ( Kleinberger et al . , 2014 ) . Microglia isolated from 3 or 6 month old APPPS1 , APP-KI and WT mice were plated onto 24 well plate at a density of 2 × 105 cells per well and cultured for 24 hr in a humidified 5% CO2 incubator at 36 . 5°C in DMEM/F12 media ( Invitrogen ) supplemented with 10% heat inactivated FCS ( Sigma ) , 1% Penicillin-Streptomycin ( Invitrogen ) and 10 ng/mL GM-CSF ( R&DSystems ) . After 24 hr , plating media were replaced with fresh media . After 5 days in culture , microglia were incubated with 50 µL of E . coli particle suspension ( pHrodo Green E . coli BioParticles , P35366 , Invitrogen ) for 60 min . Cytochalasin D ( CytoD , 10 µM , from 10 mM stock in DMSO ) was used as phagocytosis inhibitor and added 30 min prior to addition of bacterial particles . Bacteria excess was washed four times with PBS ( on ice ) and microglia attached to the plate were incubated with CD11b-APC-Cy7 antibody ( 1:200 , clone M1/70 , 557657 , BD ) in FACS buffer ( PBS supplemented with 2 mM EDTA and 1% FBS ) for 30 min at 4°C . Microglia were then washed twice with PBS , scraped off in FACS buffer and analyzed by flow cytometry . Information including mice sex and biological and technical replicates is outlined in Supplementary file 7 . For the microglial isolation quality control , around 12000 cells from a CD11b-enriched and CD11b-depleted fractions were stained in suspension with CD11b-APC-Cy7 antibody ( 1:200 , clone M1/70 , 557657 , BD ) in FACS buffer for 30 min at 4°C . After several washes with PBS , microglia were resuspended in FACS buffer for analysis . Propidium Iodide ( PI ) staining was done 10 min prior to FACS analysis . Flow cytometric data was acquired on a BD FACSverse flow cytometer by gating according to single stained and unstained samples and analyzed using FlowJo software ( Treestar ) . Mean fluorescent intensity ( MFI ) is represented as the geometric mean of the according fluorochrome . Immunohistochemical and FACS results are presented as mean ± standard deviation ( ± SD ) from at least three independent experiments with the exception of the phagocytic assay at 6 months where two independent experiments were performed ( Figure 9J and K ) . Statistical significance ( p-value ) was calculated using the unpaired two-tailed Student's t-test . Phagocytic assay was analyzed by the Dunnett’s multiple comparison test of the Two-way ANOVA . Both statistical analysis were performed in GraphPad Prism . P-value of <0 . 05 was considered statistically significant ( ; p<0 . 05 , ; p<0 . 01 , ; p<0 . 001 and **; p<0 . 0001 , n . s . =not significant ) . The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository ( Perez-Riverol et al . , 2019 ) with the dataset identifier PXD016075 .	Alzheimer’s disease is a progressive , irreversible brain disorder . Patients with Alzheimer’s have problems with memory and other mental skills , which lead to more severe cognitive decline and , eventually , premature death . This is due to increasing numbers of nerve cells in the brain dying over time . A distinctive feature of Alzheimer’s is the abnormally high accumulation of a protein called amyloid-β , which forms distinctive clumps in the brain termed ‘plaques’ . The brain has a type of cells called the microglia that identify infections , toxic material and damaged cells , and prevent these from building up by clearing them away . In Alzheimer’s disease , however , the microglia do not work properly , which is thought to contribute to the accumulation of amyloid-β plaques . This means that people with mutations in the genes important for the microglia activity are also at higher risk of developing the disease . Although problems with the microglia play an important role in Alzheimer’s , researchers still do not fully understand why microglia stop working in the first place . It is also not known exactly when and how the microglia change as Alzheimer’s disease progresses . To unravel this mystery , Sebastian Monasor , Müller et al . carried out a detailed study of the molecular ‘fingerprints’ of microglia at each key stage of Alzheimer’s disease . The experiments used microglia cells from two different strains of genetically altered mice , both of which develop the hallmarks of Alzheimer’s disease , including amyloid-β plaques , at similar rates . Analysis of the proteins in microglia cells from both strains revealed distinctive , large-scale changes corresponding to successive stages of the disease – reflecting the gradual accumulation of plaques . Obvious defects in microglia function also appeared soon after plaques started to build up . Microscopy imaging of the brain tissue showed that although amyloid-β plaques appeared at the same time , they looked different in each mouse strain . In one , plaques were more compact , while in the other , plaques appeared ‘fluffier’ , like cotton wool . In mice with more compacted plaques , microglia recognized the plaques earlier and stopped working sooner , suggesting that plaque structure and microglia defects could be linked . These results shed new light on the role of microglia and their changing protein ‘signals’ during the different stages of Alzheimer’s disease . In the future , this information could help identify people at risk for the disease , so that they can be treated as soon as possible , and to design new therapies to make microglia work again .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "tools", "and", "resources", "neuroscience" ]	2020	Fibrillar Aβ triggers microglial proteome alterations and dysfunction in Alzheimer mouse models
The pupil is primarily regulated by prevailing light levels but is also modulated by perceptual and attentional factors . We measured pupil-size in typical adult humans viewing a bistable-rotating cylinder , constructed so the luminance of the front surface changes with perceived direction of rotation . In some participants , pupil diameter oscillated in phase with the ambiguous perception , more dilated when the black surface was in front . Importantly , the magnitude of oscillation predicts autistic traits of participants , assessed by the Autism-Spectrum Quotient AQ . Further experiments suggest that these results are driven by differences in perceptual styles: high AQ participants focus on the front surface of the rotating cylinder , while those with low AQ distribute attention to both surfaces in a more global , holistic style . This is the first evidence that pupillometry reliably tracks inter-individual differences in perceptual styles; it does so quickly and objectively , without interfering with spontaneous perceptual strategies . A visual scene can be perceived at various hierarchical levels of structure , from the most local elements to the global organization . A dense patch of trees is perceived as a forest at a global level , while a progressively more local analysis reveals the individual trees , then their leaves , bark , etc . The basis of global perception is the structuring into larger units the ‘bits and pieces’ of visual information in order to perceive objects and their relations ( de-Wit and Wagemans , 2015 ) . The ability to form a whole from parts , ignoring details to form meaningful classes , is a major developmental step in childhood and a component of how we define intelligence – one of the tests for IQ in the standard Wechsler Scale . Different authors have developed relatively simple perceptual tasks that have become established indexes of the preference for local or global: the block design task ( Kohs , 1920; Wechsler , 1955 ) , the Rod-and-Frame Test ( Witkin and Asch , 1948 ) ; the Embedded Figures Test ( EFT: 5 ) ; and Navon’s hierarchical letters ( Navon , 1977 ) . The preference for local or global is also part of a general way of feeling and behaving , as described by personality traits . Perhaps , the clearest example is the tendency for local or detail-oriented perception in Autistic Spectrum Disorders ( Happé and Frith , 2006; Mottron et al . , 2006; Plaisted et al . , 1999 ) , compared with the more global or holistic perception in typical adults ( Navon , 1977 ) : autistic observers show slower responses to global structure ( Van der Hallen et al . , 2015 ) and enhanced processing of local features ( Mottron et al . , 2006; Muth et al . , 2014 ) . There is increasing interest in determining whether differences in the preference for local or global styles are also associated with autistic traits in the typical population , supporting the dimensional view of autistic disorders , where people with and without ASD diagnosis lay along a continuum and differ only quantitatively , not qualitatively ( Baron-Cohen et al . , 2001; Constantino and Todd , 2003; Ruzich et al . , 2015; Skuse et al . , 2009; Wheelwright et al . , 2010 ) . However , not all attempts have been successful at showing a local-global difference in neurotypical individuals ( Cribb et al . , 2016 ) , which may a result from the limited reliability and validity of the tests , which correlate little with each other and probably measure very different combinations of abilities and constructs ( Milne and Szczerbinski , 2009 ) . Here , we report on a pupillometry-based paradigm that yields an objective index of local-global processing . We tested 50 randomly selected neuro-typical adults with variable degrees of autistic traits , as measured by the Autism-Spectrum Quotient ( AQ ) , a tool developed to characterize the normal spectrum of subclinical autistic behaviours in the general population ( Baron-Cohen et al . , 2001 ) . Our psychophysical task – simply reporting the apparent direction of rotation of a bistable stimulus – can be performed equally well with both local and global perceptual styles: either by attending to only the front surface , or by attending to the whole global rotation . By tagging the front and back surfaces with different luminance levels , the two styles should have different modulatory effects on pupil-size , which is has shown to be modulated by cognitive and top-down factors , such as perceived rather than physical luminance , and shifts in attention from dark to bright surfaces ( Binda and Murray , 2015; Binda et al . , 2014 ) . Specifically , attending to a bright or dark surface is sufficient to evoke pupil constrictions or dilations respectively , tracking the focus of ‘feature-based’ attention ( Binda et al . , 2014 ) . We therefore hypothesized that pupil-modulations would depend on whether participants attend locally to the front surface , or globally to both – implying that pupil-modulations in this task effectively index differences in the local-global preference across our participants . We tracked pupil-size while participants viewed an ambiguous dynamic stimulus comprising 150 leftward-moving white dots and 150 rightward-moving black dots , giving the immediate impression of a cylinder rotating in depth at 10 revolutions per minute ( see Figure 1A and Video 1 ) . As the depth of the dots was ambiguous , either dot-group ( tagged by luminance and direction ) could be seen in front , resulting in either clockwise or counter-clockwise rotation . Subjects continuously reported the perceived direction of rotation by keypress or joystick for three sessions , each lasting 10 min . The 50 participants were recruited in two groups of 25 , and the two sessions were intended as self-replications of the experiment ( see below , Figure 3 ) . On average , perceived direction alternated every 5 . 5 s ( 0 . 21 ± 0 . 01 perceptual switches per second ) ; the between-subject variability of switch rate did not correlate with autistic traits ( Pearson’s r = 0 . 02 [-0 . 26 0 . 29] , p=0 . 901 , Bayes Factor [BF]=0 . 1 ) . Figure 1B shows the timecourse of pupil-size , synchronized to the perceptual switch and averaged over all subjects , separately for perceptual phases with black or white foreground ( blue and red , respectively ) . Two distinct kinds of pupil modulations are apparent . First , as reported previously for binocular rivalry ( Einhäuser et al . , 2008 ) , the pupil dilated transiently at or just after each perceptual switch; this effect is seen on both blue and red traces , meaning that the dilation occurs irrespective of whether perception switches toward a black or a white foreground . However , there is also a second kind of pupil modulation that depends on the direction of the shift: the pupil was more dilated when the foreground was black and more constricted when it was white . This is similar to the pupillary modulation that occurs during binocular rivalry between stimuli of different luminances , leading to pupillary constriction when the brighter stimulus dominates ( Naber et al . , 2011 ) . This luminance-dependent modulation is a purely perceptual effect , as the stimulus luminance is constant throughout the experiment . The constriction when the foreground was white probably results from participants attending more to the foreground than the background , as it is known that attending to white constricts the pupil ( Binda et al . , 2013 ) . On the other hand , if participants were to distribute attention evenly between the back and the front surfaces ( or rapidly switch attention between them ) , no pupil difference would be expected . As these two strategies correspond well to the local and global perceptual styles associated with high and low autistic traits ( Happé and Frith , 2006; Grinter et al . , 2009 ) , pupil modulation should be more prominent in participants with higher AQ scores . Figure 1D shows that , in our sample of typical young adults , there was considerable variation in luminance-dependent pupil change pupil modulation . The signed amplitude of the modulation was significantly positive in 20 out of the 50 participants . Crucially , the amplitude of modulation , which we suggest is an index of perceptual style , was highly correlated with AQ scores ( r = 0 . 70 , [0 . 52:0 . 82] p<10−5 , BF >105 ) . Importantly , the correlation was specific to this particular pupillometric index: AQ scores did not correlate with the general dilation that followed all perceptual switches ( r = 0 . 06 , [-0 . 22:0 . 33] , p=0 . 678 , BF = 0 . 1: Figure 1C ) . To test the idea that the variability of pupil responses emerges from different ( local-global ) perceptual styles , we performed two further experiments . Firstly , we attempted to induce changes in viewing strategy by changing the instructions given to the participants , and looked for concomitant changes of pupil modulation . We retested twice more a small subset of 10 participants , under two different instruction sets: ‘try to attend to both surfaces and see the cylinder rotate as a single unit’; or ‘focus attention on the front surface alone’ . Most subjects reported that they found the instructions difficult to follow , and that they tended to lapse into their more natural viewing style . Nevertheless , the instructions significantly affected the luminance-dependent pupil modulation ( 0 . 005 ± 0 . 007 mm when instruction favoured the global strategy; 0 . 029 ± 0 . 008 mm when they favoured the local strategy ) , which was greater when the local strategy was encouraged ( paired t-test: t: ( 9 ) = −2 . 72 , p=0 . 024 ) . This supports the suggestion that pupil modulation is driven by perceptual style , and further also shows that the pupil can be a more sensitive ( as well as more objective ) probe of perceptual style than are subjective reports . In a second experiment , we modified the procedure to encourage all participants to view the stimulus in the same local manner , attending to only one surface . Instead of the continuous presentation used before , we presented brief ( 6 s ) bursts of the stimulus and instructed subjects to which surface they should attend ( Figure 2A ) . To monitor compliance , participants reported the number of speed changes in the dots of the cued surface ( ignoring changes in the other surface ) , which they did with an average of 62 ± 2% correct responses ( uncorrelated with AQ , r = 0 . 24 [-0 . 04 0 . 49] , p=0 . 090 , BF = 0 . 5 ) . Under these conditions , there was a strong modulation of pupil size in the averaged data , depending on whether the white or the dark surface was attended ( Figure 2B ) , consistent with the reported effects of feature-based attention on pupil size ( Binda et al . , 2014 ) . However , in this experiment , where viewing style was constrained by the task , all subjects tended to show front-color dependent pupil modulation , and it was not correlated with AQ ( r = 0 . 20 [-0 . 08:0 . 46] , p=0 . 156 , BF = 0 . 3 , Figure 2C ) . Having established that the correlation between pupil difference and AQ scores is specific to the free-viewing of our bistable stimulus , we went on to check the reliability of the effect . We measured the correlation separately in the two groups of 25 participants who participated in the two sessions of the experiment ( and are combined in the plots of Figure 1B–D ) . The correlation between AQ and pupil difference was strong and significant in both groups ( Figure 3A , first subset: r = 0 . 75 , p<10−4 , BF >103; Figure 3B , second subset: r = 0 . 64 , p<0 . 001 , BF = 54 . 3 ) . We further replicated the effect in a slightly different condition , tested in 26 participants ( swapping the motion directions of white and black dots , Figure 3C , r = 0 . 66 p<0 . 001 , BF = 85 . 3 ) ; the three correlation coefficients are statistically indistinguishable ( Fisher's Z test; all p-values>0 . 23 ) , implying that the results are robust . As a further check on the reliability of the results , we also analyzed the correlations between pupil difference in the 50-participant group ( Figure 1D ) , and each of the five subscales of the AQ questionnaire ( normally distributed Figure 3D ) . All but one correlation coefficients were positive and significant ( Figure 3E ) . The strongest correlations were with the Communication and Social Skills subscales , and weakest with ‘Attention to Detail’ scores ( discussed later ) . The bottom row of Figure 3E shows the correlation of each subscale with the rest of the questionnaire ( the sum of the other four subscales ) , and also the pupil-difference index with the overall AQ score . Note that the correlation of the pupil-difference index with the total AQ score was higher than those between any subscale and the other four subscales . The results reported in Figures 1–3 show that pupil modulation reliably indexes attention to the front or both surfaces , depending on the participant’s AQ score . In principle , perceptual style should also be measurable by psychophysical techniques , with enhanced performance on the attended surface ( s ) ( Carrasco , 2011 ) . To compare our novel pupillometry approach with more standard psychophysical methods , we measured correlations between AQ and ability to detect subtle changes on the front or rear surface . As in the main experiment , participants continuously tracked the rotation of the cylinder over 10 min long sessions; but here they also reported brief changes in speed that occurred randomly on the surface formed either by white or black dots , at the front or the rear depending to the perceived rotation of the cylinder . As in the main experiment , pupils dilate more when the foreground is black compared with when it is white ( Figure 4A ) . Speed-change detection performance was generally better when targets occurred on the front than on the rear surface ( Figure 4C paired t-test on d-prime values: t ( 24 ) =4 . 43 , p<0 . 001 ) . However , AQ correlated neither with the difference in performance ( Figure 4D , r = −0 . 03 [-0 . 42 0 . 37] , p=0 . 869 , BF = 0 . 2 ) nor with the pupil difference ( Figure 4B r = −0 . 27 [-0 . 60 0 . 14] , p=0 . 187 , BF = 0 . 4 ) . This shows that the double-task setting ( necessary to obtain a psychophysical index of attention distribution ) interfered with participants’ natural viewing style , biasing all towards a ‘local’ processing style , focusing on one surface at a time to detect the speed changes . In most cases , the surface was the front one . This clearly undermined the possibility of estimating the inter-individual differences that spontaneously emerge in the free-viewing conditions of the main experiment . We measured pupil size while subjects viewed a structure-from-motion stimulus perceived as a cylinder rotating in 3D with bistable direction . In some but not all participants , pupil-size modulated in synchrony with their bistable perception , more constricted when the white surface was perceived in front . The magnitude of the modulation was tightly correlated with the Autistic Quotient scores , consistent with the view that stronger autistic traits accompany a preference for focusing on local detail , as opposed to globally attending the whole stimulus configuration . Some participants spontaneously reported their perceptual experience: of these some reported that they attending locally to only the front surface , while others reported that they had to attend globally to the whole configuration , both front and rear , in order to perceive the rotation . These spontaneous reports ( respectively from participants with high and low AQ ) provide qualitative support for our interpretation . Although preference for local versus global has been extensively investigated and has come to be a defining feature of Autistic Spectrum Disorders ( Van der Hallen et al . , 2015; Muth et al . , 2014 ) , its underlying mechanisms are essentially unknown . It could be a perceptual style , depending on differences in the bottom-up transmission of sensory information ( Mottron et al . , 2006; Mottron et al . , 2003; Plaisted et al . , 2003 ) , but it could also have a cognitive or attentional basis , depending on a preference to maintain a narrow focus of attention ( Happé and Frith , 2006; Plaisted et al . , 1999 ) or a relative inability to divide or switch attention between multiple targets ( Behrmann et al . , 2006; Frith and Happé , 1994; Happé and Frith , 1996; Happé and Booth , 2008 ) . Our results are equally supportive of all three possibilities , as are the classic tests for global-local preference , such as Navon letters or embedded figures . While our results do not distinguish between the different possible explanations of perceptual styles , they clearly show that autistic traits are associated with a more local perceptual style . There is an ongoing dispute about whether autistic traits are distributed continuously amongst the population with ASD diagnosis at an extreme , or whether the distinction is more categorical ( Baron-Cohen et al . , 2001; Constantino and Todd , 2003; Ruzich et al . , 2015; Skuse et al . , 2009; Wheelwright et al . , 2010 ) . The hypothesis of a ‘Broader Autistic Phenotype’ strongly predicts that local-global preference should be continuously distributed and correlated with autistic traits in the neurotypical population ( Cribb et al . , 2016 ) . However , many of the classic tests show very low or insignificant correlations with autistic traits ( Cribb et al . , 2016 ) . Our experiment describes a pupillometry index with the strongest correlation with autistic traits reported so far in a large group of participants , explaining about 50% of the variance in AQ scores . This correlation may appear surprisingly high , given that it is measured between two different tasks . However , each of these separate tasks has high reliability . The test-retest reliability of the autistic questionnaire has been reported to be around 0 . 75 ( Baron-Cohen et al . , 2001; Hoekstra et al . , 2008; Kurita et al . , 2005 ) . In the 22 participants who performed both the main experiment and the swapped-direction , the correlation of pupil modulation in the two tasks was 0 . 68 ( [0 . 36:0 . 86] , p<0 . 001 , BF = 62 . 7 ) . Considering the 95% confidence interval range ( r = 0 . 70 , [0 . 52:0 . 82] ) , the obtained correlations are not unrealistic , and certainly highly significant . Although the autistic questionnaire is well validated ( Baron-Cohen et al . , 2001 ) , not all items have the same validity . We also observed variability in the strength of correlation of the pupil index and the AQ subscales , which was strongest for the Communication and Social Skills subscales and , somewhat surprisingly , weakest for the ‘Attention to Detail’ scores . However , the ‘Attention to Detail’ subscale seems to have the lowest construct validity of the five subscales: it had the lowest performance in classifying individuals with and without ASD diagnosis in a sample of 350 adults , with 130 ASD diagnoses ( Lundqvist and Lindner , 2017 ) , it had the lowest correlation with the general AQ score in a sample of 2343 typical individuals ( Palmer et al . , 2015 ) , and , in our sample , scores on the ‘Attention to Detail’ subscale did not correlate significantly with the overall AQ score . It is also worth noting that among the items contributing to the ‘Attention to Detail’ subscale , only four actually relate to perceptual phenomena ( e . g . ‘I usually concentrate more on the whole picture , rather than the small details . ” ) , the others being focused on memory and cognition ( e . g . ‘I am fascinated by numbers . ” ) . All this highlights a general difficulty in measuring local-global preference , either with self-report or with laboratory tests . However , by combining the objectivity of pupillometry and the flexibility of a perceptual task involving bistable perception , we can achieve a very reliable and precise measure of inter-individual differences . There is growing interest in relating differences among individuals in their performance in perceptual tasks , with states or traits of their personality ( Tadin , 2015; Eldar et al . , 2013; Antinori et al . , 2016; Wilson et al . , 2016; Maclean and Arnell , 2010 ) . However , the amount of explained variance is usually lower than in our experiment , which is a rare case where a physiological measure obtained from a laboratory experiment explains a large portion of inter-individual differences in the way we feel and behave . To obtain the clear correlation with AQ , free-viewing of the bistable stimulus was essential: the correlation between pupillometry and AQ scores was eliminated when cueing attention , either explicitly , by instructing participants to attend globally/locally or by indicating that the surface formed by white or black dots alone is task-relevant , or implicitly , by introducing a task that is best performed by attending a single surface at a time . Although previous work has shown differences in bistable viewing that are related to autism [ ( Robertson et al . , 2013 ) ; but see also ( Said et al . , 2013 ) ] , we found that the dynamics of perceptual oscillations between the two directions of motion were uncorrelated with autistic traits . The correlation with AQ was only revealed only when taking pupillometry measures to index how attention was distributed during the bistable perception . The current study strengthens previous claims that pupil size is modulated not only by light , but also by perceptual ‘top-down’ processes ( Binda and Murray , 2015; Mathôt and Van der Stigchel , 2015; Laeng and Endestad , 2012 ) . The neural substrates of these modulatory mechanisms are beginning to be unveiled , involving pre-frontal input into the circuit of the pupillary response to light ( Ebitz and Moore , 2017 ) , which in turn may include the visual cortex ( Binda and Gamlin , 2017 ) . These multiple influences converge into a simple subcortical system that controls a unidimensional variable: pupil size . As we show here , under appropriate conditions ( free viewing a bistable stimulus with brighter and darker components ) , this variable can provide an objective physiological correlate of conscious perception , so reliable that it can successfully reveal subtle inter-individual differences that are hard to study psychophysically . The method is quick , easy , and objective , and requires only relatively simple , portable apparatus . We hope it can form the basis for tests suitable for clinical populations , possibly providing a new powerful tool for identifying anomalies in perception that are predictive of ASD . We recruited a total of 62 subjects ( 42 women; age ( mean ± SD ) : 25 . 53 ± 4 . 04 ) , in three groups ( 25 in the first , 26 in the second and 11 in the last ) . All were students from the University of Pisa or Florence , in at least their third year . All reported normal or corrected-to-normal vision , and had no diagnosed neurological condition . The number of participants recruited for the study was selected to provide a large effect size as indicated by a priori power analysis ( effect size: 0 . 50 , α = 0 . 05 , two-tail ) that reveals that in order to reach a power ( 1−β ) of 0 . 8 a sample size of 26 subjects was needed . Experimental procedures were approved by the regional ethics committee [Comitato Etico Pediatrico Regionale—Azienda Ospedaliero-Universitaria Meyer—Firenze ( FI ) ] and are in line with the declaration of Helsinki; participants gave their written informed consent . All participants completed the Autistic-traits Quotient questionnaire , self-administered with the validated Italian version ( Baron-Cohen et al . , 2001; Ruta et al . , 2012 ) . This contains 50 items , grouped in five subscales: Attention Switching , Attention to Detail , Imagination , Communication and Social Skills . For each question , participants read a statement and selected the degree to which the statement best described them: ‘‘strongly agree’’ , ‘‘slightly agree’’ , ‘‘slightly disagree’’ , and ‘‘strongly disagree’’ ( in Italian ) . Items were scored in the standard manner as described in the original paper ( Baron-Cohen et al . , 2001 ) : 1 when the participant’s response was characteristic of ASD ( slightly or strongly ) , 0 otherwise . Total scores ranged between 0 and 50 , with higher scores indicating higher degrees of autistic traits . All tested subjects scored below 32 , which is the threshold above which a clinical assessment is recommended ( Baron-Cohen et al . , 2001 ) . The mean ( SD ) of the scores was 14 . 85 ( 6 . 73 ) ; scores were normally distributed ( see Figure 3D ) , as measured by the Jarque-Bera goodness-of-fit test of composite normality ( JB = 1 . 42 p=0 . 37 ) . The experiment was performed in a quiet room with artificial illumination of 100 lux . Subjects sat in front of a monitor screen , subtending 41 × 30° at 57 cm distance , with their heads stabilized by chin rest . Viewing was binocular . Stimuli were generated with the PsychoPhysics Toolbox routines ( Brainard , 1997; Pelli , 1997 ) for MATLAB ( MATLAB r2010a , The MathWorks ) and presented on a 22-inch CRT colour monitor ( 120 Hz , 800 × 600 pixels; Barco Calibrator ) , driven by a Macbook Pro Retina ( OS X Yosemite , 10 . 10 . 5 ) . Two-dimensional eye position and pupil diameter were monitored either with a CRS LiveTrack system ( Cambridge Research Systems ) at 60 Hz , or with an Eyelink1000 Plus ( SR Research ) at 1000 Hz . We verified that although the two systems have different precision and accuracy , they yielded comparable results in our experiments . Both systems use an infrared camera mounted below the screen . Pupil diameter measures were transformed from pixels to millimeters after calibrating the tracker with an artificial 4 mm pupil , positioned at the approximate location of the subjects’ left eye . Eye position recordings were linearized by means of a standard 9-point calibration routine performed at the beginning of each session . Different subsets of participants took part in the four experiments ( main , swapped motion directions , feature-based attention , double-task ) . For the ‘main’ experiment , we recruited a total of 51 participants of which one was excluded ( see below ) . We recruited and tested them in two groups ( subjects 1–25 and 26–51 ) , intended as self-replications each with 25 participants ( after the exclusion of one participant based on criteria detailed below ) . Trials began with subjects fixating a red dot ( 0 . 15° diameter ) shown at the centre of a grey background ( 12 . 4 cd/m2 ) . The stimulus comprised a centrally positioned 8 × 14° rectangle which appeared to be a cylinder rotating about its vertical axis ( Figure 1A ) . The 3D illusion was generated by presenting a total of 300 randomly positioned dots ( each 0 . 30° diameter ) moving around a virtual vertical axis with an angular velocity of 60 deg/s ( 10 rotations per minute ) : the linear velocity followed a cosine function , 3 . 9°/s at screen centre . Dots were black ( 0 . 05 cd/m2 ) when they moved rightwards ( half at any one time ) and white ( 55 cd/m2 ) when they moved leftwards . The resulting stimulus was compatible with two perceptual interpretations: a cylinder rotating anticlockwise ( when viewed from above ) with black surface in the front and white surface at rear; or clockwise , with white surface in the front , black surface at rear . The two perceptual interpretations alternated spontaneously in all participants , who continuously reported their percept ( clockwise or anticlockwise rotation of the cylinder ) , either by holding down one of two keyboard arrow keys or by joystick . There was no response button for uncertain or mixed percepts: subjects were instructed to report which of the two percepts was dominant if in doubt . The stimulus was played for 10 trials of 59 s each , during which participants continuously reported whether the rotation was clockwise or anticlockwise . Participants were instructed to minimize blinks and maintain their gaze on the fixation spot at all times , except during a 1 s inter-trial pause , marked by a change of colour of the fixation spot ( which turned from red to black ) . Each participant completed a minimum of three runs , in a single session . A subsample of 27 participants ( 19 of the first group of participants , 4 of the second group , 4 of the last group , one excluded as explained below ) were also tested in the ‘swapped motion direction’ experiment – same as in the ‘main’ experiment , except that black dots moved leftward , and white dots moved rightward . A small subset of participants ( N = 10 ) were re-tested with the same stimuli and procedures as in the ‘main experiment’ , but different instructions . These were meant to explicitly encourage a global or a local distribution of attention . In two separate sessions ( randomized order ) , participants were either told to ‘try to attend to both surfaces and see the cylinder rotate as a single unit’ ( encouraging global viewing ) , or to ‘focus attention on the front surface alone’ ( encouraging local viewing ) . Each session lasted about 20 min and included two runs of ten trials each . Another subsample of 25 participants ( 6 from the first group , 9 from the second group , 10 from the last , chosen for the disposability for a second testing session ) took part in the ‘double-task’ experiment , with the same trial structure as the ‘main’ experiment . The primary task was unaltered , with the participant reporting whether they perceived clockwise or anticlockwise rotation of the cylinder ( using a joystick rather than the keyboard to minimize interference with the secondary task ) . Meanwhile , subtle speed increments ( 1 frame duration , 600 deg/s angular velocity increment ) occurred for either the black or the white dots ( forming the front or the rear surface depending on the participant’s perception ) , every 3 s on average ( with 2 s minimum separation between speed increments ) . Participants were asked to press the space bar as soon as they detected a speed increment ( on either surface ) . Any bar press within 2 s from a speed increment was counted as a hit; any bar press that happened more than 2 s away from any speed change was counted as a false alarm . D-prime values were computed from z-transformed hits and false alarms , separately for speed increments occurring on the front and rear surface . For each of these conditions , a minimum of two runs of 10 trials each were acquired ( approximately 20 min ) . Finally , 50 participants ( 18 from the first group , 22 from the second group , 10 from the last one ) were tested in the ‘feature-based attention’ experiment . Trials were only 10 s long . During a pre-stimulus 2 s interval , no dots were shown and a letter ( 0 . 5° wide , either ‘B’ or ‘N’ , for bianco or nero , Italian for white or black ) was shown at fixation and cued subjects to attend selectively to only white or black dots . Next came the two groups of dots , moving with the same direction and speed as in the main experiment , but lasting only 6 s . During this time , between 0 and 3 speed increments could occur on the cued and uncued surface . Upon extinction of the dots , the participant had 2 s to report by keypress how many speed increments occurred on the cued surface , ignoring speed increments on the uncued surface . In this case the participants did not report the perceived direction of rotation of the cylinder , which may or may not have perceived as a 3D object rather than two independent clouds of dots . Participants performed well above chance , with an average d-prime of 2 . 36 ± 0 . 09 . We tracked pupil diameter during the 6 s stimulus interval , separating trials where the white and the black dots were cued . This experiment was performed in two runs of 50 trials each ( approximately 15 min ) . An off-line analysis examined the eye-tracking output to exclude time-points with unrealistic pupil-size recordings ( smaller than 1 mm , likely due to blinks , or larger than 7 mm , likely due to eyelash interference ) . We further excluded perceptual phases lasting less than 1 s ( often finger errors ) and longer than 15 s ( marking trials with too few oscillations to measure bistability ) . These criteria led to the exclusion of one participant for the ‘main’ experiment , and one for the ‘swapped motion direction’ experiment ( who had less than 10 usable phases ) , leaving 50 participants for the ‘main’ experiment , for which the percentage of excluded phases is 27 . 39 ± 2 . 18% , and 26 participants for the ‘swapped motion direction’ and ‘double-task’ experiments , for which the percentages of excluded phases were similar ( 23 . 32 ± 2 . 98% and 22 . 58 ± 3 . 21% respectively ) . No trials and no participants were excluded for the ‘feature-based attention’ experiment . In all three bistable experiments ( the ‘main’ experiment , the ‘swapped motion direction’ and ‘double-task’ experiments ) , dark and white dots were equally likely seen as foreground ( percentage of time of the dark foreground percept , respectively: 51 . 85 ± 0 . 99% , 51 . 13 ± 1 . 43% , 51 . 34 ± 0 . 99% , never significantly different from 50% ) . Pupil traces were parsed into epochs locked to each perceptual switch ( when the subject changed reported perception ) . We aligned traces to the switch ( zero in Figure 1B ) , and labeled the phases according to the perceived direction of rotation . For the ‘feature-based attention’ experiment , pupil traces were time-locked to stimulus onset and separated based on whether the black or white surface was cued ( Figure 2B ) . For all experiments , we subtracted from each trace a baseline measure of pupil size , defined as the mean pupil size in the 150 ms immediately preceding or following the switch ( for negative and positive traces , respectively ) , or stimulus onset for the ‘feature based attention’ experiment . The resulting traces were averaged across trials and participants , separately for the two perceptual phases ( or attention cues for the feature-based attention experiment ) , to give , Figure 1B and Figure 2B . From these , and also for the individual traces , we defined two summary statistics: the difference of pupil traces between the two types of epochs , and overall mean pupil trace across all epochs . For the ‘main’ , the ‘swapped motion direction’ and the ‘double-task’ experiments , these indices were computed after averaging pupil measurements over the first ( or first and last ) 1 s of each epoch , the minimum phase duration , ensuring that all phases contribute equally to the mean . However , we also verified our main result ( correlations with AQ scores ) with different epoch definitions . For the ‘feature-based attention’ experiment , the difference of pupil traces between trials where the white and black dots were cued was computed in the interval between 1 and 3 s from stimulus onset ( where the effect of attention is expected to peak [Binda et al . , 2014] ) . The datasets generated during the current study and scripts used for analyses are available on Dryad , at the following address: http://dx . doi . org/10 . 5061/dryad . b3ng3	The pupils control how much light reaches the eye . They become smaller in bright light and larger in darkness to let more light in . Other factors can also affect pupil size . For example , the pupils slightly constrict when a person focuses on brighter objects and they enlarge when focusing on a darker object . Tracking changes in pupil size can tell scientists what someone is focusing on . This can be helpful because different people distribute their attention differently . Some tend to focus on the big picture , others tune into individual details . People with autism spectrum disorders ( ASD ) tend to focus on the details . Now , Turi et al . showed that measuring pupil size changes during a simple visual task could identify typical people who have milder versions of the characteristics seen in people with ASD . In the experiments , 50 young adults without a diagnosis of an ASD filled out a questionnaire designed to assess how many ASD-like traits they have . Next , the participants watched an illusion of a cylinder with a light and dark side rotating . As they watched , the pupil size of people with more ASD-linked behaviors fluctuated more than the pupil size of those with few such characteristics . The pupils of people with ASD-type traits became larger when they perceived the dark side of the cylinder to be forward , and smaller when the light side appeared . This suggests they are focusing on the front of the cylinder . Future studies are needed to see if similar pupil fluctuations occur in people diagnosed with ASD . Turi et al . predict that pupil changes will be even more dramatic in people with ASD . If this is the case , these pupil measurements could be used to help diagnose ASD or determine the severity of symptoms .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Pupillometry reveals perceptual differences that are tightly linked to autistic traits in typical adults
The spatial organization of gut microbiota influences both microbial abundances and host-microbe interactions , but the underlying rules relating bacterial dynamics to large-scale structure remain unclear . To this end , we studied experimentally and theoretically the formation of three-dimensional bacterial clusters , a key parameter controlling susceptibility to intestinal transport and access to the epithelium . Inspired by models of structure formation in soft materials , we sought to understand how the distribution of gut bacterial cluster sizes emerges from bacterial-scale kinetics . Analyzing imaging-derived data on cluster sizes for eight different bacterial strains in the larval zebrafish gut , we find a common family of size distributions that decay approximately as power laws with exponents close to −2 , becoming shallower for large clusters in a strain-dependent manner . We show that this type of distribution arises naturally from a Yule-Simons-type process in which bacteria grow within clusters and can escape from them , coupled to an aggregation process that tends to condense the system toward a single massive cluster , reminiscent of gel formation . Together , these results point to the existence of general , biophysical principles governing the spatial organization of the gut microbiome that may be useful for inferring fast-timescale dynamics that are experimentally inaccessible . The bacteria inhabiting the gastrointestinal tracts of humans and other animals make up some of the densest and most diverse microbial ecosystems on Earth ( Lloyd-Price et al . , 2017; Sender et al . , 2016 ) . In both macroecological contexts and non-gut microbial ecosystems , spatial organization is well known to impact both intra- and inter-species interactions ( McNally et al . , 2017; Tilman and Kareiva , 2018; Weiner et al . , 2019 ) . This general principle is likely to apply in the intestine as well , and the spatial structure of the gut microbiome is increasingly proposed as an important factor influencing both microbial population dynamics and health-relevant host processes ( Tropini et al . , 2017; Donaldson et al . , 2016 ) . Moreover , recent work has uncovered strong and specific consequences of spatial organization in the gut , such as proximity of bacteria to the epithelial boundary determining the strength of host-microbe interactions ( Vaishnava et al . , 2011; Wiles et al . , 2020 ) , and antibiotic-induced changes in aggregation causing large declines in gut bacterial abundance ( Schlomann et al . , 2019 ) . Despite its importance , the physical organization of bacteria within the intestine remains poorly understood , in terms of both in vivo data that characterize spatial structure and quantitative models that explain the mechanisms by which structure arises . Recent advances in the ability to image gut microbial communities in model animals have begun to reveal features of bacterial spatial organization common to multiple host species . Bacteria in the gut exist predominantly in the form of three-dimensional , multicellular aggregates , likely encased in mucus , whose sizes can span several orders of magnitude . Such aggregates have been observed in mice ( Moor et al . , 2017 ) , fruit flies ( Koyama et al . , 2020 ) , and zebrafish ( Jemielita et al . , 2014; Schlomann et al . , 2018; Wiles et al . , 2016; Schlomann et al . , 2019; Wiles et al . , 2020 ) , as well as in human fecal samples ( van der Waaij et al . , 1996 ) . However , an understanding of the processes that generate these structures is lacking . The statistical distribution of object sizes can provide powerful insights into underlying generative mechanisms , a perspective that has long been applied to datasets as diverse as galaxy cluster sizes ( Hansen et al . , 2005 ) , droplet sizes in emulsions ( Lifshitz and Slyozov , 1961 ) , allele frequency distributions in population genetics ( Neher and Hallatschek , 2013 ) , immune receptor repertoires ( Nourmohammad et al . , 2019 ) , species abundance distributions in ecology ( Hubbell , 1997 ) , protein aggregates within cells ( Greenfield et al . , 2009 ) , and linear chains of bacteria generated by antibody binding ( Bansept et al . , 2019 ) . A classic example of the understanding provided by examining size distributions comes from the study of gels . In polymer solutions , random thermal motion opposes the adhesion of molecules , resulting in cluster size distributions dominated by monomers and small clusters . Gels form as adhesion strength increases , and monomers stick to one another strongly enough to overcome thermal motion and form a giant connected cluster that spans the size of the system . This large-scale connectivity gives gels their familiar stiffness as seen , for example , in the wobbling of a set custard . Theoretical tools from statistical mechanics and the study of phase transitions relate the cluster size distribution to the inter-monomer attraction strength and the temperature ( Krapivsky et al . , 2010 ) . In addition to providing an example of the utility of analyzing size distributions , gels in particular are a ubiquitous state of matter in living systems whose physical properties influence a wide range of activities such as protection at intestinal mucosal barriers ( Datta et al . , 2016 ) and transport of molecules through amyloid plaques ( Woodard et al . , 2014 ) . Motivated by these analogies , we sought to understand the distribution of three-dimensional bacterial cluster sizes in the living vertebrate gut , aiming especially to construct a quantitative theory that connects bacterial-scale dynamics to global size distributions . Such a model could be used to infer dynamical information in systems that are not amenable to direct observation , such as the human gut . Identifying key processes that are conserved across animal hosts would further our ability to translate findings in model organisms to human health-related problems . At a finer level , validated mathematical models could be used to infer model parameters of specific bacterial species of interest , for example pathogenic invaders or deliberately introduced probiotic species , by measuring their cluster size distribution . We analyzed bacterial cluster sizes obtained from recent imaging-based studies of the larval zebrafish intestine ( Schlomann et al . , 2018; Schlomann et al . , 2019; Wiles et al . , 2020 ) . As detailed below , we find a common family of cluster size distributions with bacterial species-specific features . We show that these distributions arise naturally in a minimal model of bacterial dynamics that is supported by direct observation . The core mechanism of this model involves growth together with a fragmentation process in which single cells leave larger aggregates . Strikingly , this process can be mapped exactly onto population genetics models of mutation , with cluster size analogous to allele frequency and single-cell fragmentation analogous to mutation . The combination of growth and fragmentation generates size distributions with power law tails , consistent with the data . This process also maps onto classic network models of preferential attachment ( Barabasi and Albert , 1999 ) . Further , we show that cluster aggregation can generate an overabundance of large clusters through a process analogous to the sol-gel transition in polymer and colloidal systems , leading to plateaus in the size distribution that are observed in the data . These features of the size distribution are robust to the inclusion of a finite carrying capacity that limits growth and cluster loss due to expulsion from the intestine . In summary , we find that gut microbiota can be described mathematically as 'living gels' , combining the statistical features of evolutionary dynamics with those of soft materials . Based on the generality of our model and our observations across several different bacterial species , we predict that this family of size distributions is universal across animal hosts , and we provide suggestions for testing this prediction in various systems . We combined and analyzed previously generated datasets of gut bacterial cluster sizes in larval zebrafish ( Schlomann et al . , 2018; Wiles et al . , 2020 ) . In these experiments , zebrafish were reared devoid of any microbes , that is ‘germ-free’ , and then mono-associated with a single , fluorescently labeled bacterial strain ( Figure 1A ) . After a 24 hr colonization period the complete intestines of live hosts were imaged with light sheet fluorescence microscopy ( Keller et al . , 2008; Parthasarathy , 2018; Figure 1B ) . Bacteria were identified in the images ( Figure 1C ) using a previously described image analysis pipeline ( Jemielita et al . , 2014; Schlomann et al . , 2018 ) . Single bacterial cells and multicellular aggregates were identified separately , and then the number of cells per multicellular aggregate was estimated by dividing the total fluorescence intensity of the aggregate by the mean intensity of single cells ( Materials and methods ) . In total , we characterized eight different bacterial strains , summarized in Table 1 . Six of the strains were isolated from healthy zebrafish ( Stephens et al . , 2016 ) and then engineered to express fluorescent proteins ( Wiles et al . , 2018 ) , and two are genetically engineered knockout mutants of Vibrio ZWU0020 , defective in motility ( specifically , knockout of the two-gene operon encoding the polar flagellar motor , pomAB , referred to as ‘Δmot’ ) and chemotaxis ( specifically , knockout of the histidine kinase cheA2 , referred to as ‘Δche’ ) , as described in reference ( Wiles et al . , 2020 ) . The parent strain of these mutants , Vibrio ZWU0020 , scarcely forms aggregates at all , existing primarily as single , highly motile cells ( Wiles et al . , 2016; Schlomann et al . , 2019; Wiles et al . , 2020 ) , and so was excluded from this analysis . All strains are of the phylum Proteobacteria ( Wiles et al . , 2018 ) . A table of all cluster sizes by sample is included in Figure 2—source data 1 . We calculated for each bacterial strain the reverse cumulative distribution of cluster sizes , P ( size>n ) , denoting the probability that an intestinal aggregate will contain more than n bacterial cells . We computed P ( size>n ) separately for each animal ( Figure 2 , small circles ) and also pooled the sizes from different animals colonized by the same bacterial strain ( Figure 2 , large circles ) . There is substantial variation across fish , but the pooled distributions exhibit a well-defined average of the individual distributions . We also computed binned probability densities ( Figure 2—figure supplement 1 ) , which show similar patterns , but focus our discussion on the cumulative distribution to circumvent technical issues related to bin sizes . We find broad distributions of P ( size>n ) across all strains ( Figure 2 , bottom right panel ) . For comparison , for each strain we overlay a dashed line representing the power law distribution P ( size>n ) ∼n-1 . This P ( size>n ) is equivalent to a probability density of p⁢ ( n ) ∼n-2 since the latter is proportional to the derivative of the former . Each strain's cumulative distribution follows a similar power-law-like decay at low n , with an apparent exponent in the vicinity of -1 , and then becomes shallower in a strain-dependent manner . For example , Aeromonas ZOR0002 has a quite straight distribution on a log-log plot ( Figure 2 , top row , middle column ) , while the distribution of Enterobacter ZOR0014 exhibits a plateau-like feature at large sizes ( Figure 2 , top row , right column ) . The mutant strains Vibrio ZWU0020 Δche and Δmot follow qualitatively similar distributions to the native strains ( Figure 2 , bottom row , left and middle columns ) . We performed a sensitivity analysis and found that these two key features of the measured distributions—an initial power law-like decay with cumulative distribution exponent close to -1 and a strain-dependent plateau at large sizes—are robust to measurement error in enumeration of cluster sizes . For the initial decay of the distribution , the largest source of error is the misidentification of auto-fluorescent background as single cells . To assess the impact of our single-cell count uncertainty on the distribution , we fit a power law model to clusters sizes up to 100 cells two times: once including single cells and once considering only cells of size in the range 2–100 ( Supplementary file 1 , Materials and methods ) . In both fits we find cumulative distribution exponents consistent with −1 for most strains . The average exponent tended to decrease mildly when single cells were excluded from the fit ( the distribution decayed more slowly ) , consistent with an over-estimation of the number of single cells , but the shifts were all within uncertainties . Estimates of distribution exponents from small sizes can easily be biased ( Clauset et al . , 2009 ) , so we performed our sensitivity analysis with two different methods: a linear fit to logP ( size>n ) vs . log⁡n , and maximum likelihood estimation ( Materials and methods ) . The maximum likelihood estimate gave higher values than line-fitting , but the shifts upon removing single cells were within uncertainties for both methods . For the large-size plateau , the existence of dim cells in the center of the aggregate , perhaps due to a state of low metabolic activity , would lead to an underestimate of total cluster size . Underestimating the size of large clusters would then result in a less extreme plateau; the plateaus we observe are therefore a lower bound . In cross-sections of large aggregates , we observe mostly homogeneous fluorescence , suggesting that this effect is mild , although small dark regions do occur ( Figure 2—figure supplement 2 ) . Whether these dark regions correspond to dead or inactive bacteria , mucus , or empty space , is not clear , although we note that small clumps of dead bacteria have been observed in expelled clusters via live/dead staining ( Schlomann et al . , 2019 ) . Regardless of their origin , we conclude that these mild heterogeneities are unlikely to significantly alter the behavior of the size distributions , which span 4 orders of magnitude . In summary , we find that different bacterial strains , which exhibit a variety of swimming and sticking behaviors ( Wiles et al . , 2018; Schlomann et al . , 2018 ) , abundances ( Schlomann et al . , 2018; Wiles et al . , 2020 ) , and population dynamics ( Wiles et al . , 2016; Schlomann et al . , 2019; Wiles et al . , 2020 ) , share a common family of cluster size distributions . This observation suggests that generic processes , rather than strain-specific ones , determine gut bacterial cluster sizes . Notably , these distributions are extremely broad , inconsistent with the exponential-tailed distributions found for linear chains of bacteria ( Bansept et al . , 2019 ) . We next sought to understand the kinetics that give rise to our measured cluster size distributions . We analyzed image-derived measurements of bacterial cluster sizes from larval zebrafish intestines and discovered a common family of size distributions shared across bacterial species . These distributions are extremely broad , exhibiting a power-law-like decay at small sizes that becomes shallower at large sizes in a strain-specific manner . We then demonstrated how these distributions emerge naturally from realistic kinetics: growth and single-cell fragmentation together generate power-law distributions , analogous to the distribution of neutral alleles in expanding populations , while size-dependent aggregation leads to a plateau representing the depletion of mid-sized clusters in favor for a single large one . In summary , we found that gut bacterial clusters are well-described by a model that combines the features of evolutionary dynamics in growing populations with those of inanimate systems of aggregating particles; intestinal bacteria form a ‘living gel’ . Gels are characterized by the emergence of a massive connected cluster that is on the order of the system size . In the larval zebrafish intestine , we often find for some bacterial species that the majority of the cells in the gut are contained within a single cluster , similar to a gel-like state . While growth by cell division generates large clusters , it is the aggregation process that leads to system-sized clusters being over-represented . This enhancement of massive clusters manifests as a plateau in the size distribution and is reminiscent of a true gelation phase transition . In our model , the prominence of this plateau appears to follow the same trend as in non-living , purely aggregating systems: the plateau depends strongly on the aggregation exponent that dictates the size-dependence of aggregation , with exponents larger than 1/2 leading to strong plateaus and exponents less than 1/2 leading to weak or no plateaus . How this strong size-dependence in aggregation emerges in the intestine is unclear , although we hypothesize that active mixing by intestinal contractions , which can in fact merge multiple clusters at once ( Schlomann et al . , 2019 ) , is an important driver . We envision that the exponents for aggregation and also for fragmentation are likely generic , set by physical aspects of the intestine and the geometry of clusters , while the rates of these processes are bacterial-species dependent . In our system , we predict that differences in aggregation and/or fragmentation rates between strains underly the differences in measured size distributions . Further , it is possible that individual bacteria can tune these rates by altering their behavior , for example , modulating swimming motility ( Wiles et al . , 2020 ) , in response to environmental cues . Quantitatively understanding how the combination of intestinal fluid mechanics and bacterial behaviors determine aggregation and fragmentation rates would be a fruitful avenue of future research . More abstractly , active growth combined with different aggregation processes , for example the fractal structures of diffusion-limited aggregation , may lead to different families of size distributions that would be interesting to explore . On the experimental front , direct measurements of aggregation and fragmentation rates from time-lapse imaging would be an extremely useful next step . However , these measurements are technically challenging . Even by eye , unambiguously identifying that a single bacterium fragmented out of a larger aggregate , and did not simply float into the field of view , requires faster imaging speeds than we can currently obtain . Sparse , two-color labeling may improve reliability of detection , but would decrease the frequency of observing an event . Automatic identification of fragmentation events in time-lapse movies is a daunting task , but recent computational advances , for example using convolutional neural networks to automatically identify cell division events in mouse embryos ( McDole et al . , 2018 ) , may provide a good starting point . Aggregation is easier to observe by eye , but its automatic identification presents similar challenges in analysis . Given the general and minimal nature of the model's assumptions , we predict that the form of the cluster size distributions we described here is common to the intestines of animals , including humans . This prediction of generality could be tested in a variety of systems using existing methods . In fruit flies , live imaging protocols have been developed that have revealed the presence of three-dimensional gut bacterial clusters highly reminiscent of what we observe in zebrafish , particularly in the midgut ( Koyama et al . , 2020 ) . Quantifying the sizes of these clusters would allow further tests of our model . In mice , substantial progress has been made in imaging histological slices of the intestine with the luminal contents preserved ( Tropini et al . , 2017 ) . Intestinal contents are very dense in the distal mouse colon , however , and it is not clear how one should define cluster size . Other intestinal regions are likely more amenable to cluster analysis . Moreover , with species-specific labeling , it is possible to measure the distribution of clonal regions in these dense areas ( Mark Welch et al . , 2017 ) . One could imagine then comparing these data to a spatially-explicit , multispecies extension of the model we studied here . Our model could also be tested indirectly for humans and other animals incompatible with direct imaging by way of fecal samples . Two decades ago , bacterial clusters spanning three orders of magnitude in volume were observed in gently dissociated fecal samples stained for mucus , but precise quantification of size statistics was not reported ( van der Waaij et al . , 1996 ) . Repeating these measurements with quantification , from for example imaging or flow cytometry , would also provide a test of our model , albeit on the microbiome as a whole rather than a single species at a time . The interpretation therefore would be of an effective species with kinetic rates representing average rates of different species . To close , we emphasize that the degree of bacterial clustering in the gut is an important parameter for both microbial population dynamics and host-bacteria interactions . More aggregation leads to larger fluctuations in abundance due to the expulsion of big clusters , and also thereby increase the likelihood of extinction ( Schlomann et al . , 2019; Schlomann and moments , 2018 ) . Further , aggregation within the intestinal lumen can reduce access to the epithelium and reduce pro-inflammatory signaling ( Wiles et al . , 2020 ) . Therefore , measurements of cluster sizes may be an important biomarker for microbiota-related health issues , and inference of dynamics from size statistics using models like this one could aid the development of therapeutics . We assembled data on gut bacterial cluster sizes from three different studies on larval zebrafish ( Schlomann et al . , 2018; Wiles et al . , 2020 ) . Size data from Schlomann et al . , 2018 and Schlomann et al . , 2019 were taken directly from the supplementary data files associated with those publications . The raw size data from Wiles et al . , 2020 was not included in its associated supplementary data file , but summary statistics such as planktonic fraction were . All sizes were rounded up to the nearest integer . Details of experimental procedures can be found in the original papers . In brief , as described in Figure 1 , animals were reared germ-free , mono-associated with a single bacterial strain , each carrying a chromosomal GFP tag , and then imaged 24 hr later using a custom-built light sheet fluorescence microscope ( Jemielita et al . , 2014 ) . The gut is imaged in four tiled sub-regions that are registered via cross-correlation and manual adjustment . Imaging a full gut volume ( ≈1200 μm × 300 μm × 150 μm ) with 1 μm slices takes approximately 45 s . Laser power ( 5 mW ) and exposure time ( 30 ms ) were identical for all experiments . The image analysis pipeline used to enumerate bacterial cluster sizes is also described in detail in the original publications and in reference ( Jemielita et al . , 2014 ) . In brief , single cells ( small objects ) and multicellular aggregates ( large objects ) are identified separately . The number of cells per aggregate is then estimated as the total fluorescence intensity of the aggregate divided by the mean fluorescence intensity of a single cell . Small objects are identified in three dimensions with a combination of difference-of-gaussians and wavelet filters ( Olivo-Marin , 2002 ) and then culled using a support vector machine classifier and manual curation . Large objects are segmented in maximum intensity projections using a graph-cut algorithm ( Boykov and Kolmogorov , 2004 ) seeded by either an intensity- or gradient-thresholded mask . The total intensity of an aggregate is computed by extending the two-dimensional mask in the z-direction and summing fluorescence intensities above a threshold calculated from the boundary of the mask , with pixels detected as part of single cells removed . The boundary of the gut is manually outlined prior to image analysis and used to exclude extra-intestinal fluorescence . For the experimental data , reverse cumulative distributions were computed as ( 2 ) P ( size>n ) =numberofclusterswithsize>ntotalnumberofclusters . In combining data from different samples colonized with the same strain , we pooled together all sizes and computed the distribution in the same way . For simulations with large numbers clusters , we computed this distribution iteratively , looping through each simulation replicate and independently updating ( number clusters with size >n ) and ( total number of clusters ) , and normalizing at the end . For the binned probability densities in Figure 2—figure supplement 1 , data were similarly pooled across samples and then sorted into logarithmically spaced bins of log10 width = 0 . 4 . We estimated approximate bounds on the rate of total aggregation events as follows . For the maximum rate , we note that a typical population contains approximately 200 clusters ( mean ± std . dev of 244 ± 182 ) . In the absence of other processes , condensing this system into one cluster would require 100 aggregation events . Populations consisting of almost entirely one large cluster are rare but have been documented ( Schlomann et al . , 2018 ) . Therefore , we estimate that this complete condensation can occur no more than once an hour , leading to an upper bound on the total rate of aggregation events of 100 per hour . For the minimum rate , we start with the observation that aggregation has been directly observed between small clusters and also between small clusters and a single large cluster during a large expulsion event ( Schlomann et al . , 2019 ) . Considering just the latter process , we know that large expulsion events happen roughly once every 10 hr . If approximately 10 small clusters are grouped into the large cluster during transit out of the gut , that would correspond 10 total aggregation events in 10 hr , or , 1 per hour , which we take as a lower bound . We used three different numerical approaches for studying the models discussed here . The minimal growth-fragmentation process in Figure 3 was simulated with a Poisson tau-leaping algorithm Gillespie , 2001 with a simple fixed tau value of τ=0 . 1 hr . At each time step , the number of growth and fragmentation events was drawn from a Poisson distribution with the rates given in Figure 3B along with the constraint that clusters must be of size two or larger to fragment . For the full model including aggregation and expulsion , we used Gillespie's algorithm Gillespie , 1977 for fragmentation , aggregation , and expulsion events , while growth was updated deterministically according to a continuous logistic growth law approximated by an Euler step with d⁢t=min⁢ ( τ , 0 . 1⁢ hr ) , where τ here refers to the time to next reaction . For the Gillespie steps , if the time to next reaction exceeded the doubling time , ( ln⁡2 ) /r , the growth steps were performed and then the propensity functions were re-calculated . Finally , we compared these stochastic simulations to a model in the thermodynamic limit where individual clusters are replaced with cluster densities that evolve deterministically , which is referred to as a master equation ( Krapivsky et al . , 2010 ) . The master equation for the general model readsc˙n= α2∑m=1n[ ( n−m ) ( m ) ]νAcn−mcm−αnνAcn∑mmνAcm+r ( 1−NK ) [ ( n−1 ) cn−1−ncn]−λnνEcn+β ( 1−NK ) ( ( n+1 ) νFcn+1−nνFcn+δn , 1∑mmνFcm ) . This set of equations was solved numerically on a bounded size grid using an Euler method with step size d⁢t=0 . 0001 hr . Models that include a carrying capacity , K , are already defined on a finite domain of integers ranging from one to K and the master equation is naturally represented by a set of K ordinary differential equations . For models without a carrying capacity , we introduced a maximum size given by the average population size at the last time point , nmax=exp⁢ ( r⁢tmax ) ( rounded up to the nearest integer ) , and used reflecting boundary conditions at nmax . A distribution was deemed stationary if it was visibly unchanged after an additional 50% of simulation time . MATLAB code for simulating these models and plotting data can be found at https://github . com/rplab/cluster_kinetics ( copy archived at swh:1:rev:f55a54a9c88e4fb8376dfc91e25ac4383c4240ae , Schlomann , 2021 ) . For the simulated distributions in Figure 3 we estimated a power law exponent using the maximum likelihood-based method described in Clauset et al . , 2009 and the plfit . m code supplied therein . This model includes a minimum size as a free parameter that dictates when the power-law tail begins . The minimum size is chosen to minimize the Kolmogorov-Smirnov distance between the data and model distributions for sizes greater than the minimum size . Best fit values of the exponent and minimum size are included in Figure 3—source data 1 . For the experimentally measured distributions , we used both maximum likelihood estimation and linear fitting to the log-transformed cumulative distribution to calculate exponents .	The human gut is home to vast numbers of bacteria that grow , compete and cooperate in a dynamic , densely packed space . The spatial arrangement of organisms – for example , if they are clumped together or broadly dispersed – plays a major role in all ecosystems; but how bacteria are organized in the human gut remains mysterious and difficult to investigate . Zebrafish larvae provide a powerful tool for studying microbes in the gut , as they are optically transparent and anatomically similar to other vertebrates , including humans . Furthermore , zebrafish can be easily manipulated so that one species of bacteria can be studied at a time . To investigate whether individual bacterial species are arranged in similar ways , Scholmann and Parthasarathy exposed zebrafish with no gut bacteria to one of eight different strains . Each species was then monitored using three-dimensional microscopy to see how the population shaped itself into clusters ( or colonies ) . Schlomann and Parthasarathy used this data to build a mathematical model that can predict the size of the clusters formed by different gut bacteria . This revealed that the spatial arrangement of each species depended on the same biological processes: bacterial growth , aggregation and fragmentation of clusters , and expulsion from the gut . These new details about how bacteria are organized in zebrafish may help scientists learn more about gut health in humans . Although it is not possible to peer into the human gut and watch how bacteria behave , scientists could use the same analysis method to study the size of bacterial colonies in fecal samples . This may provide further clues about how microbes are spatially arranged in the human gut and the biological processes underlying this formation .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "physics", "of", "living", "systems", "microbiology", "and", "infectious", "disease" ]	2021	Gut bacterial aggregates as living gels
Although combinatorial regulation is a common feature in gene regulatory networks , how it evolves and affects network structure and function is not well understood . In S . cerevisiae , the phosphate starvation ( PHO ) responsive transcription factors Pho4 and Pho2 are required for gene induction and survival during phosphate starvation . In the related human commensal C . glabrata , Pho4 is required but Pho2 is dispensable for survival in phosphate starvation and is only partially required for inducing PHO genes . Phylogenetic survey suggests that reduced dependence on Pho2 evolved in C . glabrata and closely related species . In S . cerevisiae , less Pho2-dependent Pho4 orthologs induce more genes . In C . glabrata , its Pho4 binds to more locations and induces three times as many genes as Pho4 in S . cerevisiae does . Our work shows how evolution of combinatorial regulation allows for rapid expansion of a gene regulatory network’s targets , possibly extending its physiological functions . Evolution of gene regulatory networks ( GRNs ) is a major source of phenotypic diversity ( Wray , 2007; Stern and Frankel , 2013; Prud'homme et al . , 2006; Gompel et al . , 2005; Jones et al . , 2012; Wang et al . , 1999 ) . One common feature of GRNs is combinatorial regulation by multiple transcription factors ( TFs ) – for example , the co-regulation of circadian gene expression in cyanobacteria by both the cell-autonomous clock and the external conditions ( Espinosa et al . , 2015 ) , and the determination of cell fates by multiple ‘selector genes’ in animal development ( Mann and Carroll , 2002 ) . Not only is combinatorial regulation important for GRN function , it also contributes to its evolution through changes in protein-protein or protein-DNA interactions ( Kirschner and Gerhart , 2006 , chap . 4; Tsong et al . , 2003; Baker et al . , 2012; Brayer et al . , 2011 ) . The consequences of such changes can be either conserved network output ( Tsong et al . , 2006 ) or evolution of new network function ( Tuch et al . , 2008 ) . Despite a rich literature on GRN evolution , few studies have documented the evolution of combinatorial regulation and its influence on network structure and function ( Tuch et al . , 2008; Baker et al . , 2012 ) . Moreover , the existing literature on GRN evolution is strongly biased towards developmental networks ( Stern , 2010; Peter and Davidson , 2011 ) . While such networks provide attractive attributes , such as visible phenotypes and well-resolved genetic underpinning , it has been suggested that network architecture strongly influences the tempo and mode of its evolution ( Erwin and Davidson , 2009; Wittkopp , 2007 ) . Therefore , it is unclear whether all GRNs follow similar or different rules during their evolution . To approach this question we studied the regulatory divergence in the phosphate starvation ( PHO ) response network in yeast . For three reasons , this system is well suited for our question . First , starvation/stress response networks differ in architecture from developmental networks , leading us to expect differences in their evolutionary patterns . Second , the GRN controlling the PHO response has been well studied in the model yeast S . cerevisiae , providing a solid foundation for comparative analyses . Third , the two species we focus on , S . cerevisiae and a human commensal and opportunistic pathogen C . glabrata , occupy distinct ecological niches but share ~90% of their gene repertoire and have an average of ~67% protein sequence identity ( Gabaldón et al . , 2013 ) , enabling us to trace the evolution of the network by studying the function of orthologous regulatory proteins . Phosphate is an essential nutrient for all organisms . To maintain phosphate homeostasis , S . cerevisiae activates a phosphate starvation pathway in response to limitation for inorganic phosphate ( Ogawa et al . , 2000 ) . In phosphate replete conditions , the transcription factor Pho4 is phosphorylated and localized to the cytoplasm , and phosphate response genes ( PHO genes ) are not expressed ( O'Neill et al . , 1996 ) . As the concentration of extracellular inorganic phosphate ( Pi ) drops , cells activate the phosphate starvation response and the dephosphorylated Pho4 is imported into the nucleus , where it functions together with the homeodomain transcription factor Pho2 to activate PHO gene expression ( O'Neill et al . , 1996; Vogel et al . , 1989; Barbarić et al . , 1996; Barbaric et al . , 1998; Shao et al . , 1996 ) . Although Pho4 binds to ~100 locations in the S . cerevisiae genome , it regulates fewer than 30 genes ( Zhou et al . , 2011 ) . Only genes at which Pho2 and Pho4 bind cooperatively in the promoter region are activated , indicating that Pho2 increases the selectivity of the gene set induced in response to phosphate starvation ( Zhou et al . , 2011 ) . In C . glabrata , Pho4 and Pho2 orthologs ( hereinafter referred to as CgPho4 and CgPho2 ) exist , but unlike Pho4 and Pho2 in S . cerevisiae ( hereinafter referred to as ScPho4 and ScPho2 ) , CgPho4 can induce gene expression in the absence of CgPho2 ( Kerwin and Wykoff , 2009 ) . This change in the dependence on the co-activator is not due to a higher expression level of CgPho4 or changes in the promoter regions of its target genes , and therefore is likely the result of alterations in the function of CgPho4 ( Kerwin and Wykoff , 2009 ) . We investigated the evolution of the PHO pathway in a diverse group of yeast species known as Hemiascomycetes ( Knop , 2006; Diezmann et al . , 2004 ) , which includes but is not limited to S . cerevisiae , C . glabrata , K . lactis , C . albicans and Y . lypolitica , and found that PHO4 and PHO2 are conserved as single copy genes in this group . We first evaluated Pho4 orthologs from a representative set of this group of species for their ability to activate gene expression in the absence of Pho2 in the S . cerevisiae background – this allowed us to establish that the reduced dependence on Pho2 evolved in a species clade that includes C . glabrata , two other human commensal yeasts and an environmental species . We then used functional genomics to assess the consequence of reduced Pho2 dependence on gene expression . Finally , we identified the bona fide targets of Pho4 in C . glabrata and compared them to Pho4 targets in S . cerevisiae . Our results show that evolution of combinatorial regulation , in terms of dependence of the major transcription factor on the co-activator , contributes significantly to the evolution of a gene regulatory network’s targets and functions , and as a result may lead to a new physiological response to the stress . We first confirmed the previously reported result that deleting PHO2 in C . glabrata does not eliminate expression of the secreted phosphatase encoded by PMU2 ( Orkwis et al . , 2010 ) ( Figure 1A ) , and then demonstrated that , in contrast to ScPho2 , CgPho2 is dispensable for survival in phosphate-limited conditions ( Figure 1B ) . It has been shown that CgPho4 is able to induce phosphatase expression in S . cerevisiae without ScPho2 , strongly suggesting that changes in Pho4 are primarily responsible for the difference in Pho2-dependence ( Kerwin and Wykoff , 2009 ) . To understand whether dependence on Pho2 in S . cerevisiae is the ancestral or the derived state and how this property of Pho4 evolved among related species , we surveyed the activity of Pho4 orthologs from 16 species in the Hemiascomycete class . To isolate the changes in Pho4 activity from the genomic background ( e . g . promoter and Pho2 changes ) , we inserted coding sequences ( CDSs ) of the 16 Pho4 orthologs into an S . cerevisiae background lacking both the endogenous ScPho4 CDS and the negative regulator of the PHO pathway , Pho80 , under the control of the endogenous ScPho4 promoter . Deletion of PHO80 , the cyclin component of the cyclin-dependent-kinase complex , causes constitutive nuclear localization of Pho4 and de-repression of PHO genes , mimicking phosphate starvation ( O'Neill et al . , 1996; Huang et al . , 2005 ) . The use of the pho80Δ strain background allowed us to decouple the gene induction ability of Pho4 orthologs with or without ScPho2 from viability ( below ) . 10 . 7554/eLife . 25157 . 003Figure 1 . Difference among three yeast species in their dependence on Pho2 for gene induction and organismal survival under low Pi conditions . ( A ) Induction of the secreted phosphatase in each species measured by a semi-quantitative acid phosphatase assay ( Wykoff et al . , 2007 ) . The intensity of the red color indicates the total activity of the secreted acid phosphatase from the cell colony . For S . cerevisiae and C . glabrata , strains lacking the negative regulator of Pho4 – Pho80 – were spotted on synthetic medium with 10 mM Pi . For C . albicans , strains with PHO80 wild-type were spotted on synthetic medium lacking inorganic phosphate; the pho2∆ strain was not able to grow on this plate . ( B ) Colony growth phenotype of the wild-type , pho4∆ , pho2∆ strains in each of the three species , under different Pi concentrations . In both panels , two technical replicates of four-fold serial dilutions from the same culture are shown for each strain . DOI: http://dx . doi . org/10 . 7554/eLife . 25157 . 003 To determine the level of dependence on Pho2 for each Pho4 ortholog , we compared its activity , reflected by induction of the secreted phosphatase encoded by PHO5 , in the presence or absence of ScPho2 . We first evaluated whether the Pho4 orthologs can functionally compensate for ScPho4 in the S . cerevisiae background lacking the Pho4 negative regulator Pho80 and paired with ScPho2 ( Figure 2A , left three columns ) . We found that the majority of Pho4 orthologs were able to induce Pho5 to a level significantly above the background ( pho4Δ , bottom row in Figure 2A ) , although the activity declines with increasing evolutionary distance from S . cerevisiae . When we measured the activity of the Pho4 orthologs in the absence of Pho2 ( Figure 2A , right three columns ) , we found that Pho4 orthologs from the clade consisting of C . glabrata , C . bracarensis , N . delphensis and C . nivariensis ( hereafter the ‘glabrata clade’ ) were able to induce Pho5 expression in the absence of Pho2 ( Figure 2A , B ) , but Pho4 orthologs from outgroup species such as N . castellii , K . lactis , and L . waltii could not . We conclude that the common ancestor of S . cerevisiae and C . glabrata had a Pho4 that is dependent on Pho2 and that the reduced dependence was evolutionarily derived in the glabrata clade . 10 . 7554/eLife . 25157 . 004Figure 2 . Evolution of Pho4 dependence on Pho2 in the Hemiascomycetes . ( A ) Survey of Pho4 orthologs activity in the S . cerevisiae background by the semi-quantitative acid phosphatase assay with or without ScPho2 . The species phylogenetic relationship shown on the left were based on ( Wapinski et al . , 2007 ) . Species names marked in red indicate known commensal and human pathogens . All strains were constructed in an identical S . cerevisiae background lacking the PHO pathway negative regulator Pho80 . For each strain , three technical replicates in four-fold serial dilutions were assayed . A strain lacking Pho4 serves as the negative control ( pho4Δ , dotted box ) . ( B ) Quantitative phosphatase assay for the same strains in ( A ) . The bar graph shows the mean and standard deviation of the secreted phosphatase activity ( N = 2 , technical replicates ) . For Pho4 orthologs with noticeable activities ( exceeding twice the value of the pho4Δ control with and without ScPho2 , dotted lines ) , a percentage value was calculated by dividing the activity without ScPho2 by that with ScPho2 , after subtracting the negative control ( pho4∆ ) in both cases . All results are representative of multiple ( >2 ) experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 25157 . 004 Surprisingly , the Pho4 ortholog from a distantly related commensal and pathogenic yeast C . albicans ( hereinafter ‘Ca’ ) weakly induced Pho5 in a Pho2-independent manner ( Figure 2A ) . We further demonstrated that C . albicans does not require Pho2 for survival in phosphate-limited conditions , and deletion of Pho2 does not abolish the induction of the secreted phosphatase in that species ( Figure 1 ) . We were not able to infer whether the reduced Pho2 dependence in C . albicans represents the derived or the ancestral state , because the Pho4 ortholog from Y . lipolytica , an outgroup of C . albicans and S . cerevisiae , failed to complement ScPho4 in S . cerevisiae . In total , we identified five Pho4 orthologs with reduced dependence on Pho2 . Notably , the extent of the reduction varies between the glabrata clade Pho4 orthologs , suggesting that the strength of combinatorial regulation is a quantitative trait that can be fine-tuned by mutations during evolution . Since dependence on Pho2 provides additional selectivity for Pho4 induced gene expression in S . cerevisiae ( Zhou et al . , 2011 ) , we hypothesized that a reduction in Pho2-dependence would result in an increase in the number of Pho4 targets in the S . cerevisiae background . To test this prediction , we quantified the number of genes induced by different Pho4 orthologs expressed in an S . cerevisiae background lacking the negative regulator Pho80 , which allowed us to focus on Pho4-dependent genes but ignore starvation-induced , Pho4-independent genes ( Zhou et al . , 2011 ) . We identified a total of 247 genes that were significantly induced by at least one of the eight Pho4 orthologs in the presence of ScPho2 ( False discovery rate < 0 . 05 , fold change > 2 ) . Pho4 from C . glabrata , C . bracarensis , N . delphensis and C . nivariensis induced more genes than the Pho2-dependent Pho4 orthologs did ( Figure 3A , D ) , and genes induced by these Pho4 orthologs are largely unaffected when ScPho2 is absent ( Figure 3B ) . For example , 212 genes were induced by CgPho4 with ScPho2 , compared to 40 genes induced by ScPho4 in the same background . Pho4 from S . paradoxus , a close relative of S . cerevisiae , induced a smaller number of genes than ScPho4 , as did Pho4 from L . kluyveri , an outgroup of both S . cerevisiae and C . glabrata . Thus , the observed target expansion is not congruent with the phylogenetic relationship , but is a property unique to Pho4 orthologs with reduced Pho2-dependence . Moreover , differences in the mean expression levels of the Pho4 orthologs are small ( <2 . 5 fold , Figure 3—figure supplement 1 ) and do not explain the variation in their activity or dependence on Pho2 ( Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 25157 . 005Figure 3 . Pho4 orthologs that are less dependent on Pho2 induce more genes in the S . cerevisiae background . ( A ) Heatmap showing log2 fold change of genes ( rows ) induced by Pho4 orthologs ( columns ) in the S . cerevisiae background lacking PHO80 , with ScPho2 . A cutoff of 4 and −2 are used for visual presentation . The raw fold change estimates for the 247 genes by eight Pho4 orthologs were available in Figure 3—source data 1 . Species names for each of the Pho4 orthologs were abbreviated and correspond to the full names in Figure 2 . An asterisk indicates the Pho4 ortholog was shown to induce Pho5 expression in the absence of ScPho2 in S . cerevisiae . A total of 247 genes are plotted . The red box highlights a group of 16 genes that were induced by all eight Pho4 orthologs tested . ( B ) Same as ( A ) except the strains were in a pho80∆ pho2Δ background for all Pho4 orthologs . ( C ) Scatter plot comparing the levels of Pho2-dependence for each Pho4 ortholog , measured by the ratios for the 16 shared target genes between their fold changes in the absence versus in the presence of ScPho2 . The boxplots represent the interquartile range ( IQC , box ) , the mean ( thick bar in the middle ) and the highest or lowest values within 1 . 5 times of IQC ( whisker ) . The red circles highlight one of the 16 genes , PHO5 . ( D ) Bar plot showing the number of genes significantly induced more than twofold by each Pho4 ortholog in the presence of ScPho2 . DOI: http://dx . doi . org/10 . 7554/eLife . 25157 . 00510 . 7554/eLife . 25157 . 006Figure 3—source data 1 . This zip file contains four tab-delimited csv files . Two of them ( fold_with_Pho2 . csv and fold_no_pho2 . csv ) record the gene fold change estimates for each Pho4 ortholog with or without ScPho2 . The other two ( sig_with_Pho2 . csv and sig_no_pho2 . csv ) record for each Pho4 ortholog whether a gene is deemed significantly induced with or without Pho2 . DOI: http://dx . doi . org/10 . 7554/eLife . 25157 . 00610 . 7554/eLife . 25157 . 007Figure 3—figure supplement 1 . Expression levels of the Pho4 orthologs by RNA-seq . The scatter plot shows normalized transcript abundance for the PHO4 orthologs . The black bar indicates the mean for each PHO4 ortholog . The dotted line shows the average expression level over all Pho4 orthologs ( except pho4∆ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25157 . 00710 . 7554/eLife . 25157 . 008Figure 3—figure supplement 2 . Expression level of a Pho4 ortholog does not correlate with its level of Pho2-dependence or the number of genes it induces . The heatmap shows the Pearson correlation between a Pho4 ortholog’s expression level and the corresponding strain’s gene induction statistics , including ( a ) phosphatase assay activity; ( b ) PHO5 expression level with Pho2; ( c ) ratio of phosphatase assay activity without vs with Pho2; ( d ) ratio of PHO5 expression level without vs with Pho2; ( e ) number of genes induced by each Pho4 ortholog with Pho2 and ( f ) number of genes induced by each Pho4 ortholog without Pho2 . DOI: http://dx . doi . org/10 . 7554/eLife . 25157 . 008 We further investigated whether there is a quantitative relationship between the level of Pho2-dependence and the number of genes induced by Pho4 orthologs . We previously measured the level of Pho2-dependence by comparing the activity of a single Pho4 target – PHO5 – in a pair of strains differing in the presence or absence of ScPho2 . Here we made the same comparison for gene induction fold changes in a group of 16 genes induced by all eight Pho4 orthologs ( Figure 3A , red box ) . The results , measured by the mean of the ratios for the 16 genes , are largely consistent with what we observed with Pho5 alone ( Figure 3C ) , and , in general , the number of genes induced by each Pho4 ortholog increases with decreasing levels of Pho2-dependence ( Figure 3D ) . In summary , the level of Pho2-dependence is negatively correlated with the number of genes induced by the Pho4 ortholog in the S . cerevisiae background . We reasoned that the expansion of target genes for the glabrata clade Pho4 orthologs could result from Pho4 binding to more genomic locations , Pho4 activating a higher proportion of the genes to which it binds , or a combination of the two . To test if differences in Pho4 binding account for target gene expansion , we performed chromatin immunoprecipitation followed by high-throughput sequencing ( ChIP-seq ) to identify the binding locations for both ScPho4 and CgPho4 in the S . cerevisiae background lacking the negative regulator Pho80 . We identified a total of 115 ChIP-peaks for CgPho4 and 74 peaks for ScPho4 , with 72 being bound by both ( Figure 4A , Figure 4—source data 1 ) . The expansion of CgPho4 binding locations was not because it recognized new sequence motifs – 42 of the 43 CgPho4-specific peaks contain the consensus ‘CACGTG’ motif , and 69 of the 72 shared peaks contain this motif . For all four exceptions , a one-base-pair ( bp ) mismatch to the consensus motif is observed ( Figure 4A , parentheses ) . Therefore , DNA binding specificity is conserved between ScPho4 and CgPho4 . In contrast , ScPho4 and CgPho4 differ in their dependence of DNA binding on Pho2 – ScPho4 binding is significantly lower when ScPho2 is absent , but CgPho4 binding is largely unaffected by the deletion of ScPho2 ( Figure 4B ) . In summary , CgPho4 recognizes the same E-box motif as ScPho4 does , but its binding is no longer dependent on ScPho2 and CgPho4 binds to ~50% more ( 43/74 ≈ 0 . 55 ) sites than ScPho4 does . 10 . 7554/eLife . 25157 . 009Figure 4 . CgPho4 binds to more genomic locations than ScPho4 and is more likely to lead to gene activation upon binding in S . cerevisiae . ( A ) Venn diagram showing the number of and overlap between binding locations for ScPho4 and CgPho4 in the S . cerevisiae genome . The numbers in parentheses indicate binding events among the total number where the DNA sequence underlying the peak contains a suboptimal motif with one base pair mismatch to the consensus . ( B ) Scatter plot showing log2 ratio of ScPho4 or CgPho4 ChIP occupancy without vs with ScPho2 . Only sites bound by Sc or CgPho4 in the presence of ScPho2 ( N = 74 and 115 , respectively ) are plotted . The thick red bar represents the mean and the box the 95% confidence limits computed by a non-parametric bootstrapping method ( Harrell , 2016 ) . The means of the two groups are significantly different by a two-sided Student’s t-test , with a p-value < 0 . 01 . ( C ) Scatterplot for nucleosome occupancy in high Pi conditions at ‘CACGTG’ motifs either bound by neither ScPho4 nor CgPho4 ( N = 660 ) , only by CgPho4 ( N = 48 ) or by both ( N = 88 ) . The red bar and box have the same meaning as in ( B ) , and the difference between sites bound only by CgPho4 and those bound by both CgPho4 and ScPho4 is significant by a two-sided Student’s t test ( p-value = 0 . 012 ) . ( D ) Bar plot comparing the number of genome-wide binding peaks for ScPho4 and CgPho4 that are either non-functional , lead to gene induction only with ScPho2 or lead to gene induction with or without ScPho2 . The source data listing all identified ChIP peaks and the associated gene induction statistics are provided in Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25157 . 00910 . 7554/eLife . 25157 . 010Figure 4—source data 1 . List of ChIP-identified binding sites of ScPho4 and CgPho4 in S . cerevisiae , and associated gene information . Yellow highlight of chromosome names and coordinates mark the same ChIP peak associated with two downstream genes . Two genes ( YOR183W and YNL042W-B ) do not have induction fold change estimates because of low or no coverage in the RNA-seq data , and marked ‘NA’ . YDL106C ( PHO2 ) is deleted in a subset of the strains and therefore excluded from the differential gene expression analysis . Five genes ( in red ) are exceptions in that the associated CgPho4 ChIP peaks ( none of them are bound by ScPho4 ) were in their coding sequences . For two of the five genes , i . e . RIM15 and YDR089W , CgPho4 binding is associated with CgPho4-dependent gene induction . DOI: http://dx . doi . org/10 . 7554/eLife . 25157 . 01010 . 7554/eLife . 25157 . 011Figure 4—figure supplement 1 . Cbf1 enrichment in high Pi conditions in S . cerevisiae is not significantly different between sites bound by both CgPho4 and ScPho4 and sites bound by CgPho4 only . Scatter plot shows the log10 transformed Cbf1 enrichment in high Pi conditions ( Zhou et al . , 2011 ) at a subset of the E-box motifs* , grouped by whether the motif is bound by neither ScPho4 and CgPho4 , only CgPho4 or both . A two-sided t-test comparing the means of the latter two groups yielded a p-value of 0 . 53 . * To avoid the confounding nucleosome exclusion , the E-box motifs in the S . cerevisiae genome were ordered by their nucleosome occupancy in high Pi conditions ( Zhou et al . , 2011 ) from low to high , and the top 25% most accessible ones were used for this plot . DOI: http://dx . doi . org/10 . 7554/eLife . 25157 . 011 It has been reported that both nucleosomes and another transcription factor with similar binding specificity to Pho4 , Cbf1 , competitively exclude ScPho4 from the E-box motifs in the genome ( Zhou et al . , 2011 ) . It is therefore plausible that CgPho4 is able to bind to more locations because it can access sites normally occupied by nucleosomes or competitors . To test if nucleosome exclusion plays a role , we mapped the published nucleosome occupancy in phosphate-replete ( high Pi ) conditions ( Zhou et al . , 2011 ) , where Pho4 is inactive , to the binding peaks identified in this study . We find that sites bound only by CgPho4 have on average higher nucleosome occupancy in high Pi conditions than sites bound by both CgPho4 and ScPho4 ( Figure 4C , t-test for difference in the mean: p-value = 0 . 012 , two-sided test ) . We performed the same analysis for Cbf1 enrichment in high Pi conditions at the top 25% most accessible E-box motifs – those with the lowest nucleosome occupancy – to avoid confounding nucleosome competition with Cbf1 binding . We found that Cbf1 enrichment at sites bound only by CgPho4 is not significantly different from enrichment at sites bound by both ScPho4 and CgPho4 ( Figure 4—figure supplement 1 ) . It is worth noting , however , that the small sample size in the CgPho4 bound only class ( 11 ) may have precluded us from detecting small differences . In conclusion , CgPho4 binding is less dependent on ScPho2 and it competes more favorably with nucleosomes than ScPho4 does , which likely contributes to the expansion of CgPho4 binding sites in the S . cerevisiae genome . To compare the ability of CgPho4 and ScPho4 to induce gene expression upon binding to the promoter , we analyzed the transcriptional profiling data for the genes bound by the two Pho4 orthologs . We found that CgPho4 not only bound to more sites , but it also activated a higher percentage of the downstream genes upon its binding than did ScPho4 ( 64/115 = 56% vs 20/74 = 29% , Figure 4D ) . Moreover , its ability to induce gene expression is largely independent of Pho2: > 90% of CgPho4 targets ( 60/64 = 93 . 75% ) were induced in the pho2Δ background , while all ( 20/20 ) ScPho4 induced genes required ScPho2 ( Figure 4D ) . In conclusion , CgPho4 both binds DNA and activates gene expression independently of ScPho2 . Compared to ScPho4 , it is more capable of accessing nucleosome-occluded binding motifs and it is also able to activate downstream gene expression with a higher probability upon binding . We propose that the combination of these features led to the expansion in the targets of CgPho4 in S . cerevisiae and speculate that this may underlie the target expansion for Pho4 from C . bracarensis , N . delphensis and C . nivariensis . Next we asked if CgPho4 also functions independently of Pho2 in its endogenous genome . To investigate the dependence of CgPho4 binding on CgPho2 , we used the high resolution ChIP-exo technique to map CgPho4 binding in the presence and absence of CgPho2 , and CgPho2 binding , under both phosphate-replete and phosphate-limited conditions ( Materials and methods ) ( Rhee and Pugh , 2012 ) . We identified a total of 100 binding peaks for CgPho4 in the presence of CgPho2 under phosphate-limited conditions ( Figure 5—source data 1 ) . CgPho4 recognizes the same ‘CACGTG’ motif as it does in S . cerevisiae ( Materials and methods ) , and the consensus motif is present in 51 of the 100 peaks , with the rest containing a one-bp mismatch ( 46 ) or two-bp mismatches ( 3 ) . CgPho2 bound to more than 500 sites genome-wide , without a strongly enriched sequence motif ( Materials and methods ) . With respect to CgPho4 bound sites , CgPho2 binds at the same location for 77 of the 100 peaks ( Figure 5A ) . Among these shared binding peaks , only 14 ( 18% ) CgPho4 peaks showed more than two-fold reduction in peak height in the pho2Δ background ( Figure 5A ) . We hypothesized that the quality of the DNA motif underlying the peak may explain the differential requirement of Pho2 co-binding . We tested this hypothesis by comparing changes in CgPho4 binding when CgPho2 is deleted , at sites with a consensus motif vs those without ( Figure 5—figure supplement 1 ) . Although the trend matches our expectation , the difference is small and not significant at a 0 . 05 level by the Student’s t-test ( p-value = 0 . 11 ) . We conclude that CgPho4 binding to DNA is largely independent of CgPho2 in C . glabrata , but a small fraction ( 18% ) of its binding sites show CgPho2 influence . 10 . 7554/eLife . 25157 . 012Figure 5 . Identifying Pho4 targets in C . glabrata using genome-wide binding and transcriptome profiling data . ( A ) Dendrogram showing the breakdown of the 100 CgPho4 ChIP peaks based on whether CgPho2 binds next to CgPho4 , and when it does , whether CgPho4 binding is affected by the deletion of Pho2 or not ( defined as CgPho4 ChIP peak height reduced by more than twofold in the pho2∆ background ) . The graphs below the dendrogram show examples of ChIP profiles for each category of CgPho4 binding . Profiles of ChIP fold enrichment over mock are shown for CgPho2 in cyan , CgPho4 with CgPho2 in blue and CgPho4 without CgPho2 in red . The filled triangles indicate the location of a consensus "CACGTG" motif while the open triangles the one-bp-mismatches . The downstream gene is depicted as a thick bar to the right and the shortened systematic name ( remove the preceding ‘CAGL0’ ) is shown below . The fold changes in induction for the putative target gene with or without CgPho2 are shown to the right of each graph . A list of the 100 binding peaks and the associated statistics are available in Figure 5—source data 1 . ( B ) The left heat map showing the estimates of expression components for 79 genes directly bound and induced by CgPho4 . For each gene , the log2 transformed fold change is decomposed into Pho4 effect alone ( Pho4 ) + Pho2 effect alone ( Pho2 ) + Pho4/Pho2 collaborative effect ( CO ) . A cutoff of 3 and -2 is used for visual presentation . The unadjusted estimates and the associated p-values are available in Figure 5—source data 2 . The right heatmap shows the -log10 transformed p-values for the corresponding t-statistics of the estimates on the left . Three groups are defined based on their characteristic expression components: group I genes are dominated by CgPho4 main effect; group II genes depend on both CgPho4 and CgPho2 ( CO component ) ; group III genes are a mix of the first two groups , with lower fold changes ( weakly induced ) . ( C ) Bar graphs on the left showing the linear model estimates and the standard deviation of the expression defects , defined as the -log2 transformed fold changes between the mutants ( single or double ) and the wild-type . On the right are the corresponding estimates and standard deviation of the expression components for the same gene , estimated from the same data with two biological replicates per strain ( Materials and methods ) . One representative gene is plotted for each category in ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25157 . 01210 . 7554/eLife . 25157 . 013Figure 5—source data 1 . List of ChIP-identified binding sites of CgPho4 in C . glabrata . DOI: http://dx . doi . org/10 . 7554/eLife . 25157 . 01310 . 7554/eLife . 25157 . 014Figure 5—source data 2 . List of CgPho4 directly bound and induced genes and the associated expression components from the mutant cycle analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 25157 . 01410 . 7554/eLife . 25157 . 015Figure 5—figure supplement 1 . Comparison of CgPho2 influence on CgPho4 binding at sites with and without the consensus motif . Scatter plot showing log2 ratio of CgPho4 ChIP enrichment without vs with CgPho2 . The thick red bar depicts the mean and the box the 95% confidence limits computed by non-parametric bootstrapping . The difference in the means of the two groups is not significant at a 0 . 05 level by a Student’s t test , with a p-value=0 . 11 . DOI: http://dx . doi . org/10 . 7554/eLife . 25157 . 015 To evaluate whether gene induction by CgPho4 is dependent on CgPho2 , we used RNA-seq to quantify the fold changes for genes induced by CgPho4 with or without CgPho2 , in a strain lacking the negative regulator CgPho80 . Intersecting with the ChIP identified CgPho4 binding sites , we identified 79 genes that were both directly bound and induced by CgPho4 in the presence of CgPho2 ( Figure 5—source data 2 ) . We then used mutant-cycle analysis ( Capaldi et al . , 2008 ) to delineate the contribution from either CgPho4 acting alone ( Pho4 ) , CgPho2 acting alone ( Pho2 ) or the two factors acting cooperatively ( CO ) , and used unsupervised clustering ( Ward's method , Materials and methods ) on the estimated values for the three components to group the genes into three classes ( Figure 5B , C ) . In class I ( 50 genes ) , CgPho4 is the dominant contributor to gene induction , with CgPho2 either contributing to a lesser extent by itself ( Pho2 ) or through its interaction with CgPho4 ( CO ) . Class II ( 11 ) genes show a strong collaborative component , with little contribution from Pho4 acting alone . Class III ( 18 ) genes show relatively low fold changes , with the main contribution coming from either CgPho4 acting alone or its collaborative effect with CgPho2 . In conclusion , we found that >60% ( 50/79 ) of the genes bound by CgPho4 are induced primarily by CgPho4 acting alone , and that a lesser fraction ( 11/79 ) depend on the collaborative action of CgPho4 and CgPho2 . This is in contrast to S . cerevisiae , where the majority of gene induction was attributed to the cooperative interaction between ScPho4 and ScPho2 ( 23/28 genes are induced only when both ScPho4 and ScPho2 are present [Zhou et al . , 2011] ) . In S . cerevisiae , nearly all ScPho4 targets function in either regulating the PHO pathway or maintaining intracellular phosphate homeostasis ( Ogawa et al . , 2000; Zhou et al . , 2011 ) . To gain insight into the function of the PHO pathway in C . glabrata , we studied the Gene Ontology ( GO ) terms associated with the 79 genes bound and induced by CgPho4 in C . glabrata ( Figure 5—source data 2 ) . The top three enriched GO terms for CgPho4 targets are related to phosphate homeostasis , i . e . polyphosphate metabolism , phosphorus metabolism and phosphate ion transport ( Figure 6—source data 1 ) , confirming that the PHO pathway in C . glabrata is conserved in its core function . However , two observations stand out . First , the genes in this core functional group are not all conserved ( Figure 6A ) . Underlying the apparent conservation in function are non-orthologous genes that are either paralogs ( e . g . HOR2 in S . cerevisiae vs RHR2 in C . glabrata ) or evolutionarily unrelated ( e . g . the phosphatase function of PHO5 in S . cerevisiae is replaced by that of PMU2 in C . glabrata ) ( Figure 6A , Figure 6—source data 2 , [Kerwin and Wykoff , 2009 , 2012; Orlando et al . , 2015] ) . 10 . 7554/eLife . 25157 . 016Figure 6 . Functional annotation of Pho4 targets in C . glabrata . ( A ) Comparison between all 24 ScPho4 targets in S . cerevisiae and CgPho4 targets with phosphate homeostasis related functions in C . glabrata . Abbreviations: ‘Poly P’ stands for ‘Polyphosphate’ . Within each subcategory , S . cerevisiae genes in black are paired with their homologs in C . glabrata in red . N/A in either species indicates ortholog does not exist in that species . C . glabrata genes not annotated with a common name are represented by their systematic name with the preceding ‘CAGL0’ omitted . Gene names in bold indicate that they are targets of ScPho4 or CgPho4 , while gray or pink gene names indicate they are not targets of ScPho4 or CgPho4 , respectively . ( B ) Non-exclusive groups of Pho4 targets in C . glabrata based on Gene Ontology ‘Biological process’ terms and functional annotations in Candida Genome Database ( for C . glabrata or orthologs in C . albicans ) and Saccharomyces Genome Database ( for orthologs in S . cerevisiae ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25157 . 01610 . 7554/eLife . 25157 . 017Figure 6—source data 1 . Gene Ontology terms enrichment analysis results . DOI: http://dx . doi . org/10 . 7554/eLife . 25157 . 01710 . 7554/eLife . 25157 . 018Figure 6—source data 2 . Table comparing ScPho4 and CgPho4 targets with phosphate homeostasis related functions . DOI: http://dx . doi . org/10 . 7554/eLife . 25157 . 01810 . 7554/eLife . 25157 . 019Figure 6—source data 3 . All 79 CgPho4 targets in C . glabrata mapped to Gene Ontology Slim terms . DOI: http://dx . doi . org/10 . 7554/eLife . 25157 . 01910 . 7554/eLife . 25157 . 020Figure 6—source data 4 . Contains five tables listing CgPho4 target genes annotation grouped by functional categories: Table S1 – Non-phosphate related stress and starvation response; Table S2 – Response to chemicals; Table S3 – Cell wall and cell adhesion; Table S4 – Carbohydrate metabolism; Table S5 – all other functional groups . DOI: http://dx . doi . org/10 . 7554/eLife . 25157 . 020 Second , genes with functions related to phosphate homeostasis account for only 16 of 79 CgPho4 targets . To predict the functions of the remaining CgPho4 targets , we mapped all 79 genes to GO Slim terms and identified several major functional groups ( Figure 6B , Figure 6—source data 3 ) . Pho4 targets in C . glabrata are enriched in genes predicted to be involved in the response to non-phosphate-related stresses , response to chemical stresses , fungal cell wall biosynthesis and cell adhesion , and carbohydrate metabolism ( Figure 6—source data 4- Table S1-5 ) . Thus , it appears that the PHO regulon in C . glabrata has , by expanding the number of targets , likely expanded its function beyond phosphate homeostasis . It is worth mentioning that , although CgPho4 induces more genes in S . cerevisiae as well as in C . glabrata , there is virtually no overlap between its targets in S . cerevisiae and in C . glabrata , except for those involved in phosphate homeostasis ( as shown in Figure 6A ) . This is interesting because one might assume that the additional targets of CgPho4 in C . glabrata were evolutionarily acquired by exploiting existing E-box motifs in the genome that were inaccessible to the ancestral Pho2-dependent Pho4 . This may still be true for some of the targets , as it is possible that the S . cerevisiae orthologs of those C . glabrata genes lost the E-box motifs after the two species diverged . The alternative hypothesis is that the targets not involved in phosphate homeostasis were gained de novo in C . glabrata by acquiring CgPho4-recognized motifs in their promoters . In conclusion , our comparative study of the Pho network between S . cerevisiae and C . glabrata reveals a substantial target expansion in the latter species . Comparative studies of more closely related species are needed to provide the temporal resolution for reconstructing the tempo and mode of target expansion in C . glabrata . GRNs consist of regulators , targets and the connections among them that may differ in their evolutionary plasticity . Our functional genomic comparison between the S . cerevisiae and C . glabrata PHO network reveals both constraints and plasticity in its evolution . In terms of constraint , the core transcriptional regulator Pho4 is conserved as a single copy gene and regulates the phosphate starvation response in distantly related species such as S . cerevisiae and C . albicans ( Ikeh et al . , 2016 ) . Second , the mechanism for regulating Pho4 activity in response to phosphate starvation is conserved between C . glabrata and S . cerevisiae – in both species , Pho4 nuclear localization is phosphorylation-dependent , controlled by homologous cyclin-dependent kinase complexes ( Kerwin and Wykoff , 2012 ) . Third , both CgPho4 and ScPho4 recognize the E-box motif ‘CACGTG’ ( Materials and methods ) . Together , these results show that the master TF – its identity , DNA-binding specificity and the mechanism of its regulation – are highly constrained during evolution . This is consistent with previous findings showing that the identity and sequence specificity of a TF evolve much more slowly compared to its target genes ( Wilson et al . , 2008 ) . What has changed throughout evolution in the PHO response network is the combinatorial regulation by Pho4 and Pho2 , leading to changes in the network targets . Specifically , while PHO2 as a gene is conserved among all 16 species examined , its functional role in the regulation of the PHO network has been dramatically reduced in C . glabrata . This , in turn , led to a dramatic expansion of the targets of CgPho4 , potentially extending the function of the network beyond phosphate homeostasis . While the mechanism for this reduction in co-activator dependence is not clear , both transcriptional and binding assays are consistent with the hypothesis that CgPho4 evolved to be stronger in DNA-binding and in inducing gene expression ( Figure 4 ) . Compared to previous findings in GRN evolution , two differences are worth noting . Unlike previous studies which found conserved network output despite regulatory rewiring ( Tsong et al . , 2006; Kuo et al . , 2010; Habib et al . , 2012 ) , we observed a significant change in the size of the regulon ( ~20 target genes in S . cerevisiae vs . ~70 in C . glabrata ) , despite a significantly shorter evolutionary time between the two focal species in this study compared to one of the previous studies ( Tsong et al . , 2006 ) . This may be attributable to the different properties of the GRNs: previous studies have focused on developmental GRNs , whose target genes often constitute a cohesive module that function collectively to specify cell fates . By contrast , the target genes of stress response networks are more independently organized , either acting alone or in small subgroups , e . g . polyphosphate synthesis and phosphate transporters . The organization of a stress response GRN may allow it to be more evolutionarily plastic in acquiring and shedding targets as the environment shifts , creating new demands while eliminating old ones . Consistent with this view , previous studies showed that expression divergence evolves more readily among stress responsive genes compared to growth control and general metabolism genes ( Thompson and Regev , 2009 ) . A second difference worth noting is the mechanism by which the target expansion occurred . Because the DNA sequence specificity of a TF is highly constrained during evolution ( Maerkl and Quake , 2009; Struhl , 1987; Wilson et al . , 2008; Nitta et al . , 2015 ) , it is traditionally thought that gain and loss of targets occurs primarily through cis-regulatory evolution via gain or loss of DNA motifs ( Peter and Davidson , 2011; Ihmels et al . , 2005; Wittkopp and Kalay , 2011 ) . However , we find that trans evolution in CgPho4 , which conserved the core DNA binding specificity and thus the existing targets , altered its dependence on the co-activator , and resulted in a dramatic expansion in its targets . Compared to target turnover via cis regulatory changes , this trans evolutionary mechanism is likely much more rapid , and may allow natural selection to quickly sample many potential targets at once . However , it should be noted that the reduction in Pho2-dependence likely evolved gradually ( Figures 2B and 3C ) , which means cis evolution could have accompanied the trans – while Pho4 expands its targets , promoter evolution could result in fixation of beneficial targets while removing spurious , non-beneficial ones . Thus , the full picture is likely far more complex . In conclusion , we demonstrated that evolution of combinatorial regulation can lead to rapid rewiring of a gene regulatory network . In the PHO network , this could provide a convenient ‘switch’ for evolution to rapidly alter the size of the network output instead of relying on individual promoter alterations . More generally , changes affecting the interaction between transcription factors and their cofactors may be an important yet underappreciated mechanism underlying gene regulatory network evolution ( Slattery et al . , 2011 ) . Only 16 of the 79 genes directly induced by CgPho4 in C . glabrata are involved in maintaining phosphate homeostasis . Among the remaining genes , a significant number ( 25 genes ) are predicted to be involved in responses to non-phosphate related stresses , including osmotic ( seven genes ) and oxidative ( seven genes ) stresses . In addition to roles in other stress responses , a smaller group of CgPho4 targets ( 9 and 3 genes ) have potential functions in cell wall synthesis and cell adhesion , two traits that were known to be relevant for survival and virulence in the host ( Atanasova et al . , 2013; De Las Peñas et al . , 2015; Luo and Samaranayake , 2002; Jawhara et al . , 2012; Fabre et al . , 2014 ) . A similar observation has been made before in C . glabrata , where limitation of nicotinic acid was sufficient to induce genes mediating a cellular adhesion phenotype ( Domergue et al . , 2005 ) . Intriguingly , Pho4 in C . albicans has been shown to be important for survival under particular types of osmotic and oxidative stresses , suggesting a similar functional expansion as we observed in C . glabrata ( Ikeh et al . , 2016 ) . Also in C . albicans , phosphate starvation was linked to enhanced virulence , and a strain lacking Pho4 displayed extensive filamentation in response to phosphate limitation ( Romanowski et al . , 2012 ) . Combining these observations , we speculate that stress responses in the commensal species may have evolved to be more coordinated , and may have acquired new targets linked to virulence , to cope with the distinct stress profiles in the host , such as spatiotemporally overlapping challenges exerted by the host immune cells ( Kasper et al . , 2015 ) . Characterizing other stress response pathways in C . glabrata , and in other commensal species such as C . albicans , can test this hypothesis and will shed further light on the architectural differences in the stress and starvation response network compared to that in the free-living species . Phosphate-free synthetic complete medium was prepared from Yeast Nitrogen Base with ammonium sulfate , without phosphates , without sodium chloride ( MP Biomedicals , Santa Ana , California ) and supplemented to a final concentration of 2% glucose , 1 . 5 mg/ml potassium chloride , 0 . 1 mg/ml sodium chloride and amino acids , as described previously ( Lam et al . , 2008 ) . Monobasic potassium phosphate ( 1M solution , Sigma-Aldrich , St . Louis , MO ) was added to phosphate-free medium to make high phosphate ( Pi ) medium containing a final concentration of 10 mM Pi . All media were adjusted to pH 4 . 0 with HCl . Yeast strains were grown at 30°C with shaking and cell samples were collected at early/mid-logarithmic phase ( OD600 0 . 3–0 . 4 ) . To induce the phosphate starvation response , yeast cells were first grown in 10 mM Pi medium to early/mid-logarithmic phase . Cells were then harvested by filtering and washed with 2–3 volumes of no Pi medium pre-warmed to 30°C . Finally , cells were re-suspended in pre-warmed no Pi medium and grown at 30°C for 1 hr before being harvested for downstream analyses . For plate growth assays , yeast cells were grown in 10 mM Pi medium until mid-logarithmic phase , washed 2–3 times and re-suspended in sterile water . A 1:4 dilution series were made with sterile water in 96-well plates . A 48-pin tool was used to transfer ~10 µL of resuspended cell culture onto appropriate solid agar plates . After 24–48 hr of growth at 30°C , pictures were taken using a standard gel box apparatus . For the semi-quantitative assay , cells were grown overnight ( preconditioning ) , diluted to OD600 ~0 . 1 in the morning and grown to OD600 = 0 . 6–1 . The cell culture was centrifuged , washed and re-suspended in water . A four-fold serial dilution was prepared for each strain and spotted onto an agar plate using a 48-pin tool . Complete synthetic medium was used for pho80∆ strains and phosphate-free medium ( see above ) used for PHO80 wild-type strains . After overnight growth , the agar plates were overlaid with Fast Blue Salt B stain ( Sigma-Aldrich D9805 ) , 1-naphthyl phosphate ( 1 NP , Sigma-Aldrich , D5602 ) , and 1% agar in 0 . 1 m sodium acetate ( pH 4 . 2 ) ( Wykoff et al . , 2007 ) . Pictures were taken on a HP color scanner after 5 min . For the quantitative phosphatase assay , cells were preconditioned and grown the same way as above . After collection by centrifugation , cells were washed and re-suspended with sterile water to OD600 ~ 5 . Then 30 µL of the re-suspended culture was transferred to a 96-well assay plate in triplicates . The cell re-suspension was incubated with 80 µL of 10 mM p-nitrophenyl phosphate ( pNPP , Sigma-Aldrich P4744 ) dissolved in 0 . 1M sodium acetate ( pH = 4 . 2 ) for 15 min at 25°C . The reaction was quenched by adding 144 µL of saturated Na2CO3 ( pH > 11 ) followed by 5 min of centrifugation at 3000 g . Finally , 200 µL of the supernatant from each well was transferred to a new plate and OD420 was measured on a plate reader . Phosphatase activity was measured in units expressed as OD420/OD600 ( Huang et al . , 2005 ) . All transcriptional profiling experiments were done in the pho80∆ background in both S . cerevisiae and C . glabrata . Two biological replicates ( same genotype , but grown , collected and processed separately ) were obtained for each sample . Briefly , yeast cells were collected using a cold methanol quenching method ( Pieterse et al . , 2006; Zhou et al . , 2011 ) . 20 mL of mid-log phase ( OD600 = 0 . 2–0 . 5 ) cell cultures were added directly into 30 mL of pre-chilled methanol ( ~−50°C ) , and incubated in an ethanol-dry ice bath at the same temperature for at least 20 min . Cells were collected by centrifugation and quickly washed with ice-cold water to remove methanol , and resuspended in RNAlater solution ( Qiagen , Hilden , Germany ) . For each sample , 5*107 cells were collected and mechanically lysed on a Minibeadbeater ( BioSpec Products , Bartlesville , OK ) : Zirconia beads ( 0 . 5 mm , BioSpec Products #11079105z ) were added to ~600 µL of cell suspension per sample in a 2 mL screw cap tube to the meniscus . Cells were lysed by four rounds of 1 min bead beating and 2 min of cooling in an ice-water bath . An RNasy Mini kit ( Qiagen ) was used to isolate total RNA from the lysed cell . RNA-seq libraries were prepared with the TruSeq RNA Library Preparation Kit v2 ( Illumina , San Diego , CA ) with the mRNA purification option , following the manufacturer’s protocol . The resulting libraries were sequenced on an Illumina HiSeq 2000 , which produced on average 10 million 50 bp single end reads for each sample . Annotation of putative Pho4 targets in C . glabrata was based largely on information of the orthologs in the well-annotated S . cerevisiae and C . albicans genomes . Gene Ontology enrichment analysis was performed using the ‘GO term finder’ tool , and mapping genes to GO Slim terms performed using the ‘GO Slim mapper’ , both available on the CGD website ( http://www . candidagenome . org/ ) . All genomic features were used as the background set . For the enrichment analysis , a threshold of p<0 . 1 was used for selecting significant GO terms . We focused on the ‘Biological Processes’ to identify the functional groups as reported in the results .	The diversity of life on Earth has intrigued generations of scientists and nature lovers alike . Research over recent decades has revealed that much of the diversity we can see did not require the invention of new genes . Instead , living forms diversified mostly by using old genes in new ways – for example , by changing when or where an existing gene became active . This kind of change is referred to as “regulatory evolution” . A class of proteins called transcription factors are hot spots in regulatory evolution . These proteins recognize specific sequences of DNA to control the activity of other genes , and so represent the “readers” of the genetic information . Small changes to how a transcription factor is regulated , or the genes it targets , can lead to dramatic changes in an organism . Before we can understand how life on Earth evolved to be so diverse , scientists must first answer how transcription factors evolve and what consequences this has on their target genes . So far , most studies of regulatory evolution have focused on networks of transcription factors and genes that control how an organism develops . He et al . have now studied a regulatory network that is behind a different process , namely how an organism responds to stress or starvation . These two types of regulatory networks are structured differently and work in different ways . These differences made He et al . wonder if the networks evolved differently too . The chemical phosphate is an essential nutrient for all living things , and He et al . compared how two different species of yeast responded to a lack of phosphate . The key difference was how much a major transcription factor known as Pho4 depended on a so-called co-activator protein named Pho2 to carry out its role . Baker’s yeast ( Saccharomyces cerevisiae ) , which is commonly used in laboratory experiments , requires both Pho4 and Pho2 to activate about 20 genes when inorganic phosphate is not available in its environment . However , in a related yeast species called Candida glabrata , Pho4 has evolved to depend less on Pho2 . He et al . went on to show that , as well as being less dependent on Pho2 , Pho4 in C . glabrata activates more than three times as many genes as Pho4 in S . cerevisiae does in the absence of phosphate . These additional gene targets for Pho4 in C . glabrata are predicted to extend the network’s activities , and allow it to regulate new process including the yeast’s responses to other types of stress and the building of the yeast’s cell wall . Together these findings show a new way that regulatory networks can evolve , that is , by reducing its dependence on the co-activator , a transcription factor can expand the number of genes it targets . This has not been seen for regulatory networks related to development , suggesting that different networks can indeed evolve in different ways . Lastly , because disease-causing microbes are often stressed inside their hosts and C . glabrata sometimes infects humans , understanding how this yeast’s response to stress has evolved may lead to new ways to prevent and treat this infection .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "microbiology", "and", "infectious", "disease" ]	2017	Evolution of reduced co-activator dependence led to target expansion of a starvation response pathway
The brain is sensitive to the dose of MeCP2 such that small fluctuations in protein quantity lead to neuropsychiatric disease . Despite the importance of MeCP2 levels to brain function , little is known about its regulation . In this study , we report eleven individuals with neuropsychiatric disease and copy-number variations spanning NUDT21 , which encodes a subunit of pre-mRNA cleavage factor Im . Investigations of MECP2 mRNA and protein abundance in patient-derived lymphoblastoid cells from one NUDT21 deletion and three duplication cases show that NUDT21 regulates MeCP2 protein quantity . Elevated NUDT21 increases usage of the distal polyadenylation site in the MECP2 3′ UTR , resulting in an enrichment of inefficiently translated long mRNA isoforms . Furthermore , normalization of NUDT21 via siRNA-mediated knockdown in duplication patient lymphoblasts restores MeCP2 to normal levels . Ultimately , we identify NUDT21 as a novel candidate for intellectual disability and neuropsychiatric disease , and elucidate a mechanism of pathogenesis by MeCP2 dysregulation via altered alternative polyadenylation . Methyl CpG-binding protein 2 ( MeCP2 ) binds methylated cytosines ( Lewis et al . , 1992 , Guo et al . , 2014 ) and affects the expression of thousands of genes ( Chahrour et al . , 2008 ) . MECP2 loss-of-function is the predominant cause of Rett syndrome , a postnatal neurological disorder that typically manifests around 18 months of age with developmental regression ( Amir et al . , 1999; Ravn et al . , 2005; Pan et al . , 2006 ) . MECP2 mutations additionally cause other neurological disorders , such as non-specific autism ( Carney et al . , 2003 ) , Angelman-like syndrome ( Imessaoudene et al . , 2001 ) , and intellectual disability ( Orrico et al . , 2000 ) . And in the absence of MECP2 mutations , decreased MeCP2 expression has been detected in the brains of patients with autism , Angelman syndrome , and Prader–Willi syndrome ( Samaco et al . , 2004; Nagarajan et al . , 2006 ) . Duplications within Xq28 involving MECP2 account for 1% of X-linked intellectual disability ( del Gaudio et al . , 2006; Friez et al . , 2006; Lugtenberg et al . , 2006; Meins et al . , 2005; Van Esch et al . , 2005; Cox et al . , 2003 ) and triplications encompassing MECP2 lead to a more severe phenotype ( del Gaudio et al . , 2006 ) . Mouse studies have established that MECP2 alone within the duplicated and triplicated region is sufficient to cause all the neurological phenotypes of the duplication and triplication syndromes ( Collins et al . , 2004; Samaco et al . , 2012 ) . Notably , even small changes in MeCP2 protein level lead to neurocognitive deficits and behavioral abnormalities , and the severity of the phenotype correlates with the level of MeCP2 ( Chao and Zoghbi , 2012 ) . MECP2 is distinctive for its long 3′ UTR of about 8 . 5 kb that contains two predominant poly-adenylation ( p ( A ) ) sites: a proximal p ( A ) site located just after the last exon , and a distal p ( A ) site approximately 8 kb from the final exon , which result in short and long messenger ribonucleic acid ( mRNA ) isoforms , respectively ( Coy et al . , 1999; Shahbazian et al . , 2002 ) . 3′ UTRs are important for fine-tuning transcript and protein levels because they contain binding sites for regulatory molecules such as RNA-binding proteins and microRNAs ( miRNAs ) , ( Bartel , 2009 , Gennarino et al . , 2012; Gennarino et al . , 2015 ) which induce degradation of the mRNA or inhibit its translation . Thus , long isoforms allow for greater regulation due to an increased number of regulatory elements . The long 3′ UTR of MECP2 harbors more than 50 putative miRNA-binding sites , including the miRNAs known to bind MECP2—miR-483-5p , miR-132 and miR-155—and reduce MeCP2 protein abundance ( Klein et al . , 2007; Kuhn et al . , 2010; Han et al . , 2013 ) . The proximal p ( A ) site of MECP2 is increasingly used throughout postnatal development , resulting in more short mRNA isoforms and a concomitant increase in protein ( Balmer et al . , 2003 ) , which may be driven by the loss of regulatory binding sites in its 3′ UTR . These observations highlight the complex and precise regulation of MeCP2 in the human brain . Pre-messenger RNA cleavage factor Im ( CFIm ) regulates 3′ UTR length by mediating alternative poly-adenylation ( APA ) ( Millevoi and Vagner , 2010 ) . It binds pre-messenger mRNA as a heterotetramer that consists of a NUDT21-encoded CFIm25 dimer and any two of the paralogs CFIm59 or CFIm68 ( or its splice variant , CFIm72 ) ( Kim et al . , 2010; Yang et al . , 2011 ) . In vitro , CFIm25 or CFIm68 knockdown leads to a transcriptome-wide increase in proximal APA site usage , whereas knockdown of other p ( A ) proteins , such as CFIm59 , CPSF , CSTF , or CFIIm , has minimal effects on APA ( Gruber et al . , 2012; Masamha et al . , 2014 ) . Several models have been proposed to explain the role of CFIm25 in APA , but the exact mechanism remains unknown . A total of 59 genes have increased proximal p ( A ) site usage when comparing low- to high-NUDT21 expressing glioblastoma tumors , and of these , 24 show the same effect as that following CFIm25 knockdown in HeLa cells . MECP2 is the most affected , and its protein levels increase accordingly ( Masamha et al . , 2014 ) . Moreover , Clip-seq data show enriched CFIm25 binding of the MECP2 3′ UTR near its p ( A ) sites ( Gruber et al . , 2012 ) . Based on this , we hypothesized that CFIm25 regulates MeCP2 in humans by changing the ratio of short to long mRNA isoforms and that NUDT21 gain- or loss-of-function will be correlated with neuropsychiatric disease . To support our hypothesis , we sought patients with copy-number variation ( CNV ) of NUDT21 . We identified eight individuals with duplications and three with deletions spanning NUDT21 , all of whom had undergone clinical chromosome microarray testing . Of these , five consented to provide a detailed medical history ( Figure 1 and Table 1 ) , and four provided blood samples for molecular studies ( three with NUDT21-spanning duplications and one with a deletion ) . 10 . 7554/eLife . 10782 . 003Figure 1 . Subjects with NUDT21-spanning copy-number variations ( CNVs ) . ( A ) Five intrachromosomal rearrangements of chromosome 16q including the NUDT21 gene , identified by clinical array comparative genomic hybridization . Duplications shown in blue , deletion in red . Del , deletion; Dup , duplication; Mb , megabases . The striped bars indicate that the copy variant is not drawn to scale . ( B ) Array plots of oligonucleotide arrays on subjects 1 , 2 , and 4 . The array plots of subject 3 and 5 were unavailable due to the closure of the Signature Genomics microarray laboratory in June 2014 . Black dots indicate probes with normal copy-number , green dots indicate copy-number gain , and red dots indicate copy-number loss . Solid and dotted lines respectively define the minimum and maximum expected boundaries of the CNVs . DOI: http://dx . doi . org/10 . 7554/eLife . 10782 . 00310 . 7554/eLife . 10782 . 004Table 1 . Molecular and clinical characteristics of four individuals with NUDT21 duplication and one individual with NUDT21 deletionDOI: http://dx . doi . org/10 . 7554/eLife . 10782 . 004Subject 1Subject 2Subject 3Subject 4Subject 5SexFMMFFAge10 years5 years15 years13 years8 yearsDeletion/duplicationDuplicationDuplicationDuplicationDeletionDuplicationCoordinateschr16: 46 , 500 , 741-64 , 653 , 093chr16: 46 , 500 , 741-58 , 539 , 422chr16: 55 , 725 , 264-57 , 533 , 101chr16: 56 , 344 , 856-63 , 521 , 523chr16: 55 , 919 , 145-56 , 619 , 283Size18 . 15 Mb12 . 04 Mb1 . 81 Mb7 . 18 Mb0 . 70 MbZygosityHeterozygousHeterozygousMosaic marker chromosome ( present in 45% of cells ) HeterozygousHeterozygousInheritanceDe novoDe novoDe novoUnknownNot maternalAdditional CNVs52 kb duplication of chr6: 152 , 716 , 341-152 , 768 , 421 ( maternal ) NoneNone144 kb duplication of chr17: 427 , 284-571 , 275 ( inheritance unknown ) NoneDevelopmental delay/intellectual disabilityIntellectual disability ( clinical impression: moderate ) Significant developmental delayIntellectual disabilityIntellectual disability ( full scale IQ 53 ) Intellectual disabilityDevelopmental regressionYes , at 2 years of ageNoYesNoNoAutism spectrum disorderYesNoYesNoYesEpilepsySingle , isolated seizure at 5 years of ageNoNoIntractable , symptomatic partial onset seizuresNoADHDYesYesUnknownYesNoF , female; M , male; chr , chromosome; MB , megabases; CNVs , copy number variants; kb , kilobases; ADHD , attention deficit hyperactivity disorder . Father not available . Lymphoblastoid cell lines are commonly used in patient-specific molecular and functional studies of neurological disease ( Sie et al . , 2009 ) . We , therefore , established immortalized lymphoblastoid cell lines of the four individuals and tested them for relative changes in MeCP2 protein and mRNA levels , and APA site usage . As hypothesized , in the duplication patients , we found ∼60% more CFIm25 and ∼50% less MeCP2 protein ( Figure 2A ) , while in the deletion patient , we found ∼50% less CFIm25 and ∼50% more MeCP2 protein ( Figure 2B ) , when compared to 13 age-matched control subjects . In order to prove that NUDT21 alone can regulate MeCP2 levels , we performed an siRNA-mediated knockdown of NUDT21 and assessed for changes in MeCP2 protein abundance . We found that NUDT21 knockdown increases MeCP2 in patient-derived lymphoblastoid cells , and that reducing it to wild-type levels in the NUDT21-duplication patients rescues MeCP2 protein levels to that of the healthy controls ( Figure 2C ) . To our knowledge , this is the first time that patient-derived lymphoblastoid cells have been transfected or transduced with siRNAs and that the functional consequence of a candidate gene within a larger CNV has been validated in vitro . 10 . 7554/eLife . 10782 . 005Figure 2 . CFIm25 regulates MeCP2 protein levels in patient-derived lymphoblastoid cells with NUDT21 CNVs . ( A ) Representative western blot picture for three duplication patients compared to four age-matched controls showing the increase of CFIm25 and decrease of MeCP2 protein levels . ( B ) Representative western blot picture for one deletion patient compared to four age-matched controls showing the decrease of CFIm25 and increase of MeCP2 protein levels . Quantification of protein levels for both CFIm25 and MeCP2 from three duplication patients and one deletion patient compared to a total of 13 age-matched controls are shown below the corresponding western blot . Data represent mean ± SEM from a total of six technical replicates . Data were normalized to GAPDH protein levels . ( C ) Western blot and its relative quantification showing that knockdown of NUDT21 by siRNA-NUDT21 nucleofection increases MeCP2 in control and duplication subjects , and normalizing CFIm25 in duplication patients rescue MeCP2 to control levels . Data represent mean ± SEM from four age-matched control and three duplication cases . Data were normalized to GAPDH protein levels . p < 0 . 01 , p < 0 . 001 . M , mosaic patient . DOI: http://dx . doi . org/10 . 7554/eLife . 10782 . 00510 . 7554/eLife . 10782 . 006Figure 2—figure supplement 1 . siGLO nucleofection showing patient-derived lymphoblastoid cells can be transfected with small RNA . ( A ) Representative histograms of control ( i ) and siGLO-nucleofected lymphoblastoid cells showing nearly 100% efficiency using both the functionality ( ii ) and efficiency ( iii ) protocols . ( B ) Time series showing the presence of small RNA in nearly 100% of nucleofected lymphoblastoid cells up to 48 hr after nucleofection . ( C ) Time series showing cell survival following nucleofection using either the functionality or efficiency nucleofection protocol . DOI: http://dx . doi . org/10 . 7554/eLife . 10782 . 006 We then set out to understand how NUDT21 CNVs lead to altered MeCP2 protein levels by investigating their effects on MECP2 mRNA . Quantitative RT-PCR showed that total MECP2 mRNA was elevated in both the duplication cases and the deletion patient ( Figure 3A ) . However , using primer pairs that detected only the long 3′ UTR isoform of MECP2 , we found that duplication patients have increased long MECP2 ( Figure 3A ) , indicating that elevated CFIm25 increases distal APA site usage . Conversely , the deletion patient showed a decrease in the long MECP2 isoform , further indicating that NUDT21 levels correlate with MECP2 3′ UTR length ( Figure 3A ) . Northern blot analysis of RNA from the lymphoblastoid cells confirmed the switching between long and short MECP2 isoforms for both duplication and deletion patients ( Figure 3—figure supplement 1 ) . Our analysis of the MECP2 mRNA levels in NUDT21-spanning duplication and deletion subjects shows that CFIm25 regulates the ratio of short to long mRNA isoforms ( Figure 3A ) . 10 . 7554/eLife . 10782 . 007Figure 3 . NUDT21 mRNA levels correlate with inefficiently translated long MECP2 mRNA . ( A ) RNA quantification by quantitative RT-polymerase chain reaction ( qRT-PCR ) from lymphoblastoid cells of NUDT21 duplication and deletion patients . The bar graph shows the total mRNA fold change of NUDT21 , total MECP2 , and long MECP2 for the three duplication patients and one deletion patient compared to 13 age-matched controls . Data represent mean ± SEM from five independent experiments . Data were normalized to GAPDH mRNA levels . ( B ) Relative polyribosomal and non-poyribosomal enrichment of total and long MECP2 mRNA isoforms of NUDT21 duplication patients compared to age-matched controls . Data represent mean ± SEM from a total of three control and duplication cases . Data were normalized to ACTB mRNA levels . ( C ) Proposed model showing that duplication and deletion patients have more or less CFIm25 , respectively leading to a relative increase in long and short MECP2 3′ UTR isoforms . In both cases , there is an accumulation of mRNA: in the deletion patient , this leads to more MeCP2 protein , but in the duplication patients , it results in less MeCP2 protein due to a translational block from the CFIm25-mediated increase in long MECP2 isoforms and putative binding of miRNAs or RNA-binding proteins to the 3′ UTR . p < 0 . 05 , p < 0 . 01 , p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 10782 . 00710 . 7554/eLife . 10782 . 008Figure 3—figure supplement 1 . Northern blot assay from patient-derived lymphoblastoid cells . Northern blot assay of lymphoblastoid cells showing that duplication ( left panel ) and deletion ( right panel ) patients respectively have more long or short MECP2 3′ UTR isoforms . Relative northern blot quantification of three duplication patients and one deletion patient compared to 13 age-matched controls ( bottom panel ) . Data represent mean ± SEM from four independent experiments . Data were normalized to GAPDH mRNA levels . For all the experiments , p values were calculated by Student's t-test comparing controls with duplication patients . p < 0 . 05 , **p < 0 . 01 . M , mosaic patient . DOI: http://dx . doi . org/10 . 7554/eLife . 10782 . 00810 . 7554/eLife . 10782 . 009Figure 3—figure supplement 2 . Polyribosome fractionation traces of control and duplication subjects . Representative polyribosome traces from control ( A-C ) and duplication ( D-F ) subjects . UV absorption at 254 nm was plotted vs time depicting the successful resolution of the different ribosomal fractions . DOI: http://dx . doi . org/10 . 7554/eLife . 10782 . 009 To clarify how an increase in MECP2 mRNA may still lead to decreased MeCP2 protein in the NUDT21-duplication patients , we performed a polyribosome fractionation assay to assess translation efficiency of total and long MECP2 mRNA ( Figure 3—figure supplement 2 ) . We found that the NUDT21-duplication patients translate MECP2 less efficiently than healthy controls , with a ∼50% reduction in total MECP2 mRNA in the polyribosome fraction and a dramatic enrichment of the long isoform in the non-polyribosome fraction ( Figure 3B ) . These data support the hypothesis that NUDT21-duplication patients have decreased MeCP2 protein despite increased mRNA because NUDT21 promotes distal APA site usage and the elevated RNA pool consists largely of the inefficiently translated long MECP2 isoform ( Figure 3C ) . The CFIm25 subunit of pre-messenger RNA CFIm represents the first identified post-transcriptional protein regulator of MeCP2 and provides important insights into MeCP2 regulation during normal development and disease . The presented cases of individuals with NUDT21 CNVs and subsequent MeCP2 level changes suggest NUDT21 as a novel candidate gene for intellectual disability and neuropsychiatric disease . Their CNVs , however , are non-recurrent , and all affect several genes in addition to NUDT21 , so that the actual effect of the NUDT21 dosage change alone on phenotype cannot be determined at this point . CNVs affecting only NUDT21 or point mutation cases of NUDT21 would provide important insight . However , we found no NUDT21-spanning CNVs in 30 , 466 healthy controls ( Pinto et al . , 2007; Simon-Sanchez et al . , 2007; Zogopoulos et al . , 2007; International Schizophrenia C , 2008; Jakobsson et al . , 2008; Itsara et al . , 2009; Kirov et al . , 2009; Shaikh et al . , 2009; Conrad et al . , 2010; International HapMap et al . , 2010; Vogler et al . , 2010; Banerjee et al . , 2011; Campbell et al . , 2011; Cooper et al . , 2011; Genomes Project et al . , 2012 ) , while among our group of 87 , 200 patients who underwent clinical chromosome microarray analysis , mostly for neurodevelopmental disorders , we identified 11 NUDT21-spanning CNVs ( p = 0 . 025 , one-tailed Chi square test without Yates correction ) , suggesting that CNVs involving NUDT21 are indeed pathogenic . Moreover , of the four other genes common to all the patients' CNVs , only one is associated with neurological disease , BBS2 , but it is autosomal recessive and is characterized by distinctive features not seen in the patients we studied . The clinical presentation of the three individuals with increased NUDT21 copy number and decreased MeCP2 protein levels is not that of classic Rett syndrome , which is not surprising since individuals with Rett lack the protein in 50% of their cells , whereas the duplication of NUDT21 causes a partial decrease in all cells . However , these individuals suffer from intellectual disability and autism spectrum disorder , and two of the three individuals manifested significant developmental regression , which is a rare clinical phenomenon and considered a hallmark of Rett syndrome . The identification of additional patients will help us better define the phenotypic overlap and differences between NUDT21-duplication and Rett syndrome , and for NUDT21 deletion or loss-of-function and MECP2 duplication syndrome . The clinical and molecular data presented provide insight into the post-transcriptional regulation of MECP2 , and identify NUDT21 as a novel candidate for intellectual disability and neuropsychiatric disease . Further , we developed a new way of validating the role of individual genes within pathogenic CNVs by effectively transfecting patient-derived lymphoblastoid cells . Ultimately , this study provides an example of how the complexity of a CNV goes well beyond the affected chromosomal domain and the genes affected within . The better we understand gene networks , gene-protein , and protein–protein interactions , the better we will be positioned to identify the molecular bases of various neuropsychiatric disorders and design treatment strategies for the affected individuals . Of approximately 52 , 000 patients referred to the Baylor College of Medicine ( BCM ) Medical Genetics Laboratory ( MGL ) for clinical array comparative genomic hybridization ( aCGH ) analysis between April 2007 and February 2013 , six probands with copy number variants affecting NUDT21 were identified: five duplication cases and one deletion . We also screened the database of Signature Genomics Laboratories , based on approximately 35 , 200 samples submitted for clinical aCGH between November 2007 and February 2013 , and identified three duplication cases and two deletions . The search was limited to copy-number variants <20 megabases . The clinical labs shared our contact information with the referring providers of all eleven cases , and we were subsequently contacted by five families who expressed interest in participating in this research study . Their copy-number variants had been detected by clinical aCGH on the following platforms: CMA-HR + SNP V9 . 1 . 1 , Baylor College of Medicine ( subject 1 ) ; Oligo V8 . 1 . 1 , Baylor College of Medicine ( subject 2 ) ; SignatureChipOS v2 . 0 , Signature Genomics ( subject 3 ) ; and Oligo V6 . 5 , Baylor College of Medicine ( subject 4 ) . Following informed consent , approved by the Institutional Review Board for Human Subject Research at Baylor College of Medicine , we performed a comprehensive chart review of medical records and neuropsychological testing . A venous blood sample was provided by the probands in order to establish immortalized lymphoblastoid cell lines . All individuals with NUDT21 copy-number variants were enrolled in a research study approved by the Institutional Review Board of Baylor College of Medicine ( H-25466 ) . The consent form specifically allows for sharing of medical information and physical exam findings . All individuals whose lymphoblastoid cell lines were enrolled as controls had been enrolled in a research study approved by the Institutional Review Board of Baylor College of Medicine ( H-25531 ) as unaffected sibling controls . Venous blood of the recruited probands was drawn into ACD solution A tubes . Buffy coat was prepared , lymphocytes were pelleted , and transformed with Epstein–Barr virus and cyclosporin A following standard procedures . Cell lines were grown in RPMI 1640 medium ( Invitrogen , Carlsbad , CA , United States ) supplemented with 10% fetal bovine serum ( Atlanta Biological , Flowery Branch , GA , United States ) and 1% penicillin/streptomycin . Cell cultures were maintained at 37°C in a humidified incubator supplemented with 5% CO2 . Medium was renewed every 2 to 3 days to maintain the cell density between 1 × 105 and 2 × 106 cells/ml . Lymphoblastoid cells were nucleotransfected with 300 nM of either Ambion small interfering RNA siNUDT21 ( s21771 ) or siScramble ( 4390843 ) using Amaxa 4D-Nucleofector and P3 Primary Cell 4D-Nucleofector X Kit ( Lonza , Cologne , Germany , cat# V4XP-3024 ) . Cells ( 2 × 107 ) were centrifuged at 200×g for 10 min at room temperature , then resuspended in 100 μl P3 Primary Cell 4D-Nucleofector Solution with the added supplement . In microfuge tubes , 300 nM siRNA was mixed with the suspended cells and transferred to Nucleocuvette Vessels prior to nucleofection . Cells were nucleofected using the program FI-115 ( efficiency ) . Post-nucleofection , cells were resuspended with 0 . 5-ml pre-warmed media ( RPMI , 10% FBS , and 1% antibiotic–antimycotic ) and transferred to T25 flasks . After 6 hr , medium was removed and replaced with fresh media . Cells were incubated at 37°C with 5% CO2 for 48 hr prior to protein extraction and western blot analysis . The nucleofection protocol was optimized by transfecting 2 × 107 lymphoblastoid cells with 2 μM of siGLO Green transfection indicator ( Dharmacon , Lafayette , CO , United States ) or Control ( empty ) either using the program FI-115 ( efficiency ) or EO-115 ( functionality ) . Cells were later incubated at 37°C with 5% CO2 and collected at different time points from 12 to 72 hr for cytometric analysis using a BD LSR FORTESSA ( BD Biosciences , San Jose , CA , United States ) according to manufacture's instructions . We determined that the efficiency protocol was preferable because the nucleofection rate was higher , and the survival rate was acceptable . We extracted protein at 48 hr post-nucleofection because the survival was satisfactory , and it was the latest time point at which nucleofected small RNA was nearly 100% , which would allow the siRNA the most time to act ( Figure 2—figure supplement 1 ) . Lymphoblastoid cell suspension cell cultures were collected at 6 × 106 confluence and processed for protein extraction . Cell pellets were lysed with modified RIPA buffer ( 180 mM NaCl , 0 . 5% NP-40 , 0 . 5% Triton X-100 , 2% SDS , and complete protease inhibitor cocktail ( Roche , France ) ) by pipetting them up and down with a p1000 tip and then placed at room temperature rotisserie shaker for 10 min followed by boiling for 5 min . Then the lysates were repeatedly passed through a 27-G needle with a syringe to reduce viscosity due to DNA breakage , followed by centrifugation at 13 , 000 rpm at room temperature for 15 min . Proteins were quantified by Pierce BCA Protein Assay Kit ( Thermo Scientific , Waltham , MA , United States ) and resolved by high resolution NuPAGE 4–12% Bis-Tris Gel ( Life Technologies , Waltham , MA , United States ) according to the manufacturer's instruction . Antibodies: Mouse α-NUDT21 ( 1:500 , Santa Cruz Biotechnology , Dallas , TX , United States , sc-81109 ) ; rabbit α-serum MeCP2 N-terminus ( 1:5000 , Zoghbi Lab , #0535 ) ; mouse α-GAPDH ( 1:10 , 000 , Millipore , Billerica , MA , United States , CB1001-500UG ) . Lymphoblastoid cell suspension cell cultures were collected at 6 × 106 confluence and processed for RNA extraction . Total RNA was obtained using the miRNeasy kit ( Qiagen , Netherlands ) according to the manufacturer's instructions . RNA was quantified using the NanoDrop 1000 ( Thermo Fisher , Waltham , MA , United States ) . Quality of RNA was assessed by gel electrophoresis . cDNA was synthesized using Quantitect Reverse Transcription kit ( Qiagen , Netherlands ) starting from 1 μg of DNase-treated RNA . Quantitative RT-polymerase chain reaction ( qRT-PCR ) experiments were performed using the CFX96 Touch Real-Time PCR Detection System ( Bio-Rad Laboratories , Hercules , CA , United States ) with PerfeCta SYBR Green FastMix , ROX ( Quanta Biosciences , Gaithersburg , MD , United States ) . Real-time PCR results were analyzed using the comparative Ct method normalized against the housekeeping gene GAPDH ( Vandesompele et al . , 2002 ) . The range of expression levels was determined by calculating the standard deviation of the ΔCt ( Pfaffl , 2001 ) . To ensure efficacy of the genomic DNA elimination , we ran negative control samples in the qRT-PCR that did not have reverse transcriptase ( -RT ) in the cDNA synthesis reaction . Total RNA was obtained from lymphoblastoid cell suspension cell cultures as mentioned above . Total RNA ( 15 μg ) was separated on 1 . 2% formaldehyde agarose gel and transferred to Hybond-N+ nylon membrane ( Amersham , United Kingdom ) , followed by UV crosslinking with Stratalinker 2400 ( Agilent Technologies , Santa Clara , CA , United States ) . MECP2 cDNA probe ( 1431 bp ) , which contains a majority of exon1 and exon2 , was synthesized by random primed DNA labeling kit ( Roche , France ) according to the manufacturer's instructions . The blots were hybridized using ULTRAhyb hybridization buffer ( Applied Biosystems , Waltham , MA , United States ) according to the manufacturer's instructions . In brief , the blots were hybridized with radiolabeled probes ( 106 cpm/ml ) in ULTRAhyb hybridization buffer for 2 hr at 68°C . The blots were then washed with 2× SSC , 0 . 05% SDS for 20 min at room temperature , followed by washing with 0 . 1× SSC , 0 . 1% SDS at 50°C for 10 min . The blots were exposed for 72 hr for analysis . For loading control , the blots were then stripped in boiling 0 . 2% SDS and hybridized with GAPDH probe . 1 × 106 lymphoblastoid cells were treated with either 100 µg/ml cycloheximide ( Sigma–Aldrich , St . Louis , MO , United States ) for 60 min at 37°C before being washed with cold PBS containing 100 µg/ml Cycloheximide or Puromycin . The cells were then lysed in polysome lysis buffer ( 10 mM Tris-Cl , pH 7 . 4 , 5 mM MgCl2 , 100 mM KCl , 1% ( vol/vol ) Triton X-100 , 0 . 5% ( wt/vol ) deoxycholate , 1000 U/ml RNasin , 2 mM DTT and 100 µg/ml Cycloheximide in DEPC-treated water ) , incubated on ice for 10 min and centrifuged at 16 , 000×g at 4°C to obtain the clarified post-nuclear lysate . Post-nuclear clarified lysates ( 50 OD units ) were layered on top of 10–50% sucrose gradients and centrifuged at 28 , 000 rpm ( 100 , 000×g ) for 4 hr at 4°C using an SW41 rotor ( Beckman , Fullerton , CA , United States ) . Sucrose gradients were fractionated using a BR-184 tube piercer and a syringe pump ( Brandel , Gaithersburg , MD , United States ) fitted with a UA-6 UV detector ( Teledyne ISCO , Lincoln , NE , United States ) . Digital data were collected during fractionation using the DI-158U USB data acquisition device ( DATAQ Instruments , Akron , OH , United States ) . The digital data were processed using the Peak Chart Data Acquisition Software and UV absorption at 254 nm was plotted vs time ( from top to bottom or fraction number of sucrose gradients ) ( Figure 3—figure supplement 2 ) . TRIzol LS reagent ( Life Technologies , Carlsbad , CA , United States ) was used to extract RNA from the various polysome fraction and total lysate aliquots as per the manufacturer's recommendations . In order to unambiguously distinguish spliced cDNA from genomic DNA contamination , specific exon primers were designed to amplify across introns of the genes tested . The primers for all target genes tested were designed with Primer3 v . 0 . 4 . 0 ( Koressaar and Remm , 2007 ) . Primer sequences:	The X-chromosome carries a number of genes that are involved in a child's intellectual development . One of these genes encodes a protein called MeCP2 , which is important for brain function after birth . Mutations in the MECP2 gene cause a disorder known as Rett syndrome . At around 18 months of age , affected children begin to lose the cognitive and motor skills that they had previously acquired . Individuals with extra copies of this gene also show cognitive impairments . For both diseases , individuals with levels of the MeCP2 protein that are the most different from those found in healthy individuals also show the most severe symptoms . To produce the protein that is encoded by a particular gene , enzymes inside the cell must first make a copy of that gene using a molecule called messenger ribonucleic acid ( or mRNA ) . This mRNA is then used as a template to assemble the protein itself . In the case of MECP2 , two different mRNA templates are produced: a long version and a short version . A gene called NUDT21 makes a protein that regulates whether the long or short version of MECP2 mRNA is made . Gennarino , Alcott et al . have now discovered that people with too many , or too few , copies of the NUDT21 gene have intellectual disabilities and altered levels of MeCP2 protein . Specifically , individuals with extra copies of NUDT21—and thus higher levels of the corresponding protein—produce more of the long MECP2 mRNA . The production of proteins from this long mRNA is less efficient than from the short mRNA; therefore , these individuals have lower levels of MeCP2 protein . The opposite is true for individuals who lack a copy of the NUDT21 gene . To confirm these data , Gennarino , Alcott et al . grew cells in the laboratory from patients with extra copies of the NUDT21 gene and found that reducing the production of its protein returned the levels of the MeCP2 protein back to normal . These findings show that alterations in the NUDT21 gene cause changes in the level of MeCP2 protein in cells and leads to neuropsychiatric diseases .	[ "Abstract", "Main", "text", "Materials", "and", "methods" ]	[ "short", "report", "neuroscience" ]	2015	NUDT21-spanning CNVs lead to neuropsychiatric disease and altered MeCP2 abundance via alternative polyadenylation
Primary cilia are sensory organelles crucial for cell signaling during development and organ homeostasis . Cilia arise from centrosomes and their formation and function is governed by numerous factors . Through our studies on Townes-Brocks Syndrome ( TBS ) , a rare disease linked to abnormal cilia formation in human fibroblasts , we uncovered the leucine-zipper protein LUZP1 as an interactor of truncated SALL1 , a dominantly-acting protein causing the disease . Using TurboID proximity labeling and pulldowns , we show that LUZP1 associates with factors linked to centrosome and actin filaments . Here , we show that LUZP1 is a cilia regulator . It localizes around the centrioles and to actin cytoskeleton . Loss of LUZP1 reduces F-actin levels , facilitates ciliogenesis and alters Sonic Hedgehog signaling , pointing to a key role in cytoskeleton-cilia interdependency . Truncated SALL1 increases the ubiquitin proteasome-mediated degradation of LUZP1 . Together with other factors , alterations in LUZP1 may be contributing to TBS etiology . Townes-Brocks Syndrome ( TBS1 [MIM: 107480] ) is an autosomal dominant genetic disease , caused by mutations in a transcription factor called SALL1 , characterized by the presence of imperforate anus , dysplastic ears , thumb malformations and often renal and heart impairment , among other symptoms ( Botzenhart et al . , 2007; Kohlhase et al . , 1998 ) . Some of these features overlap those in the ciliopathic spectrum . It has been recently demonstrated that primary cilia defects are contributing factors to TBS etiology ( Bozal-Basterra et al . , 2018 ) . Truncated SALL1 , either by itself or in complex with full length protein ( SALL1FL ) , can interact with CCP110 and CEP97 . As a consequence , those negative regulators are reduced at the mother centriole ( MC ) and ciliogenesis is promoted ( Bozal-Basterra et al . , 2018 ) . Primary cilia are sensory organelles that have a crucial role in cell signaling and protein trafficking during development and organ homeostasis . Although several key pathways are influenced by cilia function ( Wnt , TGFbeta , PDGFRalpha , Notch ) , the best characterized is the Sonic Hedgehog ( Shh ) pathway ( Goetz and Anderson , 2010 ) . Briefly , Shh binds to its receptor PTCH1 and leads to ciliary enrichment of the transmembrane protein Smoothened ( SMO ) , with concomitant conversion of the transcription factor GLI3 from a cleaved repressor to a full-length activator form , leading to activation of Shh target genes . Two such genes are PTCH1 and GLI1 ( encoding the Shh receptor and a transcriptional activator , respectively ) , exemplifying the feedback and fine-tuning of the Shh pathway . Cilia arise from the centrosome , a cellular organelle composed of two barrel-shaped microtubule-based structures called the centrioles . Primary cilia formation is very dynamic throughout the cell cycle . Cilia are nucleated from the MC at the membrane-anchored basal body upon entry into the G0 phase , and they reabsorb as cells progress from G1 to S phase , completely disassembling in mitosis ( Rezabkova et al . , 2016 ) . Centrioles are surrounded by protein-based matrix , the pericentriolar material ( PCM ) ( Conduit et al . , 2015; Vertii et al . , 2016 ) . In eukaryotic cells , PCM proteins are concentrically arranged around a centriole in a highly organized manner ( Fu and Glover , 2012; Lawo et al . , 2012; Mennella et al . , 2012; Sonnen et al . , 2012 ) . Based on this observation , proper positioning and organization of PCM proteins may be important for promoting different cellular processes in a spatially regulated way ( Kim et al . , 2019 ) . Not surprisingly , aberrations in the function of PCM scaffolds are associated with several human diseases , including cancer and ciliopathies ( Gönczy , 2015; Nigg and Holland , 2018 ) . Cilia assembly is regulated by diverse factors . Among them , CCP110 and CEP97 form a cilia suppressor complex that , when removed from the MC , allows ciliogenesis to proceed ( Spektor et al . , 2007 ) . The actin cytoskeleton is also emerging as key regulator of cilia formation and function , with both negative and positive roles ( Copeland , 2020 ) . Ciliary dysfunction often results in early developmental problems including hydrocephalus , neural tube closure defects ( NTD ) and left-right anomalies ( Fliegauf et al . , 2007 ) . These features are often reported in a variety of diseases , collectively known as ciliopathies , caused by failure of cilia formation and/or cilia-dependent signaling ( Hildebrandt et al . , 2011 ) . In the adult , depending on the underlying mutation , ciliopathies present a broad spectrum of phenotypes comprising cystic kidneys , polydactyly , obesity or heart malformation . Truncated SALL1 likely interferes with multiple factors to give rise to TBS phenotypes . Here we focus on LUZP1 , a leucine-zipper motif containing protein that was identified by proximity proteomics as an interactor of truncated SALL1 ( Bozal-Basterra et al . , 2018 ) . LUZP1 has been previously identified as an interactor of ACTR2 ( ARP2 actin related protein two homologue ) and filamin A ( FLNA ) and , recently , as an actin cross-linking protein ( Hein et al . , 2015; Wang and Nakamura , 2019 ) . Furthermore , LUZP1 shows homology to FILIP1 , a protein interactor of FLNA and actin ( Gad et al . , 2012; Nagano et al . , 2004 ) . Interestingly , mutations in Luzp1 resulted in cardiovascular defects and cranial NTD in mice ( Hsu et al . , 2008 ) , phenotypes within the spectrum of those seen in TBS individuals and mouse models of dysfunctional cilia ( Botzenhart et al . , 2007; Botzenhart et al . , 2005; Klena et al . , 2016; Kohlhase et al . , 1998; Surka et al . , 2001; Toomer et al . , 2019 ) . Both the non-canonical Wnt/PCP ( Wingless-Integrated/planar cell polarity ) and the Shh pathways are influenced by the presence of functional cilia and regulate neural tube closure and patterning ( Campbell , 2003; Copp , 2005; Fuccillo et al . , 2006 ) . Remarkably , ectopic Shh was observed in the dorsal lateral neuroepithelium of the Luzp1-/- mice ( Hsu et al . , 2008 ) . However , in spite of the phenotypic overlaps , a link between LUZP1 and ciliogenesis has not been explored . Here we demonstrate that LUZP1 is associated with centrosomal and actin cytoskeleton-related proteins . We show that LUZP1 localizes to the PCM , actin cytoskeleton and the midbody , and also provide evidence towards its regulatory role on actin dynamics and its subsequent impact on ciliogenesis . Notably , we demonstrate that Luzp1-/- cells exhibit reduced filamentous actin ( F-actin ) , longer primary cilia , higher rates of ciliogenesis and increased Shh signaling . Furthermore , TBS-derived primary fibroblasts show a reduction in LUZP1 and actin filaments , possibly through SALL1-regulated LUZP1 degradation via the ubiquitin ( Ub ) -proteasome system ( UPS ) . As a novel regulator of ciliogenesis and the actin cytoskeleton , LUZP1 might contribute to the aberrant cilia phenotype in TBS . Using proximity proteomics , we have previously shown that a truncated and mislocalized form of SALL1 present in TBS individuals ( SALL1275 ) can interact aberrantly with cytoplasmic proteins ( Bozal-Basterra et al . , 2018 ) . LUZP1 was found among the most enriched proteins in the SALL1275 proximal interactome . We confirmed this finding by independent BioID experiments analyzed by western blot using a LUZP1-specific antibody ( Figure 1A and Figure 1—figure supplement 1 ) . To further characterize the interaction of LUZP1 with truncated SALL1 , we performed pulldowns using tagged SALL1275-YFP in HEK 293FT cells . Our results showed that endogenous LUZP1 was able to interact with SALL1275 , confirming our proximity proteomics data ( Figure 1—figure supplement 2A , lane 6 , and Figure 1—figure supplement 1 ) . The interaction with SALL1275 persisted in presence of overexpressed SALL1FL ( Figure 1—figure supplement 2A , lane 9 , and Figure 1—figure supplement 1 ) , suggesting that heterodimerization of the truncated and FL forms does not inhibit the interaction with LUZP1 . When expressed alone , we noted that SALL1FL-YFP also interacts with LUZP1 in pulldown assays ( Figure 1—figure supplement 2A and Figure 1—figure supplement 1 ) . As these proteins have distinct localizations ( nuclear and cytoplasmic , respectively ) , the interaction likely occurs in post-lysis cell extracts ( more in Discussion ) . These results show that the truncated form of SALL1 expressed in TBS individuals , either by itself or in complex with the FL form , can interact with LUZP1 . To gain insights into the function of LUZP1 , we sought to identify its proximal interactome using the TurboID approach ( Branon et al . , 2018 ) . We used RPE1 cells stably expressing low levels of FLAG-TurboID-LUZP1 ( TbID-LUZP1 ) or FLAG-TurboID ( TbID ) as control . Transduced cells showed sub-endogenous expression levels of TbID-LUZP1 ( Figure 1—figure supplement 2B , two asterisks; endogenous , two open arrowheads ) . Staining of transfected cells revealed that , proteins biotinylated by TbID are diffusely localized throughout the nucleus and cytoplasm , whereas those biotinylated by TbID-LUZP1 are localized primarily at the centrosome and actin cytoskeleton , as shown by fluorescent streptavidin ( Figure 1—figure supplement 2C ) . Total lysates from TbID-LUZP1 or TbID-expressing cells were subjected to streptavidin pulldown and isolated proteins were analyzed by liquid chromatography tandem mass spectrometry ( LC-MS/MS ) . 234 high-confidence proximity LUZP1 interactors were enriched in the TbID-LUZP1 vs the TbID proteome in three replicates ( Source Data 1 ) . Proteins enriched among the identified LUZP1 proximal interactors were centrosomal and actin cytoskeleton-related proteins ( Figure 1B–C ) . With the purpose of obtaining a functional overview of the main pathways associated to LUZP1 , a comparative Gene Ontology ( GO ) analysis was performed with all the hits enriched in TbID-LUZP1 versus TbID cells . In the Cellular Component domain , ‘actin cytoskeleton’ , ‘microtubule cytoskeleton’ , ‘centrosome’ , ‘cilium’ and ‘midbody’ terms were highlighted among others ( Figure 1D and Source Data 1 ) . In the category of Biological Process , LUZP1 proteome showed enrichment in the ‘microtubule cytoskeleton organization’ , ‘cell division’ , ‘cilium assembly’ , ‘vesicle-mediated transport’ , ‘microtubule organizing center organization’ and ‘cell adhesion’ categories among others ( Figure 1D and Source Data 1 ) . With respect to Molecular Function , LUZP1 also showed enrichment in cytoskeleton-related proteins ( ‘actin binding’ , ‘actinin binding’ and ‘microtubule binding’ terms; Figure 1D and Source Data 1 ) . 37 or 96 of the verified or potential , respectively , centrosome/cilia gene products previously identified by proteomic studies ( Alves-Cruzeiro et al . , 2014; Gupta et al . , 2015 ) were found as LUZP1 proximal interactors , supporting the enrichment of centrosome-related proteins among the potential interactors of LUZP1 . In addition , 45 of LUZP1 proximal interactors were present among the actin-localized proteins identified by the Human Protein Atlas project based on subcellular localization to actin filaments ( Uhlen et al . , 2015 ) . We examined the subcellular localization of LUZP1 in diverse cell types . First , immunostainings showed that endogenous LUZP1 surrounds both centrioles , labelled by centrin-2 ( CETN2 ) in human RPE1 cells ( Figure 2A ) and gamma-tubulin in human dermal fibroblasts ( Figure 2—video 1 ) ( for specificity of LUZP1 antibody , check Figure 5 ) . We examined LUZP1 localization at the centrosome in synchronized RPE1 cells . LUZP1 was reduced at the centrosome during G2/M and G0 phases ( Figure 2—figure supplement 1 ) . Interestingly , LUZP1 levels increased upon treatment with the proteasome inhibitor MG132 in G0 phase arrested-RPE1 cells . All together , these results indicate that LUZP1 levels are reduced at the centrosome in G2/M phase and upon starvation , and this reduction might be mediated by the UPS . The localization of LUZP1 at the centrosome was reproduced in U2OS cells expressing LUZP1-YFP ( Figure 2B–D ) . We did not observe colocalization of LUZP1 with the distal centriolar marker CCP110 ( Figure 2B ) , indicating that LUZP1 is likely found at the proximal end of both centrioles . We further imaged LUZP1 along with pericentrin ( PCNT ) and PCM1 , markers of PCM . Interestingly , we observed that LUZP1 enveloped PCNT ( Figure 2C ) , with LUZP1 itself being surrounded by PCM1 ( Figure 2D ) . These results suggest that LUZP1 might be a novel PCM associated-protein , decorating the proximal end of both centrioles . In concordance with this localization , LUZP1 was associated to PCM1 in TurboID experiments ( Source Data 1 ) . In addition to the localization at the centrosome/basal body , LUZP1 also localized to actin stress fibers ( Figure 2E ) and to the midbody ( Figure 2F ) in U2OS cells . Based on the LUZP1 interaction with truncated SALL1 , we checked its subcellular localization in fibroblasts derived from a TBS individual ( TBS275; see Materials and methods ) as well as non-TBS control . We observed that LUZP1 was markedly decreased at the centrosome of TBS275 cells compared to control cells in non-starved conditions ( Figure 3A–C ) . Furthermore , LUZP1 levels decreased in starved vs non-starved control cells ( Figure 3A–C ) , while centrosomal size remained unaltered ( Figure 3C , right panel ) . We previously found that SALL1275-YFP interacted with the centrosome-associated ciliogenesis suppressors , CCP110 and CEP97 ( Bozal-Basterra et al . , 2018 ) , so we checked whether LUZP1 could also interact with these factors . Indeed , LUZP1-YFP interacts with CCP110 and CEP97 in both WT ( 293WT ) and TBS model ( 293335 ) HEK 293FT cells ( Figure 3D , lanes 5 and 7 , respectively and Figure 3—figure supplement 1; Bozal-Basterra et al . , 2018 ) . Less CCP110 and CEP97 was recovered in LUZP1-YFP pulldowns from 293335 cells , but this is likely due to the reduced LUZP1-YFP seen in those cells ( Figure 3D , Input , lanes 1 and 2 vs lane 3 and 4 and Figure 3—figure supplement 1 ) . Beyond pulldowns , we found that immunoprecipitation of endogenous LUZP1 led to co-purification of endogenous CCP110 ( Figure 3E and Figure 3—figure supplement 1 ) and that anti-CEP97 antibodies immunoprecipitated endogenous LUZP1 ( Figure 3F and Figure 3—figure supplement 1 ) . Both CCP110 and CEP97 were also identified as proximal interactors of TbID-LUZP1 ( Source Data 1 ) . Immunofluorescent colocalization on centrioles of LUZP1 ( proximal ) and CCP110 ( distal ) was not evident , suggesting that the interaction is indirect or occurs before proteins reach their destinations . However , these key regulators of ciliogenesis are just two of multiple centrosomal proteins associated with LUZP1 , suggesting that LUZP1 may have a function at this dynamic organelle . In addition to the centrosome/basal body , LUZP1 also localized to actin stress fibers , as well as the midbody in dividing cells ( Figure 2 ) . Intriguingly , when LUZP1 levels were examined in TBS275 cells , a reduction in both actin-associated LUZP1 and phalloidin-labelled stress fibers was observed when compared to control cells ( Figure 4A–C ) . These results indicate that actin cytoskeleton might be altered in TBS cells . Using pulldown assays , we confirmed that LUZP1-YFP interacts with both actin and FLNA ( Figure 4D and Figure 4—figure supplement 1 ) . Notably , actin , FLNA , alpha-actinin , palladin , LIMA1/Eplin and other stress fiber-associated proteins are proximal interactors of TbID-LUZP1 ( Source Data 1 ) . To examine whether LUZP1 levels change upon F-actin perturbation , HEK 293FT cells were treated with Cytochalasin D ( CytoD ) , an inhibitor of actin polymerization . No changes in LUZP1 levels upon actin depolymerization were observed when cells were lysed in strong lysis conditions ( WB5 ) ( Figure 4E–F and Figure 4—figure supplement 1 ) . However , we observed increased LUZP1 levels using mild lysis conditions ( 0 . 1% Triton X-100; Figure 4E–F and Figure 4—figure supplement 1 ) . These results reflect that the integrity of the actin cytoskeleton may influence the solubility but not the stability of LUZP1 . Based on the localization of LUZP1 at the centrosome , its interaction with centrosomal proteins and the defects in ciliogenesis previously observed in TBS cells ( Bozal-Basterra et al . , 2018 ) , we hypothesized that LUZP1 might have a role in cilia formation . To examine this , we analyzed ciliogenesis in Shh-LIGHT2 cells , a cell line derived from immortalized mouse NIH3T3 fibroblasts that can display primary cilia and report on Shh pathway status using integrated luciferase reporters ( herein designated as WT ) ( Taipale et al . , 2000 ) . Using CRISPR/Cas9 gene editing directed to exon 1 of murine Luzp1 , we generated Shh-LIGHT2 mouse embryonic fibroblasts null for Luzp1 ( Luzp1-/- cells ) . For genetic rescue experiments , LUZP1 was restored to these cells by the expression of human LUZP1-YFP fusion ( +LUZP1 cells ) . To examine the effect of the Luzp1 mutation and rescue strategies , we used anti-LUZP1 antibody and checked LUZP1 localization associated with the actin cytoskeleton and the centrosome by immunofluorescence ( Figure 5A , B ) , and its levels of expression by western blot ( Figure 5C and Figure 5—figure supplement 1 ) in WT , Luzp1-/- and +LUZP1 cells . These experiments showed the effectiveness of the knockout and rescue strategies . To analyze the role of LUZP1 in ciliation , WT , Luzp1-/- and +LUZP1 cells were plated at equal densities and induced to ciliate for 48 hr by serum withdrawal ( starved; Figure 6A ) . We quantified ciliation rates and primary cilia length in non-starved and starved cells . Luzp1-/- fibroblasts displayed higher ciliation rate ( 60% ) than WT ( 10 . 5% ) and +LUZP1 ( 22 . 2% ) when the cells were not subjected to starvation ( Figure 6B ) . However , Luzp1-/- cells were not significantly more ciliated than WT or +LUZP1 fibroblasts upon 48 hr of starvation ( Figure 6B ) . In addition , primary cilia in Luzp1-/- cells were significantly longer than in non-starved WT cycling cells ( Figure 6A and C ) , while under starvation the differences were not significant ( Figure 6A and C ) . Note that , differently than the ciliation rate , cilia length was not rescued by adding human LUZP1 . Taken together , these results confirm that Luzp1-/- cells display longer and more abundant primary cilia compared to WT cells in cycling conditions and indicate that LUZP1 might affect primary cilia dynamics . One key event in ciliogenesis is the depletion of CCP110 and its partner CEP97 from the distal end of the MC , promoting the ciliary activating program in somatic cells ( Goetz et al . , 2012; Kleylein-Sohn et al . , 2007; Prosser and Morrison , 2015; Spektor et al . , 2007; Tsang et al . , 2008 ) . We analyzed the centrosomal localization of CCP110 in WT and Luzp1-/- cells by immunofluorescence . Consistent with the higher ciliogenesis rate , CCP110 was present at two centrosomal spots at a lower proportion in Luzp1-/- cells ( 19% ) compared to WT ( 84%; Figure 6D and E ) . This result suggests that the lack of LUZP1 might result in CCP110 reduction at the centrosome , leading to higher frequency of ciliogenesis in Luzp1-/- cells , and is reminiscent to the results obtained in TBS cells ( Bozal-Basterra et al . , 2018 ) . It is well-established that mammalian Shh signal transduction is dependent on functional primary cilia ( Huangfu et al . , 2003; Yin et al . , 2009 ) . Therefore , we examined whether Shh signaling is altered in Luzp1-/- cells . Cells were starved for 24 hr and incubated in the presence or absence of purmorphamine ( a SMO agonist ) for 24 hr to activate the Shh pathway . The mRNA expression of two Shh target genes ( Gli1 and Ptch1 ) was quantified by qRT-PCR ( Figure 7A , B ) . We found that Gli1 and Ptch1 expression levels in non-treated Luzp1-/- cells were higher than in WT cells ( Gli1 1 . 5 fold and Ptch1 2 . 3 fold increase in Luzp1-/- vs WT cells without purmorphamine ) ( Figure 7A , B ) . To further study the role of LUZP1 in Shh signaling , we analyzed GLI3 processing by western blot using total lysates extracted from WT vs Luzp1-/- cells . Without purmorphamine induction , we found a significantly higher ratio of GLI3 activating form vs GLI3 repressive form ( GLI3-A:GLI3-R ) in Luzp1-/- cells compared to WT ( 2 . 9 fold increase in Luzp1-/- cells vs WT ) ( Figure 7C and Figure 7—figure supplement 1 ) . After induction , the values were similar for Luzp1-/- and WT cells . We also examined the effects of lacking Luzp1 on Shh signaling by measuring the activity of the Shh-responsive Firefly luciferase reporter in the presence or absence of purmorphamine for 24 hr to activate the Shh pathway ( Figure 7D , E ) . Consistent with the Gli1 and Ptch1 qRT-PCR data , non-treated Luzp1-/- cells showed higher Shh activity compared to control or +LUZP1 cells , as observed in TBS-derived cells ( Figure 7D ) . However , the induction capacity of Luzp1-/- cells upon purmorphamine treatment was reduced compared to WT or +LUZP1 cells ( Figure 7E ) . Altogether , the observed defects in Ptch1 and Gli1 gene expression and Shh reporter misregulation point to a role for LUZP1 in Shh signaling . Based on the localization of LUZP1 to actin stress fibers and interaction with cytoskeletal proteins , we hypothesized that LUZP1 might also affect F-actin cytoskeleton . We observed a reduction in F-actin ( labelled by phalloidin ) in the Luzp1-/- cells compared to WT , which was recovered in +LUZP1 cells ( Figure 8A ) . We also note the correlation between LUZP1 levels and actin filaments in non-starved versus starved WT fibroblasts ( Figure 8B , C ) . These results suggest that LUZP1 can stabilize actin stress fibers and that starvation triggers both LUZP1 and F-actin reduction . In concordance with immunofluorescence results in Figures 3 and 4 , we confirmed a reduction in total LUZP1 levels in TBS275 cells compared to controls by western blot ( Figure 9A , B and Figure 9—figure supplement 1 ) . No transcriptional changes in LUZP1 expression were detected between control and TBS275 samples ( Figure 9—figure supplement 2 ) , so we hypothesized that truncated SALL1 might lead to UPS-mediated LUZP1 degradation . We analyzed LUZP1 levels after treatment with the proteasome inhibitor MG132 , both in control and TBS275 cells , and LUZP1 levels were increased to a higher extent in TBS275 compared to control cells ( 1 . 8 fold increase in control vs 2 . 4 fold increase in TBS275 cells ) ( Figure 9A , B ) . Moreover , we confirmed the reduction of LUZP1 levels in the CRISPR/Cas9 TBS model cell line ( 293335 ) , compared to its parental cell line ( 293WT ) ( Figure 9C , D and Figure 9—figure supplement 1 ) , and likewise in HEK 293FT cells stably overexpressing truncated SALL1 ( SALL1275-YFP ) compared to cells with YFP as control ( Figure 9E , F and Figure 9—figure supplement 1 ) . A more prominent increase in LUZP1 accumulation upon MG132 treatment was also observed in 293335 and HEK 293FT cells overexpressing SALL1275-YFP compared to controls ( Figure 9C , D and Figure 9E , F , respectively , and Figure 9—figure supplement 1 ) . Additionally , we also observed LUZP1 accumulation upon MG132 treatment by immunofluorescence in RPE1 cells , both at the actin cytoskeleton ( Figure 9G , upper panels ) and at the centrosome ( Figure 9G , lower panels ) . All together , these results show that LUZP1 levels are sensitive to degradation via the UPS pathway and suggest that truncated SALL1 may contribute to this process . Furthermore , we compared LUZP1 ubiquitination in 293WT vs 293335 cells using the BioUb strategy ( see Materials and methods ) ( Pirone et al . , 2017 ) . In the pulldowns , we could observe a prominent band in presence of BioUb , possibly corresponding to a monoubiquitinated form of LUZP1 ( Figure 9H and Figure 9—figure supplement 1 ) . This form was present in 293WT and 293335 cells , and increased in the presence of MG132 in both cell lines . In addition , we observed a smear at higher molecular weight corresponding to polyubiquitinated forms of LUZP1 ( Figure 9H , Biotin PD and Figure 9—figure supplement 1 ) . Notably , the LUZP1 ubiquitinated pool relative to the input levels was higher in 293335 compared to 293WT cells upon MG132 treatment ( Figure 9H , Biotin PD , lane 8 vs lane 11 and Figure 9—figure supplement 1 ) . These results suggest that truncated SALL1 promotes LUZP1 degradation through the UPS pathway . Our results suggest that LUZP1 could be a mediator of TBS cilia phenotype and that this could be caused , at least in part , by the increased degradation of LUZP1 triggered by truncated SALL1 . Therefore , increasing LUZP1 levels in TBS cells might affect the cilia and actin cytoskeleton phenotypes . To check whether an increase in LUZP1 levels is sufficient to repress ciliogenesis in primary human fibroblasts , Control and TBS275 cells were transduced with YFP or LUZP1-YFP using lentivirus ( Figure 10A ) . Whereas most non-transduced surrounding cells , as well as 100% of the TBS275 cells expressing YFP were ciliated , only 40% of the Control and TBS275 cells transduced with LUZP1-YFP displayed cilia ( Figure 10B ) . Furthermore , we checked the actin cytoskeleton defects observed in TBS275 cells to see the effect of overexpressing LUZP1-YFP . Immunostaining showed that LUZP1-YFP overexpression led to an increase in F-actin levels both in control and in TBS275 cells compared to the surrounding non-transfected cells or TBS275 cells overexpressing YFP ( Figure 10C ) . F-actin has a suppressive effect on ciliogenesis , and CytoD-mediated actin depolymerization has been shown to be permissive to cilia formation in cultured cells ( Kim et al . , 2010 ) . To corroborate the relationship between LUZP1 , actin and ciliogenesis , we performed CytoD treatment experiments . We observed that LUZP1-YFP overexpression can counteract the positive effects of CytoD on cilia formation ( Figure 10—figure supplement 1 ) . All together , these results support the notion that LUZP1 is a negative regulator of cilia formation and an F-actin stabilizing protein . LUZP1 was previously described as a nuclear protein , with expression limited to the mouse brain ( Lee et al . , 2001; Sun et al . , 1996 ) . We tested two different commercial antibodies against LUZP1 by immunofluorescence and , while nuclear localization was sporadic and weak , the most prominent localization of LUZP1 was observed in the actin cytoskeleton and centrosome , both in human and mouse cells . This localization is consistent with our TurboID analysis that showed an enrichment of factors associated with the actin cytoskeleton and/or centrosomes among the potential interactors of LUZP1 . The localization of LUZP1 to the actin cytoskeleton , as well as its expression in tissues beyond the brain , is consistent with independent validation in cell lines by the Human Protein Atlas ( HPA; proteinatlas . org ) and other expression databases ( e . g . EMBL EBI Expression Atlas ebi . ac . uk/gxa ) . Two independent proximity labeling studies identified LUZP1 as a proximal interactor of centriole ( Gupta et al . , 2015 ) and centriolar satellite-related proteins ( Gheiratmand et al . , 2019 ) . Moreover , LUZP1 has been recently reported as a centrosomal protein involved in cilia regulation ( Gonçalves et al . , 2019 ) . These localizations are also consistent with fluorescent LUZP1 fusion proteins ( this work; [Gupta et al . , 2015; Gonçalves et al . , 2019] #117 ) . Discrepancies with the previously reported LUZP1 localization and distribution might be due to technical differences , such as epitope specificity for the antisera used in the immunohistochemistry . Here , we report that LUZP1 surrounds the proximal end of both centrioles . Like LUZP1 , a large number of centrosomal scaffold proteins contain coiled-coil regions , and the proteins are concentrically localized around a centriole in a highly organized fashion ( e . g . Cep120 , Cep57 , Cep63 , Cep152 , CPAP , Cdk5Rap2 , PCNT ) ( Fu and Glover , 2012; Lawo et al . , 2012; Mennella et al . , 2012 ) . Furthermore , we show that LUZP1 interacts with centrosome and actin-related proteins ( Figure 3 and Figure 4 ) . LUZP1 is associated with CCP110 and CEP97 , using pull down , immunoprecipitation and proximity proteomics approaches . This association is likely to be dynamic and potentially indirect , via other bridging factors , since our immunostainings show that LUZP1 and CCP110 are located in different areas of the centrioles . LUZP1 has also been identified as an interactor of ACTR2 ( ARP2 actin related protein two homologue ) and FLNA ( Hein et al . , 2015; Wang and Nakamura , 2019 ) , and it has been recently described as an actin cross-linking protein ( Wang and Nakamura , 2019 ) . We found that LUZP1 localizes not only to centrioles and actin cytoskeleton , but also to the midbody in dividing cells , which was recently reported to influence ciliogenesis in polarized epithelial cells ( Bernabé-Rubio et al . , 2016 ) . Our data suggest that the association of LUZP1 to centrosomes and actin filaments may contribute to its overall roles . Actin dynamics coordinate several processes that are crucial for ciliogenesis . For example , positioning the MC to the appropriate area at the cell cortex is an actin-dependent process ( Boisvieux-Ulrich et al . , 1990; Euteneuer and Schliwa , 1985 ) . A reduction in cortical actin should promote ciliogenesis , since there is less physical restriction for docking of the MC . Supporting this hypothesis , several studies have found that disruptions in the actin cytoskeleton , induced either chemically or genetically , promote ciliogenesis or affect cilia length ( Cao et al . , 2012; Drummond et al . , 2018; Hernandez-Hernandez et al . , 2013; Kang et al . , 2015; Kim et al . , 2015; Kim et al . , 2010 ) . How actin regulates cilium length is not clear . Recently , a role for actin has been implicated in ectocytosis and cilium tip scission , preventing the axoneme from growing too long ( Nager et al . , 2017; Phua et al . , 2017 ) . Whether LUZP1 at the centrosome is complexed with filamin and actin is unknown , but if so , they could together serve to stabilize the basal body as the axoneme extends or is subjected to mechanical stress . TBS is caused by mutations in SALL1 gene , which give rise to truncated proteins that interfere with the normal function of the cell . We show that LUZP1interacts with truncated SALL1 and with SALL1FL , suggesting that interaction occurs through an N-terminal domain shared by both . In control cells , LUZP1 and SALL1FL likely have minimal or no interaction due to their respective localizations to the cytoplasm and nucleus . However , truncated SALL1 , alone or together with SALL1FL that is retained in the cytoplasm , can interact with cytoplasmic LUZP1 , promoting its degradation and functional inhibition . Importantly , we detected an increase in LUZP1 levels upon treatment with the proteasome inhibitor MG132 ( Figure 9 ) , suggesting that LUZP1 degradation is proteasome-dependent . Next , we demonstrated that LUZP1 is ubiquitinated , and that truncated SALL1 both increases LUZP1 ubiquitination and decreases its stability . In agreement with our findings , LUZP1 ubiquitination has been detected in several proteomic screens for ubiquitinated proteins ( Akimov et al . , 2018; Mertins et al . , 2013; Povlsen et al . , 2012; Udeshi et al . , 2013; Wagner et al . , 2012 ) . Further experiments would be required to understand the precise mechanism by which truncated SALL1 can influence LUZP1 ubiquitination , but one possibility could be de novo complexes involving specific Ub E3 ligases or de-ubiquitinases which could influence LUZP1 stability . In fact , various E3s/de-ubiquitinases , as well as other components of the UPS , were found as proximal interactors of truncated SALL1 and LUZP1 . Furthermore , regulation by the UPS system has been reported for centrosomal factors , including the cilia regulator CCP110 ( D'Angiolella et al . , 2010; Hossain et al . , 2017; Li et al . , 2013 ) . Both Luzp1-/- and TBS cells showed a reduction in F-actin accompanied by an increase in ciliation . We suggest that the reduction in F-actin in TBS cells might contribute to their higher cilia abundance , longer cilia and increased Shh signaling . An increase in LUZP1 in control and TBS275 cells is sufficient to increase F-actin levels and reduce cilia frequency , supporting that LUZP1 may have a contributing role in the TBS phenotype . Cilia formation , Shh signaling , and the actin cytoskeleton is aberrant in TBS patient-derived fibroblasts ( this work; [Bozal-Basterra et al . , 2018] ) . Changes in Shh signaling have not been reported in mouse models for TBS , nor in any other cell types or tissues derived from TBS patients . Nevertheless , the phenotypes observed in TBS individuals fall within the spectrum of those observed in ciliopathies , characterized by malformations in digits , ears , heart , brain , kidneys and urogenital anomalies , phenotypes that are consistent with misregulated Shh signaling . Preaxial polydactyly has been associated with ectopic Shh expression in limbs ( Dunn et al . , 2011; Johnson et al . , 2014; Lettice et al . , 2003; Lettice et al . , 2008; Liem et al . , 2009; Zhulyn and Hui , 2015 ) ; Anal stenosis or imperforate anus have been related to misregulation of Shh pathway ( Kang et al . , 1997; Mo et al . , 2001; Roberts et al . , 1995 ) , as well as deafness and dysplastic ears ( Driver et al . , 2008 ) . We observed primary cilia , Shh signaling and cytoskeletal defects in Luzp1-/- cells . Several studies have implicated defective primary cilia and Shh signaling in the etiology of neural tube closure defects , as well as crucial roles for the actin cytoskeleton ( Wallingford , 2005 ) . There are certain parallels between phenotypes observed in animal models of Luzp1 and Sall1 . Exencephaly and neural tube defects were detected in mice and Xenopus Sall1 mutants ( Böhm et al . , 2008; Exner et al . , 2017; Kiefer et al . , 2003 ) . Luzp1 KO mice exhibit ectopic Shh expression in the hindbrain neuroepithelium and display NTDs , however the expression of Shh-responsive genes ( such as Gli1 or Ptch1 ) was not reported ( Hsu et al . , 2008 ) . Perhaps the role of LUZP1 in Shh signaling , in spatial control of the signal or the response ( or both ) , contributes to the NTD phenotype . Exencephaly may also be caused by failure in bending at the dorsolateral hinge point of the neural folds , where cells undergo changes in apical actin architecture ( Sadler et al . , 1982 ) . Luzp1 KO embryos exhibited dorsolateral neural folds that were convex instead of the concave morphology observed in WT embryos ( Hsu et al . , 2008 ) , suggested that defective actin dynamics may contribute to the NTD phenotype . Taken together , defective actin dynamics , aberrant primary cilia and changes in Shh signaling might lead to NTDs observed in the LUZP1 mouse model , as well as other animal models of TBS and loss of Sall-related proteins . In addition to NTDs , cardiac malformations are another feature found in human ciliopathies ( Klena et al . , 2017 ) . Cardiac defects are observed in Luzp1 knockout mice ( Hsu et al . , 2008 ) , as well as TBS patients ( Kohlhase , 1993 ) . TBS cardiac defects include atrial or ventricular septal defect , the latter of which is seen in Luzp1 knockout mice . Moreover , compound Sall1/Sall4 KO mutant mice exhibit both NTDs and cardiac problems ( Böhm et al . , 2008 ) . While Luzp1 and Sall1 may both contribute to neural tube and heart development , a novel crosstalk may arise in TBS due to dominantly-acting truncated SALL1 that could derail these processes and cause deformities . In conclusion , our data indicate that LUZP1 functions as a cilia suppressor ( Figure 10D ) . It localizes to actin stress fibers and to the centrosome . In starved cells , likewise in TBS cells , overall LUZP1 levels are diminished in both structures , which facilitates the formation of the primary cilia . In TBS cells , the truncated form of SALL1 localizes to the cytoplasm , interacting with LUZP1 and other factors , leading to the degradation of LUZP1 , simulating what happens when control cells undergo starvation . As a result , the frequency of cilia formation increases , and cilia are longer than in control cells . Ciliogenesis requires communication between the actin cytoskeleton and the centrosome . Here , we propose that LUZP1 might contribute as a nexus in this complex intracellular network that is disrupted by truncated SALL1 . Our findings point to the intriguing possibility that LUZP1 might be a key relay switch in this network that , together with other factors , might contribute to the phenotypes observed in TBS . TBS-derived primary fibroblasts , U2OS ( ATCC HTB-96 ) , HEK 293FT ( Invitrogen R70007 ) , and mouse Shh-LIGHT2 cells ( Taipale et al . , 2000 ) were cultured at 37°C and 5% CO2 in Dulbecco’s modified Eagle medium ( DMEM ) supplemented with 10% fetal bovine serum ( FBS , Gibco ) and 1% penicillin/streptomycin ( Gibco ) . Human telomerase reverse transcriptase immortalized retinal pigment epithelial cells ( hTERT-RPE1 , ATCC CRL-4000; designated here as RPE1 ) were cultured in DMEM:F12 ( Gibco ) supplemented with 10% FBS and 1% penicillin and streptomycin . Dermal fibroblasts carrying the SALL1 pathogenic variant c . 826C > T ( SALL1c . 826C>T ) , that produce a truncated protein p . Leu275* ( SALL1275 ) , were derived from a male TBS individual UKTBS#3 ( called here TBS275 ) ( Bozal-Basterra et al . , 2018 ) . Adult female dermal fibroblasts ( ESCTRL#2 ) from healthy donors were used as controls . Cell lines were authenticated by commercial providers ( Invitrogen , ATCC ) . Additional validation was done by TBS allele genotyping ( primary fibroblasts ) and reporter selection ( zeoR , neoR; Shh-LIGHT2 ) . Cultured cells were maintained between 10 and 20 passages , tested for senescence by γ-H2AX staining and mycoplasma contamination and grown until confluence ( 6-well plates for RNA extraction and western blot assays; 10 cm dishes for pulldowns ) . The use of human samples in this study was approved by the institutional review board ( Ethics Committee at CIC bioGUNE , protocol P-CBG-CBBA-2111 ) and appropriate informed consent was obtained from human subjects or their parents . RPE1 cells were arrested in G1 phase by treatment with mimosine ( Sigma , 400 µM ) for 24 hr . For S phase arrest , cells were subjected to thymidine treatment ( Sigma , 2 . 5 mM ) for 16 hr , followed by release for 8 hr , and subsequently blocked again for 16 hr . For G2/M phase arrest , cells were treated with RO-3306 ( Sigma , 10 µM ) for 20 hr . For G0 phase synchronization and inducing primary cilia formation , cells were starved for 48 hr ( DMEM , 0% FBS , 1% penicillin and streptomycin ) . Treatments with the proteasome inhibitor MG132 ( Calbiochem , 5 µM ) were for 15 hr and with CytoD ( Sigma , 50 nM ) for 16 hr to stimulate actin depolymerization . HEK 293FT cells were transfected using calcium phosphate method and U2OS cells using Effectene Transfection Reagent ( Qiagen ) . CRISPR-Cas9 targeting of SALL1 locus was performed to generate a HEK 293FT cell line carrying a TBS-like allele ( Bozal-Basterra et al . , 2018 ) . The mouse Luzp1 locus was targeted in NIH3T3-based Shh-LIGHT2 fibroblasts ( Taipale et al . , 2000 ) ( kind gift of A . McGee , Imperial College ) . These are NIH3T3 mouse fibroblasts that carry an incorporated Shh reporter ( firefly luciferase under control of Gli3-responsive promoter ) . Cas9 was introduced into Shh-LIGHT2 cells by lentiviral transduction ( Lenti-Cas9-blast; Addgene #52962; kind gift of F . Zhang , MIT ) and selection with blasticidin ( 5 µg/ml ) . Two high-scoring sgRNAs were selected ( http://crispr . mit . edu/ ) to target near the initiation codon ( sg2: 5’-CTTAAATCGCAGGTGGCGGT_TGG-3’; sg3: 5’-CTTCAATCTTCAGTACCCGC_TGG-3’ ) . These sequences were cloned into px459 2 . 0 ( Addgene #62988; kind gift of F . Zhang , MIT ) , for expressing both sgRNAs and additional Cas9 with puromycin selection . Transfections were performed in Shh-LIGHT2/Cas9 cells with Lipofectamine 3000 ( Thermo ) . 24 hr after transfection , transient puromycin selection ( 0 . 5 µg/ml ) was applied for 48 hr to enrich for transfected cells . Cells were plated at clonal density , and well-isolated clones were picked and propagated individually . Western blotting was used to identify clones lacking Luzp1 expression . Further propagation of a selected clone ( #6 ) was carried out with G418 ( 0 . 4 mg/ml ) and zeocin ( 0 . 15 mg/ml ) selection to maintain expression of luciferase reporters . Genotyping was performed using genomic PCR ( MmLuzp1_geno_for: 5’-GTTGCCAAAGAAGGTTGTGGATGCC-3’; MmLuzp1_geno_rev: 5’-CGTAAGGTTTTCTTCCTCTTCAAGTTTCTC-3’ ) . We found that Luzp1-/- cells presented a homozygous deletion of the sequence: 5’-ccacctgcgatttaagttacagagcctgagccgccgcctcgatgagttagaggaagctacaaaaaacctccagagagcagaggatgagctcctggacctccaggacaaggtgatccaggcagagggcagcgactccagcacgctggctgagatcgaggtgctgcgccagcgg-3’ . This generated a truncation and a stop codon early in the N-terminal part of the protein . The resulting peptide was: MAELTNYKDAASNRY* . A rescue cell line was generated by transducing Shh-LIGHT2 Luzp1 KO clone #6 with a lentiviral expression vector carrying EFS-LUZP1-YFP-P2A-blastR , with a positive population selected by fluorescence-activated cell sorting . SALL1 truncated ( SALL1275-YFP or Myc-BirA-SALL1275 ) and FL versions ( SALL1FL-YFP , SALL1FL-2xHA or Myc-BirA-SALL1FL ) were previously described ( Bozal-Basterra et al . , 2018 ) . Human LUZP1 ORF was amplified by high-fidelity PCR ( Platinum SuperFi; Thermo ) from RPE1 cell cDNA and cloned to generate CB6-GFP-LUZP1 . This was used as a source clone to generate additional variants , including CMV-LUZP1-YFP . The LUZP1-YFP and TbID-LUZP1 lentiviral expression vectors were generated by replacing Cas9 in Lenti-Cas9-blast ( Addgene #52962 ) . All constructs were verified by Sanger sequencing . Plasmids CAG-BioUBC ( x4 ) _BirA_V5_puro ( called here BioUb ) and CAG-BirA-puro ( called here BirA ) were reported previously ( Pirone et al . , 2017 ) . Using the BioID and the TurboID methods ( Branon et al . , 2018; Roux et al . , 2012 ) , proteins in close proximity to SALL1 and LUZP1 , respectively , were biotinylated and isolated by streptavidin-bead pulldowns . For transient transfections , Myc-BirA-SALL1c . 826C>T or Myc-BirA-SALL1FL were used in HEK 293FT cells ( 10 cm dishes ) . For TurboID experiments , TbID-LUZP1-P2A-blast or TbID-P2A-blast alone were transduced in RPE1 cells and a stable population was selected . For the isolation of BioUb-conjugates 10 cm dishes were transfected with BioUb or BirA as control ( Pirone et al . , 2017 ) . Briefly , 24 hr after transfection , medium was supplemented with biotin at 50 μM . Cells were collected 48 hr after transfection , washed 3 times on ice with cold phosphate buffered saline ( PBS ) and scraped in lysis buffer [8 M urea , 1% SDS , 1x protease inhibitor cocktail ( Roche ) , 60 μM NEM in 1x PBS; 1 ml per 10 cm dish] . At room temperature , samples were sonicated and cleared by centrifugation . Cell lysates were incubated overnight with 40 μl of equilibrated NeutrAvidin-agarose beads ( Thermo Scientific ) . Beads were subjected to stringent washes using the following washing buffers ( WB ) , all prepared in PBS: WB1 ( 8 M urea , 0 . 25% SDS ) ; WB2 ( 6 M Guanidine-HCl ) ; WB3 ( 6 . 4 M urea , 1 M NaCl , 0 . 2% SDS ) , WB4 ( 4 M urea , 1 M NaCl , 10% isopropanol , 10% ethanol and 0 . 2% SDS ) ; WB5 ( 8 M urea , 1% SDS ) ; and WB6 ( 2% SDS ) . For elution of biotinylated proteins , beads were heated at 99°C in 50 μl of Elution Buffer ( 4x Laemmli buffer , 100 mM DTT ) . Beads were separated by centrifugation ( 18000 x g , 5 min ) . Lentiviral expression constructs were packaged using psPAX2 and pVSV-G ( Addgene ) in HEK 293FT cells , and lentiviral supernatants were used to transduce Shh-LIGHT2 cells , RPE1 cells , or TBS275 and control human fibroblasts . Stable-expressing populations were selected using puromycin ( 1 µg/ml ) or blasticidin ( 5 µg/ml ) . The vectors EFS-LUZP1-YFP-P2A-blastR , EFS-YFP-P2A-blastR , LL-GFS-SALL1c . 826C>T-IRES-puroR , LL-GFS-stop-IRES-puroR , EFS-TbID-LUZP1-P2A-blastR and EFS-TbID-P2A-blastR were used . Lentiviral supernatants were concentrated 100-fold before use ( Lenti-X concentrator , Clontech ) . Concentrated virus was used for transducing primary fibroblasts and RPE1 cells . Analysis was done in RPE1 cells stably expressing TbID or TbID-LUZP1 at sub-endogenous levels . Three independent pulldown experiments ( 1 . 5 × 108 cells per replicate ) were analyzed by MS . Samples eluted from the NeutrAvidin beads were separated in SDS-PAGE ( 50% loaded ) and stained with Sypro-Ruby ( Biorad ) according to manufacturer’s instructions . Entire gel lanes were excised , divided into pieces and in-gel digested with trypsin . Recovered peptides were desalted using stage-tip C18 microcolumns ( Zip-tip , Millipore ) and resuspended in 0 . 1% FA prior to MS analysis . Samples were analyzed in a novel hybrid trapped ion mobility spectrometry – quadrupole time of flight mass spectrometer ( timsTOF Pro with PASEF , Bruker Daltonics ) coupled online to a nanoElute liquid chromatograph ( Bruker ) . This mass spectrometer takes advantage of a novel scan mode termed parallel accumulation – serial fragmentation ( PASEF ) , which multiplies the sequencing speed without any loss in sensitivity ( Meier et al . , 2015 ) , providing outstanding analytical speed and sensibility for proteomics analyses ( Meier et al . , 2018 ) . Sample ( 200 ng ) was directly loaded in a 15 cm Bruker nanoelute FIFTEEN C18 analytical column ( Bruker ) and resolved at 400 nl/min with a 100 min gradient . Column was heated to 50°C using an oven . Protein identification and quantification was carried out using PEAKS software ( Bioinformatics solutions ) . Searches were carried out against a database consisting of human entries ( Uniprot/Swissprot ) , with precursor and fragment tolerances of 20 ppm and 0 . 05 Da . Only proteins identified with at least two peptides at FDR < 5% were considered for further analysis . Data were loaded onto Perseus platform ( Tyanova et al . , 2016 ) and further processed ( Log2 transformation , selection of proteins with at least two valid values in at least one condition , imputation ) . A t-test was applied in order to determine the statistical significance of the differences detected , and heatmaps were generated using this tool . Protein IDs were ranked according to the number of peptides found and their corresponding intensities . Gene ontology ( GO ) term enrichment was analyzed using g:GOSt Profiler , a tool integrated in the g:Profiler web server ( Reimand et al . , 2016 ) . GO enrichment was obtained by calculating –Log10 of the P-value . Network analysis of LUZP1 interactors was performed using the String app version 1 . 4 . 2 in Cytoscape version 3 . 7 . 2 , with a high confidence interaction score ( 0 . 7 ) . Transparency and width of the edges were continuously mapped to the String score ( textmining , databases , coexpression , experiments , fusion , neighborhood and cooccurrence ) . Color , transparency and size of the nodes were discretely mapped to the Log2 enrichment value as described in Figure 1 . The Molecular COmplex DEtection ( MCODE ) plug-in version 1 . 5 . 1 was used to identify highly connected subclusters of proteins ( degree cutoff of 2; Cluster finding: Haircut; Node score cutoff of 0 . 5; K-Core of 2; Max . Depth of 100 ) . All steps were performed at 4°C . HEK 293FT transfected cells were collected after 48 hr , washed 3 times with 1x PBS and lysed in 1 ml of lysis buffer [25 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 1 mM EDTA , 1% NP-40 , 0 . 5% Triton X-100 , 5% glycerol , protease inhibitors ( Roche ) ] . Lysates were kept on ice for 30 min vortexing every 5 min and spun down at 25 , 000 x g for 20 min . After saving 40 μl of supernatant ( input ) , the rest was incubated overnight with 30 μl of pre-washed GFP-Trap resin ( Chromotek ) in a rotating wheel . Beads were washed 5 times for 5 min each with WB ( 25 mM Tris-HCl pH 7 . 5 , 300 mM NaCl , 1 mM EDTA , 1% NP-40 , 0 . 5% TX-100 , 5% glycerol ) . Beads were centrifuged at 2000 x g for 2 min after each wash . For elution , samples were boiled for 5 min at 95°C in 2x Laemmli buffer . All steps were performed at 4°C . Cells were collected and lysates were processed as described for GFP-trap pulldowns . After saving 40 μl of supernatant ( input ) , the rest was incubated overnight with 1 µg of anti-CEP97 antibody ( Proteintech ) , or anti-LUZP1 antibody ( Sigma HPA028506 ) and for additional 4 hr with 40 μl of pre-washed Protein G Sepharose 4 Fast Flow beads ( GE Healthcare ) on a rotating wheel . Beads were washed 5 times for 5 min each with WB ( 10 mM Tris-HCl pH 7 . 5 , 137 mM NaCl , 1 mM EDTA , 1% Triton X-100 ) . Beads were centrifuged at 2000 x g for 2 min after each wash . For elution , samples were boiled for 5 min at 95°C in 2x Laemmli buffer . Cells were lysed in cold RIPA buffer ( Cell Signaling Technology ) , WB5 ( 8 M urea , 1% SDS ) or weak buffer ( 10 mM PIPES pH 6 . 8 , 100 mM NaCl , 1 mM EGTA , 3 mM MgCl2 , 300 mM sucrose , 0 . 5 mM DTT , 1% Triton X-100 ) supplemented with 1x protease/phosphatase inhibitor cocktail ( Roche ) . Lysates were kept on ice for 30 min vortexing every 5 min and then cleared by centrifugation ( 25 , 000 x g , 20 min , 4°C ) . Supernatants were collected and protein content was quantified by BCA protein quantification assay ( Pierce ) . After SDS-PAGE and transfer to nitrocellulose membranes , blocking was performed in 5% milk , or in 5% BSA ( Bovine Serum Albumin , Fraction V , Sigma ) in PBT ( 1x PBS , 0 . 1% Tween-20 ) . In general , primary antibodies were incubated overnight at 4°C and secondary antibodies for 1 hr at room temperature ( RT ) . Antibodies used: rabbit anti-LUZP1 ( Proteintech , 1:1 , 000 ) for Figure 1 and rabbit anti-LUZP1 ( Sigma HPA028506 , 1:1 , 000 ) for the rest of the experiments , rabbit anti-CCP110 ( Proteintech , 1:1 , 000 ) , rabbit anti-CEP97 ( Proteintech , 1:1 , 000 ) , mouse anti-GFP ( Roche , 1:1 , 000 ) , mouse anti-GAPDH ( Proteintech , 1:1 , 000 ) , mouse anti-FLNA ( Merck , 1:1 , 000 ) , rabbit anti-BirA ( Sino Biological , 1:1000 ) , HRP-conjugated anti-biotin ( Cell Signaling Technology , 1:2 , 000 ) , rabbit anti Myc ( Cell Signaling Technology , 1:2 , 000 ) , mouse anti-actin ( Sigma , 1:1 , 000 ) , goat anti-GLI3 ( R and D , 1:1 , 000 ) and mouse anti-SALL1 ( R and D , 1:1 , 000 ) . Secondary antibodies were anti-mouse or anti-rabbit HRP-conjugates ( Jackson Immunoresearch ) . Proteins were detected using Clarity ECL ( BioRad ) or Super Signal West Femto ( Pierce ) . Quantification of bands was performed using ImageJ software and normalized against GAPDH or actin levels . At least three independent blots were quantified per experiment . Shh-LIGHT2 cells , RPE1 , U2OS cells and primary fibroblasts from control and TBS individuals were seeded on 11 mm coverslips ( 15 , 000–25 , 000 cells per well; 24well plate ) . After washing 3 times with cold 1xPBS , cells were fixed with 100% methanol for 10 min at −20°C or with 4% PFA supplemented with 0 . 1% Triton X-100 in PBS for 15 min at RT . Then , coverslips were washed 3 times with 1x PBS . Blocking was performed for 1 hr at 37°C in blocking buffer ( BB: 2% fetal calf serum , 1% BSA in 1x PBS ) . Primary antibodies were incubated overnight at 4°C and cells were washed with 1x PBS 3 times . To label the ciliary axoneme and the basal body/pericentriolar region , we used mouse antibodies against acetylated alpha-tubulin ( Santa Cruz Biotechnologies , 1:160 ) and gamma-tubulin ( Proteintech , 1:160 ) . Other antibodies include: rat anti-Centrin-2 ( CETN2 , Biolegend , 1:160 ) , rabbit anti-LUZP1 ( Sigma HPA028506 , 1:100 ) , rabbit anti-CCP110 ( Proteintech , 1:200 ) , rabbit anti-PCM1 ( Cell Signaling Technology , 1:100 ) , rabbit anti ODF2 ( Atlas , 1:100 ) , mouse anti-CEP164 ( Genetex , 1:100 ) , mouse anti beta-tubulin ( DSHB , 1:100 ) and pericentrin ( Abcam , 1:100 ) . Donkey anti-rat , anti-mouse or anti-rabbit secondary antibodies ( Jackson Immunoresearch ) conjugated to Alexa 488 , Alexa 594 or Alexa 633 ( 1:200 ) , GFP-booster ( Chromotek , 1:500 ) , Alexa-594-conjugated Streptavidin ( Jackson Immunoresearch , 1:100 ) and Alexa 568-conjugated phalloidin ( Invitrogen , 1:500 ) , were incubated for 1 hr at 37°C , followed by nuclear staining with DAPI ( 10 min , 300 ng/ml in PBS; Sigma ) . Fluorescence imaging was performed using an upright wide-field fluorescent microscope ( Axioimager D1 , Zeiss ) or super-resolution microscopy ( Leica SP8 Lightning and Zeiss LSM 880 Fast Airyscan ) with 63x Plan ApoChromat NA1 . 4 . For cilia measurements and counting , primary cilia from at least fifteen different fluorescent micrographs taken for each experimental condition were analyzed using the ruler tool from Adobe Photoshop . Cilia frequency was obtained dividing the number of total cilia by the number of nuclei on each micrograph . Number of cells per micrograph was similar in both TBS and control fibroblasts . To estimate the level of fluorescence in a determined region , we used the mean intensity obtained by ImageJ . To obtain the signal histograms on Figure 2C–D , we used the plot profile tool in FIJI . Shh-LIGHT2 cells were starved for 48 hr . Total RNA was obtained with EZNA Total RNA Kit ( Omega ) and quantified by Nanodrop spectrophotometer . cDNAs were prepared using the SuperScript III First-Strand Synthesis System ( Invitrogen ) in 10 µl volume per reaction . LUZP1 , GAPDH , Gli1 , Ptch1 , and Rplp0 primers were tested for efficiency and products checked for correct size before being used in test samples . qPCR was done using PerfeCTa SYBR Green SuperMix Low ( Quantabio ) . Reactions were performed in 10 µl , adding 1 µl of cDNA and 0 . 5 µl of each primer ( 10 µM ) , in a CFX96 thermocycler ( BioRad ) using the following protocol: 95°C for 10 min and 40 cycles of 95°C for 10 s and 55–60°C for 30 s . Melting curve analysis was performed for each pair of primers between 65°C and 95°C , with 0 . 5°C temperature increments every 5 s . Relative gene expression data were analyzed using the ΔΔCt method . Reactions were done in triplicates and results were derived from at least three independent experiments normalized to GAPDH and Rplp0 and presented as relative expression levels . Primer sequences: LUZP1-F: 5´-GGAATCGGGTAGGAGACACCA-3´; LUZP1-R: 5´-TTCCCAGGCAGTTCAGACGGA-3; GAPDH-F: 5'-AGCCACATCGCTCAGACAC-3'; GAPDH-R: 5'-GCCCAATACGACCAAATCC-3'; Gli1-F: 5'-AGCCTTCAGCAATGCCAGTGAC-3'; Gli1-R: 5'-GTCAGGACCATGCACTGTCTTG-3'; Ptch1-F: 5'-AAGCCGACTACATGCCAGAG-3'; Ptch1-R: 5'-TGATGCCATCTGCGTCTACCAG-3' , Rplp0-F: 5'-ACTGGTCTAGGACCCGAGAAG-3'; Rplp0-R: 5'-CTCCCACCTTGTCTCCAGTC-3' . Shh-LIGHT2 cells were starved for 48 hr , and treated or not for the last 24 hr with purmorphamine ( 5 μM , ChemCruz ) to induce Shh signaling pathway . Firefly luciferase expression was measured using the Dual-Luciferase Reporter Assay System ( Promega ) according to the manufacturer's instructions . Luminescence was measured and data were normalized to the Renilla luciferase readout . For each construct , luciferase activity upon purmorphamine treatment was divided by the activity of cells before treatment to obtain the fold change value . Experiments were performed with both biological ( n = 3 ) and technical ( n = 6 ) replicates . Statistical analysis was performed using GraphPad 6 . 0 software . Data were analyzed by Shapiro-Wilk normality test and Levene´s test of variance . We used two-tailed unpaired Student´s t-test or Mann Whitney-U tests for comparing two groups , One-way ANOVA or Kruskall-Wallis and the corresponding post-hoc tests for more than two groups and two-way ANOVA to compare more than one variable in more than two groups . P values were represented by asterisks as follows: ( * ) p-value<0 . 05; ( ) p-value<0 . 01; ( * ) p-value<0 . 001; ( **** ) p-value<0 . 0001 . Differences were considered significant when p<0 . 05 . Values used for graphical representations and statistical analysis are available in Source Data 2 .	Primary cilia are the ‘antennae’ of animal cells: these small , flexible protrusions emerge from the surface of cells , where they help to sense and relay external signals . Cilia are assembled with the help of the cytoskeleton , a dynamic network of mesh-like filaments that spans the interior of the cell and controls many different biological processes . If cilia do not work properly , human diseases called ciliopathies can emerge . Townes-Brocks Syndrome ( TBS ) is an incurable disease that presents a range of symptoms such as malformations of the toes or fingers , hearing impairment , and kidney or heart problems . It is caused by a change in the gene that codes for a protein called SALL1 , and recent work has also showed that the cells of TBS patients have defective cilia . In addition , this prior research identified a second protein that interacted with the mutant version of SALL1; called LUZP1 , this protein is already known to help maintain the cytoskeleton . In this study , Bozal-Basterra et al . wanted to find out if LUZP1 caused the cilia defects in TBS . First , the protein was removed from mouse cells grown in the laboratory , which dramatically weakened the cytoskeleton . In keeping with this observation , both the number of cilia per cell and the length of the cilia were abnormal . Cells lacking LUZP1 also had defects in a signalling process that transmits signals received by cilia to different parts of the cell . All these defects were previously observed in cells isolated from TBS patients . In addition , LUZP1-deficient mouse cells showed the same problems with their cilia and cytoskeleton as the cells from individuals with TBS . Crucially , the cells from human TBS patients also had much lower levels of LUZP1 than normal , suggesting that the protein may contribute to the cilia defects present in this disease . The work by Bozal-Basterra et al . sheds light on how primary cilia depend on the cytoskeleton , while also providing new insight into TBS . In the future , this knowledge could help researchers to develop therapies for this rare and currently untreatable disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2020	LUZP1, a novel regulator of primary cilia and the actin cytoskeleton, is a contributing factor in Townes-Brocks Syndrome
The Insulin signaling pathway couples growth , development and lifespan to nutritional conditions . Here , we demonstrate a function for the Drosophila lipoprotein LTP in conveying information about dietary lipid composition to the brain to regulate Insulin signaling . When yeast lipids are present in the diet , free calcium levels rise in Blood Brain Barrier glial cells . This induces transport of LTP across the Blood Brain Barrier by two LDL receptor-related proteins: LRP1 and Megalin . LTP accumulates on specific neurons that connect to cells that produce Insulin-like peptides , and induces their release into the circulation . This increases systemic Insulin signaling and the rate of larval development on yeast-containing food compared with a plant-based food of similar nutritional content . Nutrient sensing by the central nervous system is emerging as an important regulator of systemic metabolism in both vertebrates and invertebrates ( Lam et al . , 2005; Levin et al . , 2011; Rajan and Perrimon , 2012; Bjordal et al . , 2014; Linneweber et al . , 2014 ) . Little is known about how nutrition-dependent signals pass the blood brain barrier ( BBB ) to convey this information . Like the vertebrate BBB , the BBB of Drosophila forms a tight barrier to passive transport , and is formed by highly conserved molecular components ( Bundgaard and Abbott , 2008; Stork et al . , 2008; Abbott et al . , 2010 ) . Its simple structure and genetic accessibility make it an ideal model to study how nutritional signals are communicated to the CNS . Insulin and Insulin-like growth factors are conserved systemic signals that regulate growth and metabolism in response to nutrition . Although Drosophila do not have a single pancreas-like organ , they do produce eight distinct Drosophila Insulin/IGF-like peptides ( Dilps ) that are expressed in different tissues ( Riedel et al . , 2011; Colombani et al . , 2012; Garelli et al . , 2012 ) . A set of three Dilps ( Dilp2 , 3 , 5 ) , released into circulation by Dilp-producing cells ( IPCs ) in the brain , have particularly important functions in regulating nutrition-dependent growth and sugar metabolism; ablation of IPCs in the CNS causes Diabetes-like phenotypes , slows growth and development , and produces small , long-lived adult flies ( Rulifson et al . , 2002; Broughton et al . , 2005; Partridge et al . , 2011 ) . Systemic Insulin/IGF signaling ( IIS ) increases in response to dietary sugars , proteins and lipids . Sugars act on IPCs directly to promote Dilp release ( Haselton and Fridell , 2010 ) , but other nutrients are sensed indirectly through signals from the fat body—an organ analogous to vertebrate liver/adipose tissue ( Colombani et al . , 2003; Geminard et al . , 2009; Rajan and Perrimon , 2012 ) . The Drosophila fat body produces two major types of lipoprotein particles: Lipophorin ( LPP ) , the major hemolymph lipid carrier , and Lipid Transfer Particle ( LTP ) . LTP transfers lipids from the intestine to LPP . These lipids include fatty acids from food , as well as from endogenous synthesis in the intestine ( Palm et al . , 2012 ) . LTP also unloads LPP lipids to other cells ( Van Heusden and Law , 1989; Canavoso et al . , 2004; Parra-Peralbo and Culi , 2011 ) . LPP crosses the BBB and accumulates throughout the brain . It is required for nutrition-dependent exit of neural stem cells from quiescence ( Brankatschk and Eaton , 2010 ) . Here , we investigate possible functions of LTP in the brain . Immunostaining reveals LTP on specific neurons and glia in larval brains . ( Figure 1A , C–E , Figure 1—figure supplement 1–3 , and Videos 1–4 ) . First instar brains have on average three LTP-positive neurons per brain lobe , increasing to 13 in early third instar larvae ( Figure 1B , Figure 1—figure supplement 1 ) . We used cell type-specific RNAi to distinguish whether LTP in the brain came from circulation , or whether it was produced in the CNS . Knock-down of ltp in the fat body reduces but does not eliminate LTP from circulation ( Figure 1F ) . Staining larval brains from these animals for LTP reveals reduced staining on both neurons and glia ( Videos 5 and 6 ) . To investigate this issue in more detail , we quantified LTP-positive neurons after knock-down of ltp in neurons , glia , or fat body . To ensure that we compared larvae of similar developmental stages we quantified glial cell numbers , which increase during larval development ( Figure 1B′ , G′ ) . Only fat body-specific ltp knock-down reduces neuronal LTP staining in the brain ( Figure 1G , Figure 1—figure supplement 3 ) . Thus , LTP particles secreted by the fat body cross the Blood Brain Barrier and become enriched on specific neurons . 10 . 7554/eLife . 02862 . 003Figure 1 . Circulating LTP crosses the BBB and accumulates on neurons . ( A ) Confocal section of the CNS from a wt larva reared on YF ( wtYF ) , at the level of the big commissure , stained for LTP ( green ) and Elav ( magenta ) . ( A′–A′″ ) . Magnified boxed region in ( A ) ( B–B′ ) Total numbers of LTP-positive neurons/brain lobe ( B ) or Repo-positive glia/brain ( B′ ) of wtYF larvae of different ages , quantified from 50–60 confocal sections per brain . Numbers indicate larval instar; superscripts indicate age in hours after egg collection . ( C–E ) Cartoons depict positions of all LTP-positive neurons ( black dots ) identified in confocal stack of three wtYF brains of different ages , stained for LTP and Elav . ( F ) Western blot showing equal volumes of hemolymph from controlYF larvae , and larvae with FB-specific knock-down of ltp ( ltpRNAiFB ) . Blots are probed for Apo-LTPI , Apo-LTPII , Apo-LPPII and Cv-D . ( G and G′ ) Total numbers of LTP-positive neurons/brain lobe ( G ) and Repo-positive glia/brain ( G′ ) quantified from controlYF larvae , and larvae where ltp has been knocked-down in neurons ( ltpRNAiN ) , glia ( ltpRNAiG ) , or FB ( ltpRNAiFB ) . Error bars indicate standard deviation . * = p < 0 . 001 , = p < 0 . 01 , n . s . = not significant ( Student's t test ) . Scale bars indicate 50 μm ( A and C–E ) or 5 μm ( A′–A′″ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 00310 . 7554/eLife . 02862 . 004Figure 1—figure supplement 1 . Numbers of LTP positive neurons rise over larval development . Panel shows numbers ( see also Figure 1B , B′ ) of LTP positive neurons/brain hemisphere ( red ) and Repo positive glia/brain ( blue ) from staged First , second , early and mid third instar ( 3rd-72h = 72 hr and 3rd-96h = 96 hr after egg collection ) wt larvae . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 00410 . 7554/eLife . 02862 . 005Figure 1—figure supplement 2 . Small glial subsets enrich LTP positive . ( A–B′″ ) Confocal section at the big commissure from wt ( A–A′″ ) or ltp ( B–B′″ ) staged larval brains probed for Repo ( magenta ) and LTP ( green ) . Boxed regions ( A′ and B′ ) enlarged in ( A″–A″″ ) and ( B″–B″″ ) respectively . Scale bars indicate 50 μm ( A–B′ ) and 5 μm ( A′″ and B′″ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 00510 . 7554/eLife . 02862 . 006Figure 1—figure supplement 3 . Circulating LTP crosses the BBB and enriches on neurons . Panel shows numbers ( see also Figure 1G , G’ ) of LTP positive neurons/brain hemisphere ( red ) and Repo positive glia/brain ( blue ) from ltpRNAi/+ ( controls ) and larvae upon LTP knock-down in neurons ( ltpRNAiN ) , glia ( ltpRNAiG ) and fat bodies ( ltpRNAiFB ) . Larvae reared on yeast food , superscript numbers indicate standard deviation values . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 00610 . 7554/eLife . 02862 . 007Video 1 . Confocal stack from wild type first instar larval brain probed for LTP ( green ) , Dilp2 ( red ) and Repo ( grey ) . Sections are spaced 1 . 5 µm apart , scale bars indicate 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 00710 . 7554/eLife . 02862 . 008Video 2 . Confocal stack from wild type second instar larval brain probed for LTP ( green ) , Dilp2 ( red ) and Repo ( grey ) . Sections are spaced 1 . 5 µm apart , scale bars indicate 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 00810 . 7554/eLife . 02862 . 009Video 3 . Confocal stack from wild type third instar larval brain probed for LTP ( green ) , Dilp2 ( red ) and Repo ( grey ) . Sections are spaced 1 . 5 µm apart , scale bars indicate 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 00910 . 7554/eLife . 02862 . 010Video 4 . Confocal stack from wild type third instar larval brain probed for Elav ( red ) , Repo ( grey ) , LTP ( green ) and DAPI ( blue ) . Sections are spaced 1 . 5 µm apart , scale bars indicates 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 01010 . 7554/eLife . 02862 . 011Video 5 . Confocal stack from UAS:ltpRNAi/+ third instar larval brain probed for Repo ( grey ) , Dilp2 ( red ) and LTP ( green ) . Sections are spaced 1 . 5 µm apart , scale bars indicate 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 01110 . 7554/eLife . 02862 . 012Video 6 . Confocal stack from lpp-Gal4>UAS:ltpRNAi third instar larval brain probed for Repo ( grey ) , Dilp2 ( red ) and LTP ( green ) . Sections are spaced 1 . 5 µm apart , scale bars indicate 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 012 To understand the CNS functions of LTP , we sought to identify the neurons where it accumulated . We first investigated whether LTP co-localized with Dilp2-positive neurons in early third instar larval brains . Dilp2 is present not only in IPCs , which lie dorso-lateral to the big commissure , but also on three neurons per hemisphere that do not produce Dilps . These neurons recruit Dilp2 from IPCs using IMPL2 , a Dilp2-binding protein ( Bader et al . , 2013 ) . Although each hemisphere contains 7-8 neurons that express IMPL2 at this stage , only 3 of these detectably recruit Dilp2 , and 2 of these 3 also accumulate LTP . These neurons can be unambiguously identified by the presence of both LTP and Dilp2 , by their position , by lack of expression of dilp2-GAL4 , and by expression of impL2-GAL4 ( Figure 2 , Figure 2—figure supplement 1 ) . Henceforth , we refer to these neurons as Dilp2-recruiting neurons ( DRNs ) . Starting in the third larval instar , LTP accumulates on two specific DRNs located dorsal to the big commissure ( Figure 2A , C , D , Figure 2—figure supplement 2 ) . Although LTP is found on other neurons at earlier stages , its accumulation on DRNs is developmentally regulated . We also sometimes ( 6/20 sampled CNS ) detect LTP on a subset of IPC neurons ( Figure 2B ) . Thus , upon entering the brain , LTP accumulates on IPCs and specific neurons that recruit Dilp2 . 10 . 7554/eLife . 02862 . 013Figure 2 . LTP accumulates on neurons positive for Dilp2 and IMPL2 . Confocal sections at the level of ( A–A′″ ) or dorsal to ( B–B′″ ) the big commissure from third instar wtYF larval brains stained for Dilp2 ( magenta ) and LTP ( green ) . Boxed regions show LTP/Dilp2 double-positive neurons . Scale bar = 50 μm ( A and B ) or 5 μm ( A′″ and B′″ ) . ( C ) Cartoon drawn from a single sectioned brain showing positions of IPCs ( magenta ) , IMPL2-producing cells that recruit Dilp2 ( DRNs ) ( grey with magenta rim ) , and IMPL2-producing cells negative for Dilp2 ( grey ) . LTP ( green dots ) is found on a subset of DRNs . ( D ) shows average number of neurons double-positive for Dilp2/IMPL2 ( DRNs , grey ) and the number of these that are LTP-positive ( black ) in larval brains of different stages . Error bars = standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 01310 . 7554/eLife . 02862 . 014Figure 2—figure supplement 1 . A small subset of Dilp2 positive neurons expresses IMPL2 . ( A-D’’’ ) 40 μm confocal stack projections from wt ( A ) , dilp2 ( B ) , impL2>cherryTM ( C–C’’’ ) and dilp2>cherryTM ( D–D’’’ ) larval brains probed for Dilp2 ( green ) and Cherry™ ( magenta ) . Boxed sections ( C , D ) magnified in C’-C’’’ or D’-D’’’ respectively . Cherry™ shown in C’ , D’ and Dilp2 in C’’ , D’’ . Scale bars = 50 μm ( A-D ) or 5 μm ( C’’’ , D’’’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 01410 . 7554/eLife . 02862 . 015Figure 2—figure supplement 2 . In the last larval developmental stage DRNs enrich LTP . Panel shows numbers ( see also Figure 2D ) of DRNs ( grey ) and LTP positive DRNs ( dark grey ) from 1st , 2nd or early and mid 3rd instar ( 3rd−72h = 72hr and 3rd−96h = 96 hr after egg collection ) larvae reared on yeast food . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 015 Thus far , we have described LTP localization in larvae reared on a rich medium: ‘yeast food’ ( YF ) ( Carvalho et al . , 2012 ) . Systemic IIS is high in these animals , as reflected by predominantly cytoplasmic localization of FOXO in fat body cells ( Figure 3E , E′ ) ( van der Horst and Burgering , 2007 ) . To ask whether neuronal LTP accumulation was regulated by nutrition , we asked how it responded to starvation . Late second instar larvae transferred from YF to food containing only glucose ( GF ) arrest growth and development . Systemic IIS is strongly reduced , as indicated by nuclear accumulation of FOXO in fat body cells ( Figure 3A , A′ ) . Although LTP still circulates ( Figure 3G ) , it accumulates on many fewer neurons in glucose-fed larvae , and is never detected on DRNs ( Figure 3F–F″ ) . Thus , the presence of glucose is insufficient to allow high systemic IIS or accumulation of LTP on most neurons . 10 . 7554/eLife . 02862 . 016Figure 3 . Neuronal LTP accumulation is diet-dependent . ( A–E′ ) Confocal stack projections of fat bodies from early third instar larvae raised on GF ( A and A′ ) , LDF ( B and B′ ) , LDSF ( C and C′ ) , PF ( D and D′ ) and YF ( E and E′ ) probed for FOXO ( A–E and A–E′ magenta ) and DAPI ( A′–E′ , green ) . Scale bars = 20 μm . ( F–F′″ ) average number of LTP-positive neurons/brain lobe ( F ) , glial cells/brain ( F′ ) and LTP-positive DRNs ( F″ ) in brains of larvae transferred from YF to indicated diets in the late second instar . Error bars indicate standard deviation . T-test significance: p < 0 . 01 , p < 0 . 001 . ( G ) Equal hemolymph volumes from wt larvae raised on indicated food sources Western blotted and probed for the indicated Apolipoproteins . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 01610 . 7554/eLife . 02862 . 017Figure 3—figure supplement 1 . Only yeast food promotes LTP enrichments on DRNs . Panel shows numbers ( see also Figure 3F–F’’ ) of LTP positive neurons/brain hemisphere ( red ) and Repo positive glia/brain ( blue ) , and DRNs/brain ( light grey ) or LTP positive DRNs/brain ( dark grey ) from wt larvae reared on YF till late 2nd instar and transferred for 16hr on GF , LD , LDS , PF or YF . p-values ( Students t’test ) are indicated , superscript numbers = standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 01710 . 7554/eLife . 02862 . 018Figure 3—figure supplement 2 . LTP enrichments on DRNs are reversible . Panel shows numbers ( see also Figure 1—figure supplement 3A , A′ , C ) of LTP positive neurons/brain hemisphere ( red ) and Repo positive glia/brain ( blue ) , and DRNs/brain ( light grey ) or LTP positive DRNs/brain ( dark grey ) from larvae reared on YF , PF , larva reared on PF till late 2nd instar and transferred on YF and larva reared till late 2nd instar on YF and transferred on PF . p-values ( Students t’test ) are indicated , superscript numbers = standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 018 We next examined the role of lipids and proteins in promoting neuronal LTP accumulation and systemic IIS . We compared larvae fed on YF , with those transferred at late second instar to a lipid-depleted food ( LDF ) , which contains a yeast autolysate that has been chloroform-extracted to remove lipid . LDF contains proteins and sugars , and has a similar caloric content to YF . However , larvae arrest on LDF because of the absence of sterols ( Carvalho et al . , 2010 ) , and systemic IIS is low ( Figure 3B , B′ ) . Brains of LDF-fed larvae contain more LTP-positive neurons than those of glucose-fed larvae , but fewer than YF-reared larvae . In particular , LTP is never detected on DRNs ( Figure 3F–F″ ) . To address the sterol-dependence of neuronal LTP accumulation , we supplemented LDF with 10 µM cholesterol ( LDSF ) . Larvae transferred from YF to LDSF grow slowly and give rise to viable adults of reduced size ( Carvalho et al . , 2010 ) . Consistent with this , nuclear FOXO staining in the fat body indicates that systemic IIS is low ( Figure 3C , C′; Carvalho et al . , 2010 ) . Sterol addition further increases the number of LTP-positive neurons , but still does not allow LTP accumulation on DRNs ( Figure 3F–F″ , and Figure 3—figure supplement 1 ) . Thus , a component present in yeast food , but not the chloroform-extracted yeast autolysate , is required for accumulation of LTP on DRNs , and for high-level systemic IIS . We wondered whether lipids in general were required for LTP accumulation on DRNs . Interestingly however , experiments with plant food ( PF ) suggest that bulk lipid is not sufficient . PF contains no yeast and is based entirely on plant materials . It has the same caloric content as YF , and slightly more lipid with a different fatty acid composition ( Carvalho et al . , 2012 , see ‘Materials and methods’ ) . Surprisingly , LTP is found only occasionally on DRNs in larvae transferred from YF to PF . Depletion of LTP on DRNs occurs within 16 hr of transfer and is reversible in the same time frame ( Figure 3F–F″ and Figure 4A , A′ , C and Figure 3—figure supplement 2 ) . Strikingly , despite the abundant calories derived from carbohydrates , proteins and lipids , feeding with PF specifically slows the larval growth phase without lengthening embryonic or pupal development . PF also dramatically extends adult lifespan compared to YF ( Figure 4E–G ) . This suggests that systemic IIS is reduced when larvae are fed with PF , compared to YF . Indeed , FOXO is predominantly nuclear in PF-reared larvae ( Figure 3D , D′ ) . 10 . 7554/eLife . 02862 . 019Figure 4 . Yeast food promotes fast larval development but reduces average life span . ( A , A′ , C ) Charts depict LTP positive neurons/brain hemisphere ( A ) and Repo positive glia/brain ( A′ ) , and DRNs or LTP positive DRNs/brain ( C ) from larvae reared on YF , PF , raised on PF till late second instar and transferred for 16hr on YF , and raised on YF till late second instar and transferred for 16 hr on PF . p-values ( Student's ttest ) are indicated , = p < 0 . 05 , ** = p < 0 . 01 , n . s . = not significant , error bars = standard deviation . ( B ) Equal volumes hemolymph from early third instar wt larvae reared on YF or PF probed for Apo-LTP , Apo-LPP and Cvd . ( D , E , G ) Plotted are percentages ( Y-axis ) of hatched embryos ( D; nYF = 198 , nPF = 198; parental flies kept for three generations on respective food types before embryo collection ) , pupariated larvae ( E; nYF = 163 and nPF = 133; time of pupal development was unchanged ) and of living mated females ( G; nYF = 40 and nPF = 35 ) over time ( X-axis ) . Please note , LD50 indicated with grey line ( G ) . ( F–F″ ) Exemplary photographs of staged wt larvae bred on YF ( left ) or PF ( right ) . Hours after egg collection ( AEC ) are indicated in top right corner . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 019 To investigate mechanisms underlying the differences in systemic IIS in larvae fed with YF and PF , we examined the pathway at different levels . We first focused on the neuronal activity of the IPCs . Neuronal activity correlates with higher levels of intracellular free calcium , which can be detected using the GCaMP reporter ( Reiff et al . , 2005 ) . This GFP reporter increases its fluorescence in response to free calcium . We expressed the GCaMP construct either in IPC neurons ( under the control of dilp2-GAL4 ) or in all neurons ( under the control of rab3-GAL4 ) and compared GFP fluorescence in brains from YF and PF-reared larvae . While the activity of most neurons is not affected by feeding with these different foods ( Figure 5A , B ) , the activity of the IPCs was dramatically higher on YF than on PF ( Figure 5C , D; Videos 7 and 8 ) . We quantified the number of IPCs exhibiting detectable fluorescence of the GCaMP reporter in 5 brains each of larvae fed with PF or YF . On YF , we detected activity of the GCaMP reporter in 4 . 8 ± 0 . 6 neurons per brain lobe , while on PF only 1 ± 1 . 3 neurons per brain lobe exhibited detectable reporter fluorescence . To ask whether higher neuronal activity of IPCs resulted in elevated release of Dilp2 into the hemolymph , we probed Western blots of hemolymph from YF and PF-reared larvae with antibodies to Dilp2 . YF-reared animals have higher levels of circulating Dilp2 than PF-reared animals ( Figure 5E ) . Thus , IPCs release more Dilp2 when animals are fed with YF . We next examined different molecular readouts of IIS in different tissues of PF and YF-reared larvae . YF increases PI3K activity in salivary glands ( Figure 5F , G ) , as revealed by membrane localization of the PHGFP reporter construct . Furthermore , phosphoAKT levels are higher in fat bodies of YF-reared larvae ( Figure 5H ) . Thus , feeding with YF activates IPCs , causing them to secrete more Dilp2 into circulation , thereby increasing systemic IIS . 10 . 7554/eLife . 02862 . 020Figure 5 . Yeast food components activate IPCs to release Dilp2 and induce systemic Insulin signaling . ( A–D ) show GFP fluorescence from collapsed confocal stacks of whole brains expressing GCaMP under the control of rab3-GAL4 ( A and B ) or dilp2-GAL4 ( C and D ) from larvae reared on YF ( A and C ) or PF ( B and D ) . ( E ) Hemolymph blots probed for Dilp2 and CvD from wild type ( lanes 1 and 3 ) , dilp2 mutant ( lane 2 ) and repo-GAL4>lrp1 , 2RNAi ( lane 4 ) larvae reared on yeast food ( YF ) or plant food ( PF ) as indicated . ( F and G ) show single confocal sections of salivary glands expressing the PIP3 reporter PHGFP under the direct control of the tubulin promoter . Larvae were reared on YF ( F ) or PF ( G ) . ( H ) Western blots from fat body lysates probed for phospho-AKT , LTP and CvD as indicated . Lysates are from wild type ( lanes 1 , 2 ) , repo-GAL4>lrp1 , 2RNAi ( lane 3 ) , lpp-GAL4>ltpRNAi ( lane4 ) and YFP-rab27;YFP-rab3;impl2-GAL4>gfpRNAi ( lane 5 ) . Larvae were reared on yeast food ( YF ) or plant food ( PF ) as indicated . ( I ) shows percentages of larvae that have pupariated at different days after egg laying ( dAEL ) . Larvae were reared on yeast food ( red , n = 49 ) , on plant food ( black , n = 34 ) , or on lipid-depleted food supplemented with either yeast lipids ( light red , n = 23 ) or plant lipids ( grey , n = 24 ) . ( J and K ) show GFP fluorescence from collapsed confocal stacks of whole brains expressing GCaMP under the control of dilp2-GAL4 from larvae fed on lipid-depleted food supplemented with yeast lipids ( J ) or lipid-depleted food supplemented with plant lipids ( K ) . See also Videos 7 and 8 . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 02010 . 7554/eLife . 02862 . 021Video 7 . Confocal stack from dilp2-Gal4>UAS:GCaMP larval brain reared on PF , right panel shows the same images at higher gain . Images show GFP fluorescence of the GCaMP reporter . Scale bars indicate 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 02110 . 7554/eLife . 02862 . 022Video 8 . Confocal stack from dilp2-Gal4>UAS:GCaMP larvae reared on YF . Images show GFP fluorescence of the GCaMP reporter . Scale bars indicate 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 022 To examine whether the differences between plant and yeast lipids were responsible for changes in IPC activity , we prepared chloroform extracts of PF and YF and used them to supplement LDF . Yeast lipids supported faster larval growth compared to plant lipids ( Figure 5I ) . Furthermore , the GCaMP reporter reveals that IPC neurons are only active when larvae are fed with yeast lipids ( Figure 5J , K; Videos 7 and 8 ) . Thus , IPC neurons respond differently to the types of lipids present in YF and PF . How does YF promote IPC activity and Dilp release ? We wondered whether the YF-dependent localization of LTP to the DRNs might be responsible . To investigate this , we first asked whether IMPL2-expressing neurons actually contact the IPCs . We labeled the projections of IMPL2 expressing neurons by driving a transmembrane Cherry ( Cherry ) under the control of impL2-Gal4 , and stained larval brains for both Dilp2 and IMPL2 . A subset of Cherry-labeled projections , also containing IMPL2 , extends from the DRNs towards the IPCs ( Figure 6B′ , B″ , B′″ ) . Dilp2 staining reveals a subset of projections from the IPCs extending towards the DRNs ( Figure 6A—class 2 projections and Figure 6B , B′″ ) . They meet in a region where IMPL2 begins to colocalize with Dilp2 ( Figure 6B′″ ) . Thus , it is likely that DRNs and IPCs communicate with each other directly . 10 . 7554/eLife . 02862 . 023Figure 6 . IPC and IMPL2-positive neurons are in direct contact . ( A ) Projected confocal stacks of an early third instar brain expressing transmembrane Cherry under the control of dilp2-GAL4 , stained for Cherry ( green ) and Dilp2 ( magenta ) . Most Dilp2 is detected in projections that terminate in the periphery ( as determined from examining individual sections ) . However , Cherry staining reveals other projections that enter the central neuropil ( Linneweber et al . , 2014 ) or remain in other regions of the CNS ( Levin et al . , 2011; Bjordal et al . , 2014 ) . ( B ) Shows a single confocal section from impl2-GAL4>cherryTM larval brains probed for Dilp2 ( B , green in B′″ ) , Cherry ( B′ , red in B′″ ) and IMPL2 ( B″ , blue in B′″ ) . White arrows point to colocalization between Dilp2 and IMPL2 . Scale bars indicate 50 µm ( A ) or 10 µm ( B–B′″ ) . ( C ) Shows percentages of yeast food-reared larvae of different genotypes that have pupariated at different days after egg laying ( dAEL ) . Black indicates pooled results from two control genotypes: UAS:lrp1 , 2RNAi/+ and UAS:gfpRNAi/+ ( n = 81 ) . Red indicates repo-GAL4>lrp1 , 2RNAi ( n = 109 ) . Light grey indicates YFP-rab27;YFP-rab3;impl2-GAL4>gfpRNAi ( n = 76 ) . Dark grey indicates YFP-rab27;YFP-rab3;dilp2-GAL4>gfpRNAi ( n = 120 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 02310 . 7554/eLife . 02862 . 024Figure 6—figure supplement 1 . Y-rab3 and Yrab27 are reduced by gfpRNAi . ( A , B ) show confocal sections at the level of the big brain commissure from control YFPrab27;YFPrab3;UAS:gfpRNAi ( A ) and YFPrab27;YFPrab3;rab3Gal4> UAS:gfpRNAi ( B ) larval brains . Scale bars indicate 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 024 How does the activity of IMPL2-positive neurons influence that of IPCs ? To address this question , we inhibited synaptic release by knocking down Rabs 3 and 27—two Rabs that both contribute to synaptic vesicle release ( Mahoney et al . , 2006; Pavlos et al . , 2010 ) . To do this , we used homologous recombination to generate N-terminal YFP fusions of Rab3 and Rab27 at their endogenous chromosomal loci . This renders them sensitive to knock-down by anti-GFP RNAi ( Figure 6—figure supplement 1 ) . Rabs 3 and 27 are expressed almost exclusively in neurons ( manuscript in preparation ) . Thus , even for GAL4 drivers that are active in many tissues , this approach will affect only the neurons that express the GAL4 driver . When anti-GFP RNAi is driven in the IPCs in a background homozygous for both YFPRab3 and YFPRab27 , larvae exhibit the slow growth phenotype characteristic of reduced IIS ( Figure 6C ) . Reducing levels of Rab3/Rab27 in the IMPL2-expressing neurons slows larval growth ( Figure 6C ) , and reduces phosphorylation of AKT in the fat body ( Figure 5H ) . This suggests that signals sent by IMPL2-expressing neurons promote systemic IIS . Since LTP localizes to DRNs only on yeast food , we wondered whether it was required to promote systemic Insulin signaling . We therefore asked whether IIS in YF-reared larvae depended on LTP . RNAi-mediated ltp knock-down in the fat body causes larval arrest in the second or third instar ( Palm et al . , 2012 ) . Fat body cells contain lower levels of phosphoAKT ( Figure 5H ) and predominantly nuclear FOXO ( Figure 7K , K′ ) . Thus LTP is required for high-level IIS in the fat body on YF . 10 . 7554/eLife . 02862 . 025Figure 7 . Glial LRP1 and LRP2 receptors move LTP across the BBB . ( A–F ) Confocal brain sections at the level of the big commissure stained for LRP1 ( A–C ) or LRP2 ( D–F ) from controlYF larvae ( A and D ) or larvae with neuronal ( B and E ) or glial ( C and F ) knock-down of each receptor . Glial knock down reduces LRP1 and LRP2 . ( G–G′″ ) Average number of LTP-positive neurons/brain lobe ( G ) Repo-positive glia/brain ( G′ ) or fraction of LTP-positive DRNs ( G′″ ) in larval brains with glial knock-down of LRP1 and LRP2 singly or in combination , as indicated . Error bars indicate standard deviation . T-test significance: p = 0 . 052 , p > 0 . 01 , *p < 0 . 001 . ( H–H″ ) control larvae ( right ) and larvae with glial knock-down of both LRP1/LRP2 ( left ) photographed at indicated times after egg collection . ( I ) Percent of control ( n = 216 , black ) or double LRP1/LRP2 knock down ( n = 183 , red ) larvae pupariating over time . ( J–L′ ) Confocal stack projections of fat bodies from control ( J and J′ ) , fat body ( K and K′ ) or glial ( L and L′ ) LRP1 , 2 knock down larvae stained for FOXO ( magenta ) and DAPI ( green ) . Scale bars = 20 μm . ( M ) Percent survival of control ( n = 34 , black ) or glial LRP1/2 double knock-down ( n = 33 , red ) flies fed with YF; grey line indicates 50% survival . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 02510 . 7554/eLife . 02862 . 026Figure 7—figure supplement 1 . Neuronal LTP enrichments are reduced in lrp1 , 2 . ( A , A’ ) Charts show LTP positive neurons/brain hemisphere ( A ) and Repo positive glia/brain ( A’ ) from wt , lpr1 , 2 and lrp1 , 2 larva reared on YF . = p < 0 . 05 , *** = p < 0 . 001 ( Students t’test ) , error bars = standard deviation . ( B ) Equal volumes hemolymph probed for Apo-LTP , Apo-LPP and CvD . Please note increased LTP levels present in lpr1 , 2 hemolymph ( arrowhead ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 02610 . 7554/eLife . 02862 . 027Figure 7—figure supplement 2 . Neuronal LTP enrichments are only mildly affected in lpr1 , 2 . Panel shows numbers of LTP positive neurons/brain hemisphere ( red ) and Repo positive glia/brain ( blue ) from staged wt , lrp1 , 2 and lpr1 , 2 larvae reared on YF . Superscript numbers = standard deviation , p-values ( Students t’test ) are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 02710 . 7554/eLife . 02862 . 028Figure 7—figure supplement 3 . Both receptors , LRP1 and LRP2 , promote LTP transport in glia . Panel shows numbers ( see also Figure 4G–G’’ ) of LTP positive neurons/brain hemisphere ( red ) and Repo positive glia/brain ( blue ) , and DRNs/brain ( light grey ) or LTP positive DRNs/brain ( dark grey ) from lrp1 , 2RNAi/+ ( controls ) and larvae upon glial knock-down of LRP1 ( lrp1RNAiG ) , LRP2 ( lrp2RNAiG ) , and LRP1 , 2 ( lrp1 , 2RNAiG ) protein . Larvae reared on YF , p-values ( Students t’test ) are indicated , superscript numbers = standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 028 Reduced IIS in ltp mutants might reflect a requirement for LTP on DRNs . Alternatively , loss of LTP in other tissues might affect Insulin signaling indirectly . Blocking transport of LTP across the BBB in YF-reared larvae could distinguish these possibilities . We therefore examined requirements for different LDL receptor family proteins in BBB transport of LTP . LTP localizes normally to DRNs and other neurons in larvae double mutant for LDL receptor homologues LpR1 and LpR2 ( Figure 7 , and Figure 7—figure supplement 1–3; Khaliullina et al . , 2009 ) . Thus , LpR1 and LpR2 are not dominant transporters of LTP across the BBB . However , we discovered redundant functions for LRP1 and LRP2 in BBB transport of LTP . Both proteins are predominantly expressed in glial cells ( Figure 7A–F ) . Double mutants die as third instar larvae ( not shown ) . Although LTP circulates at normal levels ( Figure 7—figure supplement 1B ) , the number of LTP-positive neurons is reduced ( Figure 7—figure supplement 1A , A′ ) . A similar phenotype is produced by glial-specific double knockdown of LRP1/2 , which halves the number of LTP-positive neurons , and completely blocks localization to DRNs . Knock-down of either LRP alone produces intermediate phenotypes , suggesting they function redundantly ( Figure 7G–G′″ , Figure 7—figure supplement 3 ) . Thus , both LRPs transport LTP across the BBB to DRNs . Transport of LTP to other neurons has a less strict requirement for LRP1/2 . If the absence of LTP on DRNs reduces systemic IIS , then losing LRP1/2 in BBB glia should reproduce this effect . Indeed , glial-specific LRP1/2 knock-down slows larval growth and delays pupariation , and emerging adults live longer—all phenotypes suggesting reduced IIS ( Figure 7H–I , M ) . Consistent with this , levels of circulating Dilp2 are lower in these animals ( Figure 5E ) , AKT is less phosphorylated ( Figure 5H ) , and FOXO is predominantly nuclear in larval fat body cells ( Figure 7L , L′ ) . These data support the idea that blocking transport of LTP to DRNs reduces the release of Dilp2 by IPCs , thereby lowering systemic IIS . What mechanisms promote the BBB transport of LTP to DRNs on yeast food ? We first wondered whether levels of LRP1 , 2 might change . However , we observed no obvious differences in LRP1 , 2 staining in the brain on YF vs PF ( data not shown ) . However we were surprised to observe dramatically higher levels of free intracellular calcium in BBB glia when larvae are fed YF compared to PF . Driving the free intracellular calcium sensor GCaMP using repo-GAL4 ( which is active in all glia ) reveals a marked reduction in GFP fluorescence in BBB glia relative to other glia . We confirmed this result by specifically driving the reporter in BBB glia using moody-GAL4 ( Figure 8A–D ) . 10 . 7554/eLife . 02862 . 029Figure 8 . Increased free Ca++ levels promote systemic Insulin signaling . ( A–D ) Show GFP fluorescence of the GCaMP reporter construct in confocal sections of larval brains at the level of the big brain commissure . GCaMP was expressed either in all glia under the control of repo-GAL4 ( A and B ) or in BBB glia under the control of moody-GAL4 ( C and D ) . Larvae were reared on yeast food ( A and C ) or plant food ( B and D ) . White arrowheads point to blood brain barrier and scale bars indicate 50 µm . ( E ) shows the average number of LTP-positive neurons per brain lobe ( E , red , n = 10 ) or the average number neurons co-staining for LTP and Dilp2 ( DRNs ) ( E′ , grey , n = 5 ) from either UAS:trpa1 ( control ) larvae or larvae expressing TRPA1 under the control of moody-GAL4 ( TRPA1BBB ) . All larvae were reared on plant food at 29C . Error bars show standard deviations . * indicates p < 0 . 02 , *** indicates p < 0 . 0002 . ( F ) Western blot of fat body lysates from control ( UAS:trpa1/+ ) larvae ( lane 1 ) or larvae expressing TRPA1 in BBB glia under the control of moody-GAL4 ( lane 2 ) , probed for phospho-AKT1 and CvD . Larvae were all reared on plant food ( PF ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02862 . 029 To investigate whether Ca++ signaling in BBB glia was sufficient to promote transport of LTP to DRNs , we asked whether ectopically inducing Ca++ influx would allow LTP to accumulate on DRNs even when larvae were fed with PF . To do this , we drove the expression of TRPA1 with moody-GAL4 in PF-reared larvae and stained their brains for LTP and Dilp2 . Strikingly , while LTP is undetectable on DRNs of wild type larvae reared on PF , it accumulates to high levels on these neurons when Ca++ influx is induced by TRPA1 expression in BBB cells ( Figure 8E , E′ ) . Accumulation is specific to DRNs and neuronal LTP localization in these larvae precisely mimics its localization on YF . Furthermore , Western blotting shows that phosphorylation of AKT in the fat body increases compared to wild type larvae fed with PF ( Figure 8G ) . Taken together , these data show that YF increases free cytoplasmic Ca++ in BBB glia . This is sufficient for the specific transport of LTP to DRNs and elevated systemic IIS . Interestingly , the accumulation of LTP on other neurons does not appear to require elevated Ca++ in BBB cells . In summary , this work demonstrates a key requirement for lipoproteins in conveying nutritional information across the BBB to specific neurons in the brain . As particles that carry both endogenously synthesized and diet-derived lipids , lipoproteins are well-positioned to perform this function . Our data suggest that transport of LTP across the BBB to DRNs influences communication between DRNs and the Dilp-producing IPCs , increasing the release of Dilp2 into circulation . Since the IPCs also deliver Dilp2 to the DRNs , this indicates that these two neuronal populations may communicate bidirectionally . How might LTP affect the function of DRNs ? One possibility is that it acts to deliver a signaling lipid to the DRNs . It could do so either directly , or indirectly by promoting lipid transfer from LPP , which is present throughout the brain ( Brankatschk and Eaton , 2010 ) . LTP enrichment on specific neurons may increase lipid transfer to these cells . This work highlights a key function for BBB cells in transmitting nutritional information to neurons within the brain . Feeding with yeast food increases free calcium in BBB glia , which then increases transport of LTP to DRNs . How might BBB cells detect the difference between yeast and plant food ? Our data suggest differences in the lipid composition of yeast and plant-derived foods are responsible . Our previous work has shown that the lipids in these foods differ in their fatty acid composition . Yeast food has shorter and more saturated fatty acids than plant food ( Carvalho et al . , 2012 ) . How could these nutritional lipids affect the activity of BBB glia ? Interestingly , differences in food fatty acid composition are directly reflected in the fatty acids present in membrane lipids of all larval tissues including the brain ( Carvalho et al . , 2012 ) . Thus , it is possible that the bulk membrane properties of BBB glia are different on these two diets . Membrane lipid composition is known to affect a variety of signaling events ( Simons and Toomre , 2000; Lingwood and Simons , 2010 ) . Alternatively , yeast food may influence the specific fatty acids present in signaling lipids that activate BBB glia . We demonstrate an unexpected functional specialization of the BBB glial network , which permits specific and regulated LTP transport to particular neurons . How this specificity arises is an important question for the future . We note that a subset of glial cells within the brain also accumulates LTP derived from the fat body . Could these represent specific transport routes from the BBB ? An alternative possibility is that transport depends on neuronal activity . Mammalian LRP1 promotes localized transfer of IGF in response to neuronal activity ( Nishijima et al . , 2010 ) . Could LTP delivery by LRP1 and LRP2 in the Drosophila brain depend on similar mechanisms ? The remarkable specificity of LTP trafficking in the Drosophila CNS provides a novel framework for understanding information flow between the circulation and the brain . To what extent might this be relevant to vertebrate systems ? While it is clear that the vertebrate brain ( unlike that of Drosophila ) does not depend on lipoproteins to supply it with bulk sterols ( Orth and Bellosta , 2012 ) , this does not rule out possible functions for these particles in nutrient sensing . The vertebrate cerebrospinal fluid is rich in many types of HDL particles , including those containing ApoA-1 , which is not expressed in the brain—this suggests that at least some lipoprotein particles in the brain may derive from the circulation ( Wang and Eckel , 2014 ) . Consistent with this idea , ApoA-I can target albumin-containing nanoparticles across the BBB in rodents ( Zensi et al . , 2010 ) . Recent work suggests that lipoproteins may be the source of specific Long Chain Fatty Acids that signal to the hypothalamus to regulate glucose homeostasis , since neuronal lipoprotein lipase is required for this process ( Wang et al . , 2011; Wang and Eckel , 2014 ) . Thus , it would be interesting to investigate whether circulating mammalian lipoproteins might reach a subset of neurons in the hypothalamus . It has been known for some time that increasing the amount of yeast in the diet of lab grown Drosophila melanogaster increases the rate of development and adult fertility , but reduces lifespan ( Sang , 1949; Leroi et al . , 1994; Mair et al . , 2005 ) . Here , we show that flies have evolved specific mechanisms to increase systemic IIS in response to yeast , independently of the number of calories in the diet or its proportions of sugars proteins and fats . What pressures could have driven the evolution of such mechanisms ? In the wild , Drosophila melanogaster feed on rotting plant material and their diets comprise both fungal and plant components . Drosophila disperse yeasts and transfer them to breeding sites during oviposition improving the nutritional resources available to developing larvae ( Markow and O'grady , 2008 ) . Yeast that are able to induce more rapid development of the agents that disperse them may propagate more efficiently . On the other hand , it has been noted that Drosophila species that feed on ephemeral nutrient sources like yeasts or flowers have more rapid rates of development ( Markow and O'grady , 2008 ) than other species . It may be that , even within a single species , the ability to adjust developmental rate to the presence of a short-lived resource is advantageous . Humans subsist on diets of both plant and animal materials that during most of evolution have differed in their availability . It would be interesting to investigate whether Insulin/IGF signaling in humans might respond to qualitative differences in the lipid composition of these nutritional components . Larval brains were dissected on ice in Graces medium and fixed with 4%PFA at room temperature . Fat bodies were dissected in 4%PFA-Graces medium at room temperature . Samples were stained in 7 . 5% NGS , 0 . 1% Triton X-100 PBS solution and antibodies were diluted as follows: anti-HA16B12 1:1500 ( Santa Cruz Biotechnology , Dallas , TX ) , anti-Elav7E8A10 1:1500 [Developmental Studies Hybridoma Bank ( DSHB ) , University of Iowa , Iowa City , IA] , anti-LTP 1:1000 ( Palm et al . , 2012 ) , anti-Repo8D12 1:1000 ( DSHB ) , anti-LpR1Sac8 and anti-LpR2Sac6 ( 1:500 ) , anti-LRP1 and anti-LRP2 ( 1:500 ) ( Riedel et al . , 2011 ) , anti-Dilp2 and anti-dFOXO ( 1:1000 , gifts from P Leopold ) , anti-IMPL2 ( 1:1000 , gift from E Hafen ) and DAPI 1:100000 ( Roche , Germany ) . For quantifications , tissues were treated in parallel and imaged under identical conditions using either a Zeiss or Olympus confocal microscope . Data were analyzed using Fiji ( Schindelin et al . , 2012 ) . If not stated otherwise , larval sections were performed in ice cold PBS , samples homogenized in 1% Triton X-100 lyses buffer and proteins measured with BCA protein standard Kit ( Invitrogen , manufacturer protocol , Germany ) . Resultant SDS-PAGE blots were probed with anti-AKT1 1:2000 ( Cell Signaling , Danvers , MA ) , anti-ApoLII 1:4000 ( Panakova et al . , 2005 ) , anti-CvD 1:1000 , anti-LTPI 1:3000 or anti-LTPII ( 1:3000 ) ( Palm et al . , 2012 ) . If not stated otherwise , flies were kept at 25°C on either PF or YF under a 12hr light/12hr dark cycle . UAS:lrp1RNAi and UAS:lrp2RNAi lines are from Vienna Drosophila RNAi Center , and oregonRC , repo-Gal4 , dilp2-Gal4 , dilp2 , UAS:gfpRNAi , UAS:GCaMP , UAS:trpA1 lines are from Bloomington Stock center . Published are UAS:ltpRNAi , ltp ( Palm et al . , 2012 ) , imp-l2-Gal4 ( Bader et al . , 2013 ) , lpp-Gal4 ( Brankatschk and Eaton , 2010 ) , lpr1 , lpr2 ( Khaliullina et al . , 2009 ) , lrp1 , lrp2 ( Riedel et al . , 2011 ) , moody-Gal4 ( gift from C Klaembt ) , impl2-Gal4 ( gift from E Hafen ) . UAS:cherry and rab3-Gal4 were generated by transforming VK37 or VK33 flies ( Venken and Bellen , 2005 ) with our cloned constructs . Yrab3 and Yrab27 are N-terminally YFPMYC tagged viable rab alleles ( rab3 and rab27 ) generated by targeted YFPMYC integration into the endogenous locus ( publication in preparation ) . Glucose food ( GF ) per liter: 10 g agar , 2 g glucose , phosphate-buffered saline buffer ( PBS ) ; lipid depleted food ( LDF , calculated calories = 784 kcal/l ) per liter: 10 g agar , 100 g glucose , 100 g chloroform extracted yeast extract , PBS; yeast extract food supplemented with Ergosterol ( LDS ) per liter: 10 g agar , 100 g glucose , 100 g chloroform extracted yeast extract , 10 g Ergosterol ( from a 5 mM EtOH solution ) , PBS; plant food ( PF , calculated calories = 788 kcal/l ) per liter: 10 g agar , 38 g peptone ( soy ) , 80 g cornmeal , 80 g malt , 2 ml cold pressed sun-flower oil , 22 g treacle , 6 . 3 ml propionic acid , 1 . 5 g nipagen; yeast food ( YF , calculated calories = 809 kcal/l ) per liter: 10 g agar , 80 g yeast ( brewers ) , 20 g yeast extract , 20 g peptone ( soy ) , 30 g sucrose , glucose 60 g , 0 . 5 g CaCl2 ( 2 ) H2O , 0 . 5 MgSO4 ( 6 ) H2O , 6 . 3 ml propionic acid , 1 . 5 g nipagen . 40 g PF or YF were homogenized in 100 ml Chloroform . The resulting homogenates were centrifuged and the lower organic phase was removed and evaporated . The residues were re-suspended in 2 ml Chloroform and 30 µl of these solutions were applied to lipid-depleted food . If not stated otherwise , embryos were collected for either 1 hr ( see Figure 4D ) or 4 hr at 25°C , and larvae raised on YF were sampled after 20–24 hr ( First instar ) , 30–34 hr ( Second instar ) and 72–76 hr ( early third instar ) . To compare PF-bred larvae to similarly staged YF-bred larvae , we staged them anatomically using mouth hook morphology . For food shift experiments , larvae were raised on YF and placed for indicated time intervals at 25°C on new food types . Larval hemolymph was prepared as described ( Brankatschk and Eaton , 2010 ) and analyzed using Western blotting . We quantified numbers of stained cells per brain from larval CNS confocal stacks comprising the entire brain lobes but excluding the ventral ganglion . Sections were spaced 1 . 5 μm in Z . Within each experiment , brains were stained and recorded in parallel using the same confocal settings . Unless otherwise stated , embryos were collected on apple-juice-agar plates for 4 hr at 25°C and transferred to the stated diet . Newly formed pupae were counted on a daily basis . Newly hatched female flies were collected in 2–3 hr intervals at 25°C , 3 flies were placed into one vial and weighed . For lifespan experiments , females and males were kept together ( 2:1 ratio ) at 25°C and flipped every 2 days . Deceased female animals were counted daily .	How does an animal sense if it is well nourished or not , and then regulate its metabolism appropriately ? This process largely relies on the animal's body deciphering signals that that are transmitted between different organs in the form of molecules and hormones . Many animals—ranging from insects to mammals ( including humans ) —also use their brains to sense and decipher these nutritional signals . A signaling pathway involving the hormone insulin controls how various different animals grow and develop—and how long they will live—based on these animals' food intake . Insulin is produced in mammals by an organ called the pancreas . But in the fruit fly Drosophila , this hormone is produced by cells within different tissues , including the insect’s brain . The fruit fly is used to study many biological processes because it is easy to work with in a laboratory . Insulin-producing cells make and release insulin-like molecules into the insect's hemolymph ( a blood-like fluid ) in response to sugar and to other nutrients ( which are detected via molecules generated in a fruit fly organ called the fat body ) . The fat body produces lipophorin , a protein which carries fat molecules in the hemolymph , and which is known to be able to move from the hemolymph to the brain and accumulate within the brain . The fat body also produces lipid transfer protein ( or LTP ) , which transfers fats absorbed or made within the insect's gut onto lipophorin , and can also unload fat molecules to other insect cells . If LTP can also enter the brain , and what it might do there , was unclear . Brankatschk et al . now discover that LTP can cross the ‘blood brain barrier’ in fruit fly larvae and can accumulate over time on their insulin-producing cells and the neurons in direct contact with these cells . This accumulation depends on the flies’ diet: flies fed a diet made from yeast cells accumulated LTP on these neurons , while those fed only on sugar and proteins did not . Furthermore Brankatschk et al . found that when they switched flies from a yeast-based to a plant-based diet , the larvae grew more slowly and the flies lived longer . Both of the diets contained abundant calories and nutrients , but contained slightly different kinds of fat molecules . The fly larvae on the plant-based diet also accumulated less LTP on their insulin-pathway neurons , and insulin signaling was reduced . Branskatschk et al . also found that fat molecules from the yeast-based diet activated the cells of the blood brain barrier , and that this encouraged LTP to be transported the brain . Blocking LTP from crossing the blood brain barrier reduced insulin signaling , slowed the growth of the fly larvae , and extended the lifespan of the flies . These findings of Brankatschk et al . thus reveal that fat-containing molecules carry information about specific nutrients to the brain . The extent to which these mechanisms operate in other animals—such as mammals—remains to be explored .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2014	Delivery of circulating lipoproteins to specific neurons in the Drosophila brain regulates systemic insulin signaling
A key challenge in antibiotic stewardship is figuring out how to use antibiotics therapeutically without promoting the evolution of antibiotic resistance . Here , we demonstrate proof of concept for an adjunctive therapy that allows intravenous antibiotic treatment without driving the evolution and onward transmission of resistance . We repurposed the FDA-approved bile acid sequestrant cholestyramine , which we show binds the antibiotic daptomycin , as an ‘anti-antibiotic’ to disable systemically-administered daptomycin reaching the gut . We hypothesized that adjunctive cholestyramine could enable therapeutic daptomycin treatment in the bloodstream , while preventing transmissible resistance emergence in opportunistic pathogens colonizing the gastrointestinal tract . We tested this idea in a mouse model of Enterococcus faecium gastrointestinal tract colonization . In mice treated with daptomycin , adjunctive cholestyramine therapy reduced the fecal shedding of daptomycin-resistant E . faecium by up to 80-fold . These results provide proof of concept for an approach that could reduce the spread of antibiotic resistance for important hospital pathogens . Vancomycin-resistant Enterococcus faecium ( VR E . faecium ) is an important cause of antibiotic-resistant infections in healthcare settings ( Arias and Murray , 2012; García-Solache and Rice , 2019; O'Driscoll and Crank , 2015 ) . The antibiotic daptomycin is one of the few remaining first-line therapies for VRE infection ( O'Driscoll and Crank , 2015 ) , but daptomycin-resistance is spreading in VRE populations ( Judge et al . , 2012; Kamboj et al . , 2011; Kinnear et al . , 2019; Woods et al . , 2018 ) . Therapeutic daptomycin use is thought to be a key driver of resistance ( Kinnear et al . , 2020; Woods et al . , 2018 ) . Managing the evolution of daptomycin-resistance in healthcare settings is crucial to future control of VRE infections . E . faecium is an opportunistic pathogen that colonizes the human GI tract asymptomatically , spreads via fecal-oral transmission , and causes symptomatic infections when introduced to sites like the bloodstream or the urinary tract ( Arias and Murray , 2012 ) . E . faecium colonizing the gut may be exposed to daptomycin during therapeutic use , potentially contributing to the transmission of daptomycin-resistant E . faecium . Daptomycin is administered intravenously to treat infections caused by pathogens including VRE and Staphylococcus aureus . Daptomycin is primarily eliminated by the kidneys , but 5–10% of the dose enters the intestines through biliary excretion ( Woodworth et al . , 1992 ) . We hypothesize that this therapeutically unnecessary intestinal daptomycin exposure could drive resistance evolution in E . faecium colonizing the gut . Increased resistance in colonizing populations is important , because gut E . faecium populations are sources for nosocomial infections and transmission between patients ( Alevizakos et al . , 2017; Olivier et al . , 2008 ) . If unintended intestinal daptomycin exposure drives resistance evolution in E . faecium , this offers an opportunity to intervene . The opportunity emerges from a key feature of this system—the bacteria causing infection are physically separated from the population contributing to transmission . If daptomycin could be inactivated in the intestine without altering plasma concentrations , daptomycin could be used to kill bacteria at the target infection site without driving resistance in off-target populations . Preventing resistance evolution in these reservoir populations could protect patients from acquiring resistant infections , and it could limit the shedding of resistant strains and so onward transmission to other patients . We hypothesized that co-administering an oral adjuvant that reduces daptomycin activity would prevent selection for daptomycin-resistance in the gut during systemic daptomycin treatment . We tested this strategy using the adjuvant cholestyramine in a mouse VR E . faecium gut colonization model . To directly test the proposition that systemic daptomycin treatment could select for resistance in the GI tract , and to generate daptomycin-resistant VR E . faecium mutants for subsequent experiments , we inoculated mice orally with daptomycin-susceptible VR E . faecium strains . Beginning one day after E . faecium inoculation , mice received daily doses of either subcutaneous daptomycin ( 50 , 100 , or 400 mg/kg ) , oral daptomycin ( 5 , 50 , 100 , or 400 mg/kg ) , or a control mock injection for 5 days . We used a range of doses and routes of administration to maximize the likelihood of observing resistance emergence in at least one of the mice . The 50 and 100 mg/kg subcutaneous doses were selected to generate pharmacokinetics similar to clinical human doses ( Mortin et al . , 2007; Samonis et al . , 2008 ) , and the 5 mg/kg oral approximates the 5–10% of a daptomycin dose that is secreted into the intestines during standard intravenous treatment ( Woodworth et al . , 1992 ) . We used two susceptible VR E . faecium strains , BL00239-1 ( MICc = 2 . 0 ( Minimum Inhibitory Concentration computed , see Methods ) ) and PR00708-14 ( MICc = 2 . 7 ) , which were originally isolated at the University of Michigan Hospital from a clinical bloodstream infection and a different patient’s clinical perirectal swab , respectively . Mouse fecal samples were collected to quantify VR E . faecium shedding and determine daptomycin susceptibility of isolated E . faecium clones . Only very high daptomycin doses ( 400 mg/kg subcutaneous , ≥50 mg/kg oral ) consistently reduced fecal VR E . faecium below the level of detection during treatment; with lower doses , VR E . faecium shedding was often detectable throughout treatment ( Figure 1A ) . For Strain BL00239-1 , E . faecium clones with increased daptomycin-resistance were isolated from two of three mice following treatment with 100 mg/kg subcutaneous daptomycin ( Figure 1B–C ) . We chose one of these daptomycin-resistant clones to use in subsequent experiments ( strain BL00239-1-R , MICc = 8 . 6 ) . Sequencing of the core genome showed that the resistant strain acquired a mutation in the major cardiolipin synthase clsA gene ( R211L , CGA→CTA ) , which has been previously described in association with daptomycin-resistance ( Adams et al . , 2015 ) , and a transposon insertion into the methionine sulfoxide reductase msrA gene ( Zhao et al . , 2010 ) . For the second strain ( PR00708-14 ) , mice were treated with subcutaneous daptomycin ( 50 , 100 , or 200 mg/kg ) or a mock injection . We screened for the emergence of increased resistance by plating mouse fecal suspensions on daptomycin-supplemented agar . Samples from two mice treated with 200 mg/kg daptomycin produced colonies on daptomycin-supplemented plates , and we isolated three clones from each of these samples . These isolated clones had increased daptomycin MICc relative to PR00708-14 by broth microdilution ( MICc = 13 . 5 , 11 . 5 , and 8 . 0 from one mouse; MICc = 12 . 0 , 29 . 8 , and 12 . 0 from the second mouse ) . We chose one of these isolated clones for use in subsequent experiments ( PR00708-14-R , MICc = 12 . 0 ) . Genome sequencing revealed that PR00708-14 and PR00708-14-R differed by two mutations , which to our knowledge have not previously been associated with daptomycin-resistance . We identified mutations in a TerC family integral membrane protein ( locus tag HMPREF0351_10759 in D0 E . faecium reference genome , I22T , CGA→CTA ) and in a hypothetical protein ( locus tag HMPREF0351_12146 in D0 E . faecium reference genome , frameshift , deleted G at 190nt ) . These experiments show that daptomycin-resistance can emerge de novo in E . faecium colonizing the GI tract following systemic daptomycin treatment . We used the de novo resistant mutants isolated above ( Figure 1 ) to test whether daptomycin therapy selects for daptomycin-resistance in intestinal VR E . faecium populations when a resistant mutant is already present . We orally inoculated mice with a 1:20 mixture of the experimentally generated daptomycin-resistant and susceptible VR E . faecium strains ( BL00239-1-R and BL00239-1 ) . Note that this approach – seeding inocula with known numbers of resistant bacteria – allows the response to selection to be measured while avoiding the experimental noise introduced by mutation waiting times . Mice were treated with subcutaneous daptomycin ( 50 , 75 , 100 , or 200 mg/kg ) for five or ten days after VRE inoculation . Control mice received either a mock saline injection or no injection . Fecal samples from Days 8 and 14 post-inoculation were plated in triplicate to quantify total VR E . faecium density , and samples were also plated on daptomycin-supplemented agar to estimate the proportion of VR E . faecium that were daptomycin-resistant . Control populations remained susceptible to daptomycin , but all doses and durations of daptomycin dramatically enriched for resistance in the GI tract ( Figure 2A–B ) . At both time points , controls had significantly lower proportions of resistant bacteria than daptomycin-treated mice ( Figure 2A–B; mixed effects negative binomial regression , p<0 . 01 , see Model one in Supplementary file 1 ) . The effect sizes ( Cohen’s d ) at Days 8 and 14 were 5 . 90 ( 95% CI 4 . 94 , 6 . 86 ) and 2 . 34 ( 95% CI 1 . 82 , 2 . 87 ) , respectively . The absolute numbers of VRE enumerated in fecal samples did not vary significantly between treatments ( mixed effects negative binomial regression , Model two in Supplementary file 1; Figure 2—figure supplement 1 ) . The dramatic enrichment for daptomycin-resistant VR E . faecium in treated mice showed that subcutaneously-administered daptomycin produced GI tract concentrations high enough to select for resistance . To quantify fecal daptomycin concentrations , we analyzed fecal samples from a subset of daptomycin-treated mice by liquid chromatography-mass spectrometry ( LC-MS ) ( Figure 2C ) . Samples from all time points tested ( Days 2 , 6 , and 8 ) contained detectable daptomycin , and concentrations generally peaked at the end of treatment ( Day 6 ) . Higher daptomycin doses generally corresponded to higher fecal concentrations , but concentrations were highly variable and overlapped between treatments . While fecal VR E . faecium densities correlated poorly with the daptomycin dose administered ( Figure 2—figure supplement 1 ) , fecal VR E . faecium densities correlated with the amount of daptomycin recovered in feces ( Figure 2D ) . These data confirmed that subcutaneously-administered daptomycin at our experimental doses generated a range of daptomycin concentrations in the GI tract that included inhibitory concentrations for the susceptible VR E . faecium strain . We ran two additional experiments to further investigate the competitive dynamics between this susceptible and resistant pair ( BL00239-1 and BL00239-1-R ) in the presence and absence of daptomycin treatment . First , we tested whether susceptible bacteria competitively suppressed resistant bacteria in the GI tract . We inoculated mice with either a mixture of 108 CFU susceptible + 103 CFU-resistant VR E . faecium , or with a resistant-only inoculum at one of two inoculum sizes ( 108 CFU or 103 CFU ) . Mice received 5 days of subcutaneous daptomycin injections at 100 mg/kg or control saline injections . Shedding of resistant and susceptible bacteria were quantified at time points throughout the experiment by plating ( Figure 3A ) . In the absence of daptomycin treatment , the daptomycin-susceptible strain remained the most prevalent in mixed populations . When mixed populations were exposed to daptomycin , resistance increased to high frequency in three populations , and the population size fell dramatically in the remaining two populations . In mice inoculated with only 103 CFU-resistant bacteria , the resistant clone was able to grow to high numbers with or without daptomycin . These data were consistent with the competitive suppression of the resistant strain by the susceptible strain in the absence of daptomycin treatment , and competitive release of the resistant strain during daptomycin treatment ( Day et al . , 2015; Wargo et al . , 2007 ) . Next , we tested whether the frequency of daptomycin-resistant VR E . faecium would decrease over time in the absence of daptomycin treatment , potentially indicating that daptomycin-resistance comes at a fitness cost . We inoculated mice with a 1:5 mixture of daptomycin-resistant and susceptible VR E . faecium . Mice received no daptomycin treatment . After 14 days , the proportion of resistant bacteria had declined ( one sample t-test , t = −22 . 42 , df = 9 , p<0 . 01 , Cohen’s d = 7 . 09 ) , consistent with a competitive disadvantage ( fitness cost ) to the daptomycin-resistance mutation ( Figure 3B ) . Control mice inoculated with only resistant or only susceptible bacteria did not have significantly different proportions of resistance between days 0 and 14 ( resistant: t = −3 . 73 , df = 2 , p=0 . 06; susceptible: t = 1 , df = 2 , p=0 . 42 ) . If an orally-administered adjuvant could reduce daptomycin activity in the GI tract , this could prevent the emergence of daptomycin-resistant E . faecium in the gut , potentially reducing transmission of resistant bacteria without impacting the effectiveness of intravenous daptomycin therapy . We identified cholestyramine , an FDA-approved bile-acid sequestrant , as a potential adjuvant for daptomycin therapy . Cholestyramine is a high-molecular weight anion exchange resin that binds with bile acids , forming an insoluble complex that is excreted in the feces ( Jacobson et al . , 2007 ) . Cholestyramine is known to interact with a number of co-administered drugs through the same mechanism , reducing their bioactivity ( Jacobson et al . , 2007 ) . We hypothesized that cholestyramine would bind daptomycin based on their chemical structures . In vitro tests were consistent with cholestyramine binding daptomycin . Daptomycin solutions were incubated with cholestyramine , and then the cholestyramine was removed by centrifugation . The resulting supernatants were analyzed for changes in daptomycin concentration and activity . Daptomycin concentrations can be measured directly by ultraviolet ( UV ) absorbance at 364 nm ( Figure 4A ) . Daptomycin concentrations were reduced in supernatants after incubation with cholestyramine in a dose-dependent manner ( Figure 4B ) . Additionally , daptomycin solutions incubated with cholestyramine had reduced antibiotic activity against E . faecium in broth microdilution ( Figure 4C ) . Together , these results were consistent with cholestyramine removing daptomycin from solution , supporting cholestyramine as a candidate adjuvant for daptomycin therapy . We conducted four experiments to test whether adjunctive therapy with cholestyramine could prevent the emergence of daptomycin-resistant VR E . faecium in the mouse GI tract . In each experiment , mice were orally inoculated with a 1:20 mixture of daptomycin-resistant and susceptible VR E . faecium and then treated with subcutaneous daptomycin injections for 5 days . Densities of total VR E . faecium and daptomycin-resistant VR E . faecium were determined by plating ( Figure 5 , Figure 5—figure supplements 1–5 ) . The experiments tested the evolutionary impact of oral cholestyramine in different mouse strains , with different VR E . faecium strains , and with different timing of cholestyramine administration . The design of the four experiments was as follows: ( A ) Swiss Webster mice with E . faecium strains BL00239-1 and BL00239-1-R , with cholestyramine started one day before daptomycin ( Figure 5—figure supplement 1 ) , ( B ) C57BL/6 mice with E . faecium strains BL00239-1 and BL00239-1-R , with cholestyramine started one day before daptomycin ( Figure 5—figure supplement 2 ) , ( C ) Swiss Webster mice with E . faecium strains PR00708-14 and PR00708-14-R , with cholestyramine started one day before daptomycin ( Figure 5—figure supplement 3 ) , and ( D ) Swiss Webster mice with E . faecium strains BL00239-1 and BL00239-1-R , with cholestyramine started the same day as daptomycin ( Figure 5—figure supplement 4 ) . Data from these experiments were analyzed together , with a block effect included in the models . Because bacterial densities were found not to correlate to daptomycin dose ( Figure 2—figure supplement 1 ) , all daptomycin doses were combined into a single group for analysis . Figure 5—figure supplements 1–4 show these data broken down by experiment and by daptomycin dose . For daptomycin-treated mice shedding detectable levels of VR E . faecium by our plating assay ( at least 20 CFU/10 mg feces at a given time point ) , the cholestyramine-supplemented diet reduced the proportion of daptomycin-resistant VR E . faecium ( mixed effects binomial regression , p<0 . 01 , Model three in Supplementary file 1 ) . The effect size ( Cohen’s d ) of cholestyramine diet on the proportion of resistant bacteria in daptomycin-treated mice was 0 . 41 ( 95% CI 0 . 01 , 0 . 82 ) at Day 2 , 0 . 56 ( 95% CI 0 . 13 , 1 . 00 ) at Day 4 , 0 . 90 ( 95% CI 0 . 45 , 1 . 35 ) at Day 6 , 1 . 08 ( 95% CI 0 . 63 , 1 . 53 ) at Day 8 , and 0 . 63 ( 95% CI 0 . 21 , 1 . 05 ) at Day 14 . In addition , to more accurately determine resistance proportions for Days 8 and 14 , we plated an estimated 200 CFU from each sample in triplicate on plates with and without daptomycin ( Figure 5—figure supplement 5 ) . This second assay confirmed that cholestyramine reduced the proportion of resistant VR E . faecium at these time points ( mixed effects binomial regression , p<0 . 01 , Model four in Supplementary file 1; Cohen’s d 1 . 25 ( 95% CI 0 . 97 , 1 . 53 ) at Day 8 , 0 . 73 ( 95% CI 0 . 48 , 0 . 98 ) at Day 14 ) . We also quantified absolute densities of daptomycin-resistant and susceptible VR E . faecium over time by plating samples from Days 0 , 1 , 2 , 4 , 6 , 8 , and 14 ( Figure 5 ) . These data showed that the cholestyramine-supplemented diet reduced fecal shedding of daptomycin-resistant VR E . faecium in daptomycin-treated mice ( AntibioticDietDay interaction p<0 . 01 , Model five in Supplementary file 1 ) . The effect size was greatest in the days after daptomycin treatment . The effect size ( Cohen’s d ) of cholestyramine diet on shedding of resistant bacteria in daptomycin-treated mice was 0 . 43 ( 95% CI 0 . 02 , 0 . 83 ) at Day 2 , 0 . 08 ( 95% CI −0 . 32 , 0 . 48 ) at Day 4 , 0 . 12 ( 95% CI −0 . 28 , 0 . 51 ) at Day 6 , 0 . 36 ( 95% CI −0 . 04 , 0 . 76 ) at Day 8 , and 0 . 57 ( 95% CI 0 . 16 , 0 . 98 ) at Day 14 . Total VR E . faecium shedding was also influenced by the addition of cholestyramine ( AntibioticDietDay interaction p<0 . 01 , Model six in Supplementary file 1 ) . Here the effect size is greatest during daptomycin treatment . The effect size ( Cohen’s d ) of cholestyramine diet on total shedding in daptomycin-treated mice was 0 . 84 ( 95% CI 0 . 43 , 1 . 23 ) at Day 2 , 0 . 36 ( 95% CI −0 . 04 , 0 . 76 ) at Day 4 , 0 . 31 ( 95% CI −0 . 09 , 0 . 71 ) at Day 6 , 0 . 04 ( 95% CI −0 . 36 , 0 . 43 ) at Day 8 , and 0 . 40 ( 95% CI 0 . 00 , 0 . 80 ) at Day 14 . If we consider only the control treatments , where there is no possibility of cholestyramine protecting against daptomycin killing , the addition of cholestyramine to the diet does not significantly influence total VR E . faecium counts alone ( p=0 . 63 , Model seven in Supplementary file 1 ) or in combination with day ( p=0 . 18 ) . Here we have shown proof of concept for an adjunctive therapy approach to prevent the emergence of daptomycin-resistant E . faecium in the GI tract during daptomycin therapy . Ideally , this approach would allow clinicians to treat bloodstream infections with intravenous daptomycin without fueling the hospital transmission of multidrug-resistant bacteria . This would be a novel approach because the desired outcome is reduced resistance evolution and reduced transmission of resistant pathogens . Colonization with VR E . faecium is a risk factor for future infection ( Alevizakos et al . , 2017; Olivier et al . , 2008 ) , presumably because GI tract populations are sources for infections . By maintaining daptomycin sensitivity GI tract populations , this strategy could reduce the risk of resistant infection for colonized patients , as well as reducing transmission of resistance between patients . With some optimization , this adjunctive approach could be implemented in hospitals . Cholestyramine is an inexpensive , FDA-approved drug with few side effects ( Beckett and Wilhite , 2015; Jacobson et al . , 2007 ) . Cholestyramine has been used clinically for over 50 years , including in hospital patient populations that would be targets for adjunctive cholestyramine therapy ( Ballantyne , 2009 ) . Cholestyramine does have potential to interfere with other orally administered drugs , which could be a risk considered on a patient by patient basis ( Scaldaferri et al . , 2013 ) . This drug–drug interaction potential is managed in practice by administering oral drugs an hour before or 4 to 6 hr after cholestyramine administration . Because the effects and risks of cholestyramine are well understood , testing cholestyramine as an adjuvant with daptomycin in human trials is appealingly low risk to patient health while offering large potential gains to hospital infection control . While cholestyramine has the potential to be repurposed in hospitals , this study also revealed potential limitations of cholestyramine therapy and areas requiring further study prior to clinical trials . Cholestyramine therapy was most effective in a narrow range of conditions ( intermediate daptomycin doses , cholestyramine started prior to daptomycin therapy ) ( Figure 5—figure supplements 1–5 ) , and the daptomycin dose at which resistance was most likely to be enriched varied between E . faecium strains ( Figure 5—figure supplements 1–5 ) . The effect of cholestyramine was also variable among mice ( Figure 5—figure supplements 1–5 ) , possibly as a result of variable daptomycin concentrations in the intestines ( Figure 2C–D ) . This is potentially a major limitation for translating this therapy , as daptomycin pharmacokinetics and pharmacodynamics also vary among human patients . Optimizing cholestyramine therapy will require detailed data on intestinal daptomycin pharmacokinetics , including temporal variation over the course of treatment and variation among individuals . From these data , one could estimate a maximum target for daptomycin capture , and optimize cholestyramine dosing accordingly . This therapeutic approach has the potential to be expanded to other pathogen-drug combinations , but there is likely no one size fits all adjunctive therapy; optimization may be required for each pathogen-drug combination . In addition to binding antibiotics , cholestyramine sequesters bile acids in the intestine , which alters the GI tract environment . This means that cholestyramine treatment could indirectly affect bacteria in the GI tract through mechanisms other than direct interactions with drugs . For example , exposure to bile acids has implications for E . faecium phenotypes . Secondary bile acids trigger a morphotype switch in Enterococcus that facilitates intestinal colonization and biofilm formation ( McKenney et al . , 2019 ) . In our experiments , we did not observe differences in colonization efficiency or VR E . faecium counts in mice treated with cholestyramine alone . In mouse models , cholestyramine has also been shown to reduce bile acid-mediated resistance to Clostridium difficile infection , which is a possible risk associated with cholestyramine treatment ( Buffie et al . , 2015 ) . Here we tested cholestyramine as an adjuvant to reduce selection for antibiotic resistance in the GI tract , but other adjuvants could be used or designed for the same purpose . At least two other drugs that reduce antibiotic activity in the GI tract are currently in development . DAV-132 , a formulation of activated charcoal encased in zinc-pectinate beads , was shown to site-specifically bind antimicrobials in the gut in a stage I clinical trial ( de Gunzburg et al . , 2018; de Gunzburg et al . , 2015 ) . DAV-132 has been shown to reduce fecal concentrations of antibiotics by 99% without affecting plasma concentrations ( Burdet et al . , 2017; de Gunzburg et al . , 2018; de Gunzburg et al . , 2015; Khoder et al . , 2010 ) . Activated charcoal could adsorb antibiotics in the gut , and could be an effective adjuvant for a broad range of antibiotic classes . Adjuvants with specific activity against particular antibiotics could also be developed . This has been successfully demonstrated with orally-administered β-lactamases given with intravenous β-lactam antibiotics . β-lactamases enzymatically inactivate β-lactams . Under the name SYN-004 , this β-lactamase treatment advanced to clinical trials in human subjects ( Kaleko et al . , 2016; Kokai-Kun et al . , 2017; Pitout , 2009; Tarkkanen et al . , 2009 ) . Data from clinical trials show the drugs successfully inactivate β-lactams in the digestive tract without adversely affecting levels of antibiotic in plasma ( Kaleko et al . , 2016; Kokai-Kun et al . , 2017; Pitout , 2009; Tarkkanen et al . , 2009 ) . These drugs have been developed with the goal of preventing infection with C . difficile after antibiotic therapy , but they could likely also prevent the emergence of antibiotic resistance in the GI tract . The orally-administered β-lactamase SYN-006 mitigated the enrichment of genes associated with antibiotic resistance in the microbiomes of pigs treated with a carbapenem antibiotic ( Connelly et al . , 2019 ) . So far as we are aware , there is no direct experimental evidence analogous to ours that those drugs can prevent resistance emergence in colonizing opportunistic pathogens , but it seems likely they could . Adjunctive therapies like the one proposed here could help manage resistance evolution in other important pathogens listed by Centers for Disease Control and Prevention ( CDC ) and World Health Organization ( WHO ) as major threats ( CDC , 2019; WHO , 2017 ) . Like VRE , many opportunistic pathogens experience substantial antibiotic exposure when they are not the targets of treatment . Opportunistic pathogens like Klebsiella pneumoniae , Escherichia coli , Staphylococcus aureus , and Enterobacter cloacae colonize the gut asymptomatically , where they can be unintentionally exposed to antibiotics ( Morley et al . , 2019 ) . According to one estimate , over 90% of the total antimicrobial exposure experienced by K . pneumoniae occurs when K . pneumoniae was not the target of treatment ( Tedijanto et al . , 2018 ) . For H . influenzae , E . coli , and Staphylococcus aureus over 80% of total exposure to antibiotics was estimated to occur when the bacteria were bystanders ( Tedijanto et al . , 2018 ) . These colonizing populations are sources for infections , so selection for resistance in bystander populations can contribute to rising rates of resistant infections ( Morley et al . , 2019 ) . Adjunctive strategies that allow intravenous antibiotics to reach target sites while reducing off-target exposure could help stem the spread of resistant pathogens listed by the CDC as urgent or serious threats ( CDC , 2019 ) , including carbapenem-resistant Enterobacteriaceae ( CRE ) , extended-spectrum beta-lactamase producing Enterobacteriaceae , vancomycin-resistant Enterococcus ( VRE ) , methicillin-resistant Staphylococcus aureus ( MRSA ) , and drug-resistant Staphylococcus aureus . The strategy presented here focuses on inactivating antibiotic in the gastrointestinal tract , the likely source of the bulk of the antimicrobial resistance in pathogens listed as top threats by the CDC , but similar strategies could be developed to shield microbiota at other sites , such as the skin and respiratory tract . While adjunctive therapies have potential to be used for a variety of pathogens , the genetic mechanism of resistance will influence the resulting evolutionary dynamics . The cholestyramine therapy described here prevents enrichment of resistant clones within a bacterial population consisting of both sensitive and resistant bacteria . Daptomycin-resistance in E . faecium emerges through point mutations in the genome , so patients colonized with daptomycin-susceptible E . faecium likely harbor low-frequency resistant clones due to spontaneous mutation . Other forms of resistance , like VanA-type vancomycin resistance in E . faecium , emerge through horizontal gene transfer rather than point mutations ( Depardieu and Courvalin , 2017 ) . For resistance to emerge in patient colonized with vancomycin-susceptible E . faecium , a horizontal transfer event would have to occur , or a new resistant colonizer would have to be introduced . If these events are rarer than spontaneous mutation , patients are less likely to be colonized with a mixture of vancomycin sensitive and resistant bacteria than with daptomycin-resistance , as studied here . However , gains and losses of transmissible resistance elements have been observed in the gut microbiomes of patients ( Kinnear et al . , 2020 ) , and altering antibiotic pressure in the gut could reduce selection for resistant bacteria , even when that resistance is encoded by horizontally transferred genes . Unless otherwise specified , mice in all experiments were female Swiss Webster . In one experiment , inbred female C57BL/6 mice were used to ensure the results were not specific to one mouse strain . Mice were fed a standard diet ( 5001 Laboratory Rodent Diet ) or a standard diet supplemented with 2% w/w cholestyramine resin . All mice were housed individually during experiments . Daptomycin-susceptible VR E . faecium strains were isolated from different patients at the University of Michigan Hospital . Strain BL00239-1 was isolated from a bloodstream infection , and strain PR00708-14 was isolated from a perirectal swab . Additional strains were isolated during these experiments from mouse fecal samples ( including BL00239-1-R and PR00708-14-R ) . These strains were isolated by streaking on agar plates for two rounds of clonal purification . All mice were pretreated with ampicillin ( 0 . 5 g/L in drinking water ) for 7 days before E . faecium inoculation . Ampicillin disrupts the natural gut flora and facilitates Enterococcus colonization ( McKenney et al . , 2019 ) . Sample sizes for mouse experiments were chosen based on previous experience with similar experiments . Mice that were co-housed during ampicillin pre-treatment were evenly allocated among experimental treatment groups . E . faecium strains were plated from glycerol stocks and then grown overnight in liquid culture in Brain Heart Infusion broth . Mice were inoculated via oral gavage with 108 CFU E . faecium suspended in saline . E . faecium inoculum counts were confirmed by plating . Following E . faecium inoculation , mice were split into individual cages with untreated water and any experimental diets . Daptomycin doses were administered daily starting one day post-inoculation via subcutaneous injection or oral gavage . Daptomycin doses were based on an average mouse weight for each experiment . In some experiments , a cholestyramine-supplemented diet ( 2% w/w ) was provided to mice starting one day prior to the first daptomycin dose ( Experiments A-C ) or starting the same day as the first daptomycin dose ( Experiment D ) . Once started , mice were maintained on the cholestyramine diet for the duration of the experiment . For stool collection , mice were placed in clean plastic cups , and fresh stool was collected using a sterile toothpick . Stool samples were suspended in PBS ( 25 uL PBS/mg stool ) and frozen with glycerol at −80°C for subsequent analysis . For measurements of UV absorbance , solutions of 5 mg/mL daptomycin in phosphate-buffered saline ( PBS ) were combined with various concentrations of cholestyramine . These mixtures were vortexed for 30 s , then allowed to incubate for 5 min at room temperature . Following incubation , cholestyramine was removed by centrifugation . Supernatants were analyzed for absorbance at 364 nm on a NanoVue Plus Spectrophotometer . A calibration curve was used to determine daptomycin concentrations from A364 values . For tests of daptomycin bioactivity , solutions of 1 mg/mL daptomycin were incubated with or without 12 mg/mL cholestyramine for 45 min at 37°C with shaking ( N = 3 per treatment ) . The cholestyramine was removed by centrifugation , and the supernatant was used in broth microdilutions with E . faecium . Saline solution incubated with cholestyramine run as a control had no effect on cell growth . VR E . faecium were enumerated by plating diluted fecal suspensions on selective plates ( Enterococcosel agar supplemented with 16 µg/mL vancomycin ) . Plates were incubated at 35°C for 40–48 hr , and colonies were counted . To quantify the proportion of these bacteria that were daptomycin-resistant , fecal suspensions were plated on calcium-supplemented Enterococcosel plates with 16 µg/mL vancomycin and 10 µg/mL daptomycin . Plates were incubated at 35°C for 40–48 hr , and colonies were counted . Serially-diluted fecal suspensions were each plated once on plates without daptomycin and once on plates containing daptomycin to estimate the proportion of daptomycin-resistant bacteria . An additional plating assay was performed to more accurately determine proportions of resistant bacteria for Days 8 and 14 in each experiment . For this assay , known sample densities were used to plate an estimated 200 CFU on plates with and without daptomycin in triplicate . This assay only included samples with high enough initial bacterial density to generate 200 CFU on each plate ( 3 × 103 CFU/10 mg in initial sample ) . Counters were not blinded to treatment groups . In some experiments , E . faecium clones were isolated from fecal samples and analyzed by broth microdilution . Clones were purified by streaking twice on Enterococcosel agar with 16 µg/mL vancomycin , and were then stored in glycerol stocks at −80°C . Broth microdilutions were performed according to Clinical and Laboratory Standards Institute ( CLSI ) guidelines ( CLSI , 2017 ) . After incubation , cell densities were measured by OD600 absorbance in a plate reader . OD values were fitted to a Hill function curve to determine the computed MIC ( MICc ) as described previously ( Kinnear et al . , 2020 ) . Whole genomic DNA preparations were submitted to the University of Michigan sequencing core for Illumina library preparation and paired end sequencing with Illumina NovaSeq 6000 ( isolates PR00708-14 and PR00708-14-R ) or Illumina HiSeq ( isolates BL00239-1 and BL00239-1-R ) . Long read data was additionally generated for strain BL00239-1 using the Oxford Nanopore MinION . The nanopore library were prepared using the Nanopore Ligation Sequencing Kit ( SQK-LSK109 ) . Quality control of sequencing reads was performed using Trimmomatic ( Bolger et al . , 2014 ) . De novo genome assembly was performed using SPAdes ( Bankevich et al . , 2012 ) and genomes were annotated using Prokka ( Seemann , 2014 ) . Trimmed reads from resistant isolates were mapped against corresponding susceptible reference genomes using Burrows-Wheeler Aligner ( BWA ) ( Li and Durbin , 2009 ) , and candidate variants were identified with The Genome Analysis Toolkit ( GATK ) ( McKenna et al . , 2010 ) or breseq ( Deatherage and Barrick , 2014 ) . Reads from the reference sample were aligned to the reference genome ( aligned to self ) to generate a list of background variants; these background variants were filtered out during variant calling . Remaining candidate variants were screened by visual inspection of alignments in Integrative Genomics Viewer ( IGV ) ( Thorvaldsdottir et al . , 2013 ) . Fecal daptomycin concentrations were measured via LC-MS at the University of Michigan Pharmacokinetics Core . A labeled daptomycin-d5 internal standard was used to generate calibration curves . Statistical analyses were run in R v1 . 2 . 1335 ( Brooks et al . , 2017 ) using the packages ‘nlme’ ( Pinheiro et al . , 2016 ) and ‘glmmTMB’ ( Brooks et al . , 2017 ) . To analyze proportions of resistant bacteria , samples were plated on agar with and without daptomycin , resulting in a count of resistant bacteria and a count of total bacteria . Due to sampling , these proportions were not bounded by zero and one , so proportion data were normalized by dividing each value by the maximum value in the data set . Proportions of resistant bacteria were analyzed using mixed binomial regression models . Absolute VRE densities were analyzed using mixed models with an autoregressive error structure as previously described ( Pollitt et al . , 2012 ) . Full model structures and output are shown in Supplementary file 1 .	Antibiotics are essential for treating infections . But their use can inadvertently lead to the emergence of antibiotic-resistant bacteria that do not respond to antibiotic drugs , making infections with these bacteria difficult or impossible to treat . Finding ways to prevent antibiotic resistance is critical to preserving the effectiveness of antibiotics . Many bacteria that cause infections in hospitals live in the intestines , where they are harmless . But these bacteria can cause life-threatening infections when they get into the bloodstream . When patients with bloodstream infections receive antibiotics , the bacteria in their intestines are also exposed to the drugs . This can kill off all antibiotic-susceptible bacteria , leaving behind only bacteria that have mutations that allow them to survive the drugs . These drug-resistant bacteria can then spread to other patients causing hard-to-treat infections . To stop this cycle of antibiotic treatment and antibiotic resistance , Morley et al . tested whether giving a drug called cholestyramine with intravenous antibiotics could protect the gut bacteria . In the experiments , mice were treated systemically with an antibiotic called daptomycin , which caused the growth of daptomycin-resistant strains of bacteria in the mice’s intestines . In the laboratory , Morley et al . discovered that cholestyramine can inactivate daptomycin . Giving the mice cholestyramine and daptomycin together prevented the growth of antibiotic-resistant bacteria in the mice's intestines . Moreover , cholestyramine is taken orally and is not absorbed into the blood . It therefore only inactivates the antibiotic in the gut , but not in the blood . The experiments provide preliminary evidence that giving cholestyramine with antibiotics might help prevent the spread of drug resistance . Cholestyramine is already used to lower cholesterol levels in people . More studies are needed to determine if cholestyramine can protect gut bacteria and prevent antibiotic resistance in people .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2020	An adjunctive therapy administered with an antibiotic prevents enrichment of antibiotic-resistant clones of a colonizing opportunistic pathogen
Transposable elements ( TEs ) , the movement of which can damage the genome , are epigenetically silenced in eukaryotes . Intriguingly , TEs are activated in the sperm companion cell – vegetative cell ( VC ) – of the flowering plant Arabidopsis thaliana . However , the extent and mechanism of this activation are unknown . Here we show that about 100 heterochromatic TEs are activated in VCs , mostly by DEMETER-catalyzed DNA demethylation . We further demonstrate that DEMETER access to some of these TEs is permitted by the natural depletion of linker histone H1 in VCs . Ectopically expressed H1 suppresses TEs in VCs by reducing DNA demethylation and via a methylation-independent mechanism . We demonstrate that H1 is required for heterochromatin condensation in plant cells and show that H1 overexpression creates heterochromatic foci in the VC progenitor cell . Taken together , our results demonstrate that the natural depletion of H1 during male gametogenesis facilitates DEMETER-directed DNA demethylation , heterochromatin relaxation , and TE activation . Large proportions of most eukaryotic genomes are comprised of transposable elements ( TEs ) , mobile genetic fragments that can jump from one location to another . For example , TEs comprise approximately 50% of the human genome ( International Human Genome Sequencing Consortium , 2001; Venter et al . , 2001 ) , and more than 85% of the genomes in crops such as wheat and maize ( Schnable et al . , 2009; Wicker et al . , 2018 ) . Regarded as selfish and parasitic , activities of TEs compromise genome stability , disrupt functional genes , and are often associated with severe diseases including cancers in animals ( Anwar et al . , 2017 ) . To safeguard genome integrity , eukaryotic hosts have evolved efficient epigenetic mechanisms , including DNA methylation , to suppress TEs ( He et al . , 2011; Law and Jacobsen , 2010 ) . Curiously , recent studies point to episodes of TE activation that occur in specific cell types and/or particular developmental stages ( Garcia-Perez et al . , 2016; Martínez and Slotkin , 2012 ) . These TE activation events provide unique opportunities to understand epigenetic silencing mechanisms , and the co-evolution between TEs and their hosts . Developmental TE activation has been shown in mammalian embryos , germlines and brain cells . In pre-implantation embryos and the fetal germline , LINE-1 retrotransposons are highly expressed despite relatively low levels of transposition ( Fadloun et al . , 2013; Kano et al . , 2009; Percharde et al . , 2017; Richardson et al . , 2017 ) . Recently , LINE-1 RNA was shown to play a key regulatory role in promoting pre-implantation embryo development in mice ( Percharde et al . , 2018 ) . LINE-1 elements have also been shown to transcribe and mobilize in neuronal precursor cells in mice and human ( Coufal et al . , 2009; Muotri et al . , 2005 ) . The underlying mechanism of such cell-specific TE activation is still unclear . Hypomethylation at LINE-1 promoters in neurons has been proposed to contribute ( Coufal et al . , 2009 ) , and possibly the availability of transcription factors ( Muotri et al . , 2005; Richardson et al . , 2014 ) . The frequency of LINE-1 retrotransposition in mammalian brain is still under debate; however , it has been speculated that LINE-1 activities might serve to promote genetic diversity among cells of a highly complex organ like the brain ( Garcia-Perez et al . , 2016; Richardson et al . , 2014; Singer et al . , 2010 ) . One of the best demonstrated cases of developmental TE activation occurs in the male gametophyte of flowering plants , pollen grains . Pollen are products of male gametogenesis , which initiates from haploid meiotic products called microspores . Each microspore undergoes an asymmetric mitosis to generate a bicellular pollen comprised of a large vegetative cell ( VC ) and a small generative cell engulfed by the VC ( Berger and Twell , 2011 ) . Subsequently the generative cell divides again mitotically to produce two sperm . Upon pollination , the VC develops into a pollen tube to deliver the sperm to meet the female cells , and subsequently degenerates . In the mature tricellular pollen of Arabidopsis thaliana , several TEs were found activated and transposed ( Slotkin et al . , 2009 ) . Enhancer/gene trap insertions into TEs showed specific reporter activity in the VC , and TE transpositions detected in pollen were absent in progeny ( Slotkin et al . , 2009 ) . These results demonstrated TE activation in pollen is confined to the VC . TE expression in the short-lived VC has been proposed to promote the production of small RNAs , which may be transported into sperm to reinforce the silencing of cognate TEs ( Calarco et al . , 2012; Ibarra et al . , 2012; Martínez et al . , 2016; Slotkin et al . , 2009 ) . However , TE transcription in the VC has not been comprehensively investigated , hindering our understanding of this phenomenon . The mechanisms underlying TE reactivation in the VC are also unknown . One proposed mechanism is the absence of the Snf2 family nucleosome remodeler DDM1 ( Slotkin et al . , 2009 ) . DDM1 functions to overcome the impediment of nucleosomes and linker histone H1 to DNA methyltransferases ( Lyons and Zilberman , 2017; Zemach et al . , 2013 ) . Loss of DDM1 leads to DNA hypomethylation and massive TE derepression in somatic tissues ( Jeddeloh et al . , 1999; Lippman et al . , 2004; Tsukahara et al . , 2009; Zemach et al . , 2013 ) . However , global DNA methylation in the VC is comparable to that of microspores and substantially higher than in somatic tissues ( Calarco et al . , 2012; Hsieh et al . , 2016; Ibarra et al . , 2012 ) . This suggests that DDM1 is present during the first pollen mitosis that produces the VC , so its later absence is unlikely to cause TE activation . A plausible mechanism underlying TE activation in the VC is active DNA demethylation . DNA methylation in plants occurs on cytosines in three sequence contexts: CG , CHG and CHH ( H = A , C or T ) . Approximately ten thousand loci – predominantly TEs – are hypomethylated in the VC , primarily in the CG context and to a lesser extent in the CHG/H contexts ( Calarco et al . , 2012; Ibarra et al . , 2012 ) . Hypomethylation in the VC is caused by a DNA glycosylase called DEMETER ( DME ) ( Ibarra et al . , 2012 ) . DME demethylates DNA via direct excision of methylated cytosine , and its expression is confined to the VC and its female counterpart , the central cell , during sexual reproduction ( Choi et al . , 2002; Schoft et al . , 2011 ) . DME demethylation may therefore cause TE transcription in the VC; however , this hypothesis has not been tested . Another plausible mechanism for epigenetic TE activation is chromatin decondensation ( Feng et al . , 2013 ) . Drastic reprogramming of histone variants and histone modifications occurs during both male and female gametogenesis , rendering the gametes and companion cells with radically different chromatin states ( Baroux et al . , 2011; Borg and Berger , 2015 ) . For example , centromeric repeats , which are condensed in sperm and other cell types , are decondensed in the VC , accompanied by the depletion of centromeric histone H3 ( Ingouff et al . , 2010; Mérai et al . , 2014; Schoft et al . , 2009 ) . Chromocenters , which are comprised of condensed pericentromeric heterochromatin and rDNA repeats ( Chandrasekhara et al . , 2016; Fransz et al . , 2002; Tessadori et al . , 2004 ) , are observed in sperm nuclei but absent in the VC nucleus , suggesting that pericentromeric heterochromatin is decondensed in the VC ( Baroux et al . , 2011; Ingouff et al . , 2010; Schoft et al . , 2009 ) . Heterochromatin decondensation in the VC is proposed to promote rDNA transcription that empowers pollen tube growth ( Mérai et al . , 2014 ) . However , the cause of VC heterochromatin decondensation remains unclear . Our previous work showed that histone H1 , which binds to the nucleosome surface and the linker DNA between two adjacent nucleosomes ( Fyodorov et al . , 2018 ) , is depleted in Arabidopsis VC nuclei ( Hsieh et al . , 2016 ) . H1 depletion in the VC has also been observed in a distantly related lily species ( Tanaka et al . , 1998 ) , suggesting a conserved phenomenon in flowering plants . In Drosophila and mouse embryonic stem cells , H1 has been shown to contribute to heterochromatin condensation ( Cao et al . , 2013; Lu et al . , 2009 ) . H1 is also more abundant in heterochromatin than euchromatin in Arabidopsis ( Ascenzi and Gantt , 1999; Rutowicz et al . , 2015 ) . However , it is unknown whether H1 participates in heterochromatin condensation in plant cells , and specifically whether the lack of H1 contributes to heterochromatin decondensation in the VC . Whether and how the depletion of H1 in the VC contributes to TE derepression is also unclear . A recent study pointed to an intriguing link between H1 and DME . In the central cell , the histone chaperone FACT ( facilitates chromatin transactions ) is required for DME-directed DNA demethylation in heterochromatic TEs , and this requirement is dependent on H1 ( Frost et al . , 2018 ) . However , DME activity in the VC is independent of FACT ( Frost et al . , 2018 ) . One attractive hypothesis is that the lack of H1 in the VC causes heterochromatin decondensation and thereby contributes to the independence of DME from FACT . H1 depletion may therefore participate in VC TE activation by promoting DME-directed demethylation . Additionally , H1 depletion may activate TE transcription independently of DNA methylation , as shown in Drosophila where DNA methylation is absent ( Iwasaki et al . , 2016; Lu et al . , 2013; Vujatovic et al . , 2012; Zemach et al . , 2010; Zhang et al . , 2015 ) . In this study , we identify heterochromatic TEs that are epigenetically activated in Arabidopsis VCs . We demonstrate that these TEs are typically subject to DME-directed demethylation at the transcriptional start site ( TSS ) , which is at least partially permitted by the depletion of H1 . However , we find that loss of H1 activates some TEs without altering DNA methylation . We also show that developmental depletion of H1 decondenses heterochromatin in late microspores and is important for pollen fertility . Our results demonstrate that H1 condenses heterochromatin in plants and maintains genome stability by silencing TEs via methylation-dependent and -independent mechanisms . To measure the extent of TE activation in the VC , we performed RNA-seq using mature pollen grains , followed by the annotation of gene and TE transcripts using Mikado and the TAIR10 annotation ( Venturini et al . , 2018 ) . We identified 114 TEs that are transcribed at significantly higher levels in pollen than rosette leaves ( fold change >2; p<0 . 05 , likelihood ratio test ) , and hence likely to be specifically activated in the VC ( Figure 1—source data 1 ) ( Slotkin et al . , 2009 ) . The VC-activated TEs are primarily located in pericentromeric regions and exhibit features of heterochromatic TEs , such as being long and GC rich ( Frost et al . , 2018 ) ( Figure 1A , B , Figure 1—figure supplement 1A ) . As is typical of heterochromatic TEs ( Zemach et al . , 2013 ) , VC-activated TEs are significantly enriched in dimethylation of histone H3 on lysine 9 ( H3K9me2 ) in somatic tissues , and are significantly depleted of euchromatin-associated modifications ( Figure 1B , Figure 1—figure supplement 1B ) . VC-activated TEs encompass diverse TE families , among which MuDR DNA transposons and Gypsy LTR-retrotransposons are significantly overrepresented ( p<10−9 and 0 . 01 , respectively , Fisher’s exact test; Figure 1C ) . To assess whether TE activation in the VC is caused by DME-mediated DNA demethylation , we examined DNA methylation in VC and sperm at the 114 activated TEs . We found that these TEs have substantially lower CG methylation in the VC than in sperm at and near the TSS ( Figure 1D , E ) , indicative of DME activity . Because TEs tend to be flanked by repeats ( Joly-Lopez and Bureau , 2018 ) , the transcriptional termination site ( TTS ) regions of activated TEs also tend to be hypomethylated in the VC ( Figure 1D , E , Figure 1—figure supplement 1C ) . Examination of DNA methylation in VCs from dme/+ heterozygous plants ( dme homozygous mutants are embryonic lethal ) , which produce a 50:50 ratio of dme mutant and WT pollen , revealed an intermediate level of methylation at TSS and TTS of VC-activated TEs ( Figure 1D , E ) . CHG and CHH methylation is also substantially increased at the TSS ( and TTS ) of VC-activated TEs in dme/+ VC ( Figure 1—figure supplement 1C ) , consistent with the knowledge that DME demethylates all sequence contexts ( Gehring et al . , 2006; Ibarra et al . , 2012 ) . Consistent with the above results , 71 of the 114 ( 62% ) VC-activated TEs overlap VC DME targets at their TSSs ( Figure 1F , Figure 1—source datas 1 and 2 ) . 92 out of the 114 TEs ( 81% ) have VC DME targets within 500 bp of the TSS ( Figure 1F , Figure 1—figure supplement 1D , Figure 1—source data 1 ) . As DNA methylation at/near the TSS has been well-demonstrated to suppress the transcription of genes and TEs in plants and animals ( Barau et al . , 2016; Eichten et al . , 2012; Hollister and Gaut , 2009; Manakov et al . , 2015; Meng et al . , 2016 ) , our results indicate that DME-directed demethylation is a major mechanism of TE activation in the VC . Consistently , RNA-seq of pollen from dme/+ heterozygous plants showed significantly reduced expression at the 114 VC-activated TEs ( Figure 1G , Figure 1—source data 1 ) . As dme/+ heterozygous plants produce half dme mutant and half WT pollen , we expect the transcription of VC-activated TEs to be reduced to roughly half in dme/+ pollen . The 114 VC-activated TEs are transcribed at levels close to this expectation ( Figure 1G ) , with the median ratio of their transcription in dme/+ versus WT pollen being 0 . 43 ( Figure 1—source data 1 ) . We next tested our hypothesis that the lack of histone H1 in the VC ( Hsieh et al . , 2016 ) allows heterochromatin to be accessible by DME . We first examined the developmental timing of H1 depletion during microspore and pollen development using GFP translational fusion lines ( Hsieh et al . , 2016; She et al . , 2013 ) . There are three H1 homologs in Arabidopsis , with H1 . 1 and H1 . 2 encoding the canonical H1 proteins , and H1 . 3 expressed at a much lower level and induced by stress ( Rutowicz et al . , 2015 ) . H1 . 1- and H1 . 2- GFP reporters exhibit the same expression pattern: present in early microspore nucleus but absent in the late microspore stage , and remaining absent in the VC nucleus while present in the generative cell and subsequent sperm nuclei ( Figure 2A ) . H1 . 3 is not detectable in either microspore or pollen ( Figure 2A ) . These results are consistent with our previous observations , confirming that H1 is absent in the VC ( Hsieh et al . , 2016 ) , and demonstrating that H1 depletion begins at the late microspore stage . To understand how H1 affects DME activity , we ectopically expressed H1 in the VC . To ensure H1 incorporation into VC chromatin , we used the pLAT52 promoter , which is expressed from the late microspore stage immediately prior to Pollen Mitosis 1 , and is progressively upregulated in VC during later stages of pollen development ( Eady et al . , 1994; Grant-Downton et al . , 2013 ) . Using pLAT52 to drive the expression of H1 . 1 tagged with mRFP ( simplified as pVC::H1 ) , we observed continuous H1-mRFP signal in the VC at the bicellular and tricellular pollen stages , while the signal was undetectable in the generative cell and sperm ( Figure 2B ) . H1-mRFP signal was also undetectable in late microspores ( Figure 2B ) , probably due to the low activity of pLAT52 at this stage ( Eady et al . , 1994 ) . Notably , we found H1 expression in VC leads to shortened siliques , a substantial proportion of malformed pollen , and reduced pollen germination rate ( Figure 2—figure supplement 1A–D ) , suggesting the depletion of H1 in the VC is important for pollen fertility . To evaluate the effect of VC-expressed H1 on DNA methylation , we obtained genome-wide methylation profiles for VC nuclei from a strong pVC::H1 line ( #2; Figure 2B ) and WT via fluorescence-activated cell sorting ( FACS ) followed by bisulfite sequencing ( Supplementary file 1 ) . CG methylation in the VC of pVC::H1 plants is largely similar to that of WT , except for a slight increase in TE methylation ( Figure 2C , Figure 2—figure supplement 2A ) . Consistently , the frequency distribution of CG methylation differences between VCs of pVC::H1 and WT at loci that are not DME targets peaks near zero , showing almost no global difference ( Figure 2D ) . However , a substantial proportion of loci that are targeted by DME show hypermethylation in pVC::H1 VC ( Figure 2D , Figure 2—figure supplement 2B ) . DME targets also show preferential hypermethylation in CHG and CHH contexts in the VC of pVC::H1 ( Figure 2—figure supplement 2C–D ) . These results indicate that H1 expression in the VC specifically impedes DME activity . Across the genome , we found 2964 differentially methylated regions ( DMRs ) that are significantly CG hypermethylated in the VC of pVC::H1 plants ( referred to as H1 hyperDMRs hereafter; ranging from 101 to 2155 nt in length , 280 nt on average; Figure 2—source data 1 ) . Most of the H1 hyperDMRs ( 1618 , 55% ) overlap DME targets in the VC ( Figure 2—source data 1 ) , and H1 hyperDMRs exhibit strong hypomethylation in WT VCs , with 81 . 4% ( 2412 sites ) having significantly more CG methylation in sperm than VC ( p<0 . 001 , Fisher’s exact test ) , indicating that most H1 hyperDMRs are DME targets ( Figure 2E–H ) . Our results demonstrate that H1 hyperDMRs are primarily caused by the inhibition of DME . However , only 3066 out of 11896 ( 26% ) VC DME targets have significantly more CG methylation in the VC of pVC::H1 than WT ( p<0 . 001 , Fisher’s exact test; Figure 1—source data 2 ) , indicating that VC-expressed H1 impedes DME at a minority of its genomic targets . These H1-impeded DME targets are heterochromatic and significantly enriched in H3K9me2 compared with H1-independent DME targets ( Figure 2I ) . To further examine the link with heterochromatin , we aligned all VC DME target loci at the most hypomethylated cytosine , and separated them into five groups by H3K9me2 levels ( Figure 2J ) . pVC::H1-induced hypermethylation peaks where DME-mediated hypomethylation peaks , but is apparent only in the most heterochromatic group ( highest H3K9me2 ) of DME target loci ( Figure 2J ) . Taken together , our results demonstrate that developmental removal of H1 from the VC allows DME to access heterochromatin . Given the importance of H1 removal for DME-directed DNA demethylation and the well-demonstrated role of DME demethylation in regulating gene expression ( Choi et al . , 2002; Ibarra et al . , 2012; Schoft et al . , 2011 ) , we investigated the contribution of H1 to gene regulation in pollen . RNA-seq was performed using pollen from the pVC::H1 line ( #2 ) , which showed strong H1 expression in VC ( Figures 2B and 3A ) . Only a small fraction of pollen-expressed genes ( 3%; 89 out of 2845 ) is differentially expressed ( fold change >2; p<0 . 05 , likelihood ratio test ) between pVC::H1 and WT ( Figure 3B , Figure 3—source data 1 ) . Among these 89 genes , 58 ( 65% ) are suppressed by H1 expression in the VC , and 31 ( 35% ) are activated ( Figure 3B , Figure 3—source data 1 ) . 85 out of these 89 genes ( 96% ) do not overlap DME targets within 500 bp of the TSS , hence the effect of H1 on their expression is probably not mediated by DME . Among the four genes that overlap DME targets within 500 bp of the TSS , two genes gain a small amount of methylation at the overlapping DME targets and are suppressed in pVC::H1 pollen ( Figure 3—figure supplement 1 , Figure 3—source data 1 ) , hence are possibly suppressed by H1 via the inhibition of DME . Because DME regulates the expression of imprinted genes in the endosperm ( Choi et al . , 2002; Hsieh et al . , 2011; Ibarra et al . , 2012 ) , we specifically examined the imprinted genes to further investigate the role of VC H1 removal in regulating pollen gene transcription . Out of the 640 known imprinted genes ( Gehring et al . , 2011; Hsieh et al . , 2011; Pignatta et al . , 2014; Wolff et al . , 2011 ) , 85 overlap endosperm DME targets within 500 bp of their TSS ( Figure 3—source data 2 ) . Because DME targets different loci in the VC and CC ( Ibarra et al . , 2012 ) , we subsequently examined if these 85 putative DME-regulated imprinted genes are also subject to DME demethylation in the VC . 63 of these genes also overlap VC DME targets within 500 bp of their TSS , none of which is differentially expressed in pVC::H1 pollen compared with WT pollen ( Figure 3—source data 2 ) , showing that transcription of these genes in pollen is unlikely regulated by H1 . This is not surprising , because only 8 out of these 63 genes are expressed in WT pollen ( Figure 3—source data 2 ) , and H1 preferentially regulates heterochromatic DME targets ( Figures 2I , J and 3G ) , whereas DME targets involved in gene regulation are typically euchromatic sites next to genes ( Figure 3C ) . Consistently , DME targets likely regulating imprinted genes are associated with significantly less H1 in somatic tissues than the VC DME targets that are dependent on H1 ( Figure 3C ) . In contrast to the small effect of H1 on gene transcription , a substantial proportion of VC-activated TEs ( 41%; 47 out of 114 ) show significant differential expression ( fold change >2; p<0 . 05 , likelihood ratio test ) due to H1 expression in VC ( Figure 3D ) . Among these differentially expressed TEs , the overwhelming majority ( 46; 98% ) are repressed ( Figure 3D , E , Figure 1—source data 1 ) . These data indicate that ectopic expression of H1 in the VC preferentially represses TEs . Quantitative RT-PCR validated our RNA-seq results and confirmed the strong suppression of TEs in pVC::H1 ( Figure 3F ) . Taking advantage of a pVC::H1 line #7 with weaker H1 expression in pollen ( Figure 3A ) , we found H1 represses TE expression in a dosage-dependent manner: TEs are suppressed to a lesser extent in line #7 compared to the strong line #2 ( Figure 3F ) . H1-repressed TEs in the VC are predominantly localized to pericentromeric regions and are overrepresented for LTR retrotransposons , including Gypsy and Copia elements ( Figure 3G–I ) . Compared to other VC-activated TEs , the H1-repressed TEs are significantly longer and enriched for H3K9me2 and H1 in somatic tissues ( Figure 3H ) , consistent with the observation that H1 precludes DME access to heterochromatin . In support of the hypothesis that H1 represses VC TE expression by blocking DME , 20 of 46 H1-repressed TEs show significant increase of DNA methylation in at least one sequence context within 500 bp of the TSS in pVC::H1 ( p<0 . 001 , Fisher’s exact test; Figure 4A , B ) . Four more TEs overlap a DME target , which is hypermethylated in pVC::H1 , within 500 bp of the TSS , and hence may also be suppressed by DME inhibition . However , 21 out of the rest of 22 TEs do not overlap any H1 hyperDMRs within 500 bp of the TSS ( Figure 4A , marked by asterisks in the lower panel ) , indicating that their suppression by H1 is not mediated by DNA methylation . Of these 21 TEs , 15 TEs overlap DME targets within 500 bp of TSS . DME maintains access to these TEs in the presence of H1 , suggesting their VC demethylation does not rely on the depletion of H1 and their repression in pVC::H1 is DME-independent as exemplified by AT3TE60310 ( Figure 4C ) . Our results demonstrate that H1 overexpression in the VC represses heterochromatic TEs via both DNA methylation-dependent and independent mechanisms . H1 depletion and TE activation in the VC are accompanied by loss of cytologically detectable heterochromatin ( Baroux et al . , 2011; Ingouff et al . , 2010; Schoft et al . , 2009 ) . We therefore tested whether H1 contributes to heterochromatin condensation in plant cells . Immunostaining of leaf nuclei showed that H1 co-localizes with H3K9me2 in highly-compacted heterochromatic foci , known as chromocenters ( Figure 5A ) . Furthermore , we found that chromocenters become dispersed in the nuclei of h1 mutant rosette leaves ( Figure 5B ) . These observations demonstrate that H1 is required for heterochromatin condensation in plants . We then examined whether ectopic H1 expression can condense the heterochromatin in VC nuclei . Consistent with previous observations ( Baroux et al . , 2011; Ingouff et al . , 2010; Schoft et al . , 2009 ) , no condensed chromocenters were detected in WT VC ( Figure 5C ) . pVC::H1 VC also showed no obvious chromocenters ( n > 500; Figure 2B ) . This suggests either that H1 expression is not strong enough in pVC::H1 , or other factors are involved in heterochromatin decondensation in the VC . Heterochromatin decondensation during male gametogenesis seems to be gradual: chromocenters are observed at early microspore stage , but become dispersed in late microspore stage , when H1 is depleted ( Figures 2A and 5C ) . We observed strong and weak chromocenters , respectively , in 27% and 59% of late microspore nuclei , whereas no chromocenters were observed in the VC at either bicellular or tricellular pollen stage ( Figure 5C , D ) . The further decondensation of VC heterochromatin after H1 depletion during the late microspore stage suggests the involvement of other factors in the VC . To test whether H1 is sufficient to induce chromatin condensation in microspores , we used the late-microspore-specific MSP1 promoter ( Honys et al . , 2006 ) to drive H1 expression ( pMSP1::H1 . 1-mRFP , short as pMSP1::H1 ) . In pMSP1::H1 , we observed strong chromocenters in the majority ( 68% ) of late microspores ( Figure 5D ) . H1 expression in pMSP1::H1 is specific to late microspores , and co-localizes with induced chromocenters ( Figure 5E ) . These results show that H1 is sufficient to promote heterochromatic foci in late microspores , thus demonstrating the causal relationship between H1 depletion and the decondensation of heterochromatin . Epigenetic reactivation of TEs in the VC of flowering plants is an intriguing phenomenon , which is important not only for understanding sexual reproduction , but also for elucidating epigenetic silencing mechanisms . Here we show that Arabidopsis VC-activated TEs are heterochromatic , and mostly subject to DME-directed demethylation at their TSS ( Figure 1F ) . Given the well-demonstrated role of DNA methylation at the TSS for transcriptional suppression ( Barau et al . , 2016; Eichten et al . , 2012; Hollister and Gaut , 2009; Manakov et al . , 2015; Meng et al . , 2016 ) , our data demonstrate that DME-mediated demethylation in the VC is the primary cause of TE activation . As DNA demethylation of TEs during reproduction also occurs in rice and maize ( Park et al . , 2016; Rodrigues et al . , 2013; Zhang et al . , 2014 ) , species that diverged from Arabidopsis more than 150 million years ago ( Chaw et al . , 2004 ) , our results suggest that TE activation in the VC is prevalent among flowering plants . DME demethylates about ten thousand loci in the VC and central cell , respectively , however , only half of these loci overlap ( Ibarra et al . , 2012 ) . It was unclear why DME targets differ in these cell types , but differences in chromatin configuration have been postulated to contribute ( Feng et al . , 2013 ) . Our finding that the access of DME to heterochromatic TEs in the VC is permitted by the lack of H1 supports this idea . H1 is presumably present in the central cell ( Frost et al . , 2018 ) but is absent in the VC ( Hsieh et al . , 2016 ) , thus rendering heterochromatic TEs more accessible in the VC . Differential distribution of other factors in the VC and central cell , such as histone variant H3 . 1 ( Borg and Berger , 2015; Ingouff et al . , 2010 ) , may also affect DME targeting . Consistently , FACT is required for DME activity in the central cell at many loci even in the absence of H1 , whereas DME is entirely independent of FACT in the VC ( Frost et al . , 2018 ) , suggesting the presence of impeding factor ( s ) other than H1 in the central cell . With distinct chromatin architectures , the vegetative and central cells are excellent systems for understanding how chromatin regulates DNA demethylation . Our finding that histone H1 affects DME activity adds to the emerging picture of H1 as an important and complex regulator of eukaryotic DNA methylation . H1 depletion causes local hypomethylation in mouse cells ( Fan et al . , 2005 ) and extensive hypermethylation in the fungi Ascobolus immersus ( Barra et al . , 2000 ) and Neurospora crassa ( Seymour et al . , 2016 ) . In Arabidopsis , loss of H1 causes global heterochromatic hypermethylation in all sequence contexts by allowing greater access of DNA methyltransferases ( Lyons and Zilberman , 2017; Zemach et al . , 2013 ) . Our results suggest that H1 may also influence DME-homologous demethylases that control methylation in somatic tissues ( He et al . , 2011 ) . By regulating both methylation and demethylation , H1 may serve as an integrator of methylation pathways that tunes methylation up or down depending on the locus . Our data also indicate that the regulatory functions of H1 extend beyond DNA methylation in plants . Activated TEs in the VC can be categorized into four groups , based on the mechanism of their activation ( Figure 6 ) . TEs in Group I are the least heterochromatic and their activation is dependent on DME but not H1 ( Figures 3H and 6 ) . Group II comprises TEs in which H1 absence is required for DME demethylation and activation ( Figure 6 ) . For TEs in Group III , H1 depletion and DME demethylation are both required for activation , but DME activity is not affected by H1 ( Figure 6 ) . Group IV TEs are activated by H1 depletion and are not targeted by DME ( Figure 6 ) . Groups III and IV demonstrate that H1 can silence TEs independently of DNA methylation . Group III also demonstrates that DNA methylation and H1 cooperate to suppress TE expression in plants . Thus , H1 regulates TEs via DNA methylation-dependent and -independent mechanisms . During the ongoing arms race between TEs and their hosts , it may be difficult to determine whether TE expression represents temporary TE triumphs or is domesticated by the host to serve a function . TE activation in the VC – a cell that engulfs the male plant gametes – has been proposed as a defense strategy , which generates small RNAs that enhance TE silencing in sperm ( Calarco et al . , 2012; Ibarra et al . , 2012; Martínez et al . , 2016; Slotkin et al . , 2009 ) . However , TEs can also use companion cells as staging grounds for invasion of the gametes ( Wang et al . , 2018 ) . Our demonstration that programmed DME demethylation , which is facilitated by developmental heterochromatin decondensation , is the predominant cause of VC TE activation is consistent with a defensive , host-beneficial model . Nonetheless , the alternative TE-driven model is also plausible . DME demethylates about ten thousand loci in the VC , most of which are small and euchromatic TEs ( Ibarra et al . , 2012 ) . However , only a hundred heterochromatic TEs are activated , at least partially permitted by developmental H1 depletion . As small euchromatic TEs tend to be next to genes , DME demethylation regulates genes and is important for pollen fertility ( Choi et al . , 2002; Ibarra et al . , 2012; Schoft et al . , 2011 ) . Our data show that developmental H1 depletion is also important for pollen fertility . Therefore , at least some TEs may be hijacking an essential epigenetic reprogramming process . TE activation in the VC may facilitate both host defense and transposition , with the balance specific to each TE family and changing over evolutionary time . The effects of VC TE activation on TE proliferation in the progeny may warrant investigation , particularly in out-crossing species with aggressive TEs and in natural populations . A . thaliana plants were grown under 16 hr light/8 hr dark in a growth chamber ( 20°C , 80% humidity ) . All plants used are of the Col-0 ecotype . pH1 . 1::H1 . 1-eGFP , pH1 . 2::H1 . 2-eGFP , dme-7 , and the h1 ( h1 . 1 h1 . 2 double ) mutant lines were described previously ( Schoft et al . , 2011; She et al . , 2013; Zemach et al . , 2013 ) . pLAT52::H1 . 1-mRFP and pMSP1::H1 . 1-mRFP were constructed with MultiSite Gateway System into the destination vector pK7m34GW ( Invitrogen ) . The BP clones pDONR-P4-P1R-pLAT52 and pDONR-P2R-P3-mRFP were kindly provided by Prof . David Twell ( Leicester University , UK ) ( Eady et al . , 1994 ) . MSP1 promoter was cloned into pDONR-P4-P1R as described previously ( Honys et al . , 2006 ) . WT plants were transformed via floral dip ( Clough and Bent , 1998 ) , and T2 or T3 plants homozygous for the transgene were used in this study . Open flowers were collected for pollen isolation in Galbraith buffer ( 45 mM MgCl2 , 30 mM sodium citrate , 20 mM MOPS , 1% Triton-X-100 , pH 7 . 0 ) by vortexing at 2000 rpm for 3 min . The crude fraction was filtered through a 40 μm cell strainer to remove flower parts , and subsequently centrifuged at 2600 g for 5 min to obtain pollen grains . RNA was extracted from pollen grains with RNeasy Micro Kit ( Qiagen ) following manufacturer’s instructions . RNA-sequencing libraries were prepared using Ovation RNA-seq Systems 1–16 for Model Organisms ( Nugen Technologies ) , and sequenced on the Hiseq 2500 ( Illumina ) instrument at the UC Berkeley Vincent J . Coates Genomics Sequencing Laboratory or on the Nextseq 500 ( Illumina ) at the John Innes Centre . Quantitative RT-PCR ( qRT-PCR ) was performed as described previously ( Walker et al . , 2018 ) , and TUA2 was used as an internal control . Primers for qRT-PCR are listed in Supplementary file 2 . The experiment was performed as described previously with some modifications ( Rodriguez-Enriquez et al . , 2013 ) . Pollen from three newly opened flowers for individual plants were brushed on cellulose membrane sitting on germination medium ( 18% sucrose , 0 . 01% boric acid , 1 mM CaCl2 , 1 mM Ca ( NO3 ) 2 , 1 mM KCl , 0 . 25 mM spermidine , pH 8 . 0 with KOH adjusted , 0 . 5% agarose ) in small petri dishes . Petri dishes were placed in a box with a piece of wet tissue at the bottom to keep humidity . The boxes were incubated in Sanyo cabinet ( MLR-351H ) at 21°C for 4 hr . 300 pollen grains were counted from each cellulose membrane and eight replicates for each genotype . Pollen germination was counted by using ImageJ . TE transcript annotation was created using RNA-seq data from four biological replicates of pollen . Tophat2 , Hisat , and STAR were used to align RNA-seq reads to the TAIR10 genome , and transcripts were assembled using CLASS2 , StringTie , and Cufflinks , respectively . Assembled transcripts were selected by Mikado using default options except that the BLAST and Transdecoder steps were disabled ( Venturini et al . , 2018 ) . As a result , 21381 transcripts ( called superloci; GSE120519 ) were identified . To identify VC-activated TEs , we first refined the list of superloci by selecting those overlapping with TAIR10 TE annotation . Subsequently to eliminate TE-like genes from the refined list , superloci with CG methylation less than 0 . 7 in rosette leaves ( Stroud et al . , 2014; Stroud et al . , 2013 ) were excluded . This gave rise to an annotation of pollen TE transcripts , which was combined with TAIR10 gene annotation for Kallisto analysis ( Bray et al . , 2016 ) . RNA-seq data from WT and dme/+ pollen ( this study ) and rosette leaves ( Walker et al . , 2018 ) , each including three biological replicates , were processed using Kallisto and Sleuth ( Bray et al . , 2016; Pimentel et al . , 2017 ) . TEs that are transcribed at least five times more in WT pollen than leaves ( with p<0 . 05 , likelihood ratio test ) are considered as activated in the VC ( refer to Figure 1—source data 1 for the list of VC-activated TEs ) . A total of 2845 genes were found to be expressed in pollen with TPM ( transcripts per million ) more than five in the Kallisto output ( Figure 3—source data 1; data used in Figure 3B ) . To identify TEs and genes that are suppressed by H1 in the VC , we analyzed RNA-seq data from WT and pLAT52::H1 . 1-mRFP line #2 ( short as pVC::H1 unless specified otherwise ) pollen using Kallisto and Sleuth as described above . Significant differential expression was defined with a fold change at least two and a p-value less than 0 . 05 . H1-repressed TEs were listed in Figure 1—source data 1 . Vegetative and sperm nuclei were isolated via FACS as described previously ( Ibarra et al . , 2012 ) . Bisulfite-sequencing libraries were prepared as previously described ( Walker et al . , 2018 ) . Sequencing was performed on Hiseq 2500 ( Illumina ) at the UC Berkeley Vincent J . Coates Genomics Sequencing Laboratory , Hiseq 4000 ( Illumina ) at Novogene Ltd . and Harvard University , and Nextseq 500 ( Illumina ) at Cambridge University Biochemistry Department and the John Innes Centre . Sequenced reads ( 100 , 75 , or 50 nt single-end ) were mapped to the TAIR10 reference genome , and cytosine methylation analysis was performed as previously described ( Ibarra et al . , 2012 ) . As all CG hypomethylation in the VC in comparison to sperm is caused by DME ( Ibarra et al . , 2012 ) , we identified VC DME targets via detecting CG differentially methylated regions ( DMRs ) that are hypermethylated in sperm in comparison to the VC . DMRs were identified first by using MethPipe ( settings: p=0 . 05 and bin = 100 ) ( Song et al . , 2013 ) , and subsequently retained if the fractional CG methylation across the whole DMR was at least 0 . 2 higher in sperm than the VC . The refined DMRs were merged to generate larger DMRs if they occurred within 300 bp . Finally , merged DMRs were retained if they cover at least 100 bp , and the fractional CG methylation across the whole DMR was significantly ( Fisher’s exact test p<0 . 01 ) and substantially ( >0 . 2 ) higher in sperm than the VC . This resulted in the identification of 11896 VC DME targets ( Figure 1—source data 2 ) . H1 hyperDMRs were identified using the same criteria , except comparing CG methylation in VCs from pVC::H1 and WT . In total , 2964 H1 hyperDMRs were identified ( Figure 2—source data 1 ) . Box plots compare the enrichment of genomic or chromatin features among TEs ( Figures 1B and 3G , Figure 1—figure supplement 1B ) or VC DME targets ( Figure 2I ) as described in corresponding figure legends . ChIP-seq data for H3K9me2 ( Stroud et al . , 2014 ) , and ChIP-chip data for H1 ( Rutowicz et al . , 2015 ) , H3K27me3 ( Kim et al . , 2012 ) , and other histone modifications ( Roudier et al . , 2011 ) were used . All DNA methylation kernel density plots compare fractional methylation within 50 bp windows . We used windows with at least 20 informative sequenced cytosines and fractional methylation of at least 0 . 5 ( Figure 2D , Figure 2—figure supplement 2 ) or 0 . 7 ( Figure 2E ) for CG context , and 0 . 4 and 0 . 1 for CHG and CHH context , respectively , in at least one of the samples being compared . Ends analysis for TEs and genes was performed as described previously ( Ibarra et al . , 2012 ) . Similarly , ends analysis of TE transcripts was performed using the annotation of VC-activated TEs described above ( Figure 1—source data 1 ) . DNA methylation data from Ibarra et al . ( 2012 ) was used . In Figure 2J , DME sites were aligned at the most demethylated cytosine , and average CG methylation levels for each 10 bp interval at both sides were plotted . To identify individual hypomethylation sites created by DME , we first obtained the 50 bp windows with a CG methylation difference larger than 0 . 5 between sperm and VC ( sperm – VC >0 . 5 and Fisher’s exact test p<0 . 001 ) . Windows were then merged if they occurred within 200 bp . Merged windows were retained for further analysis if the fractional CG methylation across the whole site was 0 . 2 greater in sperm than VC ( sperm – VC >0 . 2 and Fisher’s exact test p<0 . 0001 ) . This resulted in 13610 DME sites , which were separated into five groups according to H3K9me2 level ( Stroud et al . , 2014 ) :<2 . 5 , 2 . 5–4 . 3 , 4 . 3–6 . 5 , 6 . 5–10 . 5 , and >10 . 5 ( Figure 2J ) . The most demethylated cytosine within each site was identified if it had the greatest differential methylation in sperm than VC among cytosines in the CG context ( sperm – VC >0 . 2 , and Fisher’s exact test p<0 . 001 ) and was sequenced at least 10 times . Differential methylation at a 1000 bp region centered upon the TSS of H1-repressed TEs was calculated between VCs of pVC::H1 and WT ( Figure 4A ) . TEs whose differential methylation is significant ( Fisher’s exact test p<0 . 001 ) and larger than 0 . 2 ( in CG context ) , 0 . 1 ( in CHG context ) , or 0 . 05 ( in CHH context ) are illustrated in the upper panel in Figure 4A . Microspores and pollen were isolated as described previously ( Borges et al . , 2012 ) , stained with Hoechst or DAPI , and examined under a Leica SP8 confocal microscope . Scanning electron microscopy was performed on a Zeiss Supra 55 VP FEG . Immunofluorescence was performed as described previously with small modifications ( Yelagandula et al . , 2014 ) . Rosette leaves from 3-week-old plants were fixed in TRIS buffer with 4% paraformaldehyde ( 10 mM Tris-HCl pH 7 . 5 , 10 mM EDTA , 100 mM NaCl ) for 20 min . After being washed with TRIS buffer twice , the fixed leaves were chopped with razor blades in 1 mL of lysis buffer ( 15 mM Tris pH 7 . 5 , 2 mM EDTA , 0 . 5 mM spermine , 80 mM KCl , 20 mM NaCl , 0 . 1% Triton X-100 ) and filtered through a 35 μm cell strainer . Nuclei were pelleted via centrifugation at 500 g for 3 min and resuspended in 100 μL of lysis buffer . Next , 10 μL was spotted onto coverslips , air-dried , and post-fixed in PBS with 4% paraformaldehyde for 30 min . After being washed with PBS twice , coverslips were incubated in blocking buffer ( PBS with 1% BSA ) at 37°C for 30 min and then incubated in blocking buffer with primary antibodies at 4°C overnight ( Mouse anti-H3K9me2 Abcam ab1220 , 1:100; Rabbit anti-GFP Abcam ab290 , 1:100 ) . After being washed with PBS three times , coverslips were incubated in PBS with secondary antibodies at 37°C for 30 min , and then washed with PBS three times again before being counterstained and mounted in Vectashield mounting media with DAPI ( Vector H-1200 ) .	In most organisms , the genetic information contains DNA segments called transposable elements which are able to move around in the genome . When transposable elements insert themselves in a new location , this can lead to negative outcomes such as cell death or cancer . Animals , plants and other organisms have evolved sophisticated mechanisms to ‘lock in’ transposable elements and prevent them from jumping from place to place in the genome . For instance , adding small molecules called methyl groups onto these sequences tightly packages the DNA , which wraps itself around proteins known as histones . The resulting structure is known as heterochromatin , and it limits the movement of the transposable elements . In certain situations , cells may ‘reactivate’ some of their transposable elements: this is for example the case in plant sperm companion cells , which protect the sperm and deliver them to the egg cell . However , it was not clear how many transposable elements are reactivated in these cells , or how this process works . Here , He et al . investigate this process in the sperm companion cells of a small weed known as Arabidopsis thaliana . The experiments showed that around 100 transposable elements were reactivated , most of them when an enzyme called DEMETER removed the methyl groups found in heterochromatin . However , this enzyme alone was not enough . It could only access the methyl groups if the tightly packed structure of the heterochromatin had relaxed following the removal of a histone protein called H1 . Taken together , these results indicate that histone H1 and DEMETER cooperate to regulate the activity of transposable elements in the genome . In addition , H1 is known to prevent the addition of methyl groups onto DNA; that it also impedes their removal suggests that this protein plays a complex role in controlling the way genetic information is interpreted . The next step would now be to investigate the impact of the reactivation of transposable elements on the next generation of plants and during evolution .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "plant", "biology" ]	2019	Natural depletion of histone H1 in sex cells causes DNA demethylation, heterochromatin decondensation and transposon activation
Parvalbumin-positive ( PV+ ) γ-aminobutyric acid ( GABA ) interneurons are critically involved in producing rapid network oscillations and cortical microcircuit computations , but the significance of PV+ axon myelination to the temporal features of inhibition remains elusive . Here , using toxic and genetic mouse models of demyelination and dysmyelination , respectively , we find that loss of compact myelin reduces PV+ interneuron presynaptic terminals and increases failures , and the weak phasic inhibition of pyramidal neurons abolishes optogenetically driven gamma oscillations in vivo . Strikingly , during behaviors of quiet wakefulness selectively theta rhythms are amplified and accompanied by highly synchronized interictal epileptic discharges . In support of a causal role of impaired PV-mediated inhibition , optogenetic activation of myelin-deficient PV+ interneurons attenuated the power of slow theta rhythms and limited interictal spike occurrence . Thus , myelination of PV axons is required to consolidate fast inhibition of pyramidal neurons and enable behavioral state-dependent modulation of local circuit synchronization . GABAergic interneurons play fundamental roles in controlling rhythmic activity patterns and the computational features of cortical circuits . Nearly half of the interneuron population in the neocortex is parvalbumin-positive ( PV+ ) and comprised mostly of the basket cell ( BC ) type ( Hu et al . , 2014; Tremblay et al . , 2016 ) . PV+ BCs are strongly and reciprocally connected with pyramidal neurons ( PNs ) and other interneurons , producing temporally precise and fast inhibition ( Bartos et al . , 2002; Gonchar and Burkhalter , 1997; Tamás et al . , 1997 ) . The computational operations of PV+ BCs , increasing gain control , sharpness of orientation selectively , and feature selection in the sensory cortex ( Atallah et al . , 2012; Cardin et al . , 2009; Lee et al . , 2012; Yang et al . , 2017; Zucca et al . , 2017 ) , are mediated by a range of unique molecular and cellular specializations . Their extensive axon collaterals targeting hundreds of PNs are anatomically arranged around the soma and dendrites , and electrotonically close to the axonal output site . In addition , the unique calcium ( Ca2+ ) sensor in PV+ BCs terminals , synaptotagmin 2 ( Syt2 ) , is tightly coupled to Ca2+ channels mediating fast and synchronized release kinetics ( Chen et al . , 2017; Sommeijer and Levelt , 2012 ) , powerfully shunting excitatory inputs and increasing the temporal precision of spike output ( Hu et al . , 2014; Somogyi et al . , 1983; Tamás et al . , 1997; Thomson et al . , 1996 ) . Recent findings have shown that the proximal axons of PV+ interneurons are covered by myelin sheaths ( Micheva et al . , 2016; Peters and Proskauer , 1980; Somogyi et al . , 1983; Stedehouder et al . , 2017; Tamás et al . , 1997; Yang et al . , 2020 ) . How interneuron myelination defines cortical inhibition remains , however , still poorly understood . Myelination of axons provides critical support for long-range signaling by reducing the local capacitance that results in rapid saltatory conduction and by maintaining the axonal metabolic integrity ( Cohen et al . , 2020; Nave and Werner , 2014 ) . For PV+ BCs , however , the average path length between the axon initial segment ( AIS ) and release sites involved in local circuit inhibition is typically less than ~200 µm ( Micheva et al . , 2021; Schmidt et al . , 2017; Tamás et al . , 1997 ) , and theoretical and experimental studies indicate the acceleration by myelin may play only a limited role ( Micheva et al . , 2021; Micheva et al . , 2016 ) . Another notable long-standing hypothesis is that myelination of PV+ axons may be critical for the security and synchronous invasion of presynaptic terminals ( Somogyi et al . , 1983 ) . In support of a role in reliability , in Purkinje cell axons of the long Evans shaker ( les ) rat , which carries a deletion of Mbp , spike propagation shows failures and presynaptic terminals are disrupted ( Barron et al . , 2018 ) . Interestingly , in a genetic model in which oligodendrocyte precursor cells lack the γ2 GABAA receptor subunit , fast-spiking interneuron axons in the neocortex are aberrantly myelinated and feedforward inhibition is impaired ( Benamer et al . , 2020 ) . At the network level , PV+ BC-mediated feedback and feedforward inhibition is critical to produce local synchronization between PNs and interneuron at the gamma ( γ ) frequency ( 30–80 Hz ) , which is a key rhythm binding information from cell assemblies , allowing synaptic plasticity and higher cognitive processing of sensory information ( Buzsáki , 2006; Cardin et al . , 2009; Hu et al . , 2014; Sohal et al . , 2009; Veit et al . , 2017 ) . Here , we determined whether PV+ BC-driven neocortical rhythms require myelination by using de- and dysmyelination models , studying the cellular properties of genetically labeled PV+ BCs and examining the functional role of myelin by longitudinally examining the frequency spectrum of cortical oscillations . We investigated in vivo cortical rhythms by recording local field potential ( LFP ) in layer 5 ( L5 ) together with surface electrocorticogram ( ECoG ) signals from both primary somatosensory ( S1 ) and visual ( V1 ) areas . Freely moving mice ( C57BL/6 , 7–9 weeks at the start of the experiment ) were recorded in their home cage every second week ( 18–24 hr/week ) across an 8-week cuprizone treatment , inducing toxic loss of oligodendrocytes in white- and gray matter areas ( Clarner et al . , 2012; Hamada and Kole , 2015; Kipp et al . , 2009 ) . Remarkably , after 6 weeks of cuprizone feeding we detected high-voltage spike discharges ( approximately five times the baseline voltage and ~50–300 ms in duration , Figure 1a–c ) . These brief spike episodes on the ECoG and LFP ( Figure 1c ) occurred bilaterally and near synchronously in S1 and V1 , resembling the interictal epileptiform discharges ( also termed interictal spikes ) that are a hallmark of epilepsy ( Cohen et al . , 2002; Dubey et al . , 2018; Hoffmann et al . , 2008; Tóth et al . , 2018 ) . Automated detection of interictal spikes in the raw ECoG–LFP signal was performed with a machine-based learning classifier ( see Figure 1—figure supplement 1 and Materials and methods ) , revealing a progressively increasing number of interictal spikes , from ~5/hr at 4 weeks up to ~70/hr at 8 weeks of cuprizone treatment ( Figure 1d ) . Interestingly , interictal spikes were highly dependent on vigilance state and present exclusively during quiet wakefulness ( 30 out of 30 randomly selected LFP segments from awake or quiet wakefulness , chi-square test p<0 . 0001 , n = 6 cuprizone mice ) , with no other discernible association to specific behaviors ( Figure 1—figure supplement 1b , Figure 1—video 1 ) . Whether the pathological cortical oscillations were specific to certain frequency bands , including gamma ( γ , 30–80 Hz ) , was examined by plotting the power spectrum density of the LFP in S1 during periods of quiet wakefulness or active movement ( Figure 1e ) . During quiet wakefulness , LFP power in cuprizone-treated mice was selectively amplified in the theta frequency band ( θ , 4–12 Hz , Šidák’s multiple comparisons test , p=0 . 0013 , Figure 1e and f , Figure 1—figure supplement 1c ) . In contrast , during active states when mice were moving and exploring no differences were observed in the power spectrum , in none of the frequency bands ( Šidák’s multiple comparisons test , p>0 . 166 , Figure 1e and f , Source data 1 ) . Finally , to more firmly establish whether interictal epileptiform discharges result from the lack of myelin , we analyzed ECoG signals in the dysmyelinated shiverer mice ( MbpShi ) lacking compact myelin due to a truncating mutation in Mbp ( Readhead et al . , 1987 ) . Shiverer mice suffer progressively increasing number of epileptic seizures beginning at approximately 8 weeks of age ( Chernoff , 1981; Readhead et al . , 1987 ) . ECoG recordings at 8 weeks showed that in addition to ictal discharges interictal spikes were detected with a rate of ~1/min , comparable to cuprizone-treated mice ( Figure 1g and h , Figure 1—figure supplement 2 ) . Although the waveform of interictal spikes in shiverer was substantially longer in duration ( ~100–500 ms ) , analysis of the power across the four frequency bands around interictal spikes revealed no difference in comparison to the cuprizone-treated mice ( two-way analysis of variance [ANOVA] p=0 . 7875 , n = 6 mice for both groups , Figure 1—figure supplement 2 ) . Increased power of sensory-driven slow oscillations and epileptiform activity in the neocortex of normally myelinated brains is also observed when PV+ interneurons are optogenetically silenced ( Brill et al . , 2016; Veit et al . , 2017; Yang et al . , 2017 ) . To investigate how myelin loss affects the PV+ interneuron morphological and functional properties , we crossed the PvalbCre mouse line , having Cre recombinase targeted to Pvalb-expressing cells , with a tdTomato fluorescence ( Ai14 ) Cre reporter line ( hereafter called PV-Cre; Ai14 mice ) . The cytoplasmic fluorescence allowed quantification of PV+ cell bodies and their processes in the primary somatosensory cortex ( Figure 2a and b , Figure 2—figure supplement 1a ) , and immunofluorescent labeling with myelin basic protein ( MBP ) revealed substantial myelination of large-diameter PV+ axons ( mean ± SEM , 80 . 15% ± 9 . 95% along 83 mm of PV+ axons analyzed , n = 3 slices from two mice , z-stack with a volume of 7 . 66 × 105 μm3 , Figure 2b , Figure 2—figure supplement 1a ) . Electron microscopy ( EM ) immunogold-labeled tdTomato showed that PV+ axons possessed multilamellar compact myelin sheaths ( on average , 6 . 33 ± 0 . 80 myelin lamella ) with 10 . 8 ± 0 . 76 nm distance between the major dense lines and a mean g-ratio ( axon diameter/fiber diameter ) of 0 . 74 ± 0 . 01 ( n = 6 sheaths , Figure 2c ) . PV-Cre; Ai14 mice fed with 0 . 2% cuprizone for 6 weeks showed strongly reduced MBP in S1 and PV+ axons were largely devoid of myelin ( Figure 2a and b , Figure 2—figure supplement 1b ) while the total number of PV+ cell bodies across cortical layers remained constant ( control density , 326 ± 14 cells mm–2 vs . cuprizone density , 290 ± 48 cells mm–2 , n = 6 sections from N = 6 animals/group , Mann–Whitney test p=0 . 1649 , Figure 2d ) . Further , single-cell analysis was performed on biocytin-filled PV-Cre+ interneurons that were re-sectioned and stained for MBP to identify the location of myelin and the axon morphology ( Figure 2e , Figure 2—figure supplement 1c and d ) . Myelin was present on multiple proximal axonal segments of all control BCs ( 4/4 fully reconstructed axons , on average 2 . 8% ± 1 . 2% myelination ) . In contrast , none of the BCs from cuprizone-treated mice showed myelinated segments ( 0/6 axons ) . Furthermore , the total number of axon segments ( ~80 per axon , Mann–Whitney test p=0 . 3032 , Figure 2f ) as well as the total path length were unaffected by cuprizone treatment ( on average ~4 . 5 mm in both groups , range 2 . 0–9 . 5 mm , Mann–Whitney test p=0 . 9871 , Figure 2g , Figure 2—figure supplement 1c–f ) . To examine whether myelin loss changes the intrinsic excitability of PV+ BCs , we made whole-cell recordings in slices from PV-Cre; Ai14 mice ( Figure 2h ) . Recording of steady-state firing properties by injecting increasing steps of currents injections revealed an increase in the rheobase current ( ~90 pA , Figure 2i and j ) and an ~50 Hz reduced firing frequency during low-amplitude current injections ( two-way ANOVA , treatment p=0 . 0441 , Šidák’s multiple comparisons post hoc test at 200 pA; p=0 . 0382 , 250 pA; p=0 . 0058 , 300 pA; p=0 . 0085 ) without a change in the maximum instantaneous firing rate ( two-way ANOVA , Šidák’s multiple comparisons post hoc test , p=0 . 92 , data not shown ) . However , neither the AP half-width ( control , 290 ± 10 µs , n = 34 cells from 12 mice vs . cuprizone , 295 ± 10 µs , n = 15 cells from 7 mice , two-tailed Mann–Whitney U-test p=0 . 7113 ) nor AP amplitude was affected by cuprizone treatment ( control 78 . 12 ± 1 . 66 mV , n = 34 cells from 12 mice vs . cuprizone , 80 . 13 ± 2 . 48 mV , n = 15 cells from 7 mice , p=0 . 4358 ) . In contrast , the resting membrane potential ( VRMP ) of PV+ interneurons was on average ~4 mV significantly more hyperpolarized ( Figure 2k ) without a change in the apparent input resistance ( control , 133 . 3 ± 8 . 55 MΩ , n = 42 cells from 21 mice vs . cuprizone 125 ± 8 . 96 MΩ , 27 cells out of 13 mice , p=0 . 5952 ) . In addition to the hyperpolarization in VRMP , demyelinated PV+ interneurons also had an ~3 mV more hyperpolarized AP voltage threshold ( control , –40 . 51 ± 0 . 97 mV , n = 34 cells from 12 mice ) and cuprizone –43 . 65 ± 1 . 29 ( n = 15 cells from 7 mice , Mann–Whitney test p=0 . 0269 ) . Taken together , the results indicate that cuprizone treatment completely demyelinates proximal branches of PV+ interneuron axons , and while not affecting axon morphology , causes a net decrease in the intrinsic PV+ interneuron excitability . Is myelin required for PV+ BC-mediated inhibition ? Single PV+ BCs typically make 5–15 synapses with a PN in a range of <200 µm , forming highly reliable , fast , and synchronized release sites ( Micheva et al . , 2021; Packer and Yuste , 2011; Tamás et al . , 1997; Thomson et al . , 1996 ) . Intercellular variations in both myelin distribution and aberrant myelin patterns have been associated with conduction velocity changes ( Benamer et al . , 2020; Micheva et al . , 2021 ) . To examine the role of myelin on inhibitory transmission , we made paired recordings of PV+ BCs and L5 PNs with and without myelination , in control or cuprizone-treated PV-Cre; Ai14 mice , respectively ( Figure 3 ) . We evoked APs in PV+ BCs while recording unitary inhibitory postsynaptic currents ( uIPSCs ) under conditions of physiological Ca2+/ Mg2+ ( 2 . 0/1 . 0 mM in n = 78 pairs , Figure 3a and b ) . Concordant with optogenetic mapping of PV+ inputs onto L5 PNs in mouse S1 ( Packer and Yuste , 2011 ) , in control slices the probability of a given PV+ cell being connected to a nearby PN was high ( ~0 . 48 , Figure 3c ) . In contrast , the connection probability was significantly lower in cuprizone-treated mice ( ~0 . 23 , p=0 . 0182 , Figure 3b and c ) . In 13 stable connected pairs , we examined unitary IPSC properties including failure rate and amplitude , as well as rise- and decay time , using automated fits of the uIPSCs ( n > 80 trials per connection , Figure 3d ) . Cuprizone treatment led to a significant increase in the number of failures ( from 0 . 05 to 0 . 26 , Figure 3e ) and an ~2 . 5-fold reduction in the average uIPSC peak amplitude ( Figure 3f ) . To obtain an estimate of propagation speed , we determined on successful trials the latency between the AP peak and uIPSCs at 10% peak amplitude ( Figure 3d ) . Interestingly , both the mean latency and the trial-to-trial latency variability remained unchanged ( average ~800 µs; Mann–Whitney test p>0 . 999; SD in cuprizone 319 ± 65 µs , n = 7 pairs , SD in control , 276 ± 38 µs , n = 5 pairs , p>0 . 60 , Figure 3g ) . To further examine the properties of GABA release in demyelinated PV-BCs , we recorded uIPSCs during a train of five APs at 100 Hz ( averaging >50 trials , Figure 3h ) . Consistent with the temporary facilitation in IPSCs of adult Purkinje cells ( Turecek et al . , 2016 ) , uIPSC recordings in control PV BCs showed that paired-pulse ratios were on the second spike facilitated by 20% ( uIPSC2/uIPSC1 1 . 20 ± 0 . 060 ) and gradually depressed on the subsequent spikes ( spikes 3–5 ) . In contrast , in cuprizone-treated mice uIPSCs were depressed during the second and subsequent pulses ( two-way repeated-measures ANOVA pulse × treatment effect p<0 . 021 , Šidák’s multiple comparisons tests for uIPSC2/uIPSC1 0 . 89 ± 0 . 041 , p=0 . 0339 , Figure 3h ) . The uIPSC failures and impairment of temporary facilitation may reflect failure of AP propagation along demyelinated PV axons , changes in the GABA release probability or a lower number of active release sites ( <5 , Tamás et al . , 1997; Thomson et al . , 1996 ) . To further examine the properties of inhibition at L5 PNs , we recorded miniature inhibitory postsynaptic currents ( mIPSCs ) . In line with the uIPSC findings , the results showed that mIPSCs were significantly reduced in peak amplitude ( from ~20 to ~7 pA , p=0 . 002 ) without a change in frequency ( Figure 3—figure supplement 1c ) . Furthermore , using PV immunofluorescence staining with biocytin-filled L5 PNs the number of PV+ puncta was 40% reduced both at the soma and the primary apical dendrite , correlating with the overall reduction in immunofluorescent signals in cuprizone treatment ( Figure 3—figure supplement 1g–i ) . Interestingly , in contrast to the loss of perisomatic PV+ BC puncta , putative PV+ chandelier cell inputs , identified by co-staining with the AIS marker ßIV-spectrin , were preserved ( ~8 puncta/AIS , Mann–Whitney test p=0 . 96 , Figure 3—figure supplement 2 ) . Furthermore , staining for Syt2 , a Ca2+ sensor protein selective for PV+ presynaptic terminals ( Sommeijer and Levelt , 2012; Xu et al . , 2007 ) confirmed an ~35% reduction ( Figure 3—figure supplement 3a and b ) . Together with the reduction in uIPSC peak amplitudes ( Figure 3f ) , these data suggest that cuprizone-induced demyelination causes a loss of presynaptic PV+ terminal sites . Interestingly , Syt2+ puncta analysis in the dysmyelinated shiverer mouse line also showed a reduced number of Syt2+ puncta at L5 PN somata and a reduced frequency of mIPSCs ( p=0 . 019 , Figure 3—figure supplement 3c–g ) , indicating that compact myelin is not only required for maintaining PV+ interneuron inputs but also for PV+ BC presynaptic terminal development . Cuprizone treatment did not affect PV+ axon length ( Figure 2g , Figure 2—figure supplement 1 ) , suggesting that the density of presynaptic terminals should be reduced . To test this idea , we performed Syt2 immunolabeling of individually biocytin-filled PV+ BCs ( Figure 3i ) . Consistent with the hypothesis , cuprizone treatment significantly reduced the density of Syt2+ puncta by twofold ( cuprizone , ~1 Syt2+ puncta per 10 µm vs . 1 Syt2+ puncta per 5 µm in control , Mann–Whitney test p<0 . 0001 , Figure 3j and k ) . Interestingly , recordings of miniature excitatory postsynaptic currents ( mEPSCs ) from PV+ interneurons of control and cuprizone-treated mice showed no changes in peak amplitude nor frequency ( Figure 3—figure supplement 4 ) , in keeping with the preservation of excitatory inputs onto L5 PNs following cuprizone-induced demyelination ( Hamada and Kole , 2015 ) and suggesting that myelin loss has a significant impact on inhibitory synapse stabilization and maintenance . Thus , myelin loss reduces the number of presynaptic sites , causing an increase of GABA release failures and a frequency-dependent depression , ultimately limiting the fast component of BC to PN inhibitory transmission . To understand how myelin deficits and loss of fast PV-mediated inhibition of PNs impacts network dynamics , we used AAV1-mediated delivery of Cre-dependent channelrhodopsin-2 ( ChR2 ) into L5 of PV-Cre; Ai14 mice ( Figure 4a ) . The ChR2 transduction rate was comparable between control and cuprizone mice ( ~70% , Figure 4b and c ) . In acute slices , we voltage-clamped L5 PNs and optogenetically evoked IPSC ( oIPSC ) with full-field blue light illumination ( Figure 4d ) . Consistent with S1 L5 PNs receiving converging input from >100 PV+ interneurons ( Packer and Yuste , 2011 ) , control oIPSCs rapidly facilitated to a peak amplitude of ~700 pA followed by rapid synaptic depression ( Figure 4f and g ) . In slices from cuprizone mice , however , the oIPSC peak amplitude was significantly reduced ( approximately twofold ) while neither the steady-state amplitude during vesicle replenishment nor the total charge transfer reached a significant difference ( control , –99 . 58 ± 28 . 5 pC vs . cuprizone , –54 . 3 ± 20 . 57 pC , p=0 . 236 , n = 9 control and n = 8 cuprizone neurons , Figure 4f and h ) . Impaired phasic PV+ interneuron-mediated inhibition predicts a disrupted γ-rhythm . Experimental and computational studies show that in most cortical areas γ-rhythms are strongly shaped by electrically and synaptically coupled PV+ interneurons , which , by temporally synchronizing firing rates , synaptic inhibitory time constants ( ≅ 9 ms ) , and the recurrent excitatory feedback from PNs , give rise to network resonance in the 30–80 Hz bandwidth ( Bartos et al . , 2002; Cardin et al . , 2009; Sohal et al . , 2009; Traub et al . , 1997; Wang and Buzsáki , 1996 ) . To test the cellular and circuit properties of the γ-rhythm , we examined the extent of evoked γ-modulation by leveraging optogenetic activation of PV+ interneurons with AAV1-hChR2-YFP and introducing a laser fiber into L5 and recording the LFP ( Figure 5a ) . Evoking brief pulses of blue light ( 1 ms at a low gamma frequency of 30 Hz ) showed that local circuit currents were modulated and highly phase-locked in slices from control mice ( bandpass filter 25 and 40 Hz , Figure 5b and c , Figure 5—figure supplement 1 ) . In striking contrast , no modulation or entrainment was observed in cuprizone-treated mice , neither when using high laser power ( up to 6 . 5 mW , Figure 5b–e , Figure 5—figure supplement 1 ) . Could the diminished PV+ BC activity cause the emergence of θ rhythm and interictal spikes during quiet behavioral states of wakefulness ? To test the direct contribution of PV+ BCs , we activated ChR2 for 1 s duration pulses in PV-Cre; Ai14 mice to generate tonic GABA release ( Figure 6a ) . In cuprizone-treated mice , we found that optically driving PV+ interneurons normalized the LFP power in the θ band to control levels , without affecting δ , β , and γ rhythms ( two-way ANOVA treatment × light p=0 . 0124 , Šidák’s multiple comparisons tests in cuprizone , light on vs . off; for δ , p=0 . 9975; θ , p=0 . 0076; β , p=0 . 9481; γ , p=0 . 9998 , Figure 6a–c , Source data 1 ) . Furthermore , activation of blue light significantly reduced the frequency of interictal epileptic discharge frequency ( p=0 . 0089 , Figure 6d and e , Figure 6—video 1 ) . The normalization of cortical rhythms by elevating sustained PV+-mediated activity suggests that GABAA receptors are insufficiently activated in the demyelinated cortex . Finally , to directly examine the role of GABAA receptors agonism in dampening global interictal spikes we administered a nonsedative dose of diazepam ( 2 mg/kg i . p . ) , an allosteric modulator of postsynaptic GABAA receptors , in cuprizone-treated mice ( 7-week treatment ) . The results showed that diazepam significantly suppressed the interictal epileptiform discharges in cuprizone mice , indicating a prominent role of GABA in the deficits of circuit excitability ( Figure 6f , Figure 6—figure supplement 1 ) . The requirement of myelination of PV+ BCs to generate gamma rhythms is surprising in view of its sparse distribution in patches of ~25 µm across <5% of the total axon length ( Micheva et al . , 2021; Micheva et al . , 2016; Stedehouder et al . , 2019; Figure 2 , Figure 2—figure supplement 1 ) . The sparseness of interneuron myelination previously raised the question whether myelin speeds conduction velocity in these axon types ( Micheva et al . , 2016; Stedehouder et al . , 2019; Stedehouder and Kushner , 2017 ) . In a genetic mouse model with aberrant myelin patterns along fast-spiking interneuron axons , the inferred conduction velocity was reduced ( Benamer et al . , 2020 ) . In contrast , we found that the average uIPSC latency ( ~800 µs ) in completely demyelinated axons was normal and well within the range of previous paired recordings between myelinated PV+ BC and PNs ( 700–900 µs , Miles , 1990; Rossignol et al . , 2013 ) . Assuming a typical axonal path length of ~200 µm between the AIS and presynaptic terminals connected with a PN , combined with an ~250 µs delay for transmitter release , the calculated conduction velocity would be 0 . 4 m/s , consistent with optically recorded velocities in these axons ( ~0 . 5 m/s , Casale et al . , 2015 ) . Our paired recordings , made near physiological temperature ( 34–36°C ) , may have had a limited resolution to detect temporal differences and are not excluding changes in the order of microseconds . To study submillicecond changes , it may be necessary to employ simultaneous somatic and axonal whole-cell recording ( Hu and Jonas , 2014 ) and/or high-resolution anatomical analysis of myelin along the axon path , which recently showed a small albeit positive correlation between percentage of myelination and conduction velocity ( Micheva et al . , 2021 ) . Furthermore , conduction velocity tuning by myelination of GABAergic axons may become more readily apparent for long-range projections . Another constraint of the present study is the lack of information on the nodes of Ranvier along demyelinated PV+ BC axons . Aberrant interneuron myelin development , causing myelination of branch points , impairs the formation of nodes of Ranvier ( Benamer et al . , 2020 ) . Reorganization of nodal voltage-gated ion channel clustering also occurs with the loss of myelin or oligodendroglial-secreting factors causing deficits in action potential propagation ( Freeman et al . , 2015; Lubetzki et al . , 2020 ) . How PV+ BC interneuron myelin loss changes the nodal ion channel distribution remains to be examined . Converging evidence from the two distinct models ( shiverer and cuprizone ) showed that interneuron myelination critically determines PV+ release site number , dynamics , and connection probability ( Figure 3 , Figure 3—figure supplement 3 ) , concordant with the observed synapse loss in Purkinje axons of the les rat ( Barron et al . , 2018 ) . The molecular mechanisms how compact myelination of proximal axonal segments establishes and maintains GABAergic terminals in the higher-order distal axon collaterals are not known but may relate to its role in supplying metabolites to the axon ( Fünfschilling et al . , 2012 ) . The PV+ interneuron myelin sheath contains high levels of noncompact 2′ , 3′-cyclic nucleotide 3′-phosphodiesterase ( CNP ) protein ( Micheva et al . , 2018; Micheva et al . , 2016 ) , which is part of the inner cytoplasmic inner mesaxon ( Edgar et al . , 2009 ) . In the absence of inner cytoplasmic loops of oligodendroglial myelin the interneuron axons may lack sufficient trophic support to maintain GABAergic presynaptic terminals . Another possibility is pruning of the presynaptic terminals by microglia ( Chen et al . , 2014; Favuzzi et al . , 2021; Ramaglia et al . , 2021 ) . Microglia become increasingly activated during sub-demyelinating stages within the first week of cuprizone treatment ( Caprariello et al . , 2018; Skripuletz et al . , 2013 ) and in aged Mbp+/– mice ( Poggi et al . , 2016 ) . In future studies , it needs to be examined whether attenuation of microglia activation could protect against PV+ synapse loss and interictal epileptiform discharges . The identification of a cellular mechanism for interictal spikes may shed light on the role of PV+ axon myelination in cognitive impairments in MS ( Benedict et al . , 2020 ) and possibly other neurological disorders . In preclinical models of epilepsy and epilepsy patients , interictal spikes have been closely linked to disruptions of the normal physiological oscillatory dynamics such as ripples required to encode and retrieve memories ( Cohen et al . , 2002; Henin et al . , 2021; Kleen et al . , 2013; Kleen et al . , 2010 ) . Interictal epileptic discharges are also observed in other neurodegenerative diseases , including Alzheimer ( Lam et al . , 2017 ) . Notably , reduced gray matter myelination and oligodendroglia disruption are reported in multiple epilepsy models and recently in Alzheimer ( Chen et al . , 2021; Drenthen et al . , 2020 ) . Therefore , the cellular and circuit functions controlled by PV+ interneurons may represent a common mechanism for memory impairments in neurological disease encompassing myelin pathology . In support of this idea , neuropathological studies in MS show a specific loss of PV+ interneuron synapses in both cortex and hippocampus ( Ramaglia et al . , 2021; Zoupi et al . , 2021 ) . In MS patients , increased connectivity and synchronization in delta and theta band rhythms during resting state or task-related behavior have been reported ( Schoonheim et al . , 2013; Tewarie et al . , 2014 ) and low GABA levels in sensorimotor and hippocampal areas are correlated with impairments of information processing speed and memory ( Cawley et al . , 2015; Gao et al . , 2018 ) . Taken together with the present work , promoting PV+ interneuron myelination , and thereby strengthening fast inhibition , may provide important new therapeutic avenues to improve cognition . We crossed PvalbCre mice ( B6;129P2-Pvalbtm1 ( cre ) Arbr/J , stock no: 008069 , Jackson Laboratory , RRID:IMSR_JAX:008069 ) with the Ai14 Cre reporter line B6;129S6-Gt ( ROSA ) 26Sortm14 ( CAG-tdTomato ) Hze/J ( stock no: 007908 , Jackson Laboratory , RRID:IMSR_JAX:007908 ) . For other experiments , we used C57BL/6 mice ( Janvier Labs , Saint-Berthevin Cedex , France , RRID:MGI:2670020 ) . Shiverer mice were obtained from Jackson ( C3Fe . SWV-Mbpshi/J , stock no: 001428 , RRID:IMSR_JAX:001428 ) and backcrossed with C57BL/6 mice for >10 generations . All mice were kept on a 12:12 hr light-dark cycle ( lights on at 07:00 , lights off at 19:00 ) with ad libitum food and water . For cuprizone treatment , either PV-Cre; Ai14 or C57BL/6 male or female mice , from 7 to 9 weeks of age , were fed ad libitum with normal chow food ( control group ) or were provided 0 . 2% ( w/w ) cuprizone ( Bis ( cyclohexanone ) oxaldihydrazone , C9012 , Merck ) added either to grinded powder food or to freshly prepared food pellets ( cuprizone group ) . Cuprizone-containing food was freshly prepared during every second or third day for the entire duration of the treatment ( 6–9 weeks ) . The average maximum weight loss during cuprizone feeding was ~11% ( n = 31 ) . All animal experiments were done in compliance with the European Communities Council Directive 2010/63/EU effective from 1 January 2013 . The experimental design and ethics were evaluated and approved by the national committee of animal experiments ( CCD , application number AVD 80100 2017 2426 ) . The animal experimental protocols were designed to minimize suffering and approved and monitored by the animal welfare body ( IvD , protocol numbers NIN17 . 21 . 04 , NIN18 . 21 . 02 , NIN18 . 21 . 05 , NIN19 . 21 . 04 , and NIN20 . 21 . 02 ) of the Royal Netherlands Academy of Arts and Science ( KNAW ) . Mice were briefly anesthetized with 3% isoflurane and decapitated or received a terminal dose of pentobarbital natrium ( 5 mg/kg ) and were transcardially perfused with ice-cold artificial CSF ( aCSF ) of the composition ( in mM ) : 125 NaCl , 3 KCl , 25 glucose , 25 NaHCO3 , 1 . 25 Na2H2PO4 , 1 CaCl2 , 6 MgCl2 , 1 kynurenic acid , saturated with 95% O2 and 5% CO2 , pH 7 . 4 . After decapitation , the brain was quickly removed from the skull and parasagittal sections ( 300 or 400 µm ) containing the S1 cut in ice-cold aCSF ( as above ) using a vibratome ( 1200S , Leica Microsystems ) . After a recovery period for 30 min at 35°C , brain slices were stored at room temperature . For patch-clamp recordings , slices were transferred to an upright microscope ( BX51WI , Olympus Nederland ) equipped with oblique illumination optics ( WI-OBCD; numerical aperture , 0 . 8 ) . The microscope bath was perfused with oxygenated ( 95% O2 , 5% CO2 ) aCSF consisting of the following ( in mM ) : 125 NaCl , 3 KCl , 25 D-glucose , 25 NaHCO3 , 1 . 25 Na2H2PO4 , 2 CaCl2 , and 1 MgCl2 . L5 PNs were identified by their typical large triangular shape in the infragranular layers and in slices from PV-Cre; Ai14 mice the PV+ interneurons expressing tdTomato were identified using X-Cite series 120Q ( Excelitas ) with a bandpass filter ( excitation maximum 554 nm , emission maximum 581 nm ) . Somatic whole-cell current-clamp recordings were made with a bridge current-clamp amplifier ( BVC-700A , Dagan Corporation , USA ) using patch pipettes ( 4–6 MΩ ) filled with a solution containing ( in mM ) : 130 K-gluconate , 10 KCl , 4 Mg-ATP , 0 . 3 Na2-GTP , 10 HEPES , and 10 Na2-phosphocreatine , pH 7 . 4 , adjusted with KOH , 280 mOsmol/kg , to which 10 mg/mL biocytin was added . Voltage was analog low-pass filtered at 10 kHz ( Bessel ) and digitally sampled at 50–100 kHz using an analog-to-digital converter ( ITC-18 , HEKA Electronic ) and data acquisition software AxoGraph X ( v . 1 . 7 . 2 , AxoGraph Scientific , RRID:SCR_014284 ) . The access resistance was typically <20 MΩ and fully compensated for bridge balance and pipette capacitance . All reported membrane potentials were corrected for experimentally determined junction potential of –14 mV . Analysis for the electrophysiological properties includes PV+ interneuron recordings from cells in normal ACSF and in the presence of CNQX and d-AP5 with high chloride intracellular solution ( see below ) . Whole-cell voltage-clamp recordings were made with an Axopatch 200B amplifier ( Molecular Devices ) . Patch pipettes with a tip resistance of 3–5 MΩ were pulled from thins wall borosilicate glass . During recording , a holding potential of –74 mV was used . Both the slow- and fast pipette capacitance compensation were applied , and series resistance compensated to ~80–90% . Patch pipettes were filled with high chloride solution containing ( in mM ) 70 K-gluconate , 70 KCl , 0 . 5 EGTA , 10 HEPES , 4 MgATP , 4 K-phosphocreatine , 0 . 4 GTP , pH 7 . 3 adjusted with KOH , 285 mOsmol/kg and IPSCs isolated by the presence of the glutamate receptor blockers 6-cyano-7-nitroquinoxaline-2 , 3-dione ( CNQX , 20 µM ) , d-2-amino-5-phosphonovaleric acid ( d-AP5 , 50 µM ) and the sodium ( Na+ ) channel blocker tetrodotoxin ( TTX , 1 µM Tocris ) . Individual traces ( 5 s duration ) were filtered with a high-pass filter of 0 . 2 Hz and decimated in AxoGraph software ( RRID:SCR_014284 ) . Chart recordings of mIPSCs were analyzed with a representative 30 ms IPSC template using the automatic event detection tool of AxoGraph . Detected events were aligned and averaged for further analysis of inter-event intervals ( frequency ) and peak amplitude . For mEPSC recordings from PV+ interneurons , we filled patch pipettes with a solution containing ( in mM ) 130 K-gluconate , 10 KCl , 4 Mg- ATP , 0 . 3 Na2-GTP , 10 HEPES , and 10 Na2-phosphocreatine , pH 7 . 4 , adjusted with KOH , 280 mOsmol/kg and both gabazine ( 4 µM ) and TTX ( 1 μM ) were added to the bath solution . The mEPSCs were analyzed using events detection tool in AxoGraph . The recorded signals were bandpass filter ( 0 . 1 Hz to 1 kHz ) and recordings analyzed with a representative 30 ms EPSC template , after which selected EPSCs aligned and averaged for further analysis of inter-event intervals ( frequency ) and peak amplitude . PV+ interneurons ( visually identified in PV-Cre; Ai14 mice based on tdTomato fluorescence expression ) were targeted for whole-cell current-clamp recording within a radius of 50 µm from the edge of the L5 soma recorded in voltage-clamp configuration . APs in PV+ interneurons were evoked with a brief current injection ( 1–3 ms duration ) and uIPSCs recorded in the L5 PN from a holding potential of –74 mV . Only responses with 2× S . D . of baseline noise were considered being connected . Both fast and slow capacitances were fully compensated , series resistance compensation was applied to ~80–90% , and the current and voltage traces acquired at 50 kHz . For stable recordings with >50 uIPSCs , the episodes were temporally aligned to the AP and the uIPSCs were fit with a multiexponential function in Igor Pro . The curve fitting detected the baseline , uIPSC onset , rise time , peak amplitude , and decay time and was manually monitored . Fits were either accepted or rejected ( e . g . , when artifacts were present ) and the number of uIPSC failures was noted for each recording . 50 nL of AAV1 particles ( titer 1 × 1012 cfu/mL ) produced from pAAV-EF1a-double-floxed-hChR2 ( H134 ) -EYFP-WPRE-HGHpA ( Addgene . org #20298 , RRID:Addgene_20298 ) was injected into L5 of S1 ( coordinates from bregma; AP 0 . 15 mm , ML 0 . 30 mm , and DL 0 . 75 mm ) of 6–9-week-old PV-Cre; Ai14 mice . About 7 days after the injection , a subset of mice was placed on 0 . 2% cuprizone diet for 8–9 weeks . PV+ interneurons expressing hChR2 were identified using td-tom and YFP co-expression . Whole-cell voltage-clamp recordings were made from L5 PNs and oIPSCs were evoked with a X-cite 120Q , fluorescent lamp using filter BA460-510 ( Olympus ) in the presence of CNQX ( 50 µM ) and dAP5 ( 20 µM ) in the bath solution . The oIPSCs were evoked by illumination of large field with five light pulses of each 1 ms and 100 ms apart . Peak amplitude and area under the curve ( charge ) of oIPSC were quantified using AxoGraph . Only the first pulse was used for the quantification . Chronic ECoG and LFP recordings were performed using in-house-made electrodes of platinum-iridium wire ( 101R-5T , 90% Pt , 10% Ir , complete diameter of 200 µm with 127 µm metal diameter , Science Products ) . The perfluoroalkoxy alkanes ( PFA ) -coated wire platinum-iridium wire was only exposed at the tip to record the LFP . For placement of the recording electrode , animals were anesthetized with isoflurane ( 3% , flow rate 0 . 8 L/min with maintenance 1 . 5–1 . 8% , flow rate 0 . 6 L/min ) . A 1 cm midline sagittal incision was made starting above the interaural line and extending along the neck to create a pocket for subcutaneous placement of the transmitter along the dorsal flank of the animal . The recording electrodes in each hemisphere ( stereotaxic coordinates relative to bregma: S1; –0 . 15 mm anterior and ±0 . 30 mm lateral; for LFP; ventral 0 . 75 mm , V1; 0 . 40 mm anterior and ±0 . 30 mm lateral; for LFP; ventral 0 . 75 mm ) and ground electrode ( 6 mm posterior and 1 mm lateral ) were implanted subdurally through small holes drilled in the skull , held in place with stainless steel screws ( A2-70 , Jeveka ) , and subsequently sealed with dental cement . Mice were provided with Metachem analgesic ( 0 . 1 mg per kg ) after surgery and allowed to recover for 4–7 days before recordings . To obtain multiple hours recordings of ECoG-LFP at multiple weeks , mice remained in their home cage during an overnight recording session . ECoG-LFP data were collected using a ME2100-system ( Multi Channel Systems ) ; ECoG-LFP data were acquired at a sampling rate of 2 kHz using the multi-channel experimenter software ( Multi Channel Systems ) . An additional 0 . 1–200 Hz digital bandpass filter was applied before data analysis . Large noise signals , due to excessive locomotion or grooming , were manually removed from the data . The ECoG and LFP recordings were processed offline with the Neuroarchiver tool ( Open Source Instruments , http://www . opensourceinstruments . com/Electronics/A3018/Seizure_Detection . html ) . To detect interictal spikes , an event detection library was built as described previously ( Dubey et al . , 2018 ) . During the initial learning phase of the library , the observer , if needed , overruled the identity of each new event by the algorithm until automated detection reached a false positive rate <1% . Subsequently , the ECoG-LFP data were detected by using a single library across all ECoG-LFP recordings . For determining the interictal rate , only S1 LFP signals were used for quantification . 50 nL of AAV1 particles ( titer 1 × 1012 cfu/mL ) produced from pAAV-EF1a-double-floxed-hChR2 ( H134 ) -EYFP-WPRE-HGHpA ( Addgene #20298 , RRID:Addgene_20298 ) was injected unilaterally into the L5 of S1 ( coordinates from bregma; AP 0 . 15 mm , ML 0 . 30 mm , and DL 0 . 75 mm ) of 6–9-week-old PV-Cre; Ai14 mice . ECoG-LFP electrode ( stereotaxic coordinates relative to bregma: –0 . 15 mm anterior and ±0 . 30 mm lateral; for LFP; ventral 0 . 75 mm ) and ground electrode ( 6 mm posterior and 1 mm lateral ) were implanted through small holes drilled in the skull , held in place with stainless steel screws ( A2-70 , Jeveka ) . Through the drilled hole , a polished multimode optical fiber ( FP200URT , Thorlabs ) held in ceramic ferrule ( CFLC230-10 , Thorlabs ) was driven into layer 5 and ~50 µm above virus injection site . Once optical fiber and electrode were correctly placed , the drilled hole subsequently sealed with dental cement . A blue fiber-coupled laser ( 473 nm , DPSS Laser T3 , Shanghai Laser & Optics Co . ) was used to activate the ChR2 . Cyclops LED Driver ( Open Ephys ) , together with customized program , was used to design the on and off state of the laser . The driving signal from LED driver was also recorded at one of the empty channels in multichannel systems . This signal was used to estimate the blue light on or off condition . For gamma entrainment in S1 , 40 pulses of blue light were flashed with 1 ms on and 28 ms off pulse . To inhibit interictal spikes , 300 pulses of blue light were flashed with 1 s on and 100 ms off by manual activation of light pulses when periods of high interictal spikes were observed ( >10 interictals/min ) . Aged-matched control mice were stimulated during the resting phase of the EEG , which was estimated using online EMG signal and video observation . For interictal counts , 5 min LFP signals were used from before light stimulation , during , and post light stimulation . Interictals were detected using event detection library . For analysis of the cortical rhythms , epochs were extracted using 2 s window at the start and after 180 pulses of blue light . Epoch-containing interictals were not included in the analysis . For pharmacology experiment , continuous LFP recordings of >10–12 hr duration from the circadian quiet phase ( from 19:00 to 09:00 ) of six cuprizone mice ( 7-week treatment ) and three control mice were used for the analysis . To activate GABAA receptors in cuprizone-treated mice , we used diazepam ( Centrafarm Nederland B . V ) prepared in a 10% solution of ( 2-hydroxypropyl ) -β-cyclo-dextrin ( Sigma-Aldrich ) . A nonsedative dose of 2 mg/kg diazepam was injected intraperitoneally , and data was acquired for a period of 10 hr , starting 15 min after injection of drug in control and cuprizone mice . The automated event detection library ( Figure 1—figure supplement 1 ) was used to determine the event frequency before and after diazepam injection . Power spectral density ( PSD ) analysis was done using multitaper PSD toolbox from Igor Pro 8 . 0 ( RRID:SCR_000325 ) . The absence of high-voltage activity in the EMG electrode was classified as quiet wakefulness ( Figure 1—figure supplement 1 , Figure 1—video 1 ) . For PSD analysis during interictal activity , a 2 s window was used to extract LFP signal epochs . Epochs from control animals were selected comparing the EMG activity with cuprizone EMG activity . The interictal activity itself was excluded from the analysis . Selected LFP epochs were bandpass filtered between different frequency bands; delta , δ ( 0 . 5–3 Hz ) , theta , θ ( 4–12 Hz ) , beta , β ( 12 . 5–25 Hz ) , and gamma , γ ( 30–80 Hz ) . Multitaper PSD function ( Igor Pro 8 . 0 ) was applied to the filtered data to plot the power distribution within each frequency band . Area under the curve was measured for each frequency band to compare power density between the control and cuprizone groups . L5 PNs were filled with 10 mg/mL biocytin during whole-cell patch-clamp recording for at least 30 min . Slices were fixed for 30 min with 4% paraformaldehyde ( PFA ) and stored in 0 . 1 M phosphate buffered saline ( PBS; pH 7 . 4 ) at 4°C . Fixed 400 μm slices were embedded in 20% gelatin ( Sigma-Aldrich ) and then sectioned with a Vibratome ( VT1000 S , Leica Microsystems ) at 80 μm . Sections were preincubated with blocking 0 . 1 M PBS containing 5% normal goat serum ( NGS ) , 5% bovine serum albumin ( BSA; Sigma-Aldrich ) , and 0 . 3% Triton-X ( Sigma ) during 2 hr at 4°C to make the membrane permeable . For biocytin-labeled cells , streptavidin biotin-binding protein ( Streptavidin Alexa 488 , 1:500 , Invitrogen , RRID:AB_2315383 ) was diluted in 5% BSA with 5% NGS and 0 . 3% Triton-X overnight at 4°C . Sections including biocytin-filled cells were incubated again overnight at 4°C with primary antibody rabbit anti-ßIV-spectrin ( 1:200; gift from M . N . Rasband , Baylor College of Medicine ) , mouse anti-MBP ( 1:250; Covance ) , mouse anti-PV ( 1:1000; Swant , RRID:AB_10000343 ) , rabbit anti-syt2 ( 1:500 , Synaptic Systems , RRID:AB_108 94084 ) in PBS blocking solution containing 5% BSA with 5% NGS and 0 . 3% Triton-X . Secondary antibody were used to visualize the immunoreactions: Alexa 488-conjugated goat anti-rabbit ( 1:500; Invitrogen ) , Alexa 488 goat anti-mouse ( 1: 500; Sanbio ) , Alexa 488 goat anti- guinea pig , Alexa 555 goat anti-mouse ( 1:500; Invitrogen ) , Alexa 555 goat anti-rabbit ( 1:500; Invitrogen ) , Alexa 633 goat anti-guinea pig ( 1:500; Invitrogen ) , Alexa 633 goat anti-mouse ( 1:500; Invitrogen ) , and Alexa 633 goat anti-rabbit ( 1:500; Invitrogen ) . Finally , sections were mounted on glass slides and cover slipped with Vectashield H1000 fluorescent mounting medium ( Vector Laboratories , Peterborough , UK ) and sealed . A confocal laser-scanning microscope SP8 X ( DM6000 CFS; acquisition software , Leica Application Suite AF v3 . 2 . 1 . 9702 , RRID:SCR_013673 ) with a ×63 oil-immersion objective ( 1 . 3 NA ) and with 1× digital zoom was used to collect images of the labeled L5 neurons and the abovementioned proteins . Alexa fluorescence was imaged using corresponding excitation wavelengths at 15 units of intensity and a z-step of 0 . 3 μm . Image analysis was performed with Fiji ( ImageJ ) graphic software ( v . 2 . 0 . 0-rc-65/1 . 5w , National Institutes of Health , RRID:SCR_002285 ) . The intensity of PV+ or Syt2 immunostaining was measured with a z-axis profile , calculating the mean RGB value for each z-plane . When quantifying the axosomatic puncta , the soma was defined to extend into the apical dendrite maximally ~4 μm and a boundary was drawn around the maximum edges ( ROI ) . For counting apical dendritic puncta , a 200 µm length of apical dendrite was selected as ROI . Linear immunofluorescent signals from ßIV-spectrin were identified as AIS and used as ROI . For all analyses , the RGB images were separated into single-color channels using the color deconvolution plugin in ImageJ . The single-color channel containing boutons signals was subjected to thresholding and particle filter of 0 . 5 μm . The threshold was saved and applied to all images in the same staining group . The boutons were selected by scanning through the 3D projection of ROI with 0 . 35 µm z-steps . Trained experimenters identified the boutons either by colocalization of the ROI and PV/Syt2 or direct contact of the two . The boutons were characterized as round spots with a minimal radius of 0 . 5 μm ranging to almost 2 μm . Three experimenters blinded to the identity of the experiment group independently replicated the results . All image analyses were done in Fiji ( ImageJ ) graphic software ( v . 2 . 0 . 0-rc-65/1 . 5w , National Institutes of Health , RRID:SCR_002285 ) . For immunolabeling of biocytin-filled PV+ interneuron , 400 μm electrophysiology slices were incubated overnight at 4°C in PFA . Slices were rinsed with PBS followed by staining using streptavidin 488 ( 1:300 , Jackson ) diluted in PBS containing 0 . 4% Triton-X and 2% normal horse serum ( NHS; Gibco ) overnight at 4°C . Confocal images of 400-μm-thick slices were taken ( see ‘Confocal imaging’ ) and immediately after thoroughly rinsed with 0 . 1 M PB and 30% sucrose at 4°C overnight . Next , slices were sectioned into 40 μm thick and preserved in 0 . 1 M PB before staining . Sections were preincubated in PBS blocking buffer containing 0 . 5% Triton-X and 10% NHS during 1 hr at room temperature . Sections were stained with primary mouse anti-MBP ( 1:300 , Santa Cruz , RRID:AB_675707 ) , rat anti-syt2 ( RRID:AB_10894084 ) in 0 . 4% Triton-X , and 2% NHS with PBS solution for 72 hr . Alexa 488-conjugated secondary antibodies ( 1:300 , Invitrogen ) were added in PBS containing 0 . 4% Triton-X and 2% NHS , posterior to washing steps with PBS . Then , sections were mounted on slides and cover slipped with Vectashield H1000 fluorescent mounting medium , sealed , and imaged . Biocytin-labeled PV+ neurons were imaged using upright Zeiss LSM 700 microscope ( Carl Zeiss ) with ×10 and ×63 oil-immersion objectives ( 0 . 45 NA and 1 . 4 NA , respectively ) and 1× digital zoom with step size of 0 . 5 µm . Alexa 488 and Alexa 647 were imaged using 488 and 639 excitation wavelengths , respectively . The 10× image was taken to determine the exact location of biocytin-filled cells . Subsequently , axonal images were taken at ×63 magnification . Axons were analyzed as described previously ( Stedehouder et al . , 2019 ) and identified by their thin diameter , smoothness , obtuse branching processes , and occasionally by the presence of the axon bleb . Images were opened in Neurolucida 360 software ( v2018 . 02 , MBF Bioscience , RRID:SCR_001775 ) for reconstruction using the interactive user-guided trace with the Directional Kernels method . Axon and myelinated segments were analyzed using Neurolucida Explorer ( MBF Bioscience , RRID:SCR_001775 ) . Axonal segments were accepted as myelinated when at least one MBP-positive segment colocalized with streptavidin across the internode length . All statistical tests were performed using Prism 8 or 9 ( GraphPad Software , LLC , San Diego , CA , RRID:SCR_014284 ) . For comparisons of two independent groups , we used two-tailed Mann–Whitney U-tests . For multiple group comparisons , data were initially assessed for normality and subsequently we either used ordinary one-way ANOVA followed by Tukey’s multiple comparisons or two-way ANOVA with repeated measures followed by Šidák’s multiple comparisons tests to correct for multiple comparisons . The level of significance was set to 0 . 05 for rejecting the null hypothesis . A detailed overview of the statistical analyses performed in this study , together with the numbers used for figures and statistical testing , is provided in Source data 1 .	The brain contains billions of neurons that connect with each other via cable-like structures called axons . Axons transmit electrical impulses and are often wrapped in a fatty substance called myelin . This insulation increases the speed of nerve impulses and reduces the energy lost over long distances . Loss or damage of the myelin layer – as is the case for multiple sclerosis , a chronic neuroinflammatory and neurodegenerative disease of the central nervous system – can cause serious disability . However , a fast-firing neuron within the brain , called PV+ interneuron , has short , sparsely myelinated axons . Even so , PV+ interneurons are powerful inhibitors that regulate important cognitive processes in gray matter areas , including the outermost parts , in the cortex . Yet it remains unclear how the unusual , patchy myelination affects their function . To examine these questions , Dubey et al . used genetically engineered mice either lacking or losing myelin and studied the impact on PV+ interneurons and slow brain waves . As mice progressively lost myelin , the speed of inhibitory signals from PV+ interneurons did not change but their signal strength decreased . As a result , the power of slow brain waves , no longer inhibited by PV+ interneurons , increased . These waves also triggered spikes of epileptic-like brain activity when the mice were inactive and quiet . Restoring the activity of myelin-deficient PV+ interneurons helped to reverse these deficits . This suggests that myelination , however patchy on PV+ interneurons , is required to reach their full inhibitory potential . Moreover , the findings shed light on how myelin loss might underpin aberrant brain activity , which have been observed in people with multiple sclerosis . More research could help determine whether these epilepsy-like spikes could be a biomarker of multiple sclerosis and/or a target for developing new therapeutic strategies to limit cognitive impairments .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2022	Myelination synchronizes cortical oscillations by consolidating parvalbumin-mediated phasic inhibition
RasGRP1 and SOS are Ras-specific nucleotide exchange factors that have distinct roles in lymphocyte development . RasGRP1 is important in some cancers and autoimmune diseases but , in contrast to SOS , its regulatory mechanisms are poorly understood . Activating signals lead to the membrane recruitment of RasGRP1 and Ras engagement , but it is unclear how interactions between RasGRP1 and Ras are suppressed in the absence of such signals . We present a crystal structure of a fragment of RasGRP1 in which the Ras-binding site is blocked by an interdomain linker and the membrane-interaction surface of RasGRP1 is hidden within a dimerization interface that may be stabilized by the C-terminal oligomerization domain . NMR data demonstrate that calcium binding to the regulatory module generates substantial conformational changes that are incompatible with the inactive assembly . These features allow RasGRP1 to be maintained in an inactive state that is poised for activation by calcium and membrane-localization signals . An intriguing aspect of lymphocyte development is that the generation of a population of self-tolerant immune cells requires Ras activation by two distinct guanine nucleotide exchange factors , Ras guanine nucleotide releasing protein 1 ( RasGRP1 ) and Son-of-Sevenless ( SOS ) ( Figure 1A ) ( Dower et al . , 2000; Ebinu et al . , 2000; Kortum et al . , 2011 ) . Ras cycles between an inactive GDP-bound form and an active GTP-bound form , and in both states the nucleotide is very tightly bound ( Vetter and Wittinghofer , 2001; Rajalingam et al . , 2007; Ahearn et al . , 2012 ) . A critical regulatory function is therefore provided by the action of guanine nucleotide exchange factors , which loosen the grip of Ras on the bound nucleotide , allowing GTP loading and activation ( Bos et al . , 2007; Cherfils and Zeghouf , 2013 ) . 10 . 7554/eLife . 00813 . 003Figure 1 . Control of Ras activity in T cells . ( A ) Ras cycles between an inactive , GDP-bound form and an active GTP-bound form . In T cells , two nucleotide exchange factors , SOS and RasGRP1 enhance the removal of nucleotide from Ras , which is then replaced with GTP . Each exchange factor shares a common catalytic module but is regulated by distinct signaling inputs . SOS activity is enhanced by Ras•GTP , generated by RasGRP1 , binding to an allosteric site . The regulatory domains from each exchange factor are distinct and are represented in gray . SOS is recruited to the membrane in part by Grb2 , which interacts with phosphotyrosine residues in the adapter LAT . ( B ) The catalytic core of RasGRP1 includes the REM and Cdc25 domains , which are followed by a regulatory module containing the EF domain , membrane binding C1 domain and a predicted coiled coil . An alternate translational start site is present that leads to a RasGRP1 protein without the first 49 residues . The constructs used in this study are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00813 . 00310 . 7554/eLife . 00813 . 004Figure 1—figure supplement 1 . Domain architecture of RasGRP1 and SOS . RasGRP1 and SOS contain similar catalytic modules ( REM + Cdc25 domains ) , but differ in the flanking regulatory domains . DOI: http://dx . doi . org/10 . 7554/eLife . 00813 . 004 There are three major families of Ras-specific nucleotide exchange factors in humans . The RasGRP and Ras guanine nucleotide releasing factor ( RasGRF ) families of exchange factors have tissue-specific expression patterns whereas SOS proteins are expressed ubiquitously ( Stone , 2011 ) . RasGRP proteins have been studied most extensively in T and B lymphocytes ( Aiba et al . , 2004; Brodie et al . , 2004; Coughlin et al . , 2005; Roose et al . , 2005; Limnander et al . , 2011 ) where they activate Ras in a manner that is non-redundant with SOS ( Dower et al . , 2000; Roose et al . , 2007 ) . In addition , RasGRP proteins play important roles in squamous cell carcinoma and melanoma ( Luke et al . , 2007; Oki-Idouchi and Lorenzo , 2007; Diez et al . , 2009; Yang et al . , 2011 ) , T cell- and myeloid- leukemia ( Reuther et al . , 2002; Yang et al . , 2002; Klinger et al . , 2005; Lauchle et al . , 2009; Oki et al . , 2012; Hartzell et al . , 2013 ) and prostate cancer ( Yang et al . , 2010 ) that are distinct from those of SOS . Developing T lymphocytes pass through quality control checkpoints to generate a repertoire of protective but self-tolerant immune cells ( Starr et al . , 2003 ) and Ras signaling plays a critical role in this progression ( Swan et al . , 1995 ) . In response to T cell receptor stimulation during T lymphocyte development , both RasGRP1 and SOS are recruited to the membrane where they encounter membrane-anchored Ras . Interestingly , knockout mouse models have revealed that the requirements for RasGRP1 and SOS during different T lymphocyte checkpoints are distinct , despite the fact that they both convert Ras•GDP to Ras•GTP ( Dower et al . , 2000; Layer et al . , 2003; Kortum et al . , 2011 , 2012 ) . The importance of RasGRP1 in human diseases highlights a critical requirement for tight regulation of RasGRP1 activity . Elevated RasGRP1 mRNA expression has been reported in T cell leukemia microarray studies and is found frequently in pediatric T cell leukemia in which it stimulates the growth of this blood cancer ( Oki et al . , 2012; Hartzell et al . , 2013 ) . Conversely , reduced RasGRP1 expression has been reported for autoimmune patients with lupus erythematosus where it may play a role in aberrant DNA methylation in T cells ( Yasuda et al . , 2007; Pan et al . , 2010 ) . Additionally , single nucleotide polymorphisms in RasGRP1 have been described in genome-wide association studies of autoimmune diabetes and thyroid disease ( Qu et al . , 2009; Plagnol et al . , 2011 ) . The Ras-specific exchange factors have similar catalytic modules that contain two domains . The Cdc25 domain interacts directly with Ras and dislodges the bound nucleotide ( Boriack-Sjodin et al . , 1998 ) . The Ras exchanger motif ( REM ) domain that is associated with the Cdc25 domain is usually essential for activity but its function does not appear to be conserved in different exchange factors . Each family of Ras-specific exchange factors contains distinct regulatory domains that enable Ras signaling to be activated in response to a variety of upstream receptor stimuli . Despite the importance of the regulatory domains for controlling activation , our understanding of how these work at the structural level is limited to SOS ( Sondermann et al . , 2004; Gureasko et al . , 2008 , 2010 ) and the Rap-specific exchange factor , Epac2 ( Rehmann et al . , 2006 , 2008 ) . One important role for RasGRP1 is to prime SOS for activation by generating an initial burst of Ras•GTP ( Roose et al . , 2007 ) . This priming function of RasGRP1 potentiates SOS activity because of a feedback loop in which Ras•GTP activates SOS by binding to an allosteric site that bridges the REM and Cdc25 domains ( Margarit et al . , 2003; Boykevisch et al . , 2006; Sondermann et al . , 2004; Gureasko et al . , 2008 , 2010 ) . Ras•GTP binding to the allosteric site helps stabilize SOS at the plasma membrane and promotes the conversion of Ras•GDP to Ras•GTP . The action of RasGRP1 in initiating the positive feedback loop of SOS leads to ultrasensitive ERK activation in Jurkat T cells and has been postulated to define the sharp boundary between positively and negatively selecting ligands during thymocyte development ( Das et al . , 2009; Prasad et al . , 2009 ) . Compartmentalization of Ras signaling has also been proposed to play a role in the selection process ( Daniels et al . , 2006 ) . A complete understanding of how the interplay between RasGRP1 and SOS results in ultrasensitive activation of the ERK pathway requires mechanistic knowledge of how RasGRP1 is regulated , about which little is known . The catalytic module of RasGRP1 is followed by an EF domain with a predicted pair of EF hands ( EF1 and EF2 modules ) , a diacylglycerol-binding C1 domain , and a C-terminal segment that includes a primarily unstructured region of ∼140 residues and a predicted coiled coil ( Ebinu et al . , 1998; Beaulieu et al . , 2007; Zahedi et al . , 2011 ) ( see Figure 1B for the domain architecture of RasGRP1 ) . A portion of the C-terminal segment of RasGRP1 has been demonstrated to enhance membrane recruitment through electrostatic interactions with phosphoinositides ( Zahedi et al . , 2011 ) , and the physiological importance of this segment is illustrated by impaired T lymphocyte development in mice lacking this part of the protein ( Fuller et al . , 2012 ) . Little is known about how the regulatory domains of RasGRP1 control the activity of the catalytic module . The simplest model for RasGRP1 activation assumes that the recruitment of the protein from the cytosol to the membrane upon diacylglycerol production by phospholipase C suffices for activation by facilitating encounters with Ras . However , addition of a membrane localization tag to a fragment of RasGRP1 does not lead to constitutive Ras activation , suggesting more complexity in the regulatory mechanisms ( Beaulieu et al . , 2007 ) . The presence of two EF hands suggests that they might be responsible for the sensitivity of RasGRP1 to calcium , but there are conflicting reports as to whether calcium binding to the EF domain is coupled to the localization and activity of RasGRP1 ( Ebinu et al . , 1998; Lorenzo et al . , 2000; Tazmini et al . , 2009 ) . To identify the structural basis for the regulation of RasGRP1 , we have determined two crystal structures of RasGRP1 . Together , these structures span the folded domains of the protein and omit the N-terminal 50 residue segment and the ∼140 residue segment immediately following the C1 domain that are both predicted to be intrinsically disordered . The first structure includes the REM , Cdc25 , EF and C1 domains and suggests a structural basis for autoinhibition by the regulatory domains . Key aspects of the mechanisms we identify involve occlusion of the Ras-binding site by an interdomain linker and dimerization-mediated masking of the C1 domains . The second structure is pertinent in this regard , as it shows that the C-terminal segment forms a parallel coiled coil that facilitates dimerization . The EF domain , which we demonstrate has a single Ca2+ binding site , is likely to contribute to activation through a calcium-triggered conformational change that we have analyzed using NMR . Our results are consistent with a mechanism wherein membrane docking and calcium binding drive RasGRP1 activation by disrupting the C1 interface and removing an inhibitory linker that blocks the Ras-binding site . Our structural and functional studies on the autoinhibition and activation of RasGRP1 reveal striking contrasts between the control of RasGRP1 and SOS , reflecting how the two exchange factors respond differently to thymocyte T cell receptor inputs . RasGRP1 is maintained in an autoinhibited , dimeric state , which is released by the coordinated binding of diacylglycerol and calcium second messenger molecules ( Figure 12 ) . In contrast , SOS employs a conformational switch in the helical hairpin that is coupled to the allosteric Ras-binding site ( Margarit et al . , 2003; Sondermann et al . , 2004; Gureasko et al . , 2008 , 2010 ) . The N-terminal membrane-binding domains of SOS are coupled to this switch because they physically block access to the allosteric Ras-binding site . Another distinction between RasGRP1 and SOS is that dimeric RasGRP1 has buried membrane-interacting surfaces whereas the membrane binding sites in SOS appear to be readily accessible . SOS may not require masking of its membrane interacting surfaces because its activation requires occupation of the allosteric Ras-binding site . In contrast , RasGRP1 appears competent for Ras activation as soon as autoinhibitory restraints are released . These differences in regulatory mechanisms underlie ultrasensitive signaling by SOS , while the activation of Ras signaling by RasGRP1 occurs in a graded manner in response to stimulation . Further experiments will be necessary to understand how the different signaling inputs such as diacylglcyerol , calcium and phosphorylation are integrated by RasGRP1 to produce this signaling pattern . 10 . 7554/eLife . 00813 . 030Figure 12 . Model of RasGRP1 activation . Inactive RasGRP1 ( left ) is stabilized by the C1-dimer interface , which sequesters the membrane-interacting surface of the C1 domain , and the active-site blocking Cdc25-EF linker ( red ) . The C-terminal coiled coil stabilizes the dimer , thereby preventing inappropriate Ras activation . The autoinhibited form is activated by multiple signaling inputs that enhance nucleotide exchange activity ( right ) . Diaclyglycerol binding disrupts C1 dimerization , while Ca2+ binding to EF1 causes a conformational change that contributes to C1 reorientation , and the release of the inhibitory segment from the Ras-binding surface . Phosphorylation of the Cdc25 domain could aid in removal of the inhibitory linker . DOI: http://dx . doi . org/10 . 7554/eLife . 00813 . 030 Our finding that the Cdc25-EF linker physically blocks the Ras binding site in RasGRP1 fits into a theme for regulation of exchange factors by steric blockage of the small GTPase binding site . For example , Sec7 domain-containing Arf exchange factors are autoinhibited by an interdomain linker and a regulatory helix ( DiNitto et al . , 2007 ) . Similarly , the regulatory domains of Epac block the Rap binding site , and cAMP binding releases autoinhibition ( Rehmann et al . , 2006 , 2008 ) . SOS is an exception to this theme , although it is possible that the C-terminal segment of SOS , for which no structural information is available but which is known to inhibit nucleotide exchange activity ( Aronheim et al . , 1994; Wang et al . , 1995 ) , may play a similar role . C . elegans possess a single RasGRP protein , whereas mammals have four RasGRP proteins with distinct but often overlapping tissue distributions . It appears that these four proteins have evolved partially overlapping regulatory mechanisms . While the hydrophobic portion of the Cdc25-EF linker that docks onto the Cdc25 domain is conserved among the four human RasGRP proteins , only RasGRP1 contains a C-terminal coiled coil . This suggests that the Cdc25-EF linker is inhibitory for each protein , but indicates that RasGRP2 , 3 and 4 are either not dimeric upon activation or utilize distinct oligomerization mechanisms . The calcium-binding properties among the proteins of this family are also different . For example , examination of the EF domain sequence of RasGRP4 suggests that this protein does not bind calcium , or does so in a manner distinct from classical EF hands . It is perhaps due to these distinctions that loss of RasGRP1 is so detrimental to thymocyte development ( Dower et al . , 2000 ) and only minimally compensated for by RasGRP3 or 4 ( Zhu et al . , 2012; Golec et al . , 2013 ) . RasGRP1 also selectively appears as a common integration site in murine leukemia virus screens ( Mikkers et al . , 2002; Suzuki et al . , 2002; Akagi et al . , 2004 ) , resulting in increased expression of RasGRP1 , which is not observed for other Ras-specific exchange factors ( Hartzell et al . , 2013 ) . Nucleotide exchange factors have evolved distinct solutions to tightly tune Ras activation in response to diverse external stimuli . Germline gain-of-function SOS alleles can perturb Ras signaling , leading to Noonan syndrome , and the structures of SOS have aided the functional analysis of these variants and have further enhanced our understanding of SOS regulation ( Roberts et al . , 2007; Tartaglia et al . , 2007; Findlay et al . , 2013 ) . We anticipate that the mechanisms of RasGRP1 that we have identified will likewise provide a structural framework for understanding the connections between RasGRP1 variants and diseases such as leukemia and systemic lupus erythematosus . The genes for human RasGRP1CEC ( residues 50–607 ) , RasGRP1EF ( residues 459–540 ) RasGRP2EF ( residues 417–495 ) and RasGRP1CC ( residues 739–793 ) were cloned into the pSMT3 vector , containing N-terminal 6xHis and Sumo tags , at the BamHI and XhoI restriction sites . RasGRP1cat ( residues 50–468 ) was cloned into a pET28 derivative with a tobacco etch virus ( TEV ) protease-cleavable C-terminal 6xHis tag at the NdeI and XhoI sites . RasGRP1 and SOS mammalian expression plasmids for flow cytometry assays were constructed using the pEF6 vector as described previously ( Roose et al . , 2005 , 2007 ) . Mutations were introduced by Quikchange site-directed mutagenesis ( Agilent Technologies , Santa Clara , CA ) . RasGRP1 proteins were expressed in E . coli BL21 ( DE3 ) cells grown at 37°C in Terrific Broth with 50 μg/ml kanamycin grown to an OD of ∼1 . 0 . Cells were then induced with 1 mM IPTG at 15°C ( RasGRP1CEC , RasGRP1EF , RasGRP2EF , RasGRP1CC ) or 18°C ( RasGRP1cat ) for 14–18 hr . Growth media for cells expressing RasGRP1CEC was supplemented with 30 μM ZnCl2 . SeMet-substituted protein was expressed in cells grown in M9 minimal media with 50 μg/ml kanamycin . Before induction , the growth media was supplemented with 50 μg/ml of each amino acid ( except Met ) , 5 μg/ml Met and 50 μg/ml SeMet . 15N , 13C-labeled proteins were expressed in M9 minimal media with 1 g/L 15N ammonium chloride and 3 g/L 13C glucose ( Cambridge Isotope Laboratories , Andover , MA ) . Cell pellets were resuspended in Buffer A ( 25 mM Tris [pH 8 . 0] , 500 mM NaCl , 10% glycerol , 20 mM imidazole and 5 mM β-mercaptoethanol [βME] ) and frozen at −80°C . All purification steps were carried out at 4°C using columns from GE Healthcare ( Piscataway , NJ ) . Cells were lysed with a cell disrupter with 5 mM βME , 200 μM AEBSF , 5 μM leupeptin and 500 μM benzamidine . Clarified lysates were applied to a 5 ml HisTrap FF affinity column equilibrated in Buffer A . The column was then washed with 100 ml Buffer A , and proteins were eluted in Buffer A with 500 mM imidazole . Proteins were immediately desalted in Buffer B ( 25 mM Tris , 100 mM NaCl , 10% glycerol and 1 mM TCEP ) using a HiPrep 26/10 desalting column . For RasGRP1CEC and RasGRP1cat the buffer pH was 8 . 5 , while the pH for the buffer for all other constructs was 8 . 0 . Purification tags were removed by addition of ULP1 protease ( 6xHis-Sumo tag ) or TEV protease ( C-terminal 6xHis-tag ) and incubation at 4°C for 14–18 hr . Proteases and tags were removed by a second pass over a HisTrap FF column in Buffer A . The flow-through and wash fractions were concentrated to 2 ml and further purified on a Superdex 26/60 column equilibrated in Buffer B . Fractions containing pure protein were concentrated and frozen at −80°C until use . Ras , SOS and RasGRF constructs were expressed and purified as described previously ( Freedman et al . , 2006; Gureasko et al . , 2008 ) . Purification of RasGRP1CEC and RasGRP1cat yielded ∼1 mg of protein per liter of cells , while the other constructs yielded 5–10 mg of protein per liter of cells . Crystallization of RasGRP1CEC was carried out initially with sparse matrix screening using a Phoenix crystallization robot ( Art Robbins Instruments , Sunnyvale , CA ) , and thin hexagonal rod-shaped crystals were obtained in a single condition . The initial hit was further optimized through additive screening . Crystals used for data collection were grown by hanging drop vapor diffusion ( 500 μl reservoir volume ) by mixing 1 μl of protein ( 10 mg/ml ) with 1 μl of 0 . 15 M sodium citrate tribasic , 22% PEG 3350 and 1 mM MnCl2 . Crystals appeared at 20°C in 1–2 days and grew to a maximum length of ∼200 μm over 3–5 days . Crystals were cryoprotected in the crystallization solution with 20% glycerol and flash frozen in liquid nitrogen . RasGRP1CC ( 10 mg/ml ) was crystallized in 20 mM sodium acetate ( pH 3 . 6 ) , 22% PEG 3350 , 100 mM lithium sulfate and 0 . 4% formamide by mixing 0 . 2 μl protein with 0 . 2 μl of well solution . Square plate-like crystals were harvested after 5–7 days and cryoprotected in the crystallization solution with 20% glycerol . Diffraction data for both RasGRP1CEC and RasGRP1CC were collected at 100 K on beamline 8 . 2 . 2 at the Advanced Light Source , Lawrence Berkeley National Laboratories . X-ray data were processed with XDS ( Kabsch , 2010 ) , then Pointless and Scala from the CCP4 program suite ( Winn et al . , 2011 ) . Refinement was performed with Phenix . refine ( Adams et al . , 2010 ) . For RasGRP1CEC , an initial molecular replacement solution was found using Phaser ( McCoy et al . , 2007 ) with the RasGRF Cdc25 domain and the core of the SOS REM domain . The location of the C1 domain was identified from anomalous data from the two intrinsic Zn2+ ions and the proper orientation was defined by incremental rotation about the axis defining the two metal ions and refinement of the resulting structures . The correct sequence register was determined through identification of 14 of the expected 15 selenium sites using X-ray data for the SeMet-substituted protein ( the anomalous peak for residue Met 50 was not present ) . The position of the Cdc25-EF linker was determined from averaged kick omit maps ( Praznikar et al . , 2009 ) generated in Phenix , which aid in removing model bias . The RasGRP1CC structure was solved by molecular replacement using the APC coiled coil ( Day and Alber , 2000 ) . The structural model for RasGRP1CEC spans residues 53–593 and includes the REM , Cdc25 , EF and C1 domains . The electron density for portions of the linkers between the REM and Cdc25 domains ( residues 186–192 ) and between the Cdc25 and EF domains ( 437–448 ) is poor and therefore these residues have been excluded from the final model . The model for RasGRP1CC contains two molecules in the asymmetric unit that form the functional unit . Molecule A contains residues 745–793 , while molecule B includes residues 745–786 . Ras-coupled vesicles for in vitro nucleotide exchange experiments were generated as previously described ( Gureasko et al . , 2008 ) . Experiments with Ras-coupled vesicles were repeated at different Ras densities , which ranged from ∼1500 to 7000 Ras/μm2 . Nucleotide exchange measurements were measured by mant-dGDP ( Jena Bioscience , Jena , Germany ) fluorescence using a stopped-flow apparatus with a Fluoromax-3 fluorometer ( Horiba Scientific , Edison , NJ ) and analyzed as described ( Gureasko et al . , 2008 ) . The final Ras concentration was 500 nM . Samples were excited at 370 nm ( 5 nm slit width ) and the emission at 430 nm ( 5 nm slit width ) was followed for at least eight minutes with a 0 . 5 s sampling interval . DNA was introduced into JPRM441 cells ( Roose et al . , 2005 ) by electroporation ( 20 × 106 cells with 20 μg DNA ) ( Biorad Genepulser Xcel ) . After recovery , cells were washed in RPMI , plated in a 96 well roundbottom plate ( 0 . 4 × 106 cells /well ) , starved , and stimulated for 5 min with DMSO or with 1 μM ionomycin . Cells were fixed with prewarmed ( 37°C ) Fixation Buffer ( BD Cytofix , BD Biosciences , San Jose , CA ) and permeabilized with methanol at −20°C . Barcoding protocols were modified from described methods ( Krutzik and Nolan , 2006 ) . Pacific Blue and Alexa Fluor 488 carboxylic acid succinimidyl-esters ( Life Technologies , Grand Island , NY ) were added in methanol in serial dilutions , and cells were incubated for 30 min at −20°C . Cells were washed thoroughly in PBS containing 1% BSA and 2 mM EDTA ( FACS buffer ) , and barcoded cells were pooled and incubated with anti-phospho-ERK ( clone 197G2 ) and anti-myc ( clone 9B11; Cell Signaling , Danvers , MA ) for 1 hr at 22°C . Cells were washed in FACS buffer and incubated with allophycocyanin-Donkey anti-Rabbit IgG ( Jackson ImmunoResearch , West Grove , PA ) and Pe/Cy7 Goat anti-mouse IgG ( Biolegend , San Diego , CA ) for 30 min at 22°C . Cells were washed and analyzed using the LSR II flow cytometer ( BD Biosciences , San Jose , CA ) . Data were analyzed using Cytobank . The specific RasGRP1 mutations made for the Cdc25-EF linker variants are: Linker 2A- W454A and D453A , Linker 3D- V451D , V452D and W454D , Linker 5A- V450A , V451A , V452A , D453A and W454A . ITC experiments were carried out using a VP-ITC instrument ( GE Healthcare ) and analyzed using the Origin 7 software package . RasGRP proteins were treated with 10- to 20-fold excess EDTA for at least 12 hr at 4°C to remove bound metal ions . The sample was then dialyzed extensively against Chelex-treated ( Sigma Aldrich , St . Louis , MO ) 25 mM Tris ( pH 8 . 0 or 8 . 5 ) , 100 mM NaCl , 10% glycerol and 1 mM TCEP for 15 hr to remove the EDTA . Protein ( 20–50 μM ) was placed in the cell , and aliquots of 0 . 2–1 . 0 mM CaCl2 were titrated into the protein sample . An initial injection of 4 μl was excluded from data analysis , followed by 23 injections of 12 μl each , separated by 300 s with a filter period of 2 s . Titration curves were fit with a “one set of sites” model , with the fitting parameters N ( stoichiometry ) , Ka ( association constant ) and ΔH ( enthalpy of interaction ) . Calcium binding to RasGRP1CEC could not be analyzed by ITC due to poor stability of the protein during stirring , and difficulties removing metal ions pre-bound to the EF domain without removing Zn2+ from the C1 domain . Instead , metal binding was analyzed using a Tb3+-FRET assay by excitement of tryptophan and tyrosine residues , which can transfer energy to nearby Tb3+ ions bound to the protein . Protein ( 1 μM ) was incubated at 22°C for 30 min with varying Tb3+ concentrations , and the fluorescence emission at 543 nm was recorded after excitation at 285 nm . A separate sample was made for each point of the titration . Data were fit to a single binding site isotherm . Isotopically labeled proteins for NMR were prepared as described above . For RasGRP2EF samples , 500 μM protein was mixed with 1 . 2 mM CaCl2 or 5 mM EDTA in 25 mM Hepes ( pH 7 . 0 ) , 100 mM NaCl , 1 mM TCEP and 7% D2O . RasGRP1EF ( 250 μM ) samples were made in 25 mM Hepes ( pH 7 . 0 ) , 100 mM NaCl , 1 mM TCEP , 15 mM β-octyl glucoside and 7% D2O with 400 μM CaCl2 or 5 mM EDTA . All NMR experiments were recorded on a Bruker Avance II 800 MHz NMR spectrometer ( Bruker Biospin Corp , Billerica , MA ) equipped with a room temperature TXI probe at 304 K . Data were processed with NMRPipe ( Delaglio et al . , 1995; Goddard and Kneller ) and analyzed using Sparky 3 . 111 ( Goddard and Kneller ) . The cross-peak fit height was measured using the Gaussian line fitting protocol implemented in Sparky 3 . 111 . The 2D 15N-1H-HSQC spectra were acquired with spectral widths of 35 and 14 ppm for the 15N and 1H dimensions , respectively and with 128 ( 15N ) and 1024 ( 1H ) complex points . During all experiments the carrier frequencies of the proton and nitrogen channels were centered at 4 . 7 and 119 ppm , respectively . The chemical shifts for the proton dimension were referenced relative to 4 , 4-dimethyl-4-silapentane-1-sulfonic acid ( DSS ) , and the nitrogen and carbon dimensions were indirectly referenced using the respective gyro-magnetic ratios relative to that of a proton ( Wishart et al . , 1995 ) . Backbone chemical shifts were assigned using HNCO , HNCA , HNHA , CBCA ( CO ) NH , HNCACB and 15N HSQC-NOESY-15N HSQC experiments ( Kay et al . , 2011 ) . Side chain assignments were completed with 1H-13C HSQC , 13C ( CT ) -HSQC , H ( CC ) ( CO ) NH and CC ( CO ) NH experiments ( Grzesiek et al . , 1993 ) . The ( CT ) -HSQC spectrum was recorded with 28 ms of constant time 13C evolution . TOCSY mixing time for the H ( CC ) ( CO ) NH and CC ( CO ) NH experiments was set at 16 ms . 1H-1H distance restraints were measured from 15N edited NOESY and 13C edited NOESY spectra . 15N edited NOESY data were acquired with 80 ms mixing times , whereas 13C edited NOESY spectra were acquired with 120 ms mixing times . The weakly aligned sample for the RDC measurements was prepared by mixing the protein sample with a final concentration of 17 mg/ml of Pf1 phage ( Asla Biotech , Riga , Latvia ) ( Hansen et al . , 1998 ) . The backbone 1DNH and 1DCαHα residual dipolar couplings were measured from 2D IPAP-HSQC ( Ottiger et al . , 1998 ) and 3D IPAP-J-HNCO ( CA ) ( Yang et al . , 1998 ) experiments , respectively . The backbone dihedral angle ( ϕ and ψ ) restraints and the side chain χ1 angles were estimated from chemical shifts using TALOS+ ( Shen et al . , 2009 ) and PREDITOR ( Berjanskii et al . , 2006 ) , respectively . In total , 15 χ1 angles were selected based on a high confidence limit ( >85% ) . The proton–proton distance restraints were obtained from the peak intensities of the NOESY experiments , where NOE intensities are defined as strong ( 1 . 8–2 . 5 Å ) , medium ( 1 . 8–3 . 0 Å ) , and weak ( 1 . 8–5 . 0 Å ) . Hydrogen bond restraints were also implemented for residues in the helical segment , as judged by the presence of sequential HN-HN NOEs and high helical propensity by TALOS+ . Two distance restraints were defined for each hydrogen bond: 1 . 8–2 . 0 Å for the HN-O distance and 2 . 7–3 . 0 Å for the N-O distance ( Guntert et al . , 1998 ) . The ten lowest energy structures satisfying the NOE , dihedral angle and H-bond restraints were obtained from an extended chain using standard high-temperature annealing implemented in CNS ( v1 . 3 ) ( Brunger , 2007 ) . Next , each of these structures were refined with NH RDCs and then with Cα-Hα RDCs . During this refinement step , the temperature was decreased from 1000 to 0 K , and the NH RDC restraint potential was ramped from 0 . 1 to 5 kcal mol−1 Hz−2 . Initial estimates of the axial ( Da ) and rhombicity ( Rh ) of the alignment tensor were obtained using the program PALES ( Zweckstetter and Bax , 2000 ) . The force constant for the NOE and dihedral retraints were fixed at 50 kcal mol−1 Å−2 and 200 kcal mol−1 rad−2 , respectively . 50 structures from each of the first 10 models were calculated . The lowest energy structure obtained from each run was subsequently refined using Cα-Hα RDCs in addition to the other restraints . The Cα-Hα RDCs were normalized to the NH RDC values ( Clore et al . , 1998 ) . Sidechains coordinating the Ca2+ ion in EF hand 1 ( residues 23 , 25 , 27 and 34 ) and EF hand 2 ( O’Shea et al . , 1991; Zhang et al . , 1995; Ahmadian et al . , 2002 ) were implemented as distance restraints . A distance of 2 . 2–3 Å was defined for each of the O–Ca2+ restraints . During this step , the temperature was decreased from 500 to 0 K , and the RDC restraint potential was ramped from 0 . 001 to 0 . 2 kcal mol−1 Hz−2 . Ten representative structures were selected , and the quality of the selected structures was assessed using PROCHECK-NMR ( Laskowski et al . , 1996 ) and Protein Structure Validation Software suit v1 . 4 ( Bhattacharya et al . , 2007 ) . A summary of the structural statistics is presented in Table 4 .	Individual cells within the human body must grow , divide or specialize to perform the tasks required of them . The fates of these cells are often directed by proteins in the Ras family , which detect signals from elsewhere in the body and orchestrate responses within each cell . The activities of these proteins must be tightly controlled , because cancers and developmental diseases can result if Ras proteins are not properly regulated . Binding to the small molecule GTP activates Ras and causes conformational changes that allow it to interact with other proteins in various signaling pathways in the cell . GTP is loaded into Ras by proteins called nucleotide exchange factors , which can replace ‘used’ nucleotides with ‘fresh’ ones to activate Ras . These nucleotide exchange factors are also tightly regulated . For example , the genes for many exchange factors are only switched on after particular signals are received , which can restrict their presence to defined times and locations ( e . g . , cells or tissues ) . Also , when activating signals are absent , nucleotide exchange factors commonly reside in the cytoplasm , whereas the Ras proteins remain bound to lipid membranes inside the cell . RasGRP1 is a nucleotide exchange factor that controls the development of immune cells , and leukemia and lupus can result if it is not regulated correctly . However , many questions about RasGRP1 remain unanswered , including how it is able to remain inactive , and how it is activated by various different signals . Iwig et al . have now revealed the mechanisms through which RasGRP1 suppresses Ras signaling in immune cells by solving the structures of two fragments of RasGRP1 and then using a combination of structural , biochemical and cell-based methods to explore how it is activated . These analyses revealed that inactive RasGRP1 adopts a conformation in which one of its regulatory elements blocks access to the Ras binding site . Surprisingly , RasGRP1 can form dimers; this hides the portions of the protein that associate with the membrane and thereby keeps RasGRP1 away from Ras . Iwig et al . also found that two signals , calcium ions and a lipid called diacylglycerol , overcome these inhibitory mechanisms by changing the conformation of RasGRP1 and recruiting it to the membrane . These studies provide a framework for understanding how disease-associated mutations in RasGRP1 bypass the regulatory mechanisms that insure proper immune cell development , and will be critical for developing therapeutic agents that inhibit RasGRP1 activity .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2013	Structural analysis of autoinhibition in the Ras-specific exchange factor RasGRP1
Tumors are composed of many different cell types including cancer cells , fibroblasts , and immune cells . Dissecting functional metabolic differences between cell types within a mixed population can be challenging due to the rapid turnover of metabolites relative to the time needed to isolate cells . To overcome this challenge , we traced isotope-labeled nutrients into macromolecules that turn over more slowly than metabolites . This approach was used to assess differences between cancer cell and fibroblast metabolism in murine pancreatic cancer organoid-fibroblast co-cultures and tumors . Pancreatic cancer cells exhibited increased pyruvate carboxylation relative to fibroblasts , and this flux depended on both pyruvate carboxylase and malic enzyme 1 activity . Consequently , expression of both enzymes in cancer cells was necessary for organoid and tumor growth , demonstrating that dissecting the metabolism of specific cell populations within heterogeneous systems can identify dependencies that may not be evident from studying isolated cells in culture or bulk tissue . Tumors are composed of a heterogeneous mix of cell types , including cancer cells and stromal cells such as fibroblasts , macrophages , and other immune cells . How these different cell types interact to enable tumor growth is poorly understood . Environmental context plays an important role in determining how cancer cells use nutrients to proliferate and survive , and non-cancer cells within a tissue can alter nutrient availability ( Lyssiotis and Kimmelman , 2017; Mayers and Vander Heiden , 2017; Muir et al . , 2018; Pavlova and Thompson , 2016; Sullivan and Vander Heiden , 2019 ) . There is evidence that different cell populations within tumors can compete for limiting nutrients ( Chang et al . , 2015; Ho et al . , 2015; Zecchin et al . , 2017 ) , and metabolic cooperation between different cell types can also influence tumor phenotypes ( Linares et al . , 2017; Sousa et al . , 2016; Valencia et al . , 2014; Vander Heiden and DeBerardinis , 2017 ) . Nevertheless , technical challenges associated with studying the metabolism of individual cell types within a mixed population have limited a complete understanding of the metabolic interactions between cells in tumors . More broadly , this challenge has been a barrier to study how cells use nutrients differently within tissues to support both normal and disease physiology . Cancer cell metabolism in culture can differ from the metabolism of tumors in vivo ( Biancur et al . , 2017; Cantor et al . , 2017; Davidson et al . , 2016; Mayers and Vander Heiden , 2015; Muir et al . , 2017; Sellers et al . , 2019; Vande Voorde et al . , 2019 ) . This can at least partially be ascribed to changes in cancer cell metabolism that are driven by different nutrients present in the extracellular environment; however , another major difference between cell culture and tumors is the presence of additional cell types within tumors that are absent from most culture systems . The presence of many different cell types complicates the ability to characterize cancer cell metabolism in tumors , particularly in cases where a minority of the tumor is composed of cancer cells . For instance , cancer cells are a minority cell type in pancreatic ductal adenocarcinoma ( PDAC ) tumors ( Feig et al . , 2012 ) , and an understanding of cell metabolism in these tumors requires de-convolution of cancer-specific and stroma-specific phenotypes . Furthermore , there is evidence that the metabolism of cancer cells and different stromal cells isolated from these tumors can be different from each other when studied in culture ( Francescone et al . , 2018; Halbrook et al . , 2019; Sousa et al . , 2016 ) , and it is unknown whether the metabolic programs used by different cell populations in culture are also used within PDAC tumors in vivo where environmental conditions are different . Studies of bulk tumor metabolism fail to capture information about metabolic heterogeneity with regard to different cell types ( Xiao et al . , 2019 ) , and existing approaches are limited in their ability to study functional metabolic phenotypes in different cell populations in intact tissue . A major limitation arises from the fact that metabolic reactions take place on time scales that are faster than the turnover of many metabolic intermediates , complicating metabolite analysis after tumor digestion and cell sorting ( Shamir et al . , 2016 ) . Furthermore , cell sorting exposes cells to conditions that are different from those experienced by cells in tissues and can change metabolism in many ways . For example , sorting can induce mechanical and oxidative stress and reduce the levels of certain metabolites ( Binek et al . , 2019; Llufrio et al . , 2018; Roci et al . , 2016 ) , and even adding small amounts of bovine serum to the sorting buffer was not sufficient to prevent changes in some metabolite levels during cell sorting ( Llufrio et al . , 2018 ) . Indeed , isotopic metabolite labeling patterns can be more robust than metabolite levels when assessing metabolites from flow cytometry sorted cells , although labeling can also be affected by cell sorting ( Roci et al . , 2016 ) . Nevertheless , interpretation of metabolite labeling patterns is influenced by whether cells are at metabolic steady state ( Buescher et al . , 2015 ) , which when coupled with the rapid timescales of metabolism relative to the time needed to isolate cells suggests that new approaches are needed to better understand metabolism of individual cell types within mixed cell populations such as tumors . To overcome the challenges associated with studying cell metabolism within intact tumors and organoid co-cultures , we adapted an approach based on end-product biomass labeling ( Green et al . , 2016; Hosios et al . , 2016; Le et al . , 2017; Lewis et al . , 2014; Mayers et al . , 2016; Shankaran et al . , 2016 ) . This technique has been applied in studies of microbial metabolism and metabolic engineering to better understand mixed populations of bacteria ( Gebreselassie and Antoniewicz , 2015; Ghosh et al . , 2014; Rühl et al . , 2011; Zamboni et al . , 2005 ) . Unlike short-lived metabolic intermediates , turnover of end-product macromolecules such as protein is slow relative to the time period needed to isolate tumor cell populations . By assessing the isotope-labeling pattern of end-product biomass generated when cells are exposed to labeled nutrients in a mixed cell population , metabolic differences in nutrient use by different cells can be inferred . We used this approach to uncover a difference in glucose metabolism between cancer cells and fibroblasts in PDAC . Specifically , we find that , relative to fibroblasts , cancer cells within PDAC tumors have increased use of glucose for tricarboxylic acid ( TCA ) cycle anaplerosis through increased flux through pyruvate carboxylase ( PC ) . This phenotype is not evident when cancer cells and fibroblasts are studied as separate populations in mono-culture , even though PC is necessary for tumor growth in vivo . Furthermore , deletion of PC was insufficient to account for all pyruvate carboxylation activity within cancer cells in a mixed population , revealing that malic enzyme 1 ( ME1 ) also contributes to pyruvate carboxylation in cancer cells when fibroblasts are present and is required for PDAC tumor growth . These data argue that tracing labeled nutrients into stable biomass can be used to reveal metabolic differences between subpopulations of cells in a mixed cell system and to identify phenotypes that depend on the co-existence of multiple cell types . PDAC involves tumors where cancer cells can be a minority cell population ( Feig et al . , 2012 ) . To better understand glucose metabolism of PDAC tumor tissue in vivo , we infused U-13C-glucose into conscious , unrestrained mice ( Davidson et al . , 2016; Hui et al . , 2017; Marin-Valencia et al . , 2012 ) bearing PDAC tumors from autochthonous models that are driven by activating mutations in Kras and disruption of Trp53 function ( Bardeesy et al . , 2006; Hingorani et al . , 2005 ) . Similar to what has been observed with other mouse cancer models and in humans ( Davidson et al . , 2016; Fan et al . , 2009; Hensley et al . , 2016; Sellers et al . , 2015 ) , extensive labeling of multiple metabolic intermediates is observed from U-13C-glucose in pancreatic tumors and normal pancreas ( Figure 1 , Figure 1—figure supplements 1–3 , Figure 1—source data 1 , Figure 1—figure supplement 1—source data 1 , Figure 1—figure supplement 2—source data 1 , Figure 1—figure supplement 3—source data 1 ) . To assess U-13C-glucose labeling of metabolites in autochthonous pancreatic tumors arising in the LSL-KrasG12D; Trp53fl/fl; Pdx1-Cre ( KP-/-C ) ( Bardeesy et al . , 2006 ) and the LSL-KrasG12D; Trp53R172H/+; Pdx1-Cre ( KPC ) ( Hingorani et al . , 2005 ) mouse models , we first confirmed that plasma glucose levels were not changed over the course of the experiment ( Figure 1A ) . At this rate of glucose infusion , enrichment of 13C-glucose in plasma was around 40% in non-tumor-bearing mice , and in both KP-/-C and KPC animals ( Figure 1B ) . Glucose enrichment was not measured in tumors , as glucose is thought to be rapidly metabolized after entry into cells ( Nguyen et al . , 2011; Parikh et al . , 2015; Yeh et al . , 2018 ) . Under these conditions , labeling of pyruvate and lactate , as well as the glucose-derived amino acids alanine and serine , from U-13C-glucose was observed in normal pancreas tissue and in tumors arising in both models ( Figure 1C–F , Figure 1—source data 1 ) . Labeling of TCA cycle metabolites was also observed in normal pancreas tissue and in PDAC tumors from both models ( Figure 1G–L , Figure 1—source data 1 ) . Carbon from glucose-derived pyruvate can contribute to the TCA cycle via reactions catalyzed by pyruvate dehydrogenase ( PDH ) or pyruvate carboxylase ( PC ) , and the relative use of these routes of TCA cycle labeling has been inferred from the labeling pattern of TCA cycle intermediates from 13C-labeled glucose ( Davidson et al . , 2016; Fan et al . , 2009; Hensley et al . , 2016; Sellers et al . , 2015 ) . PDH decarboxylates three-carbon pyruvate to two-carbon acetyl-CoA , and therefore if 13C-labeled pyruvate is metabolized via PDH , a two carbon ( M+2 ) labeling pattern is observed in TCA cycle metabolites as well as in the amino acid aspartate ( Figure 1G ) . In contrast , PC carboxylates three-carbon pyruvate to four-carbon oxaloacetate , and therefore if unlabeled CO2 is added to 13C-labeled pyruvate via this enzyme , a three carbon ( M+3 ) labeling pattern is observed ( Figure 1G ) . We observed an increase in both M+2 and M+3 labeling of TCA cycle metabolites and the TCA cycle-derived amino acids aspartate and glutamate in KP-/-C tumor tissue compared to normal pancreas ( Figure 1H–N ) . Importantly , despite some aspartate labeling in the plasma of infused mice ( Figure 1—figure supplement 1J , Figure 1—figure supplement 1—source data 1 ) , prior studies of this model have shown that cancer cells cannot take up extracellular aspartate ( Sullivan et al . , 2018 ) , arguing that labeled aspartate in plasma contributes minimally to aspartate labeling in cancer cells in PDAC tumors . Proline was not extensively labeled from glucose in pancreas or PDAC tumor tissue from either model , suggesting that glucose carbon contributes minimally to the synthesis of this amino acid ( Figure 1O , Figure 1—source data 1 ) . Of note , in KPC mouse PDAC tumors , we observed higher M+2 labeling of only some TCA cycle metabolites compared to normal pancreas , and did not observe an increase in M+3 metabolite labeling compared to pancreas ( Figure 1H–O ) , illustrating that glucose labels metabolites differently in bulk tumors arising in each model . These data suggest that glucose contributes to labeling of TCA cycle carbon via reactions that involve PDH and PC in PDAC tumors , although the extent of labeling depends on the PDAC model used . To further study glucose metabolism in PDAC tumor tissue , we infused control and tumor-bearing KP-/-C mice with U-13C-glucose at a higher rate in an attempt to increase plasma enrichment of labeled glucose ( Davidson et al . , 2016; Figure 1—figure supplement 2 , Figure 1—figure supplement 3A–K , Figure 1—figure supplement 2—source data 1 , Figure 1—figure supplement 3—source data 1 ) . Plasma glucose levels were increased as a result of this higher infusion rate ( Figure 1—figure supplement 2B ) , and resulted in extensive labeling of glycolytic metabolites and glucose-derived amino acids ( Figure 1—figure supplement 2C–F , Figure 1—figure supplement 2—source data 1 ) , as well as an increase in M+2 and M+3 labeling of TCA cycle intermediates and related amino acids in pancreatic tumor tissue relative to normal pancreas ( Figure 1—figure supplement 2G–N , Figure 1—figure supplement 2—source data 1 ) . These data further support that glucose carbon can contribute label to TCA cycle intermediates via pathways that involve PDH and PC in PDAC tumors , but the relative contribution varies based on plasma glucose levels and the PDAC model examined . Despite both being driven by mutant Kras and loss of normal Trp53 function , differences in the autochthonous KP-/-C and KPC PDAC models are known and have been attributed to differences in tumor latency and p53 status , as well as differences in how stromal cell populations interact with cancer cells to support tumor growth ( Rosenfeldt et al . , 2013; Vennin et al . , 2019 ) . Thus , a difference in relative abundance of cancer and non-cancer cells , or in interactions between non-cancer cells and cancer cells , are possible explanations for why differences in glucose labeling are observed across these models . To explore whether the relative abundance of cancer cells in the tumor might affect labeling , we infused mice with pancreatic tumors derived from orthotopic injection of a syngeneic KrasG12D; Trp53-/- pancreatic cancer cell line derived from tumors arising in the KP-/-C model ( Danai et al . , 2018 ) , since cell line transplantation models are thought to result in tumors with a less dense , desmoplastic stroma compared to autochthonous models ( Baker et al . , 2016; Olive et al . , 2009 ) . When compared to adjacent normal pancreas , we observed an increase in M+2 and M+3 labeling of TCA metabolites and aspartate in this orthotopic tumor model ( Figure 1—figure supplement 3L–O , Figure 1—figure supplement 3—source data 1 ) . Regardless , tumors arising from orthotopic transplantation of murine PDAC cells still contain stroma ( Danai et al . , 2018 ) . Thus , tumors in all models considered consist of multiple cell types , and cancer cells are known to be a minority cell population in both autochthonous PDAC tumor models . In all cases , metabolite labeling will reflect a weighted average of labeling in all cell types present in the tissue sample and this heterogeneity in cell types in all tissues is a limitation to the use of labeled nutrient infusions to understand the metabolism of cancer cells , or any individual cell population , in tumors or other tissues . Dissecting the metabolism of individual cell types within a mixed cell population is a barrier to identifying cancer cell-specific liabilities via functional metabolic measurements in tumors such as PDAC . Therefore , we sought to better understand the contribution of cancer cells to the labeling of TCA cycle intermediates from 13C-glucose in PDAC tumors . We focused on M+3 labeling of TCA cycle intermediates from glucose , reflective of pyruvate carboxylation activity , because this is a metabolic phenotype observed in tumors that is less prominent in cancer cells in culture ( Davidson et al . , 2016 ) . One existing approach to determine which cell type within the tumor contributes to this activity is to evaluate expression of an enzyme known to catalyze this reaction . Indeed , immunohistochemistry ( IHC ) analysis of tumors arising in KP-/-C mice revealed higher PC expression in cancer cells ( Figure 2A ) . Analysis of a human pancreatic tumor tissue by IHC shows that human tumors exhibit a range of PC expression levels ( Figure 2—figure supplement 1A–D ) , although higher PC expression is observed in cancer cells compared to stroma ( Figure 2B ) , similar to findings in human lung tumors ( Sellers et al . , 2015 ) . However , while IHC can be useful to determine relative expression in tumor sections , it does not prove lack of expression by non-cancer cells . Indeed , qPCR analysis of mRNA isolated from sorted cell populations derived from KP-/-C tumors trended toward higher PC expression in cancer cells relative to fibroblasts , although PC mRNA was detected in fibroblasts ( Figure 2C ) . Furthermore , metabolic fluxes can be more dependent on metabolite concentrations than enzyme expression levels ( Hackett et al . , 2016 ) . Therefore , increased enzyme expression may not reflect increased activity in tissues , and argues that while suggestive , expression analysis is not definitive for identifying which cell population is responsible for M+3 TCA cycle labeling from glucose in pancreatic tumors . Another approach that has been used to determine which cell type ( s ) within the tumor contribute to a metabolic activity is isolating distinct cell populations and studying them in culture ( Dalin et al . , 2019; Francescone et al . , 2018; Linares et al . , 2017; Sousa et al . , 2016; Valencia et al . , 2014; Yang et al . , 2016 ) . Pancreatic stellate cells ( PSCs ) are a type of resident fibroblast in the pancreas which can become activated during tumorigenesis and impact the tumor microenvironment ( Bynigeri et al . , 2017; Dunér et al . , 2011 ) . When PDAC cells or PSCs alone are cultured in the presence of 13C-glucose in vitro , PSCs exhibit similar or higher M+3 TCA cycle metabolite labeling than cancer cells ( Figure 2D–E ) , even though the fibroblast cell population exhibited lower PC expression in tumors ( Figure 2A–B ) . These data further highlight the challenges associated with ascribing functional metabolic phenotypes using enzyme expression alone . Nevertheless , isolated cell populations in culture also may not retain the same functional metabolic phenotypes found within tumor tissue where many different cells compete for available nutrients . To develop new approaches to study the phenotype of individual cell types in a mixed cell population , we first sought to generate a more tractable system that only involves interactions between two different cell types . One approach is to use organoid cultures involving PDAC cancer cells and fibroblasts ( Öhlund et al . , 2017 ) where nutrient conditions are modified such that cancer cells rely on the presence of the fibroblasts to proliferate . To do this , we generated pancreatic cancer organoid cultures from KP-/-C and KPC tumors ( Figure 2—figure supplement 1E; Boj et al . , 2015 ) , and found that when exposed to a more minimal medium than is commonly used ( Boj et al . , 2015 ) , pancreatic cancer organoid growth becomes dependent on including PSCs in the culture system ( Figure 2F–G , Figure 2—figure supplement 1F–G ) . Relevant to the M+3 labeling of TCA cycle-associated metabolites from glucose observed in pancreatic tumors , when sorted from this co-culture organoid system , both cancer cells and PSCs expressed higher levels of PC mRNA compared to PSCs and pancreatic cancer cells in standard monoculture ( Figure 2H ) . In addition , when U-13C-glucose is provided to whole organoid co-cultures comprised of both cancer cells and PSCs , and TCA cycle intermediate labeling is assessed after rapid quenching and extraction of metabolites , M+3 labeling of aspartate and malate was observed ( Figure 2I–J , Figure 2—figure supplement 1H–K , Figure 2—source data 1 ) . These data argue that this organoid co-culture system may provide a model to explore the relative contribution of each cell population to the pyruvate carboxylation phenotype observed when both cell types are present . To dissect PDAC cancer cell versus other cell type contributions to specific metabolic activities , we reasoned it would be necessary to isolate each cell type for analysis after exposure to labeled nutrients . To experimentally evaluate the effect of sorting cells from the organoid co-culture system on metabolite levels and labeling from glucose , we cultured AL1376 murine PDAC cells in U-13C-glucose and incubated the cells in buffer on ice for various lengths of time to simulate conditions the cells would experience during separation by flow cytometry ( up to 240 min ) or other antibody based methods , which require a minimum of 10–12 min ( Abu-Remaileh et al . , 2017; Chen et al . , 2016 ) . Metabolite levels and labeling were then measured over time using mass spectrometry , allowing comparison to that observed when metabolism is rapidly quenched ( the zero time point ) . Consistent with the known rapid turnover of metabolites ( Shamir et al . , 2016 ) , the levels ( Figure 3A–D , Figure 3—figure supplement 1A–C , Figure 3—source data 1 ) and/or labeling from U-13C-glucose ( Figure 3E–H , Figure 3—figure supplement 1D–F , Figure 3—source data 1 ) , of many metabolites changed over the time required to separate cells using antibodies and/or flow cytometry . These changes indicate that metabolism is not at metabolic steady-state where levels and labeling of metabolites are stable over time and could complicate interpretation of some differential isotope labeling patterns ( Buescher et al . , 2015 ) . In fact , changes in metabolite levels and labeling may be even greater when using flow cytometry to sort cells in practice because temperature as well as factors such as mechanical stress are less easily controlled ( Binek et al . , 2019 ) . While this does not absolutely preclude an ability to gain information from metabolite measurements in sorted cells , assessment of M+3 labeling of TCA cycle intermediates in sorted cell populations from organoids or tumors may not fully portray the contribution of each cell type to the pyruvate carboxylation phenotype observed when material containing multiple cell types is analyzed . The turnover of protein and nucleic acid is slow relative to metabolites ( Shamir et al . , 2016 ) , which allows gene expression and proteomic analysis in separated cell types to better reflect the state of cells within a mixed population . Because metabolites contribute to protein , lipid , and nucleic acid biomass , and isotope-labeled nutrients can be traced into this biomass ( Gebreselassie and Antoniewicz , 2015; Ghosh et al . , 2014; Green et al . , 2016; Hosios et al . , 2016; Le et al . , 2017; Lewis et al . , 2014; Mayers et al . , 2016; Rühl et al . , 2011; Shankaran et al . , 2016; Zamboni et al . , 2005 ) , we reasoned that 13C-labeling patterns in biomass might be used to infer the contribution of glucose to different metabolic pathways within a mixed cell population relevant to pancreatic cancer . We confirmed that glucose labeling of protein was stable over the time period needed to sort cells by flow cytometry ( Figure 3I–J , Figure 3—source data 1 ) . We also confirmed that amino acids from protein hydrolysates were detectable in sorted cells from murine PDAC tumors , and were within the linear range of detection by GC-MS even when low cell numbers of less abundant cell populations were recovered from tumors ( Figure 3—figure supplement 1G–L ) . Therefore , examining 13C label in amino acids from hydrolyzed protein may be informative of the labeling of free amino acids in tumor cell subpopulations that existed prior to sorting the cells . To facilitate sorting of PSCs and PDAC cancer cells from organoid co-cultures and tumors , a LSL-tdTomato reporter allele was bred to the KP-/-C and KPC PDAC models as a source of tdTomato+ cancer cells for both organoid and tumor models . PSCs were isolated from pancreata from mice bearing a β-actin-GFP allele to enable sorting of GFP+ PSCs for labeling of the PSC population in the organoid co-culture model ( Figure 4A ) . To determine the relative contribution of 13C-glucose to M+3-labeled aspartate in cancer cells and PSCs in the organoid-fibroblast co-culture model , we exposed organoid co-cultures containing tdTomato+ cancer cells and GFP+ PSCs to U-13C-glucose for 1–4 days prior to sorting cancer cells and PSCs , and then hydrolyzed protein for amino acid analysis from each cell population . Over time , similar M+2 protein aspartate labeling was observed between the two cell types , while higher M+3 aspartate labeling was observed in cancer cells as compared to PSCs , suggesting that while the two cell types have similar labeling via reactions involving PDH , the cancer cells appear to have higher pyruvate carboxylation activity ( Figure 4B–C , Figure 4—source data 1 ) . This higher M+3 level was also reflected in the other TCA cycle-derived amino acids glutamate and proline ( Figure 4D–G , Figure 4—source data 1 ) , whereas labeling of the glucose-derived amino acids alanine and serine was not higher in cancer cells ( Figure 4—figure supplement 1A–B , Figure 4—source data 1 ) . We also exposed organoid co-cultures to U-13C-glutamine over 4 days and traced the fate of labeled carbon into protein in each cell population . Of note , we observed slightly higher labeling of aspartate from glutamine in protein in PSCs ( Figure 4—figure supplement 1C , Figure 4—figure supplement 1—source data 1 ) , matching the lower fractional labeling we observed from glucose . We did not observe other appreciable differences in fractional labeling of glutamate or proline from glutamine in protein between cancer cells and PSCs ( Figure 4—figure supplement 1D–E , Figure 4—figure supplement 1—source data 1 ) . Taken together , these data suggest a differential fate for glucose in these cell types , with increased M+3 labeling of aspartate from glucose carbon in cancer cells relative to PSCs . Because the labeling of amino acids in protein is unlikely to reach steady-state even after multiple days of labeling , one explanation for the difference in aspartate labeling from labeled glucose in the cancer cells relative to the PSCs is a higher rate of protein synthesis in the cancer cells , although this is unlikely to differentially affect only M+3 labeled species . Nevertheless , to examine this possibility , we measured protein synthesis rates in each cell type using a fluorescent protein synthesis reporter in which BFP is fused to an unstable E . coli dihydrofolate reductase ( DHFR ) domain . Upon addition of the DHFR active site ligand trimethoprim ( TMP ) , the reporter is stabilized and the rate of fluorescence accumulation reflects the synthesis rate of the fluorescent protein ( Han et al . , 2014 ) . Consistent with previous reports , this reporter produced similar results compared to an assessment of protein synthesis through incorporation rates of the aminoacyl tRNA analog puromycin ( Darnell et al . , 2018 ) when BFP accumulation after TMP addition was assayed over time in PDAC cancer cell and PSC mono-cultures and compared to cells with no TMP added as a negative control ( Figure 4—figure supplement 1F–G ) . The BFP reporter is suitable for use in sorted cells from a mixed cell system , and thus was used to assess protein synthesis rates in cancer cells and PSCs in organoid co-cultures . Interestingly , even though cancer cells and PSCs exhibited a similar protein synthesis rate in monoculture ( Figure 4—figure supplement 1F–G ) , accumulation of BFP fluorescence was slower in cancer cells compared to PSCs in 3D co-cultures ( Figure 4H ) . This argues that protein synthesis rates are higher in PSCs in organoid co-cultures , and that the higher M+3 aspartate labeling observed in cancer cells cannot be explained by a higher rate of protein synthesis in the cancer cells in this co-culture system . M+3 labeling from U-13C-glucose is often used as a surrogate for pyruvate carboxylation activity , but can also occur from multiple rounds of TCA cycling ( Alves et al . , 2015 ) . To more directly assess pyruvate carboxylation activity in additional experiments with multiple replicates , we traced 1-13C-pyruvate or 3 , 4-13C-glucose fate in organoid-PSC co-cultures . 1-13C-pyruvate or 3 , 4-13C-glucose can only label aspartate via pyruvate carboxylation , because the 13C-label is lost as carbon dioxide if pyruvate is metabolized to acetyl-coA via PDH prior to entering the TCA cycle ( Figure 4I ) . Compared to U-13C-glucose labeling , a greater difference and significantly higher M+1 aspartate labeling in protein was observed using 1-13C-pyruvate or 3 , 4-13C-glucose in sorted cancer cells compared to PSCs from organoid-PSC co-cultures , further supporting that pyruvate carboxylation activity is higher in these cells ( Figure 4J–K , Figure 4—figure supplement 1H ) . Taken together , these data argue that cancer cells within PDAC organoid-PSC co-cultures have higher pyruvate carboxylation activity than PSCs . To investigate whether PDAC cancer cells also exhibit higher pyruvate carboxylation activity in tumors in vivo , we first verified that the tdTomato fluorescence in tumors arising in KP-/-C mice bearing a LSL-tdTomato allele did not co-localize with staining for the fibroblast-specific marker alpha-smooth muscle actin ( α-SMA ) ( Figure 5A ) , but did co-localize with Cytokeratin 19 ( CK19 ) , a marker of pancreatic cancer cells ( Figure 5B ) . This verifies that tdTomato labeling can be used to isolate cancer cells from α-SMA-positive fibroblasts , and a combination of tdTomato fluorescence and an antibody for the pan-hematopoietic marker CD45 allowed efficient sorting of cancer cells , fibroblasts , and hematopoietic cells as verified by expression of relevant mRNAs using qPCR ( Figure 5C–F , Figure 5—figure supplement 1A–B ) . To label protein in PDAC tumors in vivo , autochthonous tumor-bearing mice were infused with U-13C-glucose for 24 hr ( Figure 5—figure supplement 1C–D ) , with aspartate labeling observed in protein hydrolysates from bulk tumors in this time frame ( Figure 5G ) . Cell populations were sorted from tumors , and labeling of amino acids was determined in protein hydrolysates from each cell population as well as from protein obtained from the bulk digested tumor ( unsorted ) . In agreement with labeling patterns from organoid-co-cultures , tdTomato+ cancer cells from PDAC tumors in mice had the highest M+3 protein aspartate labeling in protein , as well as higher M+2 aspartate labeling ( Figure 5H–I , Figure 5—source data 1 ) . This labeling pattern was also reflected in higher M+2 and M+3 labeling in glutamate but not in other glucose-labeled amino acids in protein ( Figure 5—figure supplement 1E–J , Figure 5—source data 1 ) . Taken together , these data are consistent with increased pyruvate carboxylation , as well as increased glucose oxidation via PDH , in the cancer cells relative to the stromal cell populations analyzed in PDAC tumors in vivo . To test whether PC is responsible for the observed pyruvate carboxylation activity and is functionally important for cells to proliferate in organoid co-cultures and tumors , PC expression was disrupted in murine PDAC cell lines , organoids , and PSCs using CRISPR/Cas9 . First , CRISPRi was used to knock down PC expression in a PDAC cancer cell line derived from KP-/-C mice ( Figure 6—figure supplement 1A; Horlbeck et al . , 2016 ) . The ratio of M+1 aspartate to M+1 pyruvate derived from 1-13C-pyruvate or 3 , 4-13C-glucose has been used as a way to approximate pyruvate carboxylation activity ( Davidson et al . , 2016 ) . As expected , PC knockdown in these PDAC cells resulted in a decrease in aspartate labeling from 1-13C-pyruvate and relative pyruvate carboxylation activity compared to control cells as assessed by the ratio of M+1 labeled aspartate to M+1 labeled pyruvate ( Figure 6—figure supplement 1B–C ) , but did not affect proliferation in culture ( Figure 6—figure supplement 1D ) . However , knockdown of PC in PDAC organoids reduced growth of these cells in organoid-PSC co-cultures ( Figure 6—figure supplement 1E–G ) . PC expression level and aspartate labeling from 1-13C-pyruvate were increased by exogenous PC expression in PDAC PC knockdown cells ( Figure 6—figure supplement 1H–J ) . When transplanted subcutaneously , PDAC cell lines with PC knockdown formed tumors that grew similarly to control cells ( Figure 6—figure supplement 1K ) ; however , PC expression was similar or increased in the tumors formed from PC knockdown cells compared to control tumors ( Figure 6—figure supplement 1L ) . These data suggest that over time , cells that grew into tumors were selected for reversal of PC knockdown and that PC is required for PDAC tumor growth in vivo even though it is dispensable in culture , as has been observed previously in lung cancer ( Davidson et al . , 2016; Fan et al . , 2009; Sellers et al . , 2015 ) . To further test the requirement for PC in PDAC tumors , we generated cancer cell clones with complete CRISPR/Cas9 disruption of PC expression ( Figure 6—figure supplement 1M–N ) . Similar to knockdown experiments , loss of PC had no effect on proliferation of PDAC cells in culture ( Figure 6A ) , whereas loss of PC reduced the growth of organoid co-cultures ( Figure 6B–C ) . CRISPR/Cas9 was also used to knockout PC in PSCs , and despite loss of PC expression and reduced pyruvate carboxylation activity ( Figure 6—figure supplement 2A–C ) , PC knockout PSCs retained the ability to enhance PDAC organoid growth or growth of PDAC cancer cells as tumors in subcutaneous transplants , although the effect was reduced compared to sgControl PSCs ( Figure 6—figure supplement 2D–F ) . Consistent with a requirement for PC expression in cancer cells to form PDAC tumors , PC-null cancer cells did not form tumors when transplanted into syngeneic mice subcutaneously or orthotopically ( Figure 6D–E ) . However , surprisingly , PC-null cancer cells still displayed M+1 aspartate labeling from 1-13C-pyruvate with similar or only a slight decrease in pyruvate carboxylation activity compared to control cells ( Figure 6F–G ) . Taken together , these data argue that loss of PC in cancer cells can impact tumor growth , but another enzyme must also contribute to pyruvate carboxylation activity in these cells . A candidate for the pyruvate carboxylation activity observed in PC-null cells is malic enzyme , since this enzyme catalyzes the interconversion of pyruvate and CO2 with malate , another 4-carbon TCA cycle intermediate . Malic enzyme is typically assumed to catalyze malate decarboxylation as a source of NADPH in cells ( Cairns et al . , 2011; Hosios and Vander Heiden , 2018 ) , but has previously been shown to be reversible and produce malate from pyruvate and CO2 in purified enzyme assays ( Ochoa et al . , 1947; Ochoa et al . , 1948 ) . Thus , we tested whether malic enzyme activity could sustain M+1 labeling of aspartate in PDAC cancer cells lacking PC by using CRISPR/Cas9 to knock out malic enzyme 1 ( ME1 ) . After knockout of both PC and ME1 in PDAC cell lines , aspartate labeling from 1-13C-pyruvate is virtually abolished , suggesting that ME1 activity can contribute to pyruvate carboxylation activity in these cells ( Figure 7A–B , Figure 7—figure supplement 1A ) . This aspartate labeling was also increased after exogenous expression of ME1 in PC and ME1 double knockout cells ( Figure 7A–B ) . We used CRISPR/Cas9 to knockout or knockdown both PC and ME1 in organoids , which also resulted in decreased M+1 aspartate labeling from 1-13C-pyruvate and decreased pyruvate carboxylation activity ( Figure 7—figure supplement 1B–D ) . We also used CRISPR/Cas9 to knockout ME1 alone in PDAC cell lines and organoids ( Figure 7—figure supplement 1E–H ) . Reduction or loss of ME1 alone in PDAC cell lines resulted in lower aspartate M+1 labeling and pyruvate carboxylation activity from 1-13C-pyruvate ( Figure 7—figure supplement 1F–K ) , further suggesting a role for ME1 in anaplerosis . This aspartate labeling was also increased after exogenous expression of ME1 in ME1 knockdown cells ( Figure 7—figure supplement 1L–N ) . Interestingly , ME1 expression in KP-/-C mouse and human PDAC tumors and organoids mimics that of PC in that it is more highly expressed in cancer cells compared to stroma , suggesting that ME1 could also contribute to the higher pyruvate carboxylation seen in cancer cells compared to PSCs ( Figure 7C–E , Figure 7—figure supplement 2A–E ) . We next assessed whether ME1 was essential for tumor and organoid growth . Similar to loss of PC , loss of ME1 had minimal effect on cancer cell proliferation in monoculture ( Figure 7F , Figure 7—figure supplement 1F ) , but reduced growth of organoid co-cultures compared to controls ( Figure 7G–H , Figure 7—figure supplement 1E ) . Transplantation of ME1-null cancer cells in vivo also resulted in reduced tumor growth ( Figure 7I ) , consistent with published data ( Son et al . , 2013 ) . Taken together , these data argue that both PC and ME1 are important enzymes for PDAC cancer cells in tumors and can contribute to the pyruvate carboxylation activity observed in pancreatic cancer . Metabolism can differ between cancer cells in culture and tumors , and understanding how nutrients are used by cancer cells in vivo has been an area of interest for developing cancer therapies . Tumor metabolic phenotypes have been assumed to reflect the metabolism of cancer cells within a tumor; however , in many tumors such as in PDAC , cancer cells are a minority cell population . Metabolic interactions between cell types have been described in normal tissues ( Bélanger et al . , 2011 ) , and some metabolic phenotypes observed in cancer cells such as increased glucose utilization are also prominent in other cell types including fibroblasts and immune cells that can be abundant in some tumors ( Lemons et al . , 2010; Vincent et al . , 2008; Zhao et al . , 2019 ) . Therefore , methods to deconvolute which cell types in a tumor are responsible for observed tissue metabolic phenotypes are needed . We find that pancreatic tumors exhibit evidence of glucose metabolism , with carboxylation of glucose-derived pyruvate being more active in cancer cells than in other tumor cell types . However , because glucose will label both pyruvate and lactate , and these nutrients can be exchanged between cell types , it cannot be concluded that the cancer cells necessarily derive TCA cycle metabolites directly from glucose in a cell autonomous manner . In fact , rapid exchange of labeled intracellular and extracellular pyruvate and lactate among cell types is likely , making it difficult to address the original cellular source of labeled TCA metabolites with these methods . Thus , while this approach could be used to understand differential pathway use between cell types , in many cases it will not be able to distinguish the exact source of carbon that labels metabolites in individual cells . Another important caveat to interpreting labeling patterns in protein or other macromolecules in cells within tissues is that labeling is unlikely to reach steady state , particularly for analysis of cells in tissues in vivo . This failure to reach steady state means that differences in label delivery or uptake could cause differences in biomass labeling even when the pathway involved in labeling is similarly active in both cell types . Thus , controlling for variables such as biomass synthesis rates between cell types can help with data interpretation . The ability to reach a pseudo-metabolic steady state facilitates interpretation of labeling data; however , this requires that both circulating nutrient levels and labeling patterns are relatively constant ( Buescher et al . , 2015; Jang et al . , 2018 ) . Glucose infusion rates and techniques can vary across studies ( Davidson et al . , 2016; Faubert et al . , 2017; Hui et al . , 2017; Ma et al . , 2019; Marin-Valencia et al . , 2012 ) , and these differences may affect whether circulating nutrient levels are constant . While this may be one reason why differences in labeling were observed across PDAC models evaluated in this study , additional factors such as tumor initiation and growth rates , cells of origin , p53 status , and different composition of cancer and stromal cells are known to exist as well ( Rosenfeldt et al . , 2013; Vennin et al . , 2019 ) . While we did not directly assess the effects of cell sorting on metabolite levels or labeling in this study , we observed that for some metabolites , their levels and/or labeling patterns are not stable over the time needed to sort cells . However , the labeling patterns of some metabolites were maintained despite changes in levels , and might still be used to derive information about cell-specific metabolism , particularly when appropriate controls and orthogonal evidence support the conclusions . We find that whether cells are grown in 2D cultures , in 3D organoid co-culture with PSCs , or as orthotopic or subcutaneous tumors impacts whether pyruvate carboxylation is important for proliferation , with the organoid and tumor models showing a similar dependency on this activity . Tumor organoid-stromal co-cultures represent a tractable model for metabolic characterization , and thus may be useful for exploration of other symbiotic metabolic relationships between pancreatic cancer cells and fibroblasts . However , while the difference in M+3 aspartate labeling seen in vivo was recapitulated by organoid-fibroblast co-cultures , other differences such as higher M+2 aspartate and glutamate labeling observed in vivo were not observed in the co-culture model . Therefore , some aspects of cell type-specific metabolism are not recapitulated even in co-culture organoid systems . We find that PC and ME1 expression in cancer cells are both important for PDAC tumor growth in vivo . A dependence on pyruvate carboxylation seems to be a characteristic of both PDAC and lung tumors in vivo that is not prominent in standard cell culture systems ( Christen et al . , 2016; Davidson et al . , 2016; Fan et al . , 2009; Hensley et al . , 2016; Sellers et al . , 2015 ) . Why this is the case is not known , but PC is an important anaplerotic pathway for the TCA cycle , contributing to biosynthesis of macromolecules such as protein , nucleotides , and lipids in cancer cells . Glucose metabolism and increased glucose uptake have been shown to be important for biosynthesis in PDAC tumors ( Santana-Codina et al . , 2018; Ying et al . , 2012 ) , but it has also been suggested that some PDAC tumors rely less on glucose for fuel and instead on alternative nutrient sources such as circulating lactate and glutamine ( Hui et al . , 2017 ) , or alanine from stromal autophagy ( Sousa et al . , 2016 ) . We did not observe differences in protein alanine labeling from glucose in either PSCs or organoids , although it remains possible the cells differentially utilize alanine acquired from a source other than glucose . For example , macropinocytosis to catabolize extracellular protein can be an important source of amino acids for cells in PDAC tumors ( Commisso et al . , 2013; Davidson et al . , 2017 ) . Nevertheless , the findings that PDAC tumors are FDG-PET positive ( Nguyen et al . , 2011; Parikh et al . , 2015; Yeh et al . , 2018 ) and that levels of glucose are depleted in tumor interstitial fluid relative to plasma in PDAC mouse models ( Sullivan et al . , 2019 ) , are consistent with glucose being consumed by at least some cell types within the tumor . Glutamine is also a source of TCA anaplerotic carbon that may contribute to biosynthesis differentially between cancer cells and stroma . Previous work has suggested that utilization of ME1 to produce pyruvate from glutamine can be important for PDAC cells to maintain redox balance , specifically via NADPH generation in vitro ( Son et al . , 2013 ) , and that glutamine can be a major contributor to TCA metabolites in PDAC tumors ( Hui et al . , 2017 ) . A potential role for malic enzyme in pyruvate carboxylation suggests use of this enzyme to produce malate could be another pathway for TCA cycle anaplerosis . Of note , this reaction would require NADPH , and may be more favored in cancer cells that exhibit a reduced redox state ( Hosios and Vander Heiden , 2018 ) . Furthermore , other pathways produce NADPH in cancer cells , including the pentose phosphate pathway , the one-carbon cycle , or isocitrate dehydrogenase ( Chen et al . , 2019 ) . We also considered phosphoenolpyruvate carboxykinase ( PEPCK ) or malic enzymes 2 and 3 as possible contributors to pyruvate carboxylation activity , although these reactions are less energetically favorable in the reverse direction in comparison to malic enzyme 1; malic enzyme 1 is cytosolic , which is thought to be a more reducing environment than the mitochondria where malic enzymes 2 and 3 are localized ( Hu et al . , 2008 ) . We did not see evidence for differential glutamine utilization in our organoid-PSC co-cultures , and PDAC tumors are resistant to glutaminase inhibitors ( Biancur et al . , 2017 ) , but further work is needed to assess how glutamine metabolism and other anaplerotic pathways might be differentially active in cancer cells and non-cancer cells in PDAC tumors . In pancreatic β-cells , PC and ME are thought to be part of a coordinated metabolic cycle that regulates insulin secretion ( Pongratz et al . , 2007 ) . In this pyruvate cycle , ME1 generates NADPH and produces pyruvate from malate in the cytosol , which can then be used by PC to generate oxaloacetate in the mitochondria ( Pongratz et al . , 2007 ) . While loss of ME activity might be expected to impact pyruvate carboxylation activity when both enzymes are present , the fact that residual pyruvate carboxylation activity is observed in the absence of PC , and that this is lost upon ME1 disruption argues that PC and ME may have redundant metabolic functions under some conditions . Surprisingly , isotope labeling in cells with loss of ME1 alone showed a larger decrease in labeling consistent with pyruvate carboxylation than was observed with PC knocked out and ME1 left intact . However , these data should not be used to conclude that flux through ME1 is higher than PC , particularly in cells where both enzymes are expressed and pyruvate cycling can occur . PC and ME1 are each essential for pancreatic tumors in vivo , despite a possible redundancy in pyruvate carboxylation activity . This may be because pyruvate cycling is important for tumor growth , or the need for anaplerosis to support tumor growth is more constrained in tumors than in cell culture . Indeed , a dependence on pyruvate carboxylation seems to be a characteristic of tumors in vivo that is not observed in culture across many cancer models ( Christen et al . , 2016; Davidson et al . , 2016; Fan et al . , 2009; Hensley et al . , 2016; Sellers et al . , 2015 ) . PC has been targeted with antisense oligonucleotides ( Kumashiro et al . , 2013 ) and relatively non-specific chemical inhibitors ( Bahl et al . , 1997; Zeczycki et al . , 2010 ) ; however , inhibiting PC may have deleterious effects on whole body metabolism by interfering with gluconeogenesis or glucose-stimulated insulin secretion . Whether malic enzyme can compensate sufficiently for PC inhibition in those tissues to allow therapeutic targeting , or if malic enzyme is a viable alternative target , remains to be determined . Nevertheless , our data suggest that stable isotope tracing into macromolecules can be utilized to deconvolute complex tracing patterns in mammalian tissues and identify increased pathway activity in a particular cell type . Understanding the metabolic similarities and differences between cancer cells and stroma within PDAC and other tumors will be important in further delineating cancer-specific dependencies . All animal studies were approved by the MIT Committee on Animal Care under protocol #0119-001-22 . For autochthonous models , LSL-KrasG12D/+; Trp53flox/flox; Pdx1-Cre; LSL-tdTomato ( KP-/-CT ) ( Bardeesy et al . , 2006 ) and LSL-KrasG12D/+; Trp53R172H/+; Pdx1-Cre; LSL-tdTomato ( KPCT ) ( Hingorani et al . , 2005 ) , mice from a mixed 129/Sv and C57Bl6/J genetic background as well as pure C57Bl6/J mice were used . C57Bl6/J mice were used for allografts . Both sexes of mice were included in experiments . Animals were housed under a 12 hr light and 12 hr dark cycle , and cohoused with littermates with ad libitum access to water and food unless otherwise stated . Infusion of U-13C-glucose ( Cambridge Isotope Laboratories ) was performed as previously described ( Davidson et al . , 2016 ) . Surgery was performed to implant a catheter into the jugular vein of animals 3–4 days prior to infusion . For 4–6 hr infusions , mice were fasted for 4 hr prior to beginning the infusion . For 24 hr infusions , mice were not fasted prior to infusion . Infusions were performed in conscious , free-moving animals for 4 or 24 hr at a rate of 30 mg/kg/min . For 6 hr infusions , each animal , regardless of body weight , was infused with a fixed volume of 300 µl of a 500 mg/ml glucose solution over 6 hr , which is an infusion rate of 0 . 4 mg/min . Tumors were either digested for FACS or rapidly frozen using a Biosqueezer ( BioSpec Products ) and stored at −80°C prior to metabolite extraction . 100 , 000 adherent cells were plated in six-well plates , or organoids and organoid-PSC co-cultures were plated on plastic coverslips ( Thermo 174985 ) in 24-well plates . The following day , the cells were washed three times with PBS and then isotope-labeled media was added for the specified length of time ( 24–72 hr ) . For U-13C-glucose or 3 , 4-13C-glucose tracing , DMEM without glucose and pyruvate was used , supplemented with 25 mM U-13C-glucose or 3 , 4-13C-glucose , 10% dialyzed FBS , and penicillin-streptomycin . For 1-13C-pyruvate tracing , DMEM with glucose and without pyruvate was used , adding 2 mM 1-13C-pyruvate and supplementing with 10% dialyzed FBS and penicillin-streptomycin . Adherent cells were washed once with ice-cold saline on ice and then extracted with a 5:3:5 ratio of ice-cold HPLC-grade methanol:water:chloroform . Mouse tissue or coverslips containing organoids and organoid-PSC co-cultures were washed once with saline prior to extraction . Tissue or matrigel domes containing the organoids and organoid-PSC co-cultures were then rapidly frozen using a Biosqueezer ( BioSpec Products ) and stored at −80°C prior to metabolite extraction . Snap frozen tissues or organoids were extracted with a 5:3:5 ratio of ice-cold HPLC-grade methanol:water:chloroform . For mouse plasma , 10 μL of plasma was extracted with 600 μL ice-cold methanol . All samples were vortexed for 10 min at 4°C followed by centrifugation for 5 min at maximum speed on a tabletop centrifuge ( ~21 , 000 xg ) at 4°C . An equal volume of the aqueous phase of each sample was then dried under nitrogen gas and frozen at −80°C until analysis . For organoid samples , two rounds of extraction were done to eliminate excess protein from matrigel . Acid hydrolysis of protein was performed as described previously ( Mayers et al . , 2016; Sullivan et al . , 2018 ) . Frozen tissue or cell pellets were boiled for 24 hr at 100°C in 500 μL ( cell pellets ) −1 mL ( tissue ) 6M HCl for amino acid analysis ( Sigma 84429 ) . 50 μL ( tissue ) −100 μL ( cell pellets ) of HCl solution was then dried under nitrogen gas while heating at 80°C . Dried hydrolysates were stored at −80°C until derivatization . Polar metabolites were analyzed as described previously ( Lewis et al . , 2014 ) . Dried free metabolite extracts were dissolved in 16 μL methoxamine ( MOX ) reagent ( ThermoFisher TS-45950 ) and incubated at 37°C for 90 min followed by addition of 20 μL N–methyl–N– ( tert–butyldimethylsilyl ) trifluoroacetamide + 1% tert–Butyldimethylchlorosilane ( Sigma 375934 ) and incubated at 60°C for 1 hr . Dried protein hydrolysates were re-dissolved in 16 μL HPLC grade pyridine ( Sigma 270407 ) prior to derivatization with 20 μL N–methyl–N– ( tert–butyldimethylsilyl ) trifluoroacetamide + 1% tert–Butyldimethylchlorosilane ( Sigma 375934 ) at 60°C for 1 hr . Following derivatization , samples were analyzed using a DB-35MS column ( Agilent Technologies ) in an Agilent 7890 gas chromatograph coupled to an Agilent 5975C mass spectrometer . Helium was used as the carrier gas at a flow rate of 1 . 2 mL/min . One microliter of sample was injected at 270°C . After injection , the GC oven was held at 100°C for 1 min and increased to 300°C at 3 . 5 °C/min . The oven was then ramped to 320°C at 20 °C/min and held for 5 min . at this 320°C . The MS system operated under electron impact ionization at 70 eV and the MS source and quadrupole were held at 230°C and 150°C , respectively . The detector was used in scanning mode , and the scanned ion range was 100–650 m/z . Data were corrected for natural isotope abundance . Cell lines were cultured in DMEM ( Corning 10–013-CV ) supplemented with 10% fetal bovine serum and penicillin-streptomycin . Cell lines were regularly tested for mycoplasma contamination using the MycoAlert Plus kit ( Lonza ) or the Mycoprobe Mycoplasma Detection Kit ( R and D Systems ) . PSCs were isolated from β-actin-GFP mice in a C57Bl6/J background ( 006567 ) as previously described ( Apte , 2011; Danai et al . , 2018 ) : 3 mL of 1 . 3 mg/mL cold collagenase P ( Sigma 11213865001 ) and 0 . 01 mg/mL DNAse ( Sigma D5025 ) in GBSS ( Sigma G9779 ) were injected into the pancreas . The tissue was then placed into 2 mL of collagenase P solution on ice . Cells were then placed in a 37°C water bath for 15 min . The digested pancreas was filtered through a 250 μm strainer and washed with GBSS with 0 . 3% BSA . A gradient was created by resuspending the cells in Nycodenz ( VWR 100356–726 ) and layering in GBSS with 0 . 3% BSA . Cells were then centrifuged at 1300 x g for 20 min at 4°C . The layer containing PSCs was removed , filtered through a 70 μm strainer , washed in GBSS with 0 . 3% BSA , and plated for cell culture in DMEM with 10% FBS and penicillin-streptomycin . PSCs were immortalized with TERT and SV40 largeT after several passages . Organoids were isolated from mice bearing PDAC tumors and cultured as previously described ( Boj et al . , 2015 ) . Tumors were minced and digested overnight with collagenase XI ( Sigma C9407 ) and dispase II ( Roche 04942078001 ) and embedded in 50 μL domes of growth factor reduced ( GFR ) matrigel ( Corning 356231 ) covered with 500 μL of complete media . Complete media consisted of Advanced DMEM/F12 ( Thermo Fisher 12634 ) containing GlutaMAX ( Thermo Fisher 35050 ) , penicillin-streptomycin , HEPES ( Thermo Fisher 15630 ) , 0 . 5 μM TGF-b inhibitor A-83–01 ( TOCRIS 2939 ) , 0 . 05 μg/mL EGF ( Thermo Fisher PMG8041 ) , 0 . 1 μg/mL FGF ( Peprotech 100–26 ) , 0 . 01 μM Gastrin I ( TOCRIS 3006 ) , 0 . 1 μg/mL Noggin ( Peprotech 250–38 ) , 10 . 5 μM Rho Kinase Inhibitor Y-27632 ( Sigma Y0503 ) , 1 . 25 mM N-Acetylcysteine ( NAC ) ( Sigma A9165 ) , 10 mM Nicotinamide ( Sigma N0636 ) , 1X B-27 supplement ( Thermo Fisher 17504 ) , and 1 μg/mL R-spondin . R-spondin was purified from 293 T cells engineered to produce it using a Protein A Antibody Purification Kit ( Sigma PURE1A ) . Organoids were grown in complete media when passaging . For organoid-PSC co-culture experiments , co-cultures were grown in DMEM without pyruvate ( Corning 10–017-CV ) supplemented with 10% dialyzed FBS and penicillin-streptomycin . Organoids were regularly tested for mycoplasma contamination using the MycoAlert Plus kit ( Lonza ) or the Mycoprobe Mycoplasma Detection Kit ( R and D Systems ) . Organoids were digested to single cells by incubating with 2 mg/mL dispase in Advanced DMEM/F12 with penicillin-streptomycin , HEPES , and GlutaMAX at 37°C for 20 min . Organoids were then triturated with a fire-polished glass pipette and enzymatically digested with 1 mL TrypLE Express ( Thermo Fisher 12605–010 ) for 10 min rotating at 37°C , followed by addition of 1 mL of dispase containing media and 10 μL of 10 mg/mL DNAse ( Sigma 4527 ) and digested rotating at 37°C for 20 min or until single cells were visible under a microscope . Cells were counted and plated in GFR matrigel at a concentration of 2000 cells/well . 50 , 000 cells were seeded in six-well plates in 2 mL DMEM with 10%FBS and penicillin-streptomycin . The next day , cells were counted for day 0 and media was changed on remaining cells . 8 mL of media was added and cells were left to proliferate for 3 days . On day 3 , cells were trypsinized and counted . Alternatively , proliferation was measured using sulforhodamine B staining as previously described ( Vichai and Kirtikara , 2006 ) . Cells were fixed on day 0 and day 3 with 500 µl of 10% trichloroacetic acid ( Sigma T9159 ) in 1 mL media and incubated at 4°C for at least 1 hr . Plates were washed under running water and cells were stained with 1 mL sulforhodamine B ( Sigma 230162 ) and incubated at room temperature for 30 min . Dye was removed and cells were washed three times with 1% acetic acid . Plates were then dried and 1 mL of 10 mM Tris pH 10 . 5 was added to each well to solubilize the dye . 100 µL of each sample was then transferred to a 96-well plate and absorbance was measured at 510 nm on a microplate reader . A fluorescent reporter in which BFP is fused to an unstable E . coli dihydrofolate reductase ( DHFR ) degron domain which is stabilized by trimethoprim ( Han et al . , 2014 ) was used to determine global protein synthesis rate as previously described ( Darnell et al . , 2018 ) . Briefly , PDAC/PSC cell lines and PDAC organoids expressing the reporter were generated by lentiviral transduction and puromycin selection followed by flow cytometry-based sorting for populations that were BFP-positive after TMP addition for 24–48 hr . For each experiment , the reporter protein was stabilized upon addition of 10 uM trimethoprim ( TMP ) and fluorescence accumulation was measured in cells or organoids by flow cytometry over several time points within 12 hr of TMP addition . Data were normalized to no TMP controls . Puromycin incorporation assays were performed as previously described ( Schmidt et al . , 2009 ) . 10 µg/mL puromycin was spiked into the medium of cells grown in 6 cm plates . Plates were kept at 37°C for indicated pulse times ( spanning 2 . 5 to 20 min ) . As a negative control , 100 µg/mL cycloheximide was added to a plate of cells for 45 min before the addition of puromycin . At the end of the pulse , plates were washed once with ice cold PBS on ice and flash frozen in liquid nitrogen . Cells were harvested from frozen plates by scraping into RIPA buffer containing cOmplete Mini EDTA-free Protease Inhibitor Cocktail ( Roche 11836170001 ) and PhosSTOP Phosphatase Inhibitor Cocktail Tablets ( Roche 04906845001 ) and protein concentration was quantified using the Pierce BCA Protein Assay Kit ( Pierce 23225 ) . 2 µL of lysate ( approximately 2 µg ) was spotted directly onto 0 . 2 µm nitrocellulose membranes and blotted with primary antibodies against puromycin ( Sigma MABE343 1:25 , 000 dilution ) and vinculin ( Abcam ab18058 , 1:1000 dilution ) as a control . CRISPRi knockdown cell lines of PC and ME1 were generated by transfecting cells expressing modified dCas9-KRAB fusion protein , as previously described ( Horlbeck et al . , 2016 ) . The target sequences used for PC sgRNAs were ( PC1- GCGGCGGCCACGGCTAGAGG , PC2- GTGGAGGCAGGGGCCGTCAG ) , the sequence for non-targeting control was GCGACTAGCGCCATGAGCGG , and the target sequence of ME1 sgRNA was GCCGCAGTGGCCTCCCGGGT . After transfection , cells were selected under 5 ug/ml puromycin . Rescue of CRISPRi knockdown cell lines of PC was performed by re-expressing the cDNA of the rescued gene under a CMV promoter using a custom lentiviral construct generated on VectorBuilder and selected in 500 ug/ml blasticidin . CRISPR knockout cell lines for PC and ME1 were generated using the LentiCRISPRv2 system , as previously described ( Sanjana et al . , 2014 ) , with guides against the target sequence 5’ CGGCATGCGGGTCGTGCATA 3’ for PC and 5’ GTTTGGCATTCCGGAAGCCA 3’ for ME1 . After transfection , cells were selected under 5 µg/ml of puromycin , single-cell cloned , and knockout validation performed using western blot . For organoids , the same vector systems and guide sequences were used . Organoids were transfected with concentrated virus by spinfection for 45 min at room temperature . For CRISPR knockout organoids , organoid cultures were selected under 5 µg/ml of puromycin , digested to single cells , and then single organoids were picked , expanded , and validated using western blot . Double knockout cell lines for PC and ME1 were generated using pUSPmNG ( U6 sgRNA PGK with mNeonGreen , Li et al . , 2019 ) incorporated into cells via electroporation ( Amaxa VPI-1005 ) , and selected by FACS using NeonGreen expression . For organoids , double knockout organoids were generated using the LentiCRISPRv2 system and spinfected for 45 min at room temperature . After transfection , cells were selected under 500 ug/ml of blasticidin , digested to single cells , and then single organoids were picked , expanded , and validated using western blot . Tumors were dissected , minced , and digested rotating for 30 min at 37°C with 1 mg/mL Collagenase I ( Worthington Biochemical LS004194 ) , 3 mg/mL Dispase II ( Roche 04942078001 ) , and 0 . 1 mg/mL DNase I ( Sigma D4527 ) in PBS . Following digestion , cells were incubated with EDTA to 10 mM at room temperature for 5 min . Cells were then filtered through a 70 μm strainer and washed twice with PBS . Single cell suspensions were resuspended in flow cytometry staining buffer ( Thermo Fisher 00-4222-57 ) and first stained with 10 μL of CD16/CD32 monoclonal antibody ( Thermo Fisher 14-0161-82 ) for 15 min to block Fc receptors and then stained using with antibodies to CD45-APC-Cy7 ( BD 557659 ) at 1:100 dilution followed by SYTOX Red Dead Cell Stain ( Life Technologies S34859 ) at 1:1000 dilution to visualize dead cells . All antibodies were incubated for 15–20 min on ice and then washed . Cell sorting was performed with a BD FACS Aria and data was analyzed with FlowJo Software ( BD ) . 100 , 000 PDAC cells or 100 , 000 PDAC cells plus 100 , 000 PSCs in 100 μL PBS were transplanted subcutaneously into the flanks of C57BL/6J mice ( 000664 ) . Tumors were measured using calipers and volume was calculated using the formula V = ( π/6 ) ( L*W2 ) . For orthotopic transplants , 100 , 000 PDAC cells in 50 μL PBS were transplanted into the pancreas of C57BL/6J mice ( 000664 ) as previously described ( Mayers et al . , 2014 ) . Tumors were fixed in 4% paraformaldehyde ( PFA ) rotating overnight at 4°C followed by incubation in 30% sucrose in PBS rotating overnight at 4°C . Tumors were then embedded in optimal cutting temperature ( OCT ) compound and stored at −80°C until sectioning . Sections were stained with antibodies against α-SMA ( Sigma F3777 , 1:500 dilution ) , CK19 ( Abcam ab133496 1:100 ) , and tdTomato ( Rockland 600-401-379 , 1:500 dilution ) using DAPI as a nuclear stain . Sections from formalin fixed paraffin embedded mouse tissue or a human pancreatic cancer tissue microarray ( Biomax PA961e ) were stained with antibodies against PC ( Santa Cruz sc-67021 , 1:50 dilution ) or ME1 ( Proteintech 16619–1-AP , 1:200 dilution ) . The human pancreatic cancer tissue microarray was scored independently by a pathologist ( O . H . Y . ) , and assigned scores of 0–4 for both staining intensity and percent of cells positive for expression of the indicated protein . RNA was isolated from cells using the RNAqueous-Micro Total RNA Isolation Kit ( Life Technologies AM1931 ) and cDNA was made using the iScript cDNA Synthesis Kit ( Bio-Rad Laboratories 1708890 ) . qPCR reactions were performed using SYBR Green Master Mix ( Sigma L6544 ) and primers for pyruvate carboxylase ( Forward: 5’- GGG ATG CCC ACC AGT CAC T −3’ , Reverse: 5’- CAT AGG GCG CAA TCT TTT TGA −3’ ) , malic enzyme 1 ( Forward: 5’- TGT GGG AAC AGA AAA TGA GGA GTT −3’ , Reverse: 5’- TCA TCC AGG AAG GCG TCA TAC T −3’ ) , tdTomato ( Forward: 5’- AGC AAG GGC GAG GAG GTC ATC −3’ Reverse: 5’- CCT TGG AGC CGT ACA TGA ACT GG −3’ ) , α-sma ( Forward: 5’- TCC CTG GAG AAG AGC TAC GAA −3’ Reverse: 5’- TAT AGG TGG TTT CGT GGA TGC C −3’ ) , vimentin ( Forward: 5’- GTA CCG GAG ACA GGT GCA GT- 3’ , Reverse: 5’- TTC TCT TCC ATC TCA CGC ATC −3’ ) , Trp53 ( Forward: 5’- CTC TCC CCC GCA AAA GAA AAA −3’ , Reverse: 5’- CGG AAC ATC TCG AAG CGT TTA −3’ ) , CD3ε ( Forward: 5’- ATG CGG TGG AAC ACT TTC TGG −3’ , Reverse: 5’- GCA CGT CAA CTC TAC ACT GGT −3’ ) , F4/80 ( Forward: 5’- TGA CTC ACC TTG TGG TCC TAA −3’ , Reverse: 5’- CTT CCC AGA ATC CAG TCT TTC C −3’ ) , or E-Cadherin ( Forward: 5’- GCT CTC ATC ATC GCC ACA G- 3’ , Reverse: 5’- GAT GGG AGC GTT GTC ATT G- 3’ ) using 18S ( Forward: 5’- CGC TTC CTT ACC TGG TTG AT −3’ , Reverse: GAG CGA CCA AAG GAA CCA TA −3’ ) or 36B4 ( Forward: 5’- TCC AGG CTT TGG GCA TCA −3’ , Reverse: 5’- CTT TAT CAG CTG CAC ATC ACT CAG A −3’ ) as controls . Primer sequences are also included in Supplementary file 1 . Cell lines were washed with ice-cold PBS and scraped into RIPA buffer containing cOmplete Mini EDTA-free Protease Inhibitor Cocktail ( Roche 11836170001 ) and PhosSTOP Phosphatase Inhibitor Cocktail Tablets ( Roche 04906845001 ) . Lysates were then rotated at 4°C for 20 min and centrifuged for 5 min at max speed in a tabletop centrifuge at 4°C . Organoids were resuspended in ice-cold PBS containing cOmplete Mini EDTA-free Protease Inhibitor Cocktail and PhosSTOP Phosphatase Inhibitor Cocktail Tablets ( PBS-PPI ) . Organoids were then centrifuged at 3000xg for 3 min at 4°C and washed two times in ice-cold PBS-PPI . Cell pellets were resuspended in TNET buffer ( 1% Triton X-100 , 150 mM NaCl , 5 mM EDTA , and 50 mM Tris ph 7 . 5 ) containing cOmplete Mini EDTA-free Protease Inhibitor Cocktail and PhosSTOP Phosphatase Inhibitor Cocktail Tablets , incubated on ice for 10 min , and passed through a 26 gauge needle three times . Lysates were centrifuged for 10 min at max speed in a tabletop centrifuge at 4°C . Protein concentration was quantified using the Pierce BCA Protein Assay Kit ( Pierce 23225 ) . Western blots were performed using primary antibodies against PC ( Santa Cruz sc-271493 , 1:100 dilution ) , ME1 ( Proteintech 16619–1-AP , 1:250 dilution ) , or β-actin ( Cell Signaling Technologies 8457 , 1:10 , 000 dilution ) . GraphPad Prism software was used for statistical analysis . All statistical information is described in the figure legends .	Tumors contain a mixture of many different types of cells , including cancer cells and non-cancer cells . The interactions between these two groups of cells affect how the cancer cells use nutrients , which , in turn , affects how fast these cells grow and divide . Furthermore , different cell types may use nutrients in diverse ways to make other molecules – known as metabolites – that the cell needs to survive . Fibroblasts are a subset of non-cancer cells that are typically found in tumors and can help them form . Separating fibroblasts from cancer cells in a tumor takes a lot longer than the chemical reactions in each cell of the tumor that produce and use up nutrients , also known as the cell’s metabolism . Therefore , measuring the levels of glucose ( the sugar that is the main energy source for cells ) and other metabolites in each tumor cell after separating them does not necessarily provide accurate information about the tumor cell’s metabolism . This makes it difficult to study how cancer cells and fibroblasts use nutrients differently . Lau et al . have developed a strategy to study the metabolism of cancer cells and fibroblasts in tumors . Mice with tumors in their pancreas were provided glucose that had been labelled using biochemical techniques . As expected , when the cell processed the glucose , the label was transferred into metabolites that got used up very quickly . But the label also became incorporated into larger , more stable molecules , such as proteins . Unlike the small metabolites , these larger molecules do not change in the time it takes to separate the cancer cells from the fibroblasts . Lau et al . sorted cells from whole pancreatic tumors and analyzed large , stable molecules that can incorporate the label from glucose in cancer cells and fibroblasts . The experiments showed that , in cancer cells , these molecules were more likely to have labeling patterns that are characteristic of two specific enzymes called pyruvate carboxylase and malic enzyme 1 . This suggests that these enzymes are more active in cancer cells . Lau et al . also found that pancreatic cancer cells needed these two enzymes to metabolize glucose and to grow into large tumors . Pancreatic cancer is one of the most lethal cancers and current therapies offer limited benefit to many patients . Therefore , it is important to develop new drugs to treat this disease . Understanding how cancer cells and non-cancer cells in pancreatic tumors use nutrients differently is important for developing drugs that only target cancer cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2020	Dissecting cell-type-specific metabolism in pancreatic ductal adenocarcinoma
Motor-skill learning can be accompanied by both increases and decreases in brain activity . Increases may indicate neural recruitment , while decreases may imply that a region became unimportant or developed a more efficient representation of the skill . These overlapping mechanisms make interpreting learning-related changes of spatially averaged activity difficult . Here we show that motor-skill acquisition is associated with the emergence of highly distinguishable activity patterns for trained movement sequences , in the absence of average activity increases . During functional magnetic resonance imaging , participants produced either four trained or four untrained finger sequences . Using multivariate pattern analysis , both untrained and trained sequences could be discriminated in primary and secondary motor areas . However , trained sequences were classified more reliably , especially in the supplementary motor area . Our results indicate skill learning leads to the development of specialized neuronal circuits , which allow the execution of fast and accurate sequential movements without average increases in brain activity . The human brain has a remarkable ability to learn complex motor skills . However , the neural changes that underlie this ability remain largely unknown . Although functional magnetic resonance imaging ( fMRI ) studies have shown that learning can lead to both activity increases and decreases in primary and secondary motor areas ( for reviews , see Dayan and Cohen , 2011; Penhune and Steele , 2012; Hardwick et al . , 2013 ) , these average activity changes remain hard to interpret . On the one hand , motor skill acquisition may lead to increased neural recruitment for trained behaviors , thereby increasing average activation ( Grafton et al . , 1995; Karni et al . , 1995; Hazeltine et al . , 1997; Floyer-Lea and Matthews , 2005; Lehericy et al . , 2005; Penhune and Doyon , 2002; Penhune and Doyon , 2005 ) . Alternatively , learning may result in the development of representations that produce the same behavior with higher neural efficiency , thereby reducing activity ( Jenkins et al . , 1994; Toni et al . , 1998; Ungerleider et al . , 2002; Poldrack et al . , 2005; Xiong et al . , 2009; Ma et al . , 2010; Penhune and Doyon , 2005 ) . Finally , motor practice may induce simultaneous signal increases ( due to increased neural recruitment ) and signal decreases ( due to more efficient encoding ) , making motor learning difficult to detect using traditional fMRI paradigms ( Steele and Penhune , 2010 ) . Here we test the idea that the learning of fast motor sequences , a commonly-used task in the study of skill learning , causes the development of more distinct cortical activation patterns associated with each individual sequence , independent of average activity changes . Consider the neural activation patterns for two different finger sequences in areas with different types of movement representations ( Figure 1A–C ) . A region that controls elementary hand movements independent of their sequential context may comprise five separate neuronal populations ( Figure 1A ) , each of which a dynamical system ( Churchland et al . , 2012 ) , whose activation causes the production of an individual finger press . Because each sequence contains each finger once , the two sequences will activate the same cortical patches , albeit in a different temporal order . Due to the low temporal resolution of fMRI , in particularly when averaging , as done here , over three executions of the same sequence , the two sequences would be therefore associated with identical activity patterns . 10 . 7554/eLife . 00801 . 003Figure 1 . Hypothetical learning-related changes in activity patterns associated with two sequences of five finger-presses . ( A ) If a region consists of units that are preferentially activated for single finger-presses , both sequence A ( blue ) and sequence B ( red ) activate the same units in a different temporal order . ( B ) As a region develops units that preferentially encode specific finger transitions , sequence A and B will activate partly separate and partly overlapping components of the network . ( C ) In a network that is highly specialized for specific multiple-finger transitions , the two sequences activate independent parts of the network . ( D ) In the experimental paradigm , a sequence was cued , and then executed three times from memory; fMRI activity was averaged over instruction and execution phase . DOI: http://dx . doi . org/10 . 7554/eLife . 00801 . 003 Single cell recording studies have demonstrated that supplementary motor area ( SMA; Tanji and Shima , 1994 ) and primary motor cortex ( M1; Matsuzaka et al . , 2007 ) are sensitive to the sequential context and that individual neurons are preferentially active for transitions between specific movements . The development of such neuronal populations with training is thought to underlie the faster production of trained sequences ( Matsuzaka et al . , 2007 ) . Importantly , the existence of such units would cause the two sequences to rely on partly separate neuronal populations , that is they should be associated with slightly different activity patterns ( Figure 1B ) . After prolonged training , some neurons even appear to code for longer fragments , firing preferentially at the beginning of specific sequences ( Tanji and Shima , 1994 ) . Areas containing many of such units would therefore activate even more distinct neuronal populations for particular trained sequences ( Figure 1C ) . This idea therefore predicts that trained sequence should be associated with the activation of more distinct neuronal populations than comparable untrained sequences . If these neuronal populations differ sufficiently in the spatial distribution of their presynaptic activity , then each sequence should be associated with a distinct spatial activity pattern that may be detectable using fMRI . We tested this idea using ‘multi-voxel pattern analysis’ ( MVPA ) , which detects differences in voxel-by-voxel activity patterns in an area of cortex , even if these patterns are highly overlapping and defined by neuronal differences on a small spatial scale ( Kamitani and Tong , 2005; Swisher et al . , 2010; Freeman et al . , 2011 ) . Participants were trained for 4 days to produce four different movement sequences with their left hand . After training , participants underwent two fMRI scans , performing either four trained or four untrained sequences . Using MVPA , we tested whether activity patterns of trained sequences could be discriminated from each other more easily than could untrained sequences , and how the representational structure changes with learning . Movement sequences were instructed using a string of five numbers , indicating from left to right the digits to be pressed with one referring to the thumb and five to the little finger ( Figure 1D ) . Each sequence consisted of the same five isometric finger presses in a different order . Participants memorized the sequence and then executed it once as fast as possible triggered by a go-cue . This execution was repeated either 5 ( training ) or 3 times ( scanning ) . Initially , participants executed the sequences slowly and deliberately with pauses between individual presses ( Figure 2A ) . After 4 days of training on four selected sequences with the left hand , movement times ( MTs; Figure 2C , blue line ) reduced by approximately half . At the end of training individual finger presses overlapped considerably ( Figure 2B ) . 10 . 7554/eLife . 00801 . 004Figure 2 . Behavioral consequences of sequence learning . ( A ) Before learning , the finger sequence was executed in a slow , deliberate fashion . The force traces of one exemplary trial ( sequence: 3-2-5-1-4 ) are shown . ( B ) After training , the same sequence is produced much faster , with individual finger presses overlapping . ( C ) Group-average MT for the left hand ( blue line ) reduces during training . In the pre- and post-test , the left ( blue ) and right ( red ) hand was tested on trained ( filled circle ) and untrained ( empty square ) sequences . The results show general learning ( reduction in MT for all conditions ) , and sequence- and limb-specific learning ( stronger reduction for the trained sequences on the left hand ) . Stars indicate significant differences after correction for pre-test differences . DOI: http://dx . doi . org/10 . 7554/eLife . 00801 . 004 To assess generalization , participants were probed before and after training on both the trained sequences and on four novel , untrained sequences . The results show that one aspect of the acquired skill was sequence-specific: the left-hand MTs were 237 ms ( SE: ±42 ms ) faster for trained than for untrained sequences . This difference was highly significant , also when correcting for small difference between the sequences at pre-test , t ( 14 ) = 5 . 749 , p<0 . 0001 . MTs for the untrained sequences dropped by 796 ms ( ±97 ms ) compared to pre-test , t ( 15 ) = 8 . 23 , p<0 . 001 , suggesting that a considerable part of the acquired skill was general . For example , participants may have become faster at movement transitions between specific finger pairs . This ability would also enhance the production of untrained sequences , which shared 59 . 6% of digit transitions with the trained sequences . Therefore , the comparison of neural representations of trained and untrained sequences after learning will reveal correlates of the sequence-specific rather than general skill acquired during learning . We also tested the mirror-symmetric versions of the trained sequences on the untrained , right hand to determine the degree of intermanual transfer . The right hand showed a pre- to post-test drop of 570 ms ( ±80 ms ) for untrained sequences , t ( 15 ) = −7 . 13 , p<0 . 001 , but also a sequence-specific advantage of 95 ms ( ±31 . 45 ms ) , again significant after correcting for pre-test differences , t ( 14 ) = 2 . 802 , p=0 . 014 . Thus , participants acquired a sequence-specific skill representation that also enhanced performance of the untrained hand . Participants were scanned twice between the end of training and the post-test ( Figure 2C ) . During one session participants performed the four trained sequences , and during the other four untrained sequences , in both cases with their left hand . The order of scans was counterbalanced between participants . Within each imaging run , the sequences were executed in pseudo-random order , with each trial lasting 13 . 5 s ( Figure 1D ) . This design allowed us to measure repeatedly the activity pattern for each individual sequence . Using traditional univariate analysis , we first determined the regions that showed sequence-related activation increases compared to rest , average over the four sequences . As expected , we found significant activation ( Figure 3A ) in contralateral primary motor ( M1 ) and sensory cortices ( S1 ) . Additional bilateral activation was found in secondary motor areas , including dorsal and ventral premotor cortex ( PMd , PMv ) , supplementary motor areas ( SMA/pre-SMA ) , in regions on the medial bank of the intraparietal sulcus ( IPS ) , and in the occipital-parietal junction ( OPJ ) ( Culham and Valyear , 2006 ) . All activations were highly significant after correction for multiple tests . The pronounced ipsilateral activation during non-dominant hand movements is consistent with the suggested special role of the left hemisphere in complex movements ( Verstynen et al . , 2005 ) . 10 . 7554/eLife . 00801 . 005Figure 3 . Neural differences between trained and untrained sequences . ( A ) Percent signal change compared to rest displayed on an inflated lateral surface of the left and right hemisphere , and on the superior aspects of the medial surfaces ( insets ) . Maps show group-averaged data thresholded at 0 . 5% , superimposed for trained ( red ) and untrained ( blue ) sequences . Purple areas are activated for both . Figure 3—figure supplement 1 shows the maps separately . CS , central sulcus; PoCS , post central sulcus; SF , Superior frontal sulcus; CinS , cingulate sulcus; IPS , intraparietal sulcus . ( B ) Direct statistical contrast ( t-values ) of trained sequences against untrained sequences , thresholded at t ( 15 ) > 3 . 39 , p<0 . 002 . Red areas indicate higher values for trained , blue areas higher values for untrained sequences . ( C ) Group-averaged classification accuracy maps ( threshold at 45% correct , Z = 2 . 57 , p<0 . 005 ) indicate regions in which the four sequences are associated with significantly different local patterns of activity . ( D ) Direct statistical contrast for classification accuracy , displayed as in B . DOI: http://dx . doi . org/10 . 7554/eLife . 00801 . 00510 . 7554/eLife . 00801 . 006Figure 3—figure supplement 1 . Separate activity and accuracy maps for trained and untrained sequences . ( A and C ) Group-averaged percent signal change ( all four sequences vs rest ) for ( A ) untrained and ( C ) trained sequences . ( B and D ) Group-average classification accuracy for distinguishing between the four untrained ( B ) and between the four trained ( D ) sequences . Guessing rate is 25% . DOI: http://dx . doi . org/10 . 7554/eLife . 00801 . 006 We then compared the average activity for trained and untrained sequences . Our task instructions regarding speed were designed such that the error rate for the two sequence types during scanning was exactly matched ( see methods ) , therefore equating difficulty and attentional demands . As a consequence , trained sequences were produced at a slightly faster pace . In addition , participants executed the trained sequences with higher peak forces than the untrained sequences ( see Table 1 ) . 10 . 7554/eLife . 00801 . 007Table 1 . Behavioral performance variables during the fMRI sessions for the trained and untrained sequences . A paired t-test for the difference between sessions is reportedDOI: http://dx . doi . org/10 . 7554/eLife . 00801 . 007TrainedUntrainedt ( 15 ) pMT ( ms ) 1209 ( 297 ) 1341 ( 286 ) −3 . 370 . 004Force ( N ) 4 . 44 ( 0 . 63 ) 4 . 0 ( 0 . 49 ) 3 . 090 . 007Error rate ( % ) 12 . 26 ( 6 . 75 ) 12 . 22 ( 4 . 57 ) 0 . 020 . 986ACC ( % ) 58 . 98 ( 17 . 86 ) 60 . 55 ( 11 . 11 ) −0 . 350 . 732MT is the total movement time in ms . The Force is the maximal force produced for each finger , averaged across the fingers . Error rate ( % ) indicates the percentage of trials containing an incorrect finger press . Accuracy is the classification obtained when distinguishing the four trained or untrained sequences based on MT , force , and error rate . Despite these faster and harder presses , trained sequences were not associated with more activity . Rather , a direct contrast only revealed areas with less activity for trained vs untrained sequences ( Figure 3B and Table 2 ) . Lower activity for trained sequences was found in both hemispheres in PMd and in areas along the IPS . This may indicate that these regions were less involved in the production of trained sequences , or that they were equally involved , but encoded trained sequences more efficiently . Interestingly , the signal decreases ( averaged over all fronto-parietal motor regions ) were more pronounced in the left than in the right hemisphere , t ( 15 ) = 2 . 689 , p=0 . 017 . No cortical or subcortical area ( the cerebellum was not covered ) showed significantly more activity during the production of the trained compared to untrained sequences . 10 . 7554/eLife . 00801 . 008Table 2 . Cortical areas showing significantly less average activation for trained than untrained sequences . The opposite contrast did not result in any significant areasDOI: http://dx . doi . org/10 . 7554/eLife . 00801 . 008AreaArea ( cm2 ) PclusterPeak t ( 15 ) MNIXYZLeft hemisphere PMd7 . 72<0 . 0026 . 02−27−148 OPJ1 . 43<0 . 0026 . 74−10−6759 IPS0 . 160 . 0207 . 55−42−4340Right hemisphere PMd0 . 140 . 0204 . 25371651 IPS0 . 130 . 0285 . 2819−4961 IPS0 . 27<0 . 0025 . 4829−5249Table shows the result of a surface-based random effects analysis ( N = 16 ) . The uncorrected threshold was p<0 . 002 , t ( 15 ) > 3 . 39 , and Pcluster is the p-value corrected for multiple comparisons over the whole cortical surface using the area of the cluster ( Worsley et al . , 1996 ) . The coordinates reflect the location of the cluster’s peak in MNI space . We predicted that in regions involved in the production of motor sequences , individual movement sequences would be associated with distinct patterns of cortical activity . We tested this idea using a surface-based searchlight ( see ‘Materials and methods’ ) in which we sequentially selected small areas of neocortex ( circular regions of ∼10 mm diameter ) and examined classification accuracy in each . A linear classifier determined whether the spatial fMRI activation patterns in each area reliably differed between the four trained sequences—and between the four untrained sequences . We then assigned the cross-validated classification accuracy to the center of the region , thereby building a map of sequence representation across the whole cortical surface . The resultant classification map shows a widespread above-chance accuracy , even for untrained sequences ( Figure 3C , blue/purple , Figure 3—figure supplement 1 ) . Classification was best in right M1 , bilaterally in PMd , and on the medial bank of the IPS , where group-averaged accuracies reached ∼55% ( chance performance was 25% ) . We also found significant classification in PMv , SMA , and pre-SMA , each of which was highly significant after correcting for multiple tests , even though the classification accuracies were somewhat lower in these regions . Does above-chance classification accuracy imply that different sequences were represented as distinct spatial activation patterns as hypothesized ( Figure 1B , C ) ? One may argue that a region that contains neuronal populations that encode single finger movements ( Figure 1A ) may have been activated with slightly different time-courses for the different sequences , which may have been picked up by the classifier . This explanation , however , is unlikely for two reasons: first , the spatial distribution of classification accuracy for sequences differs substantially from that for single finger movements ( Diedrichsen et al . , 2012 ) , especially in the ipsilateral , left hemisphere ( Figure 4A ) . Here , single finger movements are represented close to the central sulcus , whereas sequences are represented more prominently in premotor and parietal areas . Secondly , to test the idea of different temporal activation profiles directly , we defined six bilateral regions of interest ( ROIs; see Figure 5A and ‘Materials and methods’ ) . Within each region , we identified the main components of the temporal response ( see ‘Materials and methods’ ) . The first component yielded the canonical activation profile , while the additional components described temporal variations around this mean time course ( Figure 4B ) . Classification accuracy was highest when using the weight of the first temporal component for each voxel as input to the classifier . When we added the voxel-by-voxel weights for any further temporal component , classification accuracy reduced markedly , indicating that these components did not carry any additional information ( Figure 4C ) . 10 . 7554/eLife . 00801 . 009Figure 4 . Temporal aspects of sequence representations . ( A ) Spatial distribution of classification accuracy in the left hemisphere for trained left-hand sequences ( red ) differs from left-hand single digit movements ( dashed , Diedrichsen et al . , 2012 ) . Shown is a cross-section through the surface map , from the rostral end of dorsal premotor cortex to the posterior superior parietal cortex . The lower curve indicates the average sulcal depth , showing the location of the central sulcus ( CS ) , postcentral sulcus ( PoCS ) and precentral sulcus ( PreCS ) . ( B ) The four most informative temporal components of the BOLD response , shown exemplary for right M1 . ( C ) Classification accuracy in three ROIs , using either only the first temporal component , or using the first and one additional temporal component . Adding further temporal components does not improve classification accuracy . ( D ) Normalized time course of average activation across the time course of a trial for six bilateral ROIs . Length of instruction and execution phase are indicated as gray bars . ( E ) Normalized classification accuracy over the time course of the trial . DOI: http://dx . doi . org/10 . 7554/eLife . 00801 . 00910 . 7554/eLife . 00801 . 010Figure 5 . Spatial dimensionality analysis . ( A ) Hypothetical distribution of activity patterns in the space of spatial pattern components . One set of activity patterns ( blue dots ) may differ mostly in the intensity of a common pattern component , and should be distinguished relatively well along this single component . Another set ( red dots ) may consist of four unique activity patterns and should therefore differ also along the second and third pattern component . ( B and C ) Classification accuracy in right PMd , using 1 , 2 , or 3 pattern components ( red line ) for untrained ( B ) and trained ( C ) sequences . Expected accuracies are derived from simulations using patterns randomly spaced in 1–3 dimensional space ( light-dark gray ) and evenly spaced in three dimensions ( dashed line ) . Each simulation matched the data for the accuracy of the one-dimensional classifier ( Diedrichsen et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00801 . 010 The temporal analysis also allowed us to determine how activation and classification accuracy evolved over the course of a trial ( Figure 4D , E ) . While the experimental was not designed to distinguish between instruction- and execution related activity , two insights can be gathered from the analysis . First , classification accuracy for S1 was already above chance for the second TR , which was acquired 2 . 7–5 . 4 s after the onset of the number cue , t ( 15 ) = 3 . 335 , p=0 . 002 . All other regions showed significant classification accuracy at the third TR , all t ( 15 ) > 5 . 34 , p<0 . 0001 . Thus , it is likely that the informative patterns were caused at least partly by instruction-related activity . Secondly , even areas such as OPJ and PMd exhibited sustained classification accuracy over the whole trial . Thus , these areas did not only represent the string of numbers presented on the screen; rather they exhibited a sustained representation of the sequence during the execution phase , during which the number cue was not visible anymore . Importantly , our hypothesis predicted that each sequence would be associated with a unique spatial activation pattern . To investigate this directly , we employed a newly developed method that consists of a set of classifiers , each using a different number of informative spatial pattern components ( Diedrichsen et al . , 2013 ) . To illustrate this method , assume that each sequence activated a unique set of voxels . To describe the differences between these four patterns , one would need three linear components ( or contrasts ) , each of which would capture an equivalent amount of variance . Thus , the patterns would be distributed evenly in the space spanned by the three pattern components ( Figure 5A , red dots ) . Consequently , classification accuracy should be highest when taking all three components into account . In contrast , if sequences could be distinguished because they differed on a single variable ( e . g . , difficulty ) , then they should be associated with a single activity pattern that is scaled in intensity by different amounts . The patterns would therefore mostly differ along a single pattern component ( Figure 5A , blue dots ) . In this case , classification accuracy should be highest when using only the most informative spatial component ( one-dimensional simulation , Figure 5B , C ) . While we have shown that this is indeed the case when the same movement is executed at different force levels ( Diedrichsen et al . , 2013 ) , we found here that all sequence representations showed the highest accuracy when using all three available linear components ( Figure 5B , C , red line ) . This was the case in all ROIs and for both untrained sequences ( all t ( 15 ) > 2 . 496 , p<0 . 025 ) and trained sequence ( all t ( 15 ) > 5 . 261 , p<0 . 001 ) . Thus , we can conclude that sequences are encoded in unique spatial activation patterns , rather than in differently scaled versions of a single pattern . Our central prediction was that the distinctiveness of cortical activation patterns , measured by classification accuracy , should increase with training . This is indeed what we found . We first tested this idea globally , averaging over all fronto-parietal cortical regions . For trained sequences , classification accuracy ( Figure 3C , red/purple , Figure 3—figure supplement 1 ) reached 60%—and was on average significantly higher for trained ( 38 . 4% , ±1 . 39% ) than for untrained sequences ( 34 . 2% , ±1 . 76% ) , t ( 15 ) = 2 . 203 , p=0 . 022 . Furthermore , the cortical surface area which encoded the sequences better than chance ( within subject threshold: acc > 37 . 5% , Z > 1 . 64 ) , increased from 35 . 12 cm2 ( ±5 . 38 cm2 ) to 46 . 93 cm2 ( ±4 . 65 cm2 ) , t ( 15 ) = 1 . 963 , p=0 . 034 . Finally , we also conducted a map-wise comparison between trained and untrained sequences for the classifier using all three spatial components ( Figure 3D ) . After correcting for multiple comparisons over the whole cortical surface , only the increase in left SMA/pre-SMA was significant ( uncorrected threshold p=0 . 002 , t ( 15 ) > 3 . 39; p-corrected < 0 . 012; clustersize = 0 . 19 cm2 ) . In this area , the accuracy increased from 36% to 46% ( Figure 5C ) . The dimensionality analysis ( Figure 5; Diedrichsen et al . , 2013 ) allows further insight into the representational changes that occur with training . Specifically , by looking at how the sequence-specific patterns are distributed in the space of pattern components , we can determine whether sequences activated separate or overlapping sets of voxels . Our data indicates that untrained sequences were associated with unique , but basically random activity patterns . This means that any two activation patterns exhibit some non-overlapping features , but also share a specific amount of shared activity . The predicted accuracy curve when classifying four simulated random patterns using 1–3 spatial components is shown in Figure 5B , C ( black line ) . In each simulation we matched the accuracy for the three-dimensional classifier to the empirically observed accuracy . The resulting prediction for the 1- and two-dimensional classifier characterized the data for untrained sequences quite well ( Figure 5B ) . Only in the IPS was the observed classification accuracy for the one-dimensional classifier lower than predicted , t ( 15 ) = 2 . 247 , p=0 . 040; for all other ROIs , the prediction and data did not differ significantly ( all t ( 15 ) < 1 . 206 , p>0 . 247 ) . For the trained sequences , we reasoned that practice should lead to the development of dedicated neuronal populations for trained sequence transitions . This development would cause each sequence to activate a unique subset of the network , that is reduce the overlap between patterns . In the extreme case , in which each sequence activates an exclusive set of voxels , the patterns would become evenly distributed in the space of possible patterns ( red dots , Figure 5A ) . The accuracy curves based on such an arrangement is shown as the dashed line in Figure 5C ( for details , see Diedrichsen et al . , 2013 ) . While the observed classification accuracy curves for trained sequences did not fully reach this extreme prediction , the activity patterns were more evenly distributed than expected for spatially random patterns: The classification accuracy for the one-dimensional classifier deviated significantly from the prediction based on random patterns for S1 , M1 , PMd , SMA/pre-SMA , and IPS , all t ( 15 ) > 2 . 429 , p<0 . 028 . We then compared the accuracy curves for trained and untrained sequences directly ( Figure 6 ) . For a classifier relying only on a single dimension , no significant differences were found . However , when considering also the second and third component , classification accuracy increased more for the trained sequence . When correcting for the multiple tests over the six ROIs , the effect was significant in the SMA/pre-SMA , S1 , and in IPS ( F ( 2 , 30 ) > 5 . 64 , p<0 . 0083 ) , but we also found a similar trend in the remaining regions ( F ( 2 , 30 ) > 3 . 32 , p<0 . 05 , uncorrected ) . The effect was bilateral–in none of the tested regions was there a significant interaction of this effect with the hemisphere ( all F ( 2 , 30 ) < 2 . 072 , p>0 . 144 ) . 10 . 7554/eLife . 00801 . 011Figure 6 . Region of interest ( ROI ) analysis . ( A ) ROI definition on the average cortical surface ( shown in a dorsal and medial view ) , based on anatomical criteria ( see ‘Materials and methods’; Fischl et al . , 2008 ) . Regions of the right hemisphere are defined in a symmetric fashion . ( B–G ) Left panel shows the average percent BOLD signal change for the left and right hemisphere . Stars indicate a significant difference between trained ( red ) and untrained ( blue ) sequences ( ** corresponds to p<0 . 05/6 , or p<0 . 05 corrected for multiple comparisons , [*] indicates p<0 . 05 , uncorrected ) . Right panel shows classification accuracy for the classifiers using 1–3 of the most informative spatial dimensions ( Diedrichsen et al . , 2013 ) . Stars indicate a significant interaction effect of number of spatial components and sequence type ( trained/untrained ) , p-levels as above . DOI: http://dx . doi . org/10 . 7554/eLife . 00801 . 011 Could these accuracy differences be an artifact of performance differences during the scan ? While trained and untrained sequences were executed at slightly different speeds and average forces ( Table 1 ) , the individual differences in overall classification accuracy did not correlate with the difference in MT ( p=0 . 75 ) , force ( p=0 . 51 ) , or error rate ( p=0 . 21 ) . Furthermore , we tested whether the increase in discriminability of cortical activation patterns could have been caused by the reduced behavioral variability that normally accompanies learning . In considering this idea , it is important to keep in mind that classification accuracy is determined both by within-sequence variability and by between-sequences differences . To evaluate both together , we tested how well a linear classifier could discriminate between each set of four sequences based on MT , average force , or error rate , either considered in isolation or in any combination as separate features . None of the seven combinations showed a significant difference between trained and untrained sequences ( all t ( 15 ) < 1 . 15 , p>0 . 269; see Table 1 ) . Furthermore , when including the differences in classification accuracy based on all behavioral variables as a covariate , the accuracy difference based on fMRI patterns remained significant , t ( 15 ) = 2 . 56 , p=0 . 01 . Thus , the higher classification accuracy for trained sequences was not a simply a consequence of more stable behavioral performance . In summary , our study shows for the first time that the sequence-specific component of the activity patterns increases in strength with training . While a recent report ( Huang et al . , 2013 ) showed increases in split-half correlation of the activity pattern in primary motor cortex for a trained compared to an untrained finger sequence , these authors only used a single trained sequence , and therefore could not distinguish between the component of the activity pattern that is common to any possible trained sequence , and the pattern component that is specific to a single sequence . This , however , is an important distinction: when decomposing our activity patterns ( Diedrichsen et al . , 2011 ) , we found that 99% of the voxel-by-voxel variance was explained by a pattern component that is common to all trained sequences , with less than 1% being attributable to the sequence-specific component ( Figure 7 ) . In most areas , the sequence-specific component was larger for trained sequences , and therefore followed the pattern observed in the classification accuracy . These differences , however , failed to become significant . In contrast , the common component was smaller for trained than untrained sequences ( significant for PMd , t ( 15 ) = 2 . 469 , p=0 . 026 , and IPS and OPJ , t ( 15 ) = 3 . 573 , p<0 . 003 ) and therefore followed the pattern found for the average activity ( Figure 6 ) . 10 . 7554/eLife . 00801 . 012Figure 7 . Pattern decomposition analysis for 3 of the 6 anatomically defined ROIs . Activity patterns are decomposed into a component that is common to all four trained or untrained patterns , a sequence specific component , and a noise component ( Diedrichsen et al . , 2011 ) . Plotted is the estimate of the voxel-by-voxel variance for each component , relative to the variance explained by noise . The variance explained by the common pattern outstrips the variance associated with the sequence specific component by a factor of 100 or more . While the strength of the common component follows the mean activation ( Figure 5 ) , the sequence-specific component shows the same advantage for the trained compared to the untrained sequences evident in the classification accuracy . DOI: http://dx . doi . org/10 . 7554/eLife . 00801 . 012 Here we did not find larger split-half correlations for the patterns associated with trained compared to untrained sequences . We believe that this discrepancy is caused by the fact that a split-half correlation of a single activation pattern mostly depends on the much stronger common component , which was found to be smaller for trained than untrained sequences in our study , but slightly higher in the study by Huang et al . , ( 2013 ) . This underlines the importance of using experimental methods that allow a separation between the unspecific ( common ) and information-carrying aspects of neural activity patterns . Together , our results show that different sequences of the same five finger presses are associated with overlapping but discriminable patterns of activity , and that these differences are more pronounced when the specific sequences are trained for multiple days . The changes likely reflect the formation of specialized neuronal circuits that enable fast and accurate sequential movements . We hypothesize that the differential activation patterns reflect , at least partly , populations of neurons with preferential tuning for the sequential context of movements . Neurons in M1 ( Matsuzaka et al . , 2007 ) and SMA and pre-SMA ( Tanji and Shima , 1994 ) have been reported to be preferentially active during performance of specific movement transitions . The latter study also found cells that discharged at the beginning of specific longer sequence , but not for other orderings of the same movements . Regions that contain either type of neuron will therefore activate slightly different neuronal populations depending on movement order . To be visible in fMRI , the patterns of presynaptic input to these neuronal populations must differ on a relatively coarse spatial scale . Although MVPA is able to decode information that relies on fine-grained spatial details of activity patterns ( Swisher et al . , 2010; Freeman et al . , 2011 ) , it was not clear a priori that sequential tuning would be clustered enough to reveal such representations at standard fMRI resolution . However , we found a set of areas , highly replicable across subjects , with activity patterns that distinguished reliably between sequences . In these areas we could significantly distinguish even between the four untrained sequences . This finding may be related to the substantial generalization of learning to the untrained sequences . For example , classification may have relied on populations of neurons that encode for the sequential transitions between pairs of fingers , many of which were shared between trained and untrained sequences . More likely , however , the unique activity patterns for untrained sequences relied on preexisting neuronal coding of sequential context: whether typing on a keyboard or playing an instrument , humans engage in many sequential motor behaviors , and such general representations would discriminate between trained and untrained sequences alike . We found sequential representations in motor areas in which single-cell neurophysiology has revealed such encoding before ( i . e . , M1 and SMA; Matsuzaka et al . , 2007; Tanji and Shima , 1994 ) , but also in a set of parietal and lateral premotor areas . While other imaging studies have reported increased activity in some of these regions with increased sequential processing demands ( Farooqui et al . , 2012; Heim et al . , 2012 ) , our study is the first to show that individual voxels are tuned for different sequence of movements . Together with other neurophysiological recording study ( Fogassi et al . , 2005 ) , our results argue that sequential encoding is not unique to the SMA , but is rather widespread throughout the cortical motor hierarchy . The dimensionality analysis ( Figure 5 ) indicated that each of these areas does not simply discriminate between sequences based on one single variable ( difficulty , force , etc ) , but that each sequence is associated with a unique activity pattern . The exact response characteristics of the underlying neuronal populations remain unclear , however . Some of these regions could represent the sequences phonologically as string of number during subvocal rehearsal ( Hartwigsen et al . , 2013 ) , abstractly in a spatial reference , or motorically in an intrinsic references frame ( Keele et al . , 1995 ) . Furthermore , neural populations could code for the order of finger presses or their relative timing ( Kornysheva et al . , 2013 ) , and representations could be effector-specific or independent of hand that is used to execute the movement ( Gallivan et al . , 2013 ) . Our experimental design also did not allow us to determine conclusively to what degree sequential decoding relied only on activity related to the execution , or to what degree these patterns were also present during the instruction phase . Further dedicated experiments are needed to identify the exact nature of the sequential representations in identified areas . Our main hypothesis , however , was that training promotes the formation of specialized neural representations and hence leads to more distinguishable activity patterns . Indeed , averaged across all involved areas , we were better able to classify between trained than between untrained sequences . Furthermore , we could show that these training-related changes became most evident when looking at multiple spatial components of the activity patterns . This finding implies two things . First , it shows that the changes in classification accuracy were not due to larger between-sequence differences , or lower within-sequence variability , of a single behavioral factor—otherwise the difference would have been apparent using a classifier that uses only the most informative spatial pattern component . Secondly , it shows that training leads to the formation of a unique spatial activity pattern for each sequence and therefore a more even spacing of the activity patterns in the space of possible patterns . While it is still possible that this observation reflects changes in a combination of multiple other behavioral variables , none of the possible combinations of the behavioral measures was able to explain the observed accuracy difference between trained and untrained sequences . This lends some credibility to the idea that learning caused by the formation of specialized neuronal circuits that encodes different finger transitions or whole sequences . The largest ( and in a whole-brain comparison most reliable ) difference in classification accuracies was observed in the left SMA and pre-SMA . This likely reflects neuronal circuits specialized for longer sequence segments , a representation that would be highly specific to the trained sequences . Other areas , such as M1 , also showed good discrimination for the untrained sequences , possibly reflecting representations of shorter sequence elements . In the framework of hierarchical sequence representations ( Kiebel et al . , 2009 ) , sequence-specific units in SMA may trigger the shorter sequence elements in M1 . A number of previous studies have highlighted the SMA and pre-SMA as critical structures for skillful production of motor sequences . Disruption of the mesial frontal cortex caused increased errors during performance of complex sequential finger movements ( Gerloff et al . , 1997 ) , while leaving performance of simple finger sequences unaltered . A second study showed slowing in the transition between different sequences after pre-SMA stimulation ( Kennerley et al . , 2004 ) . In non-human primates , muscimol injections into SMA and pre-SMA lead to errors in memory-guided performance of arm movement sequences ( Shima and Tanji , 1998 ) . Although we trained and tested the left hand , learning-related changes were either bilateral , especially in regions that appear to code for actions with both hand ( Gallivan et al . , 2013 ) , or in some cases even more pronounced in the ipsilateral , left hemisphere . The latter finding is consistent with specialization of the left hemisphere for complex movements , executed with either the dominant or non-dominant hand ( Verstynen et al . , 2005 ) . Indeed , our behavioral measures indicate inter-manual transfer of the sequence-specific knowledge to the untrained right hand . The increases in discriminability of cortical activation patterns contrasted drastically with changes in average activity . Generally , we observed less activity for trained than untrained sequences , despite the fact that the two conditions were matched for error rate ( and hence difficulty ) and the trained sequences were executed slightly faster and with higher peak forces . In areas that showed the largest increases in classification accuracy , no changes in average activation were found . Average activity changes during learning are likely complicated by the overlap of two competing tendencies . First , training likely leads to increased recruitment of neuronal populations , as has been shown with expansion of the cortical hand area following skill training ( Nudo et al . , 1996 ) . Involving larger populations of neurons in motor planning and execution may reduce neuronal noise and lead to less variable performance . Second , the recruited neurons likely become more specialized for specific trained behaviors . Therefore , fewer non-specialized units must be recruited , decreasing the activity for the majority of neurons in a region ( Poldrack , 2000 ) . The overlap of increased neuronal recruitment and increased efficiency through specialization ( Steele and Penhune , 2010 ) may explain previously inconsistent results , which suggest that activity in sequence-related regions increases ( Grafton et al . , 1995; Karni et al . , 1995; Floyer-Lea and Matthews , 2005; Hazeltine et al . , 1997 ) , decreases ( Wu et al . , 2004; Poldrack et al . , 2005 ) , or changes non-linearly over the course of learning ( Xiong et al . , 2009; Ma et al . , 2010 ) . In contrast , the representational analysis employed here directly reflects the increased specialization of cortical circuits for trained behaviors . Although average activity decreased in most areas , we found increased bilateral encoding of the trained sequences . This finding has important implications for models of skill development . Examining average activity alone , one might be tempted to conclude that many secondary motor areas play a reduced role in the production of highly trained skills , which may be encoded in a few execution-related areas such as the primary motor cortex ( Penhune and Steele , 2012 ) . In contrast , our results indicate that skill learning generates increasingly specialized representations , which are distributed widely across both primary and secondary cortical motor areas . Eight female and eight male healthy , right-handed volunteers ( 22 . 4 years , SD = 2 . 6 ) , participated in the experiment . The UCL Ethics Committee approved all study procedures . We used an fMRI-compatible response box , equipped with five piano-like keys , each incorporating a sensor ( FSG-15N1A; Sensing & Control Honeywell Inc . , Morristown , NJ ) that continuously measured isometric forces during sequence production . The force measurements were transmitted to a control computer outside the scanning room through a set of filtered cables to prevent RF-signal leakage into the MR-environment . During both training and scanning participants lay supine on a ( mock- ) scanner bed , with the keyboard firmly placed on their lap at a 45° angle . Participants received visual instructions and feedback though a back-projection screen , viewed though a mirror . A central asterisk served as fixation cross throughout training and scanning . Each trial consisted of either five ( training ) or three ( scanning ) repetitions of the same sequence . At the beginning of each trial , the sequence was announced by five centrally presented numbers for 2 . 7 s ( Figure 1D ) . Each number referred to a digit ( one for thumb , five for little finger ) . During the announcement , participants were instructed to memorize the sequence . The digit string was then replaced by the fixation cross . Simultaneously , five white asterisks were presented above the box . This display served as the starting signal to produce the sequence as fast as possible . A keypress was recognized when the force of a finger exceeded a threshold of 2 . 5 N , while the other fingers were below 2 . 2 N . If the correct finger was pressed , the corresponding asterisk in the sequence turned green . If a participant pressed the wrong finger , the asterisk turned red instead . Participants were instructed to complete the sequences even if they made an error and to keep their fingers placed on the respective keys at all times . After completion of a single sequence ( five presses ) , the central fixation-cross changed color . ‘Green’ indicated correct sequence production ( one point ) , ‘red’ that one or more errors occurred ( -one point ) , and ‘blue’ that the sequence was produced 20% slower than the median MT in the previous run ( zero points ) . To motivate participants during the training , we also presented three green stars if the sequence was produced 20% faster than the median MT ( three points ) . After the end of this short feedback phase ( 800 ms ) , all asterisks turned white again to signal the start of the execution . After the required number of sequences was performed , the trial ended and the next sequence was announced . During scanning only error feedback and feedback for slow executions were provided , and the frequency of each feedback type was matched between trained and untrained sequences . Based on a pilot experiment with N = 5 independent participants , we selected a total of 12 different finger sequences with approximately the same difficulty . Each sequence contained each of the five fingers once and differed only in their order . None of the sequences contained an ascending or descending sub-sequence of more than three neighboring fingers . For each participant , these 12 sequences were randomly divided into three sets of four sequences: one to be trained , one to be used as untrained control sequences for the pre- and post-test , and one as the untrained control sequences for the scan . The experiment started with a short familiarization phase , which was followed by a pre-test . Here we measured how well participants performed eight sequences with the left and right hands . For right-hand performance , the sequences were performed in mirror-symmetric fashion , for example , pressing the thumb for the number 1 , etc . The pre-test contained 36 trials , with each sequence repeated two times per hand ( 10 executions total ) . We counterbalanced sequence order by presenting the same sequences in the reverse order in the second half of the pre-test . Participants were then trained to perform the sequences in the training set with the non-dominant left hand . On each of the four separate training sessions , they performed 24 runs ( 96 trials , and 480 sequence executions ) . The sessions were usually separated by 24 hr , with a few exceptions in which there was a 48-hr gap . A day after the second scanning session ( see below ) we conducted a post-test that had exactly the same format as the pre-test . After each run , feedback about error rate , average MT and points was presented . We instructed participants to decrease their MT if they had an error-rate of less than 20% and to focus on accuracy if the error rate was larger than 20% . This speed-accuracy instruction served to keep error-rates stable across the experiment . To assess sequence-specific learning , we tested whether trained sequences were performed faster than untrained sequences at post-test . To correct for possible pre-test differences between sequences , we calculated a regression between the pre-test difference ( x-variable ) and the post-test differences ( y-variable ) and tested whether the intercept was significantly different from zero . After 4 days of sequence training , participants underwent two sessions of fMRI scanning on separate days . During one session , participants performed the four trained sequences , and during the other , four novel sequences that were not tested in pre- or post-test . The order of these sessions was counterbalanced between participants . Each imaging session comprised 8 runs of 16 randomly ordered trials , 4 per sequence . Each trial consisted of an announcement phase of 2 . 7 s and three sequence executions ( Figure 1D ) . To synchronize the paradigm to image acquisition , participants had a maximum of 2 . 8 s to complete each sequence . There were also four , randomly interspersed rest phases ( 13 . 5 s each ) in each run . Imaging data were acquired on a 3T Siemens Trio MRI scanner using a 32-channel head coil . For each participant we also obtained an anatomical image ( 3D MPRAGE sequence , 1 mm isotropic ) . Functional data were acquired using a two-dimensional echo-planar sequence ( TR = 2 . 72 ) . Each functional scanning session consisted of 8 runs of 110 vol each . The first three images of each sequence were excluded from the analysis . We acquired 32 slices with 2 . 15-mm thickness in an interleaved sequence ( 0 . 15 mm gap , 2 . 3 × 2 . 3 mm2 in-plane resolution ) in an axial orientation . This arrangement covered the dorsal part of both cerebral hemispheres , but not the inferior temporal lobe or the cerebellum . To correct for distortions arising from field inhomogeneities , we also acquired a B0 field-map with the same slice prescription as the functional data ( Hutton et al . , 2002 ) . The imaging data were analyzed using SPM8 ( http://www . fil . ion . ucl . ac . uk/spm/ ) , and custom written MATLAB routines ( The MathWorks , Inc . , Natick , MA ) . Preprocessing consisted of correction for field inhomogeneities ( Hutton et al . , 2002 ) , motion realignment , high-pass filtering ( cut-off frequency of 1/128 s ) , and co-registration between functional and individual anatomical data . To measure the signal changes for each voxel during sequence performance , we modeled the unsmoothed data using a general linear model . We defined a unique regressor for each sequence per run . These regressors were boxcar-functions ( length 13 . 5 s ) , convolved with a standard hemodynamic response function approximated by the sum of two Gamma-functions ( spm_hrf . m in SPM8 ) . The regression-coefficients were then estimated using robust linear regression ( Diedrichsen and Shadmehr , 2005 ) , correcting for movement artifacts by down-weighting noisy images . The regression coefficients indicated the size of the activity change for each specific sequence and were used as the input to both the traditional univariate analysis and MVPA . To determine whether a specific area of cortex showed reliably different patterns of activity for the four tested sequences , we sequentially selected 160 voxels contained within a spherical patch of cortex ( see surface-based searchlight below ) and then submitted these to a linear discriminant analysis ( LDA; Duda et al . , 2001 ) . The input data ( xi ) consisted of 4 ( sequences ) × 8 ( runs ) activation estimates for the p=160 neighboring voxels . Using the data from seven runs , we calculated the mean activation vector for each sequence , and the average PxP within-class covariance matrix Σ , which was regularized by adding 1% of the diagonal mean to all diagonal elements . The activation vectors from the remaining eighth run were then classified by assigning them to the class with the highest likelihood p ( x ) ( for details , see Diedrichsen et al . , 2011 ) . By retraining and cross-validating the classifier with all possible training and test sets , we obtained an average classification accuracy . If the neural activation patterns did not differ systematically between sequences , accuracy should be 25% ( guessing rate ) . Systematically higher classification rates indicate that a region showed differential activation patterns for the four sequences , and the size of the classification accuracy served as a measure of the strength of the sequence representation in that region . For between-subject analysis , we z-transformed the classification accuracies , using a normal approximation to the binomial distribution . To determine whether differences in classification accuracy could be caused by lower variability in the behavioral performance—or by larger behavioral differences between individual sequences , we also performed a LDA on the behavioral data . As for the neural activation , we calculated the average MT , force , and error rate for each sequence and run , z-standardized these , and submitted these in isolation or as separate features ( treating each behavioral variable like an individual voxel ) to the same cross-validated LDA . We also applied LDA to different temporal components of the BOLD response . We modified our first-level analysis , such that each trial was modeled with an arbitrary finite impulse response of 24 . 3 s ( 9 TRs ) length , using 9 boxcar regressors , one for each TR , for each sequence and run . Using single value decomposition we then determined the main temporal components of the response for each anatomical ROI ( Kay et al . , 2008 ) . Instead of the beta-estimate for the canonical hemodynamic response , we submitted the component weights for each voxel and trial to the LDA classifier . The first temporal component reflected the main response and led to the highest classification accuracy in all ROI . We then added each of the other temporal components by using the 160 voxel-weights for first component and the 160 voxel-weights for the additional component as 320 independent features . We also used the finite-response function model over 9 TRs to determine the time course of activation and information over a single trial ( Figure 4D , E ) . For this we either averaged the beta weights for each TR to obtain the average activation time course , or submitted the beta weights for each TR to a cross-validated LDA classifier to obtain a classification accuracy time course . To compare different time courses across regions , we normalized each individual time-series by the L2-norm of their average . To determine the number of pattern components underlying the difference between patterns , we used a one- , two- and three-dimensional classifier ( for details see Diedrichsen et al . , 2013 ) . In short , to distinguish four unique patterns , at most three linear dimensions ( pattern components ) are needed . If the patterns only differ in the intensity scaling of a common pattern , then best classification accuracy should be achieved with a classifier that uses only the most informative pattern component of the training data set . If , however , the patterns are evenly distributed in the pattern space , classification accuracy should increase for each additional spatial dimension that the classifier uses . Simulations of classification accuracies under different assumptions are conducted so that the classification accuracies of the three-dimensional classifier matched those of the data ( Figure 4E; for details , see Diedrichsen et al . , 2013 ) . Finally , we also decomposed the activity patterns of each ROI into a common activity component that is shared between the sequences , a component that is specific to each of the four sequences , a noise component that varies trial-by-trial , and a noise component that is common to all trials within a run . The employed method directly estimates the variability ( or strength ) of each pattern component across voxels ( Diedrichsen et al . , 2011 ) . Classification accuracy relates tightly to the ratio of the sequence-specific component over the noise component . The decomposition analysis was performed separately for trained and untrained sequences . To detect sequence representations anywhere in the cortex , we used a surface-based searchlight approach ( Oosterhof et al . , 2011 ) . We first reconstructed cortical surfaces for each participant using Freesurfer ( Dale et al . , 1999 ) and aligned these to a template surface using spherical registration . For each surface node , we selected a surrounding circular region such that p=160 partly touched , or lay between , the pial and grey-white matter surface patches . This resulted in a searchlight radius of 10 . 4 mm on average . The corresponding classification accuracy was then assigned to the center node . By sequentially selecting each node of the cortical surface , we built up a map of where and how well sequences were represented in the neocortex . We defined six bilaterally defined regions of interest to cover the main anatomical areas that showed encoding for sequences in general ( Figure 6A ) . Using probabilistic cytoarchitectonic maps ( Fischl et al . , 2008 ) , only surface nodes that belonged to Brodman area ( BA ) 4 with maximal probability were included into the M1 ROI . To exclude mouth and leg representations , we further excluded all nodes that had a distance of more then 2 . 5 cm from the hand knob ( Yousry et al . , 1997 ) . S1 was similarly defined as the hand-related aspect of BA 1 , 2 , and 3 . BA 6 was divided into a medial aspect ( SMA/pre-SMA ) and the lateral aspect superior to the crest of the middle frontal gyrus ( PMd ) . The posterior parietal cortex was divided into an anterior region , including anterior , medial , and ventral IPS , and a posterior region , including the medial and lateral OPJ ( Culham and Valyear , 2006 ) . For the analysis of the mean activity , we averaged all voxels within each ROI . For classification and decomposition analysis , we selected within each anatomical ROI and participant the 800 most activated surface nodes , causing each ROI to have a size of approximately 260 voxels . Because the MVPA measures are independent of the mean activity , this selection does not bias the results under the null-hypothesis . Classification analysis within each selected part of the ROI was performed using randomly drawn groups of 160 voxels , repeating this process 5000 times , and averaging the accuracy over all draws . This random-subspace approach increases the reliability of accuracy for ROI-based analyses ( Diedrichsen et al . , 2013 ) . To compare the representation of trained and untrained sequences we employed three levels of inference , each using a random-effect analysis ( N = 16 ) . On a whole-system level , we summarized the classification accuracy averaged over all fronto-parietal regions . Because only one test was conducted for each measure , no correction for multiple tests was necessary . Within each ROI we conducted a repeated measurement ANOVA with the factors hemisphere ( left vs right ) , sequence condition ( trained vs untrained ) , and classifier dimensionality ( 1–3 ) . All F-test were corrected for the number of ROIs using Bonferroni-correction . Thirdly , we also tested difference between trained and untrained sequences using a map-wise contrast . The uncorrected threshold was set to t ( 15 ) > 3 . 39 , p<0 . 002 , and family-wise error was controlled by calculating the critical size of the largest super-threshold voxel that would be expected by chance , using Gaussian Field theory as implemented in the fmristat package ( Worsley et al . , 1996 ) . The same threshold was applied for the test of overall activity differences ( Table 2 ) .	Functional magnetic resonance imaging ( fMRI ) is a widely used technique that makes it possible to observe changes in a person’s brain activity as they perform specific tasks while lying in a scanner . These could range from listening to music or looking at images , to recalling words or imagining a scene , and each will produce a distinct pattern of neural activity . However , fMRI data can be difficult to interpret . Say a particular area of the brain is very active when a subject is trying to perform a new task , but becomes less active as the subject becomes better at the task and performs it more easily . Does this mean that the brain region is used for learning the task , but not for performing once it has been learned ? Or alternatively , does it show that the brain area is involved in carrying out the task , but that it becomes more efficient with practice , and so shows less activity in later scans ? Now , Wiestler and Diedrichsen have obtained data that help to distinguish between these alternatives . Subjects were trained to carry out four specific sequences of finger movements and then asked either to reproduce these ‘trained’ sequences or to perform four ‘untrained’ sequences while in the fMRI scanner . All eight sequences produced high levels of activity in the areas of motor cortex that control finger movements . However , closer analysis showed marked differences between the patterns of activity produced during the ‘trained’ sequences and those seen during ‘untrained’ sequences that involved moving the same fingers . Wiestler and Diedrichsen proposed that when subjects train to perform specific movement sequences , this should lead to the development of neural circuits that are specialized to carry out those specific movements—and that detailed analysis of the fMRI data would allow them to identify patterns of activity that correspond to these circuits . Sure enough , when they analysed the fMRI scans , Wiestler and Diedrichsen found that the activation patterns associated with ‘trained’ movement sequences were more readily distinguishable from each other than those associated with the ‘untrained’ movement sequences , even in areas where training led to an overall reduction in activity . As well as showing that movement sequences become associated with specific spatial patterns of activation as they are learned , this study provides a new way to study learning in fMRI that should be useful for many future studies .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2013	Skill learning strengthens cortical representations of motor sequences
The segregation of the trophectoderm ( TE ) from the inner cell mass ( ICM ) in the mouse blastocyst is determined by position-dependent Hippo signaling . However , the window of responsiveness to Hippo signaling , the exact timing of lineage commitment and the overall relationship between cell commitment and global gene expression changes are still unclear . Single-cell RNA sequencing during lineage segregation revealed that the TE transcriptional profile stabilizes earlier than the ICM and prior to blastocyst formation . Using quantitative Cdx2-eGFP expression as a readout of Hippo signaling activity , we assessed the experimental potential of individual blastomeres based on their level of Cdx2-eGFP expression and correlated potential with gene expression dynamics . We find that TE specification and commitment coincide and occur at the time of transcriptional stabilization , whereas ICM cells still retain the ability to regenerate TE up to the early blastocyst stage . Plasticity of both lineages is coincident with their window of sensitivity to Hippo signaling . The first two lineages to segregate during mammalian development are the inner cell mass ( ICM ) and the trophectoderm ( TE ) . The TE is an extraembryonic tissue , giving rise to the trophoblast lineages of the placenta . The ICM will form two additional lineages before implantation - the pluripotent epiblast ( EPI ) , giving rise to all germ layers of the embryo , and the primitive endoderm ( PE ) , largely forming the endoderm layers of the yolk sacs ( Cockburn and Rossant , 2010 ) . At the blastocyst stage , the TE forms a monolayer tight junction-coupled epithelium enclosing the blastocoelic cavity , at one end of which lies the ICM . Inside and outside cell populations first form following the 8 to 16 cell divisions ( Anani et al . , 2014; Dietrich and Hiiragi , 2007; Watanabe et al . , 2014 ) . During the 16 cell stage and during the 16 to 32 cell divisions , division-independent and dependent cell internalization leads to dynamic morphological rearrangements ( Anani et al . , 2014; Maître et al . , 2016; Morris et al . , 2010; Samarage et al . , 2015; Watanabe et al . , 2014; Yamanaka et al . , 2010 ) . From the 32 cell stage onwards , apart from the relatively rare event of division-independent cell internalization , inside and outside positioning is generally a good indicator of ICM and TE lineage fates , respectively ( McDole et al . , 2011; McDole and Zheng , 2012; Pedersen et al . , 1986; Watanabe et al . , 2014; Toyooka et al . , 2016 ) . During the 8 to 16 cell divisions , cells inherit varying amounts of apically localized proteins from the apical domain – the polarized outside surface forming at the 8 cell stage ( Anani et al . , 2014; Johnson and Ziomek , 1981; Korotkevich et al . , 2017; Watanabe et al . , 2014 ) . Inside cells are apolar , while outside cells can either be apolar or polar . Interestingly , outside apolar cells were identified as the cells that internalize during the 16 cell stage in a division-independent manner ( Anani et al . , 2014; Maître et al . , 2016 ) . Although there is some evidence that individual blastomeres show variation in gene expression and epigenetic marks prior to the 8 cell stage , and that these differences may bias their future fate ( Biase et al . , 2014; Burton et al . , 2013; Goolam et al . , 2016; Plachta et al . , 2011; Torres-Padilla et al . , 2007; White et al . , 2016 ) , it is also clear that polarity differences are key to final assignment of cell fate . Polarity differences result in differential activation of the Hippo signaling pathway: active Hippo signaling in apolar cells sequesters the transcriptional co-activator Yap into the cytoplasm , while inactive Hippo in polar cells allows nuclear accumulation of Yap , and its interaction with the transcription factor Tead4 ( Hirate et al . , 2013; Nishioka et al . , 2009 ) . Nuclear Yap/Tead4 complexes are required for TE formation and are upstream regulators of Cdx2 , a key TE-specific transcription factor ( Kaneko and DePamphilis , 2013; Nishioka et al . , 2009 , 2008; Rayon et al . , 2014; Yagi et al . , 2007 ) . Activation of Cdx2 expression in TE progenitors leads to downregulation of the pluripotent factors , Oct4 and Nanog ( Chen et al . , 2009; Niwa et al . , 2005; Strumpf et al . , 2005 ) , while in ICM progenitors Hippo signaling leads to upregulation of Sox2 , a co-factor with Oct4 in pluripotency ( Wicklow et al . , 2014 ) . Thus differential Hippo activity is a key driver of ICM-TE lineage segregation . However , the exact time when differential Hippo signaling is instructive to establish cell fate is not known . Moreover , the timing of cell fate commitment to ICM or TE is also not fully understood . A number of early studies attempted to define this timing by isolating inside and outside cells at different stages of development and then assessing cell potential in chimeras or re-aggregated embryos . However , these experiments yielded conflicting results . For example , inner cells were reported to have lost TE-forming potential by the 32 cell stage ( Tarkowski et al . , 2010 ) , the early-mid blastocyst ( ~64 cell ) stage ( Handyside , 1978; Rossant and Lis , 1979; Stephenson et al . , 2010; Suwińska et al . , 2008 ) or the late blastocyst stage ( Grabarek et al . , 2012; Hogan and Tilly , 1978; Spindle , 1978 ) . Outside cells were found to retain some plasticity up to the 32 cell stage ( Rossant and Vijh , 1980; Tarkowski et al . , 2010 ) but lost ICM potential when cavitation occurred during 32 cell stage ( Suwińska et al . , 2008 ) . Using inside/outside position during cleavage as a marker for ICM and TE progenitors cannot be entirely accurate , given the dynamic rearrangements that take place , and so caution is needed in interpreting the conclusions of most of these experiments . This has led to considerable discrepancy between studies regarding the timing of restriction of developmental potential of the ICM . In this study , we use single-cell RNA sequencing to reveal the temporal dynamics of gene expression during lineage segregation and identify known , as well as novel , markers of the process . We show that quantitative Cdx2-eGFP protein expression is an accurate readout of Hippo signaling activity and thus of the process of ICM-TE specification . We could then experimentally assess the potential of individual blastomeres scored for their level of Cdx2-eGFP expression and correlate with single-cell transcriptional profiles . We were able to resolve discrepancies in the literature and provide novel insights into the dynamics of lineage segregation . Using a Cdx2-eGFP fusion knock-in mouse line ( McDole and Zheng , 2012 ) we confirmed that eGFP expression faithfully mimics endogenous Cdx2 expression dynamics ( Figure 1A ) . Expression was first clearly detectable in early 16 cell embryos and gradually became restricted to the outer layer of the TE . We found that eGFP and endogenous Cdx2 fluorescent signals in Cdx2-eGFP heterozygous embryos showed a significant correlation from the 16 cell stage onwards . 10 . 7554/eLife . 22906 . 003Figure 1 . Cdx2-eGFP is an early marker of the developing TE lineage , governed by Hippo signaling differences from the early 16 cell stage . ( A ) Immunofluorescence staining against Cdx2 and eGFP in Cdx2-eGFP heterozygous embryos at different stages . Representative images of 10 8 cell , 39 16 cell , 35 32 cell and 11 64 cell embryos stained and imaged in two independent experiments . Scale bar: 25 μm . Correlation between eGFP and endogenous Cdx2 signals was calculated by measuring fluorescence intensities in individual cell nuclei and performing Pearson’s correlation ( r indicates coefficient ) . p-values are also given for each embryonic stage . ( B ) Mean fluorescence intensity of eGFP ( Y-axis ) in individual inside and outside cell nuclei of different stage Cdx2-eGFP embryos . Position was determined by co-staining embryos with phalloidin ( F-actin ) and cells with any surface membrane exposure were classified as outside . n indicates number of embryos . * and ** note how eGFP/Dapi measurements segregate in individual embryos . Statistical significance was calculated by Mann-Whitney test and significant p-values are indicated . Error bars: s . d . of mean . ( C ) Mean eGFP intensity relative to nuclear/cytoplasmic Yap ratio in individual inside ( red ) and outside ( blue ) cells in Cdx2-eGFP embryos at different stages . Representative measurements from 5 8 cell , 8 early 16 cell , 5 late 16 cell , 5 early 32 cell and 4 late 32 cell embryos are shown . All embryos were stained and imaged in one experiment . Correlation was calculated using Pearson’s correlation ( r indicates correlation coefficient ) and p-value is given . ( D–E ) Mean fluorescenceintensity of eGFP ( Y-axis ) in single cells in different cell populations , in early and late 16 cell stage Cdx2-eGFP embryos . ( D ) Inside apolar , outside apolar and outside polar cell populations . ( E ) Inside cells , outside cells with low nuclear/cytoplasmic Yap ratio and outside cells with high nuclear/cytoplasmic Yap ratio . Polarity was determined by phospho-ezrin staining . n indicates number of embryos analyzed . Statistical significance was calculated by Kruskal-Wallis test and significant p-values are indicated . Error bars: s . d . of mean . Cells in M-phase are not included . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 00310 . 7554/eLife . 22906 . 004Figure 1—figure supplement 1 . Developmental staging of Cdx2-eGFP embryos . Cdx2-eGFP embryos were always staged based on cell number , which we determined in live embryos based on the number of Cdx2-eGFP positive cells present . An additional layer of ‘early’ and ‘late’ sub-staging was included , which refers to the time of embryo isolation . For example ‘early 16 cell’ embryos were harvested at E2 . 5 – a time point when the population of embryos are between 8 and 16 cell stages – but only 16 cell embryos were used ( embryos with average of 12 visible Cdx2-eGFP positive cells ) . Or ‘late 16 cell’ embryos were harvested at E2 . 75 -when embryos are between 16- and 32 cells - however only strictly 16 cell embryos ( embryos with average of 12 visible Cdx2-eGFP positive cells ) were used from this time point . We established criteria for staging using the number of Cdx2-eGFP positive cells in live embryos . Graph above shows average number of Cdx2-eGFP positive cells in live staged embryos at each stage ( 8 cell n = 10 , early 16 cell n = 14 , late 16 cell n = 19 , early 32 cell n = 21 , late 32 cell n = 24 , ~64 cell n = 11 and ~80 cell n = 14 ) . A subset of staged embryos were fixed and total cell numbers were determined by Dapi staining ( 8 cell n = 10 , early 16 cell n = 14 , late 16 cell n = 15 , early 32 cell n = 21 , late 32 cell n = 24 , ~64 cell n = 11 and ~80 cell n = 11 ) . Error bars indicate standard deviation of mean . Using this guide , only carefully staged embryos were used at each time point for all experiments in the study . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 004 To relate Cdx2-eGFP levels to cell position - used in previous studies to sort ICM and TE progenitors - we quantified eGFP in inside and outside cells ( Figure 1B ) in carefully staged embryos at different developmental times ( Figure 1—figure supplement 1 ) . We found that even from the early 16 cell stage onward , inside cells expressed on average significantly lower eGFP levels than outside cells . However , initially some inside and outside cells had overlapping eGFP levels , which gradually segregated by the late 32 cell stage . Cdx2 expression is initiated in a heterogeneous , Tead4-independent manner at the morula stage , whilst later expression requires nuclear Yap/Tead4 activity in TE progenitor cells ( Nishioka et al . , 2009 , 2008 ) . Yap localization is in turn regulated by Hippo signaling in the preimplantation embryo ( Cockburn et al . , 2013; Nishioka et al . , 2009 ) . To visualize how Hippo signaling differences take control of Cdx2 expression , we correlated nuclear/cytoplasmic ( n/c ) Yap ratios and Cdx2-eGFP levels in embryos at different stages ( Figure 1C ) . We found that as soon as shuttling of nuclear Yap to the cytoplasm was initiated at the early 16 cell stage , Cdx2-eGFP levels started to show a positive correlation with n/c Yap ratios . This positive correlation progressively increased up to the early 32 cell stage . Thus , emerging Hippo signaling differences , starting at the early 16 cell stage , rapidly seize control of Cdx2 expression and up-regulation in TE progenitors is layered over initial heterogeneous , Tead4-independent signals . At the 16 cell stage differential Hippo signaling has been shown to be dictated by differences in cell polarity , rather then position per se ( Anani et al . , 2014; Maître et al . , 2016 ) . While most outside cells are polarized and have nuclear Yap/Tead ( inactive Hippo signaling ) , a population of apolar outside cells with cytoplasmic Yap ( active Hippo signaling ) has been reported . Moreover apolar outside cells have been shown to be ICM progenitors , which will eventually contribute to the inside compartment ( Anani et al . , 2014; Maître et al . , 2016 ) . To examine Cdx2-eGFP levels in different outside cell populations at the 16 cell stage we co-stained embryos with a polarity marker ( Figure 1D ) and found that apolar outside cells express low eGFP , similar to levels measured in inside cells . Similar results were obtained when n/c Yap ratios were used to distinguish between different outside cell populations ( Figure 1E ) . These results indicate that upregulation of Cdx2-eGFP is downstream of polarity-induced Hippo inactivation ( nuclear Yap localization ) , rather than position per se at the 16 cell stage . Overall , this suggests that Cdx2-eGFP expression level , which is a downstream readout of nuclear Yap/Tead4 , rather than position is the more appropriate way to sort putative ICM and TE progenitor cells at different stages of development . To explore the molecular dynamics underlying lineage segregation we performed single-cell RNA sequencing of individual cells isolated from Cdx2-eGFP embryos ranging from early 16 cell to 64 cell stages ( Figure 2A ) . Cdx2-eGFP protein levels were measured in each cell prior to sequencing using quantitative fluorescence microscopy . After quality control , we retained 262 single-cell transcriptomes ( 70 early 16 cells , 43 late 16 cells , 49 early 32 cells , 39 late 32 cell and 61 64 cells ) from 36 embryos , with an average of 7267 expressed genes per cell ( RPKM > 1; Spearman pair-wise sample correlation ≥ 0 . 8 ) . To examine how cells cluster with each other in an unbiased manner , we performed Principal Component Analysis ( PCA ) using the top 100 most variable genes across all cells ( Figure 2B–D and F ) . The primary factor segregating cells was developmental time along PC1 , where a clear progression towards two different cell populations was observed ( Figure 2B ) . PC2 was strongly associated with known lineage markers – such as Cdx2 mRNA ( Figure 2C ) . We found that diversity among cells increased drastically between the late 16- and early 32 cell stages , suggesting that emergence of ICM and TE lineages at these stages . 10 . 7554/eLife . 22906 . 005Figure 2 . Single–cell RNA sequencing reveals gradual emergence of ICM and TE lineages . ( A ) Experimental outline for harvesting single cells for RNA sequencing . ( B–D ) Principal component analysis using top 100 variable genes across all cells , where each cell is annotated for ( B ) developmental time ( C ) corresponding expression level ( log10 RPKM ) of Cdx2 mRNA ( D ) corresponding Cdx2-eGFP protein ( measured prior to RNA sequencing ) . ( E ) Heatmap showing log10 RPKM expression level of early 32 cell lineage signature genes in all 262 cells . Cells were annotated for developmental time , corresponding Cdx2-eGFP values and lineage identity , assigned based on Spearman’s rank correlation clustering . ( F ) Principal component analysis using top 100 variable genes across all cells , showing TE , ICM and co-expressing ( CO ) lineage assignment of cells . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 00510 . 7554/eLife . 22906 . 006Figure 2—source data 1 . Excel file of differentially expressed genes from SCDE analysis between ‘Cdx2-low’ and ‘Cdx2-high’ cell populations ( based on PCA groupings ) at the early 32 cell stage . Top 50 ‘Cdx2-low’ genes and top 50 ‘Cdx2-high’ genes used as ICM- and TE-specific gene signature , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 006 We found a significant overall correlation ( Pearson correlation coefficient r = 0 . 6605 , p<0 . 0001 ) between Cdx2 mRNA levels ( Figure 2C ) and Cdx2-eGFP protein levels ( Figure 2D ) in the same cell . Moreover , both align with the major transcriptional differences separating ICM and TE progenitors , further verifying Cdx2 expression levels as a means to read out the progression of lineage segregation . To obtain ICM and TE progenitor-characteristic gene expression profiles we performed single-cell differential expression ( SCDE ) analysis ( Kharchenko et al . , 2014 ) between the two clearly distinct cell populations at the early 32 cell stage and found 135 ICM-specific and 207 TE-specific differentially expressed genes ( DEGs ) ( Figure 2—source data 1 ) . In order to assign lineage identity to all cells , we applied Spearman’s rank correlation clustering using the top 50 genes for each lineage from the early 32 cell DEGs as input and visualized groupings ( Figure 2E ) . We observed clear ICM and TE populations , as expected , at early 32- , late 32- and 64 cell stages . Interestingly , we found a subset of the 16 cell stage cells clustered with the ICM ( 33 out of 113 cells ) and TE ( 24 out of 113 cells ) groupings , whereas the remaining ( 56 out of 113 cells ) where positioned in a third cluster categorized as ‘co-expressing’ ( CO ) ( Figure 2E ) . This indicates that some cells at the 16 cell stage already initiated transcriptional linage divergence . Both Spearman’s rank correlation clustering ( Figure 2E ) and PCA clustering ( Figure 2F ) revealed that the ICM and CO populations clustered closer together and further from the TE group indicating a closer relationship between ICM and CO cells . To assess transcriptional differences underlying lineage segregation we performed SCDE analysis between ICM , TE and CO groups at each developmental stage ( Figure 3A and Figure 3—source data 1 ) and found an increasing difference between ICM and TE lineages with developmental time . At the 16 cell stage we found 55 DEGs between ICM and TE profiles , with a more extensive gene network ( 42 genes ) expressed in TE cells . These included known TE markers Cdx2 ( Figure 2B ) Id2 , Dppa1 , Ptges , Krt8 and Krt18 ( Figure 3C ) as well as genes previously not or less-associated with TE development , such as Lrp2 , Dab2 ( Figure 3C ) , Bmyc and Dusp4 . Thus the first transcriptional differences arising among lineage progenitors are the activation of TE-specific genes . 10 . 7554/eLife . 22906 . 007Figure 3 . Different gene expression dynamics during development of ICM and TE lineages . ( A–B ) Summary of the number of genes differentially expressed from SCDE analysis ( A ) between lineages within each developmental time point and ( B ) between developmental time points within each lineage . Due to the low number of ICM and TE cells at the early and late 16 cell stages , these time points were pooled . ( C ) Principal component analysis using top 100 variable genes across all cells , annotated for the expression level ( log10 RPKM ) of early TE-associated genes Krt8 , Krt18 , Lrp2 and Dab2 . ( D ) TE-specific genes identified by single-cell RNA sequencing associated with at least one Tead4 binding site in trophoblast stem cells ( Home et al . , 2012 ) . Gene association with Tead4 binding sites was defined as in Home et al . – genes overlapping with Tead4 peaks and genes nearest to Tead4 peaks in 5’ and 3’ directions are considered . Core trophoblast genes ( Ralston et al . , 2010 ) are also shown ( red line ) . ( E ) Representative immunofluorescence stainings of control ( Tead4m-z+ ) and Tead4 maternal/zygotic mutant ( Tead4m-z- ) embryos for Krt8 and Krt18 ( 16 cell stage embryos ) and Lrp2 and Dab2 ( 32 to 64 cell stage embryos; Lrp2 and Dab2 were not detected in earlier stage embryos ) . n indicates total number of embryos analyzed . Scale bar: 25 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 00710 . 7554/eLife . 22906 . 008Figure 3—source data 1 . Excel file of differentially expressed genes from SCDE analysis between lineages within each developmental time point . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 00810 . 7554/eLife . 22906 . 009Figure 3—source data 2 . Excel file of differentially expressed genes from SCDE analysis between developmental time points within each lineage . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 00910 . 7554/eLife . 22906 . 010Figure 3—figure supplement 1 . Examples of lineage and stage specific gene expression patterns identified by single-cell RNA sequencing . Principal component analysis using top 100 variable genes across all cells , annotated for the expression level ( log10 RPKM ) of ( A ) TE-associated genes Dppa1 , Ptges , Id2 and Anxa6 and ( B ) ICM-associated genes Sox2 , Nanog , Upp1 , Klf4 and Spp1 . Sox2 , Nanog , Upp1 were upregulated between 16 cell/early 32 cell stage ICM , Klf4 was upregulated between early 32 cell/late 32 cell stage ICM and Spp1 was upregulated between late 32 cell/64 cell stage ICM . Dppa1 , Ptges , Id2 were upregulated in as early as the 16 cell stage TE and Anxa6 was upregulated between late 32 cell/64 cell stage TE . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 01010 . 7554/eLife . 22906 . 011Figure 3—figure supplement 2 . ICM commitment coincides with initiation of epiblast and primitive endoderm segregation . Heatmap showing the expression level ( log10 RPKM ) of epiblast ( EPI ) and primitive endoderm ( PE ) markers for 64 cell ICM ( 33 cells ) and late 32 cell ICM ( 18 cells ) . A panel of known EPI and PE lineage markers were used as input ( Guo et al . , 2010; Ohnishi et al . , 2014 ) . Lineage identities ( ICM , EPI and PE ) were assigned to cells based on groupings by Spearman’s rank correlation clustering . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 011 To assess whether TE-specific genes identified by our single-cell RNA sequencing are candidates for being direct targets of Yap/Tead4 activity , we compared previously published Tead4 ChIP sequencing data in trophoblast stem cells ( Home et al . , 2012 ) - in vitro derivatives of the TE lineage - with our TE-specific genes ( Figure 3D ) . We found that 162 out of 404 TE-specific genes were associated with at least one Tead4 binding site ( p-value<0 . 038; hypergeometric test ) . Additionally , we were able to detect protein expression of 4 early TE-specific genes ( Krt8 , Krt18 , Lrp2 and Dab2 ) in wild-type embryos , but found no or very little protein expression in Tead4 maternal/zygotic mutant embryos ( Figure 3E ) . All four genes were expressed in the TE as early as the 16 cell stage and were associated with Tead4 peaks in trophoblast stem cells . These findings identify Krt8 , Krt18 , Lrp2 and Dab2 as likely direct target genes of Yap/Tead4 activity in the early embryo . To assess temporal gene expression dynamics within each lineage , we conducted SCDE analysis between different time points within ICM and TE lineages and between CO/ICM and CO/TE lineages ( Figure 3B and Figure 3—source data 2 ) . We found that the largest change in gene expression in both lineages occurred at the 16 to early 32 cell transition , when most CO cells specified either into ICM or TE profiles . It is here that we saw upregulation of known early markers in the ICM , such as Sox2 and Nanog ( Figure 3—figure supplement 1 ) . Within the ICM lineage we found 177 DEGs between the 16 cell/early 32 cell stages , 112 DEGs between early 32 cell/late 32 cell stages and only 22 DEGs between the late 32 cell/64 cell stages . In contrast , within the TE lineage a relatively mild maturation of the transcriptional profile was seen with time , suggesting a sharp specification event when CO profiles resolve into TE profiles at the 16 and early 32 cell stages . Only 21 , 13 , and 43 DEGs were found between 16 cell/early 32 cell , early 32 cell/late 32 cell and late 32 cell/64 cell TE , respectively . Examples of genes showing lineage and stage specific expression patterns are found in Figure 3—figure supplement 1 . To correlate developmental potential of single cells with their transcriptional profiles , we tested the lineage contribution of cells with varying Cdx2-eGFP levels from different developmental stages in a morula aggregation assay . As outlined ( Figure 4A ) , we harvested embryos at different stages from Cdx2-eGFP x CAG-DsRed ( Vintersten et al . , 2004 ) crosses , dissociated them to single cells and measured Cdx2-eGFP in individual cells . These individual donor cells were then aggregated to a wild-type host morula and resulting chimeras were cultured to embryonic day 4 . 5 ( E4 . 5 ) . Chimeras were immunostained for DsRed to visualise the progeny of the aggregated single donor cell and for ICM ( Klf4 ) and TE ( Cdx2 ) lineage markers ( Figure 4B ) . Donor cells isolated from 8 cell embryos did not express Cdx2-eGFP and contributed to ICM , TE or both lineages in chimeras ( Figure 4C ) . Some donor cells from early 16 cell embryos started to express Cdx2-eGFP , and a small , yet significant bias was detected of Cdx2-eGFP high cells contributing to the TE and Cdx2-eGFP low cells contributing to ICM . This bias progressively increased with the developmental stage of the donor cell . We also noted that the ability of a single cell to give rise to both lineages sharply decreased at the 16 to 32 cell transition . Interestingly , Cdx2-eGFP low cells from the early 32 cell stage exclusively contributed to the ICM , while it took until the late-32 cell stage for Cdx2-eGFP high cells give rise to solely TE , thus revealing a different time line of fate restriction of Cdx2-eGFP low and high cells . 10 . 7554/eLife . 22906 . 012Figure 4 . Individual cells with different Cdx2-eGFP levels aggregated to host morulae show gradual and differential loss of developmental potential over time . ( A ) Experimental outline of morula aggregation assay . ( B ) Examples of chimeras with TE , ICM and TE and ICM contributions , analyzed at E4 . 5 by immunofluorescence staining . ( C ) Plot showing all aggregation chimera results . Each data point represents a single donor cell isolated from different stage embryos ( X axis ) with the level of mean Cdx2-eGFP measured in each cell before aggregation ( Y axis ) . Donor cells are color coded for the lineage their progeny contributed to in the chimera . Statistically significant differences in Cdx2-eGFP intensities between different contributions ( ICM vs TE and TE vs ICM-TE ) were calculated using Mann-Whitney test and significant p-values are indicated . ( D ) Plot showing lineage identities assigned to single cells based on RNA sequencing profiles . Each data point represents a single cell isolated from different stage embryos ( X axis ) with the level of mean Cdx2-eGFP measured in each cell before sequencing ( Y axis ) . Cells are color coded according to their lineage profiles . Statistically significant differences in Cdx2-eGFP intensities between different lineage groups ( ICM vs TE , TE vs CO and ICM vs CO ) were calculated using Mann-Whitney test and significant p-values are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 01210 . 7554/eLife . 22906 . 013Figure 4—figure supplement 1 . Live imaging chimera formation following single early 32 cell stage donor cell aggregation to host morula . ( A–B ) Single plane snapshots from live imaging movies of chimera formation with ( A ) Cdx2-eGFP low and ( B ) Cdx2-eGFP high donor cells . To aid scoring of inside and outside positions in chimeras , wild-type host embryos were microinjected with membrane-localized RFP mRNA at the 2 cell stage . Single donor cells were isolated from early 32 cell stage embryos and were identified by the ubiquitous DsRed label . In order to allow imaging on a flat glass surface , donor cells in these experiments were microinjected under the zona pellucida of the host . Time scale indicates hours and minutes after aggregation . Scale bar: 20 μm . ( C–D ) Quantification of Cdx2-eGFP during chimera formation with ( C ) Cdx2-eGFP low and ( D ) Cdx2-eGFP high donor cells and their progeny throughout live imaging movies . Clear symbol indicates outside position of donor cell or its progeny , filled symbol indicates full internalization . Arrows indicate when donor cell or its progeny undergo mitosis . Position of daughter cell marker by * could not be determined as it was in a portion of the embryo that protruded though the hole made in the zona pellucida during donor cell injection . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 013 Remarkably , when ICM , TE or CO cells , annotated based on RNA sequencing profiles , were plotted in a similar manner to lineage contribution results from the morula aggregation experiments , the plots were strikingly similar ( Figure 4C and D ) . This may suggest that cell behavior in the morula aggregation assay reads out the gradual process of ICM and TE specification as judged by dynamic transcriptional profiles . We observed the same fate outcomes of donor cells isolated from early 32 cell stage embryos in live imaging aggregation experiments as in our end-point analysis . All Cdx2-eGFP low donor cells imaged ( n = 5 ) ( Video 1; Figure 4—figure supplement 1 ) fully internalized in chimeras between 8 to 10 hr after aggregation , with Cdx2-eGFP levels remaining low throughout and donor cells dividing only after taking up an ICM position . In majority of chimeras with Cdx2-eGFP high donor cells ( n = 4/6 ) , cells remained on the outside , maintained high Cdx2-eGFP expression and underwent division with all progeny also remaining in TE position ( Video 2; Figure 4—figure supplement 1 ) . In remaining chimeras with Cdx2-eGFP high donor cells ( n = 2/6 ) , cells were internalized while in the process of downregulating Cdx2 and finally divided in an ICM position ( Video 3; Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 22906 . 014Video 1 . Live imaging a single Cdx2-eGFP low donor cell from a 32 cell stage embryo aggregating with a host morula - donor cell moves in and contributes to the ICM . Related to Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 01410 . 7554/eLife . 22906 . 015Video 2 . Live imaging a single Cdx2-eGFP high donor cell from a 32 cell stage embryo aggregating with a host morula – donor cell stays on the surface and contributes to the TE . Related to Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 01510 . 7554/eLife . 22906 . 016Video 3 . Live imaging a single Cdx2-eGFP high donor cell from a 32 cell stage embryo aggregating with a host morula - donor cell moves in and contributes to the ICM . Related to Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 016 We thus found that division was not a driving force of cell internalization and that complete downregulation of Cdx2 is not a prerequisite for internalization . This supports previous observations that some Cdx2 positive cells can internalize even in intact embryos at the 32 cell stage , downregulate Cdx2 and integrate into the ICM ( McDole and Zheng , 2012; Toyooka et al . , 2016 ) . It further suggested that the morula aggregation assay read out the state of specification of the isolated blastomeres and not necessarily their full developmental potential . Next we tested the developmental potential of cells using a different assay , in which we reconstructed entire embryos from cells sorted by levels of Cdx2-eGFP expression ( Figure 5A ) , thus removing the influence of the host embryo environment . At each developmental stage examined , the appropriate number of Cdx2-eGFP low or Cdx2-eGFP high cells , or a random mixture of cells was re-aggregated in groups . Such re-constructed embryos were cultured to E4 . 5 and analyzed for ICM ( Sox2 and Gata4 ) and TE ( Cdx2 ) lineage markers . We found that embryos reconstructed solely from cells with low or high Cdx2-eGFP levels isolated from late 16 cell embryos readily formed both ICM and TE lineages by E4 . 5 ( Figure 5B; Figure 5—figure supplement 1 ) . Similarly all embryos made from Cdx2-eGFP low and most embryos from Cdx2-eGFP high early 32 cell stages still recapitulated both ICM and TE lineages ( Figure 5C; Figure 5—figure supplement 1 ) , although we observed a non-significant decrease in the number of ICM cells formed in embryos from Cdx2-eGFP high cells compared to embryos made from Cdx2-eGFP low or random cells ( Mann-Whitney test , p-value=ns ) . Embryos made from late 32 Cdx2-eGFP low cells still mostly reconstituted both ICM and TE lineages , while Cdx2-eGFP high cells by this stage completely lost their ability to form ICM and only produced Cdx2 positive TE ( Figure 5D; Figure 5—figure supplement 1 ) . It was only by the 64 cell stage that the majority of Cdx2-eGFP low cells lost their ability to make TE ( Figure 5E; Figure 5—figure supplement 1 ) . Out of eight embryos made from Cdx2-eGFP low cells , only three showed re-expression of Cdx2 in a few cells ( 4 , 6 and 3 cells per embryo ) , two without cavitating and one with a tiny cavity . A fourth embryo developed a small cavity but without any re-expression of Cdx2 . All other embryos ( n = 4/8 ) remained as tightly packed clusters without Cdx2 re-expression or cavitation . 10 . 7554/eLife . 22906 . 017Figure 5 . Embryos reconstructed entirely from Cdx2-eGFP low or high cells loose their potential to recapitulate ICM and TE lineages at different times . ( A ) Experimental outline to reconstruct embryos entirely of Cdx2-eGFP low , high or random cells . ( B–E ) Plots showing embryo reconstructions from single cells isolated from ( B ) late 16 cell , ( C ) early 32 cell , ( D ) late 32 cell and ( E ) 64 cell stages . Each embryo ( X axis , labeled with letters ) was reconstructed from Cdx2-eGFP-quantified ( Y axis ) single cells . Color-coding below indicates the presence of Cdx2 positive TE ( green ) and Sox2 or Gata4 positive ICM ( red ) cells in reconstructed embryos at E4 . 5 . Grey indicates the absence of a lineage; white ( N/A ) indicates the embryo was lost during immunofluorescence staining , thus information is only available of the TE lineage from the live Cdx2-eGFP marker before fixation . * embryo visually only consisting of trophoblast vesicles . ** embryo morphology like B-E embryos , likely contains both Gata4 and Sox2 positive cells . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 01710 . 7554/eLife . 22906 . 018Figure 5—figure supplement 1 . Representative images of embryo reconstructions from single cells at different stages . Single cells used to reconstruct embryos ( left panel ) are either Cdx2-eGFP low , high or randomly mixed . Reconstructed embryos live at embryonic day 4 . 5 ( E4 . 5 ) ( middle panel ) . Immunofluorescence staining for lineage markers in reconstructed embryos at E4 . 5 ( right panel ) . Scale bar: 40 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 018 Interestingly , this assay revealed a different timeline for fate restriction than the morula aggregation assay . Cdx2-eGFP high cells completely lost their ICM forming ability by the late 32 cell stage , while Cdx2-eGFP low cells retained TE potential at late 32 cell stage and did not lose TE potential until the 64 cell stage . Differential Hippo signaling plays a key role in ICM-TE lineage specification ( Cockburn et al . , 2013; Hirate et al . , 2013; Lorthongpanich et al . , 2013; Nishioka et al . , 2009 ) . However , the exact period over which Hippo signaling can influence cell fate has not been addressed . We used an inhibitor of Rho-associated protein kinase ( ROCKi ) and inducible expression of dominant negative Lats2 to activate or block Hippo signaling , respectively at different times during development to address this issue . Treatment of embryos from the 2 cell stage on with ROCKi was shown to enhance ICM and suppress TE characteristics through activation of Hippo signaling ( Kono et al . , 2014 ) . We treated embryos with ROCKi for a 24 hr period , starting at different stages ( Figure 6A ) , and analyzed Cdx2 expression following treatment as a measure of cell fate restriction . We found that ROCKi treatment beginning at the 16 cell stage significantly reduced the percent of Cdx2 positive cells compared to controls ( 35% and 64% , respectively ) ( Figure 6B ) . We confirmed that this coincided with ectopic Hippo activation as assessed by loss of nuclear YAP and an increase in phosphorylated cytoplasmic Yap ( Figure 6—figure supplement 1 ) . When embryos began treatment at stages later than the 16 cell stage , there was a gradual resistance to the effect of ROCKi , with no effect of activating Hippo signaling on Cdx2 expression by the late 32 cell stage . 10 . 7554/eLife . 22906 . 019Figure 6 . ICM and TE progenitors show loss of responsiveness to Hippo signaling manipulation at the same time as they loose responsiveness to positional changes . ( A ) Overview of Hippo signaling activation time course . Each bar represents 24 hr of 50 µM ROCKi treatment . ( B ) Percent of Cdx2 positive cells per embryo cultured for 24 hr in control or ROCKi conditions . Label on top indicates the stage embryos started treatment . n indicates number of embryos analyzed . Statistical significance was calculated using t-test and significant p-values are indicated . Error bars: s . d . of mean . ( C ) Strategy for inducible Hippo signaling inactivation . Mostly mosaic Dox-inducible DN Lats2-IRES-mCherry transgenic embryos were generated . Each bar represents 24 hr of Dox treatment . ( D ) Dox-inducible DN Lats2-IRES-mCherry transgenic embryos were imaged before Dox treatment ( top panel ) and the same embryo was imaged following 24 hr of Dox live ( middle panel ) and fixed/stained for lineage markers ( bottom panel ) . A representative embryo is shown for each stage . Live mCherry is shown as an extended focus image , immunofluorescence stainings shown as single plane images . mCherry positive ICMs in mosaic transgenic embryos are circled with a dotted line . Arrow points to a rare ICM cell in a 64 cell stage-induced embryo with weak Cdx2 expression , which also co-expressed an ICM marker . Scale bar: 25 µm . n indicates number of transgenic embryos analyzed . ( E ) All mCherry negative ( non-transgenic control ) and mCherry positive ( DN Lats2-mCherry transgenic ) ICM cells were scored in mosaic embryos for presence or absence of lineage markers following 24 hr of Dox treatment by immunofluorescence staining . Cells with different lineage marker expression are shown as percent of all mCherry negative or mCherry positive ICM cells analyzed . n ( cell ) indicates number of cells analyzed at each stage and n ( embryo ) indicates number of embryos cells were pooled from . Chi-squared test was used to test whether cell fate was affected by DN Lats2-mCherry expression . 16 cell p-value=8 . 48491E-18; early 32 cell p-value=5 . 50841E-34; late 32 cell p-value=6 . 32116E-35; 64 cell p-value=0 . 004103716; >64 cell p-value=0 . 588416983 . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 01910 . 7554/eLife . 22906 . 020Figure 6—figure supplement 1 . Effect of ROCKi treatment on cell number and Hippo signaling . ( A ) Total cell numbers in control and 50 µM ROCKi treated embryos at different stages . n indicates number of embryos analyzed . Statistical significance was calculated using t-test and significant p-values are indicated . Error bars: s . d . of mean . ( B ) Immunofourescence staining of control and 50 µM ROCKi treated embryos for TE marker ( Cdx2 ) , ICM marker ( Klf4 ) and Yap . 24 hr treatment was started at the 16 cell stage . A total of 4 control and 4 ROCKi-treated embryos were imaged in one experiment . Scale bar: 25 µm . ( C ) Immunofourescence staining of control and 50 µM ROCKi treated embryos for TE marker ( Cdx2 ) , ICM marker ( Klf4 ) and phospho-Yap ( form of Yap sequestered into the cytoplasm due to active Hippo signaling ) . 24 hr treatment was started at the 16 cell stage . A total of 4 control and 3 ROCKi-treated embryos were imaged in one experiment . Scale bar: 25 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 02010 . 7554/eLife . 22906 . 021Figure 6—figure supplement 2 . Expression of mCherry only does not influence cell fate in the embryo . 2 cell stage embryos were injected with H2O ( wild-type control ) or a cocktail of PB-TAC-mCherry-IRES-mCherry , PB-CAG-rtTA and PBase mRNA ( mCherry control ) . Embryos were treated with Dox for 24 hours starting at the 32 or 64 cell stages . Following Dox treatment cell fate of ICM cells was analyzed by immunofluorescence staining for lineage markers . Cell fates shown as percent of all H2O injected ICM cells ( in H2O injected embryos ) or all mCherry positive ICM cells ( in mCherry control embryos ) . n ( cell ) indicates number of cells analyzed at each stage and n ( embryo ) indicates number of embryos cells were pooled from . Chi-squared test was used to test whether cell fate was affected by mCherry expression . 32 cell p-value= 0 . 139370244 , 64 cell p-value= 0 . 07551351 . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 021 For the reverse experiment , to determine when ICM progenitors cease responding to inactivation of Hippo signaling , we employed an inducible genetic approach . Transgenic embryos were generated with a doxycycline ( Dox ) -inducible dominant-negative Lats2 ( DN Lats2 ) tagged with mCherry ( Figure 6C ) and induced at different stages for 24 hr . Following treatment , embryos expressing high levels of mCherry were analyzed for ICM ( Sox2 and Sox17 ) and TE ( Cdx2 ) lineage markers ( Figure 6D and E ) . Our approach to generate transgenic embryos typically resulted in mosaic integration of inducible DN Lats2-mCherry , which allowed us to analyze both transgenic ( mCherry positive ) and non-transgenic ( mCherry negative ) ICM cells within the same embryo . Up to the early 32 cell stage , DN Lats2 expression in inner cells could induce fate change as indicated by gain of Cdx2 and loss of ICM markers in most mCherry positive inner cells . At the late 32 cell stage we observed some mCherry positive inner cells ( 14% ) co-expressing Cdx2 and ICM markers , while majority ( 80% ) of cells still fully converted to expressing only Cdx2 . When DN Lats2 was induced at the 64 cell stage , however , most inner cells maintained ICM marker expression and did not re-express Cdx2 . Only 1% of mCherry positive inner cells switched to expressing only Cdx2 and 4% expressed markers of both lineages . All inner cells in embryos induced at >64 cell stage showed commitment to ICM . Additionally , we noted that expression of only mCherry ( without DN Lats2 ) did not influence cell fate at any stage ( Figure 6—figure supplement 2 ) . These results are in good agreement with our findings from embryo reconstruction experiments , indicating that on a mechanistic level , loss of ICM potential of Cdx2-eGFP high cells and loss of TE potential of Cdx2-eGFP low cells corresponds to the time when cells become refractory to Hippo signaling activity or inactivity , respectively . While existing expression profiling datasets provide only limited or partial coverage of ICM-TE lineage segregation ( Deng et al . , 2014; Goolam et al . , 2016; Graham et al . , 2014; Guo et al . , 2010; Ohnishi et al . , 2014; Tang et al . , 2011 ) , our study offers a large , comprehensive single-cell global transcriptional dataset spanning the entire window of lineage segregation with remarkable temporal resolution . We show a gradual separation of ICM and TE lineages starting at the 16 cell stage , where the first transcriptional changes to distinguish cell populations are the activation of TE-specific genes , including Cdx2 . Cdx2 is the only known downstream target of nuclear Yap/Tead4 in the embryo ( Rayon et al . , 2014 ) . We now present evidence that a number of the additionally identified early TE genes , such as Krt8 , Krt18 , Dab2 and Lrp2 are likely targets of nuclear Yap/Tead4 as well . Whether these transcriptional differences at the 16 cell stage relate to earlier heterogeneities among blastomeres , as reported by other groups ( Biase et al . , 2014; Burton et al . , 2013; Goolam et al . , 2016; Plachta et al . , 2011; Torres-Padilla et al . , 2007; White et al . , 2016 ) , is not clear , although we did not observe correlated differential expression of genes previously suggested to be involved in generating such heterogeneities ( e . g . Carm1 , Prdm14 or Sox21 ) ( Burton et al . , 2013; Goolam et al . , 2016; Torres-Padilla et al . , 2007 ) . While we detected the initiation of lineage segregation in some cells at the 16 cell stage , others were still in a state of co-expression . The division to the early 32 cell stage marked a drastic transcriptional change , resolving the co-expressing state into ICM or TE profiles . Interestingly , we found that as the TE profile appeared , it also stabilized . On the other hand , the emerging ICM profile underwent considerable maturation until the 64 cell stage , at which point segregation into EPI and PE lineages is initiated ( Guo et al . , 2010; Kurimoto et al . , 2006; Ohnishi et al . , 2014 ) ; and in this study [Figure 3—figure supplement 2] ) . Importantly , we link global gene expression patterns to functional measures of cell fate and potential , by charting different experimental readouts as a function of developmental time and level of Cdx2-eGFP expressed in individual cells . We experimentally tested the potential of single cells to contribute to developing lineages of the early embryo using a morula aggregation assay and found that cell behaviors observed in this test reflected the progress of lineage separation apparent from transcriptional profiling . We suggest that in this assay , a single cell finds itself in a competitive host environment and thus can act according to its intrinsic lineage identity: a specified TE progenitor will remain on the outside and contribute to the TE , while a specified ICM progenitor will move in and contribute to the ICM . Live imaging of chimera formation further supported such dynamic behaviors of ICM and TE progenitor cells . Thus we propose that despite the experimental manipulations involved , the morula aggregation assay reports cell fate specification . In contrast , when an entire embryo is reconstructed from only ICM or only TE progenitor cells , competition with the host cells is removed and some cells are inevitably forced inside and others outside . As such , this assay reveals the full potential of cells to change fate when forced into different positions . In this assay , ICM cells showed an extended period of lineage plasticity when compared with the results of the morula aggregation assay . We further showed that loss of responsiveness towards Hippo signaling manipulation coincided with the timing of fate restriction in ICM and TE progenitors , as read out by the reconstruction assay . Thus in the intact embryo and even in the morula aggregation assay , transcriptional profiles predict TE and ICM fates , while in the reconstruction assay and the Hippo manipulation experiments , the full potential of cells to respond to positional changes and associated signaling environments is revealed . Importantly , in previous studies only one type of assay was used - either morula aggregation or reconstruction ( Grabarek et al . , 2012; Handyside , 1978; Hogan and Tilly , 1978; Rossant and Lis , 1979; Spindle , 1978; Stephenson et al . , 2010; Suwińska et al . , 2008; Tarkowski et al . , 2010 ) . By performing these two assays side by side , we can observe that the morula aggregation assay largely tests cell fate , while the reconstruction assay reveals full developmental potential , thus reconciling differences reported among the previous studies . When comparing the dynamics of fate restriction shown by cells in the different assays ( summarized in Figure 7 ) , we observed that TE progenitors exhibited similar timing of fate restriction in all assays employed . A majority of Cdx2-eGFP high cells at the early 32 cell stage produced only TE in all three assays , which also corresponded to the time by which most co-expression profiles resolved , as shown by RNA sequencing . On the other hand , ICM progenitors showed fate restriction earlier in the morula aggregation test ( by the early 32 cell stage ) than they did in the embryo reconstruction and Hippo-inactivation assays ( by the 64 cell stage ) , revealing a time window between lineage specification and commitment . Correspondingly , gene expression dynamics in the ICM lineage also reflected this prolonged maturation . 10 . 7554/eLife . 22906 . 022Figure 7 . Graphical summary of specification and commitment of ICM and TE progenitors . DOI: http://dx . doi . org/10 . 7554/eLife . 22906 . 022 Why should there be asynchronous lineage commitment in the developing embryo ? During the 16 to 32 cell divisions a number of outside cells are internalized as a result of the cleavage plane orientation ( Morris et al . , 2010; Yamanaka et al . , 2010 ) . However , a recent study revealed that daughter cells pushed inwards during mitosis often sort back to the surface ( Watanabe et al . , 2014 ) . We propose that the timely commitment of majority of TE cells that we observe between the late 16 and early 32 cell stages may be the driving force of this sorting process , allowing generation of a differentiated polarized epithelial layer that would stabilize the inside compartment . Cell divisions do not cause spatial perturbation in the ICM , thus functional commitment to this lineage may not be needed at this stage . Instead , we find that the ICM loses the ability to become TE when the first heterogeneities in gene expression demarcating the start of the second lineage segregation arise . This is in line with an earlier study , which showed that outer cells of isolated ICMs of early blastocysts mostly formed trophoblast outgrowths , while isolated ICMs from later stage blastocysts mostly formed PE-like outgrowths ( Nichols and Gardner , 1984 ) . We suggest a possible functional relationship between loss of ICM plasticity and initiation of EPI and PE differentiation programs , which requires further investigation . In conclusion , we propose that asynchronous lineage commitment may be a mechanism contributing to the regulative nature of the preimplantation embryo , ensuring that the correct number of cells is allocated to inside and outside compartments . ICR ( Crl:CD1 ( ICR ) breeding stock from Charles River , Montreal , Canada ) , Cdx2-eGFP ( knock-in fusion to endogenous locus ) ( RRID:IMSR_KOMP:VG12984 ) ( McDole and Zheng , 2012 ) , DsRed ( RRID:IMSR_JAX:005441 ) ( Vintersten et al . , 2004 ) , Tead4 fl/fl ( RRID:MGI:3763368 ) ( Yagi et al . , 2007 ) and Zp3-cre ( RRID:IMSR_JAX:003651 ) ( de Vries et al . , 2000 ) mouse lines were used in this study . Crosses to produce experimental embryos are described in figures . To obtain maternal/zygotic Tead4 mutant embryos Tead4 fl/fl , Zp3-cre females were crossed to Tead4 fl/- males and genotype of embryos was determined by immunofluorescense staining against Tead4 . Preimplantation embryos were flushed from oviducts or uteri with EmbryoMax M2 Medium ( EMD Milipore , Etobicoke , Canada ) . Embryos were collected at appropriate time points from 5–8 week old hormone-primed ( 5 IU each pregnant mare serum gonadotropin ( Sigma , Oakville , Canada ) and human chorionic gonadotropin ( Sigma , Oakville , Canada ) , 48 hr apart ) and mated females , followed by careful staging based on morphology and number of Cdx2-eGFP positive cells ( Figure 1—figure supplement 1 ) . Zygotes were washed clean of cumulus cells by brief treatment with 300 μg/ml hyaluronidase ( Sigma Oakville , Canada ) . If not immediately used , embryos were cultured in small drops of KSOM supplemented with amino acids ( EMD Milipore ) under mineral oil ( Zenith Biotech , Guilford , CT ) at 37°C , with 5% CO2 for specified times . ROCK inhibitor ( Y-27632 , Sigma ) was used at 50 µM concentration . Doxycycline ( Sigma ) was used at 1 µg/ml . All animal work was carried out following Canadian Council on Animal Care Guidelines for Use of Animals in Research and Laboratory Animal Care under protocols ( protocol number: 20–0026H ) approved by The Centre for Phenogenomics Animal Care Committee . The zona pellucida was removed using acid Tyrode’s solution ( Sigma , Oakville , Canada ) and embryos were washed in M2 . Dissociation was performed by incubating embryos in TrypLE Select ( Gibco , Thermo Fisher Scientific , Waltham , MA ) for 3–6 min at 37°C followed by pipetting through fine pulled glass capillaries . Individual cells were picked , aligned in rows in M2 media under oil and imaged using a Zeiss Axiovert 200 inverted microscope equipped with a Hamamatsu C9100-13 EM-CCD camera , a Quorum spinning disk confocal scan head ( Quorum Technologies Inc . , Guelph , Canada ) and Volocity acquisition software version 6 . 3 . 1 ( Perkin Elmer , Santa Clara , CA ) . Cells were not used for experiments for 60 min after imaging to ensure image was not acquired during mitosis ( Yamagata and FitzHarris , 2013 ) . Cells dividing during this time were excluded from the analysis . Cdx2-eGFP was quantified by measuring average pixel intensities from single plane images of individual cells , focusing on the plane with maximum eGFP intensity . Average nuclear Cdx2-eGFP measurements were corrected for cytoplasmic background . The cut off between ‘Cdx2-eGFP low’ and ‘Cdx2-eGFP high’ was set at 500 fluorescent intensity units and was based on the clearly distinct Cdx2-eGFP low cell population at the 64 cell and >64 cell stages . Single-cells were directly dispensed in lysis buffer and cDNA libraries were generated using Smart-seq2 as previously described ( Petropoulos et al . , 2016; Picelli et al . , 2013 , 2014 ) . Reads were mapped to the mouse genome ( mm10 ) using STAR with default settings ( RRID: SCR_004463 ) ( Dobin et al . , 2013 ) and only uniquely mapped reads were kept . Gene expression levels ( RefSeq annotations ) were estimated in terms of reads per kilobase exon model and per million mapped reads ( RPKM ) using our established pipeline , rpkmforgenes . py ( Ramsköld et al . , 2009 ) . Read counts from regions where different RefSeq genes overlapped were excluded and cells with ~ ≥40% uniquely mapped reads were retained . Genes were filtered , keeping 15 , 713 out of 24 , 490 genes that were expressed in ≥2 cells with an expression cutoff of 1 . Cells were quality-filtered based on Spearman’s correlation , percent uniquely mapped ( ~ ≥40% ) and the minimum number of expressed genes per cell ( 3500 ) . Then PCA dimensionality reduction was conducted for each individual time point and additional outlier cells were identified . Batch or embryo effects were not observed in the dataset . Data for this study is available at NCBI Gene Expression Omnibus ( RRID: SCR_007303; http://www . ncbi . nlm . nih . gov/geo/ ) under accession number GSE84892 . This analysis was conducted as previously described ( Petropoulos et al . , 2016 ) . Briefly , gene-variability statistic was calculated that adjusted for the mean-variance relationship present in single-cell RNA-sequencing data . This was done by assuming that the expression distribution of a gene follow a negative binomial for which the variance depends on the mean , v = m + m2/r , where r is the overdispersion , implying that cv2 = v/m2 = 1/m + 1/r . To estimate the technical variability we fitted such a model to our ERCC spike-in read counts and a gene-variability statistic was then obtained by adjusting for the technical variability present when conditioning on the mean expression level ( Brennecke et al . , 2013 ) . To determine the number of variable genes used , we tested 100 , 250 , 500 and 1000 of the most variable genes , and visually assessed clusters obtained using principal component analysis ( PCA ) . Similar profiles were obtained regardless of input . Temporal separation of cells was obtained by applying dimensionality reduction technique , PCA . Similar profiles were obtained using t-Distributed Stochastic Neighbour Embedding ( t-SNE ) , another dimensional reduction algorithm ( data not shown ) ( Hinton and van der Maaten , 2008 ) . Cells were stratified by the developmental time that they were collected and their corresponding Cdx2-eGFP values . Principal components of interest were identified by both observing a separation of developmental time and Cdx2-eGFP profile of the cells . Single-cell differential expression analysis ( SCDE ) ( Kharchenko et al . , 2014 ) was then performed between the ‘ICM’ ( corresponding to the low Cdx2-eGFP group ) and ‘TE’ ( corresponding to the high Cdx2-eGFP ) of the early 32 cells to determine lineage signatures for the ICM and TE . The top 50 genes obtained for both the ICM and TE were then used as input to determine the segregation of cells at the late and early 16 cell stage , by using Spearman’s rank correlations . Following lineage classification , SCDE was then performed for all groups ( developmental stage and lineage ( ICM , TE and ‘co-expressed’ ) ) for each time point . The SCDE algorithm requires non-normalized integer values , as such , the raw read counts were provided as input . Genes with zero reads across the samples being compared were discarded . Two-sided p-values were calculated from the Benjamini-Hochberg multiple testing corrected Z-score ( cZ ) using the normal distribution as null hypothesis , and a significance level of 0 . 05 was used to deem genes as significantly differentially expressed . Embryos were fixed in 4% formaldehyde at room temperature for 15 min , washed once in PBS containing 0 . 1% Tween-20 ( PBS-T ) , permeabilized for 15 min in PBS 0 . 2% Triton X-100 and then blocked in PBS-T with 2% BSA ( Sigma ) and 5% normal donkey serum ( Jackson ImmunoResearch Laboratories Inc . , West Grove , PA ) at room temperature for 2 hr , or overnight at 4°C . Primary and secondary antibodies were diluted in blocking solution , staining was performed at room temperature for ~2 hr or overnight at 4°C . Washes after primary and secondary antibodies were done three times in PBS-T . F-actin was stained using Alexa Flour 546-conjugated phalloidin ( A22283 , Life Technologies , Waltham , MA ) diluted 1:200 and added during secondary antibody incubation . Embryos were mounted in Vectashield containing Dapi ( Vector Laboratories Canada Inc . , Burlington , Canada ) in wells of Secure Seal spacers ( Molecular Probes , Thermo Fisher Scientific ) and placed between two cover glasses for imaging . Primary antibodies: chicken anti-mCherry 1:600 ( RRID: AB_2636881 , NBP2-25158 , Novus Biologicals , Littleton , CO ) ; mouse anti-mCherry 1:500 ( RRID: AB_2307319 , 632543 , Clontech , Takara Bio USA , Inc . , Mountain View , CA , USA ) ; chicken anti-GFP 1:400 ( RRID: AB_2534023 , A10262 , Invitrogen , Thermo Fisher Scientific , ) ; mouse anti-Tead4 1:100 ( RRID: AB_2203086 , sc-101184 , Santa Cruz Biotechnology Inc . , Mississauga , Canada ) ; mouse anti-Yap 1:100 ( RRID: AB_1131430 , sc-101199 , Santa Cruz Biotechnology Inc . ) ; rabbit P-ezrin 1:200 ( RRID: AB_330232 , 3141S , Cell Signaling Technologies Inc . , Danvers , MA ) ; rabbit anti-Cdx2 1:600 ( RRID: AB_1523334 , ab76541 , Abcam , Cambridge , United Kingdom ) ; mouse anti-Cdx2 1:100 ( RRID: AB_2335627 , MU392-UC , Biogenex , Fremont , CA ) ; goat anti-Sox2 1:100 ( RRID: AB_355110 , AF2018 , RandD Systems , Minneapolis , MN ) ; goat anti-Sox17 1:100 ( RRID: AB_355060 , AF1924 , RandD Systems ) ; rabbit anti-Gata4 1:100 ( RRID: AB_2247396 , sc-9053 , Santa Cruz Biotechnologies Inc . ) ; goat anti-Klf4 1:100 ( RRID: AB_2130245 , AF3158 , RandD Systems ) ; mouse anti-Rfp 1:100 ( RRID: AB_1141717 , ab65856 , Abcam ) ; mouse anti-Lrp2 1:100 ( RRID: AB_1260798 , NB110-96417 , Novus Biologicals ) ; mouse anti-Dab2 1:100 ( RRID: AB_397837 , 610464 , BD Biosciences , San Jose , CA , USA ) ; rat anti-Krt8 1:10 ( RRID: AB_531826 , TROMA-I antibody , Developmental Studies Hybridoma Bank , Iowa City , IA , USA ) ; mouse anti Krt18 1:100 ( RRID: AB_305647 , ab668 , Abcam ) . Secondary antibodies: ( diluted 1:500 ) 448 , 549 or 633 conjugated donkey anti-mouse , donkey anti-rabbit or donkey anti-goat DyLight ( Jackson ImmunoResearch ) or Alexa Fluor ( Life Technologies ) . Images were acquired using a Zeiss Axiovert 200 inverted microscope equipped with a Hamamatsu C9100-13 EM-CCD camera , a Quorum spinning disk confocal scan head and Volocity aquisition software version 6 . 3 . 1 ( RRID: SCR_002668 ) . Single plane images or Z-stacks ( at 1 μm intervals ) were acquired with a 40x air ( NA = 0 . 6 ) or a 20x air ( NA = 0 . 7 ) objective . Images were analyzed using Volocity or Imaris software version 8 . 3 ( RRID: SCR_007370 , Bitplane , South Windsor , CT ) . Time-lapse imaging was performed on the same microscope equipped with an environment controller ( Chamlide , Live Cell Instrument , Seoul , South Korea ) . Embryos were placed in a glass-bottom dish ( MatTek , Ashland , MA ) in KSOM covered with mineral oil . A 20x air ( NA = 0 . 7 ) objective lens was used . Images were taken every 20 min for 20 hr at 4 µm Z intervals . Cdx2 , eGFP and Yap measurements from fixed whole embryo specimens were quantified using the spot function of Imaris . Cdx2 and GFP intensities were normalized against Dapi , while for Yap the average nuclear intensity over average cytoplasmic intensity was calculated . Time-lapse movies were also analyzed using Imaris and average nuclear Cdx2-eGFP intensities were measured with the spot function . ICR morulae ( 8 or 16 cell embryos ) were used as host embryos . The zona pellucida was removed and embryos were washed in M2 . Cdx2-eGFP was quantified in individual donor cells as described before . A single donor cell and a single host morula were then brought together in a micro-well generated by pressing a blunt end needle into the bottom of a plastic tissue culture dish ( Falcon , Thermo Fisher Scientific ) in drops of KSOM under oil . Such aggregation chimeras were cultured for 48 hr . Number of chimeras generated per donor cell stage: 15 8 cell stage ( one experiment ) , 63 early 16 cell stage ( two experiments ) , 71 late 16 cell stage ( two experiments ) : 74 early 32 cell stage ( two experiments ) , 34 late 32 cell stage ( three experiments ) , 23 64 cell stage ( three experiments ) and 13 > 64 cell stage ( two experiments ) . For time-lapse imaging host embryos were injected with membrane-RFP mRNA ( see below ) and single early 32 cell stage donor cells were introduced into the host embryo through a hole in the zona generated by a laser ( XYRCOS , Hamilton Thorne Inc . , Beverly , MA ) using a micromanipulator ( TransferMan NK2 , Eppendorf Canada , Mississauga , Canada ) . Micromanipulations were performed in M2 under oil . Number of chimeras live imaged: five with Cdx2-eGFP low donor cells and six with Cdx2-eGFP high donor cells . mRNA was synthesized from pCS2 membrane-RFP plasmid ( Megason and Fraser , 2003 ) using the mMESSAGE mMACHINE SP6 Kit ( Invitrogen ) and resuspended in RNase-free water . Microinjection was performed using a Leica microscope and micromanipulators ( Leica Microsystems Inc . , Richmond Hill , Canada ) . Injection pressure was provided by a FemtoJet ( Eppendorf ) and negative capacitance was generated using a Cyto721 intracellular amplifier ( World Precision Instruments , Sarasota , FL , USA ) . Injections were performed in an open glass chamber in M2 medium . Both blastomeres of ICR 2 cell embryos were injected with 200 ng/μg mRNA . After injection embryos were cultured to the morula stage ( 8 to 16 cell ) and used as host embryos in aggregation experiments , which were live imaged . Cdx2-eGFP was quantified in individual cells as described before , cells were grouped based on eGFP levels and re-scanned as a group to avoid errors . 16 , 32 or 64 single cells ( the same number as the original embryo ) were brought together in a in a micro-well ( same as in the single cell aggregation assay ) in drops of media under oil . Some 64 cell reconstructions were performed in an emptied zona pellucida: a mid blastocyst embryo was selected , a hole was made in the zona using a laser and the embryo was suctioned out of the zona using a micro-suction pipette . 64 single cells were then reintroduced into the zona through the hole . Embryos aggregated slightly better in emptied zonas than in micro-well; however no difference was observed on cell fate outcomes between methods . KSOM was used to culture embryos made at 16 and most 32 cell stages , however we found that at the 64 cell stage using embryonic stem cell media ( DMEM ( Invitrogen ) , 2 mM GlutaMAX ( Invitrogen ) , 0 . 1 mM 2-mercaptoethanol ( Sigma ) , 0 . 1 mM MEM non-essential amino acids ( Invitrogen ) , 1 mM sodium pyruvate ( Invitrogen ) , 50 U/ml each penicillin/streptomycin ( Invitrogen ) , 15% fetal bovine serum ( Gibco ) , 1000 U/ml LIF ( EMD Millipore ) ) greatly improved cell survival . We confirmed that fate outcomes were not affected by the choice of media , as reconstructions at the 32 cell stage in embryonic stem cell media gave the same results as using KSOM . Reconstructed embryos were cultured until embryonic day 4 . 5 . Number of reconstructed embryos generated for each developmental stage: 11 late 16 cell stage ( two experiments ) , 19 early 32 cell stage ( three experiments ) , 21 late 32 cell stage ( four experiments ) and 14 64 cell stage ( five experiments ) . Pronuclear injections were performed on zygotes or cytoplasmic injections on 2 cell stage embryos . Typically only one blastomere of the 2 cell embryo was microinjected . ICR or Cdx2-eGFP ( to aide embryo staging at the time of doxycycline addition ) embryos were used . Microinjection was performed using a Leica microscope , micromanipulators ( Leica Microsystems ) and a FemtoJet ( Eppendorf ) . A cocktail of two PiggyBAC ( PB ) plasmids and one mRNA was injected . The injection cocktail consisted of ( i ) PB-TAC-DNLats2-IRES-mCherry or PB-TAC-mCherry-IRES-mCherry ( mCherry only control ) ( ii ) PB-CAG-rtTA and ( iii ) PBase mRNA . For pronuclear injections ( i ) 5–10 ng/µl , ( ii ) 5–10 ng/µl and ( iii ) 20–40 ng/µl concentrations were used . For 2 cell cytoplasmic microinjections ( i ) 15 ng/µl , ( ii ) 15 ng/µl and ( iii ) 160 ng/µl concentrations were used . Transgenic embryos were collected using both microinjection methods with the same cell fate outcomes observed . However , we found that embryo development , as well as PiggyBAC integration efficiency was highly improved when 2 cell injections were used . Both microinjection methods typically generated mosaic embryos , in which mCherry positive ( DN Lats2-mCherry transgenic ) and mCherry negative ( non-transgenic control ) cells could be analyzed . PB-TAC-DNLats2-IRES-mCherry: PB – PiggyBAC terminal repeat; TAC - tamoxifen activated minimal CMV promoter; DN Lats2 - dominant negative Lats2 ( previously referred to as kdLats2 [Nishioka et al . , 2009] ) . DN Lats2 was generously provided by Hiroshi Sasaki and was cloned into the PB-TAC IRES-mCherry construct , generously provided by Andras Nagy . PB-TAC mCherry-IRES-mCherry and PB-CAG-rtTA plasmids were also provided by Andras Nagy . For producing PBase mRNA , the PBase coding region was cloned into the pCS2+ vector . In vitro transcription was performed as described before . Injected embryos were cultured to desired stage , imaged to ensure no leaky expression of DN Lats2-mCherry was present and induced with 1 µg/ml doxycycline ( Sigma ) for 24 hr . Transgenic embryos were identified by the presence of mCherry . A total of 42 DN Lats2-mCherry and 20 mCherry control transgenic embryos were made in eight independent experiments that had strong mCherry expression in inner cells . No statistical method was used to predetermine sample size . For each experiment 10–15 females were used to harvest embryos . Pre-established criteria were used for all experiments to stage embryos ( reported in Figure 1—figure supplement 1 ) . Randomization was achieved by pooling all staged embryos in one experiment , with the exception of RNA sequencing experiments . For RNA sequencing experiments embryo information was preserved , however no embryo batch effect was noted when single cell expression profiles were analyzed . The investigator was not blinded to group allocation during experiments and outcome assessment , with the exception of RNA sequencing experiments . Cdx2-eGFP measurements of single cells and RNA sequencing of the same cells were performed by different investigators and outcomes were thus blindly determined . Data were analyzed by two-sided t-test or one-way ANOVA ( Graphpad Prism ) . Data were tested for normality of residues ( D'Agostino and Pearson omnibus normality test ) and homogeneity of variances ( F test or Bartlett's test ) for each variable . When data failed to satisfy those premises a non-parametric test was chosen ( Mann-Whitney or Kruskal-Wallis test ) . For Figure 6E and Figure 6—figure supplement 2 Chi-squared test was used to test the null hypothesis that cell fate is not affected by DN Lats2-IRES-mCherry or mCherry-IRES-mCherry expression . Expression of DN Lats2-IRES-mCherry or mCherry-IRES-mCherry was considered the independent variable and cell fate was the dependent variable . Significance of overlap between TE-specific genes ( all stages of TE development considered , n = 404 ) and genes associated with Tead4 binding sites in mouse trophoblast stem cells ( n = 8190 , union of overlapping , 3’ and 5’ genes ) as reported by Home et al . ( 2012 ) were determined using hypergeometric test , assuming 18 , 388 genes with experimentally-based functional annotations ( MGI ) . Figure 3D was created using data from Home et al . overlapped with TE-specific genes in NAViGaTOR 3 ( http://ophid . utoronto . ca/navigator ) ( RRID: SCR_008373 ) ( Brown et al . , 2009 ) .	In female mammals , conception is a complex process that involves several stages . First , an egg is released from the ovary and travels along a tube called the oviduct , where sperm from a male may fertilize it . If the egg is fertilized , the newly formed embryo moves into the womb , where it will then implant into the walls . In mice , it takes around four days for the embryo to implant and during this time , the cells in the embryo divide several times and start to specialize to form distinct cell types called lineages . The first two lineages to form are known as the inner cell mass and the trophectoderm . The inner cell mass forms a ball of cells within the embryo and contains the precursors of all cells that build the animal’s body . The trophectoderm forms a layer that surrounds the inner cell mass and will become part of the placenta ( the organ that supplies the embryo with nutrients while it is in the womb ) . The embryo can organize these lineages without any instructions from the mother . However , it is still not clear when the cells start to differ from each other , and when they ‘commit’ to stay in these lineages . Cells in the inner cell mass and trophectoderm have different gene expression profiles , meaning that many genes display different levels of activity in these two lineages . Posfai et al . use a technique called single-cell RNA sequencing to analyse gene activity as the inner cell mass and trophectoderm form in mouse embryos . By measuring changes in gene activity , it is possible to track their development and show which genes change expression levels when each lineage specifies and commits . The experiments reveal that the inner cell mass and trophectoderm lineages develop at different times . As the inner cell mass forms , cells adopt the inner cell mass ‘identity’ before they commit to remaining in this lineage , revealing a window of time where different signals could still change the fate of the cells . However , when the early trophectoderm cells show the first signs of specialization , they also commit to their new identity at the same time . These findings suggest that the different timings at which these cell lineages form might provide embryos with the means to organize their own cells . An important future challenge is to understand exactly how the cells commit to their fate .	[ "Abstract", "Introduction", "Results", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2017	Position- and Hippo signaling-dependent plasticity during lineage segregation in the early mouse embryo
Regulated nuclear translocation of the PER/CRY repressor complex is critical for negative feedback regulation of the circadian clock of mammals . However , the precise molecular mechanism is not fully understood . Here , we report that KPNB1 , an importin β component of the ncRNA repressor of nuclear factor of activated T cells ( NRON ) ribonucleoprotein complex , mediates nuclear translocation and repressor function of the PER/CRY complex . RNAi depletion of KPNB1 traps the PER/CRY complex in the cytoplasm by blocking nuclear entry of PER proteins in human cells . KPNB1 interacts mainly with PER proteins and directs PER/CRY nuclear transport in a circadian fashion . Interestingly , KPNB1 regulates the PER/CRY nuclear entry and repressor function , independently of importin α , its classical partner . Moreover , inducible inhibition of the conserved Drosophila importin β in lateral neurons abolishes behavioral rhythms in flies . Collectively , these data show that KPNB1 is required for timely nuclear import of PER/CRY in the negative feedback regulation of the circadian clock . Nearly all organisms have circadian clocks , internal timing systems that anticipate and adapt physiology and behavior to daily changes in the light–dark cycle . In eukaryotes , the clock is an interlocked transcriptional–translational feedback loop , which drives rhythmic expression of both core clock genes that make up the oscillator and clock-output genes that drive biological rhythms . These include rhythms in the sleep wake cycle , body temperature , hormone secretion , and energy metabolism ( Reppert and Weaver , 2002; Bass and Takahashi , 2010 ) . Many components and the molecular mechanism of the circadian clock are conserved between flies and mammals . Conserved components include the bHLH-PAS transactivators , CLOCK/NPAS2 ( fly Clk ) and BMAL1 ( fly Cyc ) . This heterodimer binds E-box elements in the promoters of two families of transcriptional repressors , the PER1/PER2/PER3 ( fly Per ) and CRY1/CRY2 ( fly Cry ) genes . Once translated , PERs and CRYs ( Per and Tim in flies ) accumulate in a cytoplasmic complex and then translocate into the nucleus to repress the CLOCK-BMAL1 heterodimer , thereby repressing their own transcription ( Panda et al . , 2002 ) . This primary negative-feedback loop is complemented by an ancillary loop , which while conserved , has different components in mammals and flies . In mammals , the ROR element binding repressors REV-ERBα/β and activators RORα/β/γ drive rhythmic expression of BMAL1 . In flies , vrille ( dVRI ) or PAR-domain protein-1 ( dPDP1 ) make up this ancillary loop and stabilize the Drosophila clock by driving rhythmic expression of Clk ( Yu and Hardin , 2006 ) . Further modulation of these feedback circuits occurs at the post-translational level as core clock proteins undergo phosphorylation , nuclear translocation , and protein degradation by several conserved kinases ( CKIδ/CKIε , GSK3β; double-time [dDBT] and shaggy [dSGG] ) , phosphatases ( PP2A; dPP2A ) , and E3 ubiquitin ligases ( FBXL3 , FBXL21; dSLMB1 ) ( Gallego and Virshup , 2007; Hirano et al . , 2013; Yoo et al . , 2013 ) . KPNB1 is a nuclear import receptor that mediates docking of nuclear localization signal ( NLS ) -containing cargo bound to importin α to the nuclear envelope , thereby facilitating nuclear entry of the target protein ( van der Watt et al . , 2013 ) . KPNB1 can also directly recognize the cargo and facilitate nuclear transport independently of importin α ( Lam et al . , 1999; Forwood et al . , 2001; Ghildyal et al . , 2005 ) . In several intensive molecular analyses , PER1/2 and CRY1/2 , have been found to contain a functional NLS for their nuclear entry and transcriptional feedback function ( Vielhaber et al . , 2000; Miyazaki et al . , 2001; Hirayama et al . , 2003; Zhu et al . , 2003 ) . Furthermore , nucleocytoplasmic shuttling of PER proteins , in conjunction with protein dimerization , phosphorylation and turnover , is critical for clock function ( Yagita et al . , 2000 , 2002 ) . However , the specific carrier molecules responsible for nuclear translocation of the clock proteins have not been fully elucidated . Interestingly , a previous biochemical and cellular study reported that nuclear transport of mouse Cry2 is mediated by its direct interaction with importin α family members ( RCH1 , QIP-1 , NPI-2 ) ( Sakakida et al . , 2005 ) . However , the functional relevance of these factors in the clock was not evaluated ( Sakakida et al . , 2005 ) . Moreover , the role of importin β was not investigated . In recent studies , KPNB1 was identified as a key nuclear importin component of ncRNA repressor of nuclear factor of activated T cells ( NFAT ) ( NRON ) ( lncRNA repressor of the NFAT ) RNA-protein-scaffold complex containing several kinases , CKIε , GSK3β , and DYRK2 that regulate NFAT phosphorylation and nuclear translocation in human and Drosophila cells ( Gwack et al . , 2006; Liu et al . , 2011; Sharma et al . , 2011 ) . Notably , most of these kinases are also well-known to regulate phosphorylation-related proteolysis and nuclear entry of circadian clock proteins . These data suggested that the clock may share a common molecular mechanism of nuclear translocation with NFAT . Here , we tested this hypothesis and report that KPNB1 mediates circadian nuclear translocation and feedback repression activities of the PER/CRY complex through direct complex formation , independently of importin α . Loss of KPNB1 abolished not only rhythmic gene expression in human cells , but also circadian behavior in Drosophila . Collectively , these results highlight the evolutionarily conserved role of the KPNB1 in regulating nuclear translocation and function of the circadian clock . The NRON complex is composed of a variety of proteins involved in signal transduction , proteolysis , and nuclear transport of NFAT ( Willingham et al . , 2005; Sharma et al . , 2011 ) . We noted that three of these components , CK1ε ( dco , Dbt ) , GSK3β ( sgg ) , and DYRK1a ( mnb ) , are clock components in both flies and mammals . Moreover , NFAT nuclear translocation has been suggested to be mediated by KPNB1 ( Willingham et al . , 2005; Sharma et al . , 2011 ) . Thus we hypothesized that KPNB1 would also be closely involved in nuclear entry of the PER/CRY complex and thus circadian clock function . To test this , first , we explored how KPNB1 knockdown affects cellular localization of the core clock factors . Interestingly , in immunofluorescence ( IF ) analysis of cells expressing several flag-tagged clock proteins , knockdown of KPNB1 completely blocked ( PER1 , PER2 ) , partly blocked ( CRY1 ) , or didn't significantly block ( CRY2 , REVERBα , CLOCK , CHRONO [Anafi et al . , 2014] ) nuclear accumulation ( Figure 1—figure supplement 1 ) . Similar to the flag-tagged proteins , microscopy and immunoblot analysis of cells expressing the Venus ( V ) -tagged clock proteins ( PER1-V , PER2-V , CRY1-V , CRY2-V ) , KPNB1 knockdown markedly blocked PER1 and PER2 localization and slightly or negligibly effected CRY1 and CRY2 nuclear localization ( Figure 1A , B ) . In many previous studies , PER and CRY proteins have been shown to work together ( Tamanini et al . , 2005 ) . Hence , we looked into the effect of KPNB1 knockdown on the subcellular localization of various PERs/CRYs complexes using bimolecular fluorescence complementation ( BiFC ) ( Shyu et al . , 2006 ) . The BiFC assay showed that KPNB1 depletion significantly impaired nuclear localization of various combinations of PER and CRY proteins ( Figure 1C ) . Notably , nuclear accumulation of the PER2/CRY1 complex was totally blocked by KPNB1 knockdown ( Figure 1C ) . In addition to ectopically expressed clock proteins , KPNB1 depletion substantially increased cytoplasmic accumulation of endogenous PER1/2 and CRY1/2 proteins thus reducing their relative levels in nuclei in most cells ( Figure 1D , Figure 1—figure supplement 2 ) . Taken together , these evidence show that KPNB1 serves as an integral nuclear transport receptor of the PER/CRY complex . 10 . 7554/eLife . 08647 . 003Figure 1 . KPNB1 knockdown blocks nuclear translocation of PERs/CRYs complex . ( A ) Fluorescence microscopic analysis of the effect of KPNB1 depletion on subcellular localization of PER1 , PER2 , CRY1 , and CRY2 . Representative fluorescence images of U2 OS cells expressing Venus-tagged PER1 , PER2 , CRY1 , and CRY2 ( PER1-V , PER2-V , CRY1-V , CRY2-V ) in the presence of control and KPNB1 siRNA ( left ) . Quantification of the subcellular distribution of the imaged cells ( right ) . Subcellular localization is categorized as nuclear ( Nuc . ; orange ) , cytoplasmic and nuclear ( Cyto . /Nuc . ; gray ) , and cytoplasmic ( Cyto . ; blue-green ) . More than 100 fluorescent cells for each of the images were evaluated . Scale bar: 30 µm . ( B ) Immunoblot analysis of nuclear and cytoplasmic extracts of U2 OS cells expressing PER1-V , PER2-V , CRY1-V , and CRY2-V in the presence of control and KPNB1 siRNA using the indicated antibodies to GFP for Venus fused proteins ( αGFP ) , KPNB1 ( αKPNB1 ) , Tubulin for cytoplasmic fraction ( C ) marker ( αTubulin ) , and TATA box binding protein ( TBP ) for nuclear fraction ( N ) marker ( αTBP ) . Representative images from two independent experiments are shown . ( C ) Bi-molecular fluorescence complementation analysis of the effect of KPNB1 depletion on subcellular localization of PER1/2-CRY1/2 dimeric complex in various combinations in U2 OS cells . Representative fluorescence images of cells expressing Venus N terminal ( VN ) or C-terminal ( VC ) -tagged PER1 , PER2 , CRY1 , and CRY2 ( PER1-VN , PER2-VN , CRY1-VC , CRY2-VC ) as indicated combinations in the presence of control ( siCTL ) and KPNB1 siRNA ( siKPNB1 ) ( left ) . Quantification of the subcellular distribution of the imaged cells ( right ) . Subcellular localization is categorized as nuclear ( Nuc . ; orange ) , cytoplasmic and nuclear ( Cyto . /Nuc . ; gray ) , and cytoplasmic ( Cyto . ; blue-green ) . More than fluorescent 100 cells for each of the images were evaluated . Scale bar: 30 µm . ( D ) Immunoblot analysis of nuclear and cytoplasmic extracts of U2 OS cells in the presence of control and KPNB1 siRNA using the indicated antibodies for endogenous PER1 ( αPER1 ) , PER1 ( αPER2 ) , CRY1 ( αCRY1 ) , CRY2 ( αCRY2 ) , KPNB1 ( αKPNB1 ) , Tubulin for cytoplasmic fraction ( C ) marker ( αTubulin ) , and Lamin for nuclear fraction ( N ) marker ( αLamin ) . Representative images from three independent experiments are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08647 . 00310 . 7554/eLife . 08647 . 004Figure 1—figure supplement 1 . Inhibition of KPNB1 alters nuclear localization of core clock repressor proteins . ( A ) Quantitative RT-PCR analysis of KPNB1 knockdown efficiency in control ( siCTL ) and KPNB1 siRNA ( siKPNB1 ) -treated U2 OS cells . The data presented are the means ± S . E . of triplicate samples . ( B ) Representative images of IF detection of endogenous expression of KPNB1 using anti-KPNB1 ( αKPNB1 ) in control and KPNB1 siRNA-treated U2 OS cells . Scale bar: 10 μm . ( C ) Fluorescence microscopic analysis of KPNB1 knockdown effect on subcellular localization of PER1 , PER2 , CRY1 , and CRY1 in U2 OS cells . Representative images were captured after immunostaining cells expressing flag-tagged PER1 , PER2 , CRY1 , CRY2 , CLOCK , REV-VERBα , CHRONO ( PER1Flag , PER2Flag , CRY1Flag , CRY2Flag , CLOCKFlag , REV-VERBαFlag , CHRONOFlag ) in the presence of control or KPNB1 siRNA ( left panels ) . The images were taken with fluorescence imaging microscopy using specific filter sets for TRITC ( Red; Flag tagged protein ) and DAPI ( Blue; Nuclei ) . DAPI ( Blue ) merged images was presented in each of the image panels . For quantification analysis , subcellular localization is categorized as nuclear ( Nuc . ; orange ) , cytoplasmic and nuclear ( Cyto . /Nuc . ; gray ) , and cytoplasmic ( Cyto . ; blue-green ) ( right panels ) . More than 100 cells for each of the images were evaluated . Scale bar: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08647 . 00410 . 7554/eLife . 08647 . 005Figure 1—figure supplement 2 . IF analyses of KPNB1 depletion effect on subcellular localization of endogenous PER1 , PER2 , CRY1 , and CRY1 . Representative images were captured after immunostaining endogenous PER1 , PER2 , CRY1 , and CRY2 with anti-PER1 ( αPER1 ) , anti-PER2 ( αPER2 ) , anti-CRY1 ( αCRY1 ) , or anti-CRY2 ( αCRY2 ) in control ( siCTL ) and KPNB1 siRNA ( siKPNB1 ) -treated U2 OS cells . DAPI ( Blue ) merged images was presented below each of the image panels . Scale bar: 20 μm . For quantification of subcellular localization , nuclear ( Nuc . ; orange ) and cytoplasmic ( Cyto . ; blue-green ) signals in cells ( CTL , n = 10; siKPNB1 , n = 10 ) for each of the image panels were measured using densitometer and presented in the bottom . Asteriks denote the statistical significance ( *p < 0 . 0005 , by Student's t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08647 . 005 To investigate the detailed mechanism of the PER/CRY nuclear entry by KPNB1 , we performed co-immunoprecipitation experiments with cells expressing the Venus-tagged PER and CRY proteins . KPNB1 interacted more strongly with PER1 and PER2 proteins than either CRY1 or CRY2 ( Figure 2 , Figure 2—figure supplement 1A , B ) . This suggested that PER proteins may play the leading role in PERs/CRYs nuclear translocation with KPNB1 . In support of this , by IF KPNB1 co-localized with both cytoplasmic and nuclear PER2 ( Figure 2B , Figure 2—figure supplement 1C ) . Correspondingly , KPNB1 strongly interacted with the endogenous PER2 in co-immunoprecipitation analysis using mouse liver tissue extracts ( Figure 2C ) . To investigate the circadian role of KPNB1 in nuclear localization of PER/CRY , we prepared cytoplasmic and nuclear extracts from mouse liver tissues collected at 4 hr intervals for 24 hr in constant darkness . By immunoblot analysis , KPNB1 exhibited a circadian pattern of nucleocytoplasmic localization with its nuclear abundance peaking at circadian time ( CT ) 14–18 , when the intensive nuclear accumulation of PER2 and CRY1 occurred ( Figure 2D ) . Concomitantly , we observed the significant decrease of mRNA expression of E-box dependent clock genes ( Per1 , Dbp , Nr1d1 , Nr1d2 ) targeted by the clock repressors in their nuclear accumulation phase ( Figure 2E ) . Thus , these data clearly suggest that KPNB1 is responsible for circadian nuclear import and repressor activity of the PER/CRY complex in clock gene expression in mouse liver . 10 . 7554/eLife . 08647 . 006Figure 2 . KPNB1 interacts with PER2 for circadian nuclear entry and repressor activity of PER/CRY . ( A ) Immunoprecipitation ( IP ) analysis of interactions of KPNB1 with PER2 . U2 OS cells were transfected with Venus-tagged PER2 , and then immunoprecipitated with anti-GFP antibody ( IP: αGFP ) . The immunoprecipitates were analyzed by immunoblotting with anti-KPNB1 ( IB: αKPNB1 ) or anti-GFP antibody ( IB: αGFP ) . Representative images from three independent experiments are shown . ( B ) Immunofluorescence ( IF ) analyses of subcellular colocalization of PER2 with KPNB1 . U2 OS cells expressing PER2-V were fixed and immunostained with antibody to endogenous KPNB1 . The representative images were captured by fluorescence imaging microscopy using specific filter sets for FITC ( green; PER-V ) , TRITC ( Red; αKPNB1 ) , and DAPI ( Blue; Nuclei ) . See Figure 2—figure supplement 1 . Scale bar: 10 µm . ( C ) Coimmunoprecipitation of PER2 with endogenous KPNB1 . Liver extracts were immunoprecipitated ( IP ) with anti-KPNB1 or control IgG antibodies , and the immunoprecipitated proteins were probed with antibodies to PER2 , KPNB1 as well as GAPDH as negative control and IgG light chain as a positive control for the IP . Representative images from three independent experiments are shown . ( D ) Immunoblotting analysis using cytoplasmic and nuclear extracts prepared from mouse liver tissues collected at 4 hr interval as indicated for 24 hr in constant darkness ( CT: Circadian time ) . Anti-PER2 , anti-CRY1 , and anti-KPNB1 antibodies were used for detecting endogenous PER2 , CRY1 , and KPNB1 proteins . Anti-Tubulin antibody ( αTubulin ) for cytoplasmic fraction ( Cyto ) marker and anti-Lamin ( αLamin ) for nuclear fraction ( Nuc ) marker were used for loading controls respectively . Representative images from two independent experiments are shown . ( E ) Circadian expressions of endogenous clock gene mRNAs ( Per1 , Dbp , Nr1d1 , Nr1d2 ) were determined by quantitative RT-PCR analysis of mouse liver tissue samples collected at 4 hr interval as indicated for 24 hr in constant darkness ( CT; Circadian time ) . The data presented are the means ± S . E . of triplicate samples . DOI: http://dx . doi . org/10 . 7554/eLife . 08647 . 00610 . 7554/eLife . 08647 . 007Figure 2—figure supplement 1 . KPNB1 interacts and colocalizes with PER2 . ( A ) IP analysis of interactions of KPNB1 with PER1 , PER2 , CRY1 , and CRY2 . U2 OS Cells were transfected with Venus-tagged PER1 , PER2 , CRY1 , and CRY2 ( PER1-V , PER2-V , CRY1-V , CRY2-V ) and then immunoprecipitated with anti-GFP antibody ( IP: αGFP ) . The immunoprecipitates were analyzed by immunoblotting with anti-KPNB1 ( IB: αKPNB1 ) or anti-GFP antibody ( IB: αGFP ) . Representative images from three independent experiments are shown . ( B ) The intensity of the immunoprecipited protein bands in the data shown in ( A ) was quantified by densitometry ( ImageJ ) , and the data are shown as means ± S . E . of three independent experiments . ( C ) IF analyses of subcellular colocalization of PER2 with KPNB1 . U2 OS cells expressing PER2-V were fixed and immunostained with antibody to endogenous KPNB1 . The representative images were captured by fluorescence imaging microscopy using specific filter sets for FITC ( green; PER2-V ) , TRITC ( Red; αKPNB1 ) , and DAPI ( Blue; Nuclei ) . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08647 . 007 To investigate the role of KPNB1 in clock function , we used RNAi to knockdown KPNB1 in U2 OS cells and found substantially elevated Per1 promoter activity , consistent with the hypothesis that this knockdown inhibited the localization and function of the PER/CRY complex ( Figure 3A , Figure 3—figure supplement 1 ) . In contrast , the parallel depletion of the other NRON complex nucleocytoplasmic transport proteins , TNPO1 , CSE1L , did not significantly affect their repressor activities ( Figure 3A , Figure 3—figure supplement 1 , [Willingham et al . , 2005] ) . In support of this , chromatin immunoprecipitation ( ChIP ) analysis targeting the first E-box region of human PER1 ( hPer1 ) promoter revealed significantly less recruitment of both PER1 and PER2 to the CLOCK/BMAL1 responsive element in the KPNB1-depleted U2 OS cells ( Figure 3B–D ) . Moreover , the overall expression levels of several E-box dependent genes ( PER1 , CRY1 , DBP , REV-ERBβ ) were upregulated as a result of KPNB1 knockdown ( Figure 3E ) , consistent with loss of repressor function . Conversely , overexpression of KPNB1 markedly down-regulated CLOCK/BMAL1-activated PER1 transcription , probably by facilitating PER/CRY localization to the nucleus ( Figure 3—figure supplement 2 ) . Correspondingly , KPNB1 overexpression significantly down-regulated endogenous mRNA expression of these E-box-regulated genes with no significant effect on BMAL1 expression ( Figure 3E ) . 10 . 7554/eLife . 08647 . 008Figure 3 . KPNB1 mediates PERs/CRYs-regulated transcription of E-box dependent clock genes . ( A ) HEK293T cells were transiently transfected with Per1-Luc reporter construct alone or cotransfected with plasmids expressing CLOCK , BMAL1 , PER1 , PER2 , CRY1 , CRY2 , in the presence of control ( siCTL: black ) , CSE1L ( siCSE1L: light green ) , KPNB1 ( siKPNB1; gray red ) , and TNPO1 siRNA ( siTNPO1: violet ) as indicated . After 24 hr , the cells were lysed and Per1 promoter-driven luciferase activities were measured and normalized with pRL-TK activity . Results of one representative experiment of three independent experiments are shown . ( B ) Schematic diagram of the human PER1 promoter and primers used for ChIP assay . ( C , D ) ChIP assay of PER1 or PER2 binding to the E box in hPER1 . Control ( siCTL: black ) and KPNB1 siRNA ( siKPNB1; gray red ) -treated cells ( U2 OS ) were subjected to ChIP assays using anti-PER1 ( αPER1 ) or anti-PER2 ( αPER2 ) antibody . ChIP DNA samples were quantified by quantitative real-time RT-PCR . The data presented are the means ± S . E . of triplicate samples ( p < 0 . 005 , p < 0 . 05 , by Student's t-test ) . ( E ) Quantitative real-time RT-PCR analysis of expression of endogenous PER1 , CRY1 , DBP , REVERBβ , and BMAL1 mRNAs in control ( siCTL: black ) , KPNB1-depleted ( siKPNB1; gray red ) or overexpressed ( KPNB1OV; blue green ) cells ( U2 OS ) . The data presented are the means ± S . E . of triplicate samples ( p < 0 . 005 , p < 0 . 05 , by Student's t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08647 . 00810 . 7554/eLife . 08647 . 009Figure 3—figure supplement 1 . Depletion of KPNB1 reduces repressional activity of PER and CRY proteins . ( A ) Effects of knockdown of TNPO1 , KPNB1 , CSE1L on Per1 promoter activity . HEK293T cells were transiently transfected with Per1-Luc reporter construct in the presence of control siRNA ( siCTL: dark blue ) or four siRNAs targeting each of CSE1L ( siCSE1L 1–4: light green ) , KPNB1 ( siKPNB1 1–4; gray red ) , and TNPO1 siRNA ( siTNPO1 1–4: violet ) as indicated . After 24 hr , cells were lysed and Per1 promoter-driven luciferase activities were measured and normalized with pRL-TK activity . Results of one representative experiment of three independent experiments are shown . ( B ) Quantitative RT-PCR analysis of knockdown efficiencies of four siRNAs targeting each of CSE1L ( siCSE1L 1–4: light green ) , KPNB1 ( siKPNB1 1–4; gray red ) , and TNPO1 siRNA ( siTNPO1 1–4: violet ) introduced in HEK293T cells as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 08647 . 00910 . 7554/eLife . 08647 . 010Figure 3—figure supplement 2 . Deletion of KPNA2 , importin α binding domain is not required for regulatory function of KPNB1 in clock gene transcription . ( A ) Schematic illustration of full-length ( 1–876 ) and mutant ( 1–771 ) KPNB1 engineered with a deletion of the importin-α binding domain on the C-terminus . ( B ) Deletion of the importin-α binding domain does not affect the down-regulation of Per1 promoter activity by KPNB1 . HEK293T cells were transiently transfected with Per1-Luc reporter construct alone ( left ) or cotransfected with plasmids expressing full-length and mutant KPNB1 ( KPNB1-WT , KPNB1-[1–771] ) as indicated combination . After 24 hr , cells were subjected to lysis and Per1 promoter-driven luciferase activities were measured and normalized with pRL-TK activity . The data presented are the means ± S . E . of triplicate samples . ( C ) Deletion of the importin-α binding domain does not affect the down-regulation of CLOCK-BMAL1-mediated Per1 transcription by KPNB1 . HEK293T cells were transiently transfected with Per1-Luc reporter construct alone ( left ) or cotransfected with plasmids expressing CLOCK , BMAL1 , full-length or importin-α binding domain deficient mutant of KPNB1 ( KPNB1-WT , KPNB1-[1–771] ) as indicated combination . Similar transcriptional analysis was performed , as in ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08647 . 010 Notably , the KPNB1 activity was not dependent on importin α , a well-known nuclear transport partner , as deletion of the importin α binding domain of KPNB1 did not significantly affect the transcriptional regulation ( Figure 3—figure supplement 2A–C ) ( Kutay et al . , 1997 ) . Importantly , knockdown of KPNB1 in U2 OS cells severely disturbed circadian bioluminescence reporter activity driven by the Per2 promoter , whereas depletion of importin α1 ( KPNA2 ) or importin α5 ( KPNA1 ) did not affect the circadian rhythmicity ( Figure 4A , B , Figure 4—figure supplement 1A , B , Figure 4—figure supplement 3A , B ) . These knockdown phenotypes correspond to previous reports showing that most of the other importin α isoforms were not critical for circadian rhythmicity ( Wu et al . , 2009; Zhang et al . , 2009 ) . In conjunction with these results , fluorescence microscopy analysis revealed that KPNA2 or KPNA1 depletion did not alter PER2 nuclear localization , which was completely blocked by KPNB1 depletion in U2 OS cells stably expressing PER2-Venus ( Figure 4—figure supplement 1C , D , Figure 4—figure supplement 3C , D ) . On the other hand , KPNB1 depletion severely disrupted circadian transcription of several clock genes ( PER1 , PER2 ) in dexamethasone ( Dex ) -synchronized cells with altered nuclear accumulation of PER and CRY proteins ( Figure 4C , D ) . Taken together , these data show that KPNB1 gates PER and CRY localization to the nucleus independently of importin α and plays an indispensable role in negative feedback regulation of the circadian clock in human cells . 10 . 7554/eLife . 08647 . 011Figure 4 . KPNB1 is required for circadian gene expression . ( A ) Immunoblot analysis of KPNB1 knockdown efficiency with anti-KPNB1 ( αKPNB1 ) in control ( siCTL ) and KPNB1 siRNA ( siKPNB1 ) -treated U2 OS cells . Anti-GAPDH antibody ( αGAPDH ) was used as a loading control . ( B ) Bioluminescence recordings of dexamethasone ( Dex ) -synchronized control ( siCTL ) and increasing dose of KPNB1 siRNA ( siKPNB1 0 . 5×/1× ) -treated U2 OS cells expressing Per2 promoter-driven destabilized luciferase ( left; pPer2-dLuc ) . The Bioluminescence recordings were detrended by a 24-hr moving average subtraction ( right ) . ( C ) Altered rhythmic expression of KPNB1 , PER1 , PER2 , and CRY1 transcripts by KPNB1 depletion . mRNA expressions of the target genes were determined by quantitative RT-PCR in control or KPNB1-depleted U2 OS cells over the course of 44 hr after 24 hr upon Dex treatment . The data are shown with the mean value of triplicate samples in each time point . ( D ) KPNB1 knockdown alters rhythmic nuclear accumulation of PER and CRY proteins . Immunoblotting analysis using cytoplasmic and nuclear extracts prepared from control or KPNB1-depleted U2 OS cells collected at 4 hr interval over the course of 44 hr after 24 hr of Dex treatment . Anti-PER2 , anti-CRY1 , anti CRY2 , and anti-KPNB1 antibodies were used for detecting endogenous PER2 , CRY1 , CRY2 and KPNB1 proteins . Anti-Tubulin antibody ( αTubulin ) for cytoplasmic fraction ( Cyto ) marker and anti-TATA binding protein ( αTBP ) for nuclear fraction ( Nuc ) marker were used for loading controls respectively . Representative images from two independent experiments are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08647 . 01110 . 7554/eLife . 08647 . 012Figure 4—figure supplement 1 . Inhibition of KPNA2 does not affect PER2 nuclear localization and cellular clock rhythmicity . ( A ) Immunoblot analysis of KPNA2 ( importin-α ) knockdown efficiency with anti-KPNA2 ( αKPNA2 ) in U2 OS cells treated with control siRNA ( siCTL ) or each of three KPNA2 siRNAs ( siKPNA2-1 , siKPNA2-2 , siKPNA2-3 ) . Anti-GAPDH antibody ( αGAPDH ) was used as a loading control . ( B ) KPNA2 depletion does not affect circadian promotor activity rhythm . 48 hr after treatment with control siRNA ( siCTL ) or each of three KPNA2 siRNAs ( siKPNA2-1 , siKPNA2-2 , siKPNA2-3 ) in pPer2-dLuc reporter cells ( U2 OS ) , the cells were synchronized with Dex and monitored in real-time bioluminescence recorder . ( C ) Representative images were captured after immunostaining endogenous KPNA2 with anti-KPNA2 ( αPER1 ) in control ( siCTL ) and KPNB1 siRNA ( siKPNB1 ) -treated U2 OS cells . ( D ) KPNA2 depletion does not affect subcellular localization of PER2 . Fluorescence microscopic analysis of the effect of KPNA2 or KPNB1 depletion on subcellular localization of PER2 . Representative fluorescence images of U2OS cells stably expressing PER2-Venus ( PER2-V ) in the presence of control ( siCTL: upper panels ) , KPNA2 ( siKPNA2: middle panels ) , or KPNB1 siRNA ( siKPBN1: lower panels ) were captured by fluorescence imaging microscopy using specific filter sets for FITC ( green; PER2-V ) and and DAPI ( Blue; Nuclei ) . ( E ) Effect of KPNA2 overexpression on PER2 localization . Representative fluorescence images of U2 OS cells stably expressing PER2-Venus ( PER2-V ) with the transfection of control ( CTL-Sport6-OV: upper panels [20×] ) or KPNA2 encoding plasmids ( KPNA2-Sport6-OV: middle [20×] and bottom panels ) . The bottom panels show highly magnified images representative of middle panel images . The images were captured by fluorescence imaging microscopy using specific filter sets for FITC ( green; PER2-V ) , TRITC ( Red; αKPNA2 ) , and DAPI ( Blue; Nuclei ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08647 . 01210 . 7554/eLife . 08647 . 013Figure 4—figure supplement 2 . Inhibition of KPNA2 does not affect CRY2 nuclear localization . ( A ) Immunoblot analysis of nuclear and cytoplasmic extracts of U2 OS cells in the presence of control and KPNA1 siRNA using the indicated antibodies for endogenous CRY2 ( αCRY2 ) and KPNB1 ( αKPNB1 ) , Tubulin for cytoplasmic fraction ( C ) marker ( αTubulin ) , and TATA binding protein ( TBP ) for nuclear fraction ( N ) marker ( αTBP ) . ( B ) Fluorescence microscopic analysis of the effect of KPNA2 depletion on subcellular localization of CRY2 . Representative fluorescence images of subcellular localization of endogenous CRY2 ( upper panels ) , or ectopically expressed Flag-CRY2 ( middle panels ) , CRY2-Venus ( CRY2-V; lower panels ) in U2 OS cells treated with control siRNA ( siCTL; left panels ) or KPNA2 siRNA ( siKPNA2; right panels ) . The images were taken using specific filter sets for TRITC ( Red; αCRY2 , Flag-CRY2 ) , FITC ( green; PER2-V ) , and DAPI ( Blue; Nuclei ) . DAPI ( Blue ) merged images was presented in each of the image panels . DOI: http://dx . doi . org/10 . 7554/eLife . 08647 . 01310 . 7554/eLife . 08647 . 014Figure 4—figure supplement 3 . Inhibition of KPNA1 ( importin α5 ) does not significantly affect PER2 and CRY2 nuclear localization and cellular clock rhythmicity . ( A ) Quantitative RT-PCR analysis of knockdown efficiencies of three siRNAs targeting KPNA1 ( siKPNA1-1; gray red , siKPNA1-2; light green , siKPNA1-3; violet ) , introduced in U2 OS cells as indicated . ( B ) KPNA1 depletion does not affect circadian promotor activity rhythm . 48 hr after treatment with control siRNA ( siCTL ) or each of three KPNA1 siRNAs ( siKPNA1-1 , siKPNA1-2 , siKPNA1-3 ) in pPer2-dLuc reporter cells , the cells were synchronized with Dex and monitored in real-time bioluminescence recorder . ( C ) KPNA1 depletion does not affect subcellular localization of PER2 . Fluorescence microscopic analysis of the effect of KPNA1 or KPNB1 depletion on subcellular localization of PER2 . Representative fluorescence images of U2 OS cells stably expressing PER2-Venus ( PER2-V ) in the presence of control ( siCTL: upper panels ) , KPNA1 ( siKPNA2: middle panels ) , or KPNB1 siRNA ( siKPBN1: lower panels ) were captured by fluorescence imaging microscopy using specific filter sets for FITC ( green; PER2-V ) and and DAPI ( Blue; Nuclei ) . ( D ) Fluorescence microscopic analysis of the effect of KPNA1 depletion on subcellular localization of CRY2 . Representative fluorescence images of subcellular localization of endogenous CRY2 ( upper panels ) , or ectopically expressed Flag-CRY2 ( middle panels ) , CRY2-Venus ( CRY2-V; lower panels ) in U2 OS cells treated with control siRNA ( siCTL; left panels ) or KPNA1 siRNA ( siKPNA1; right panels ) . The images were taken using specific filter sets for TRITC ( Red; αCRY2 , Flag-CRY2 ) , FITC ( green; PER2-V ) , and DAPI ( Blue; Nuclei ) . DAPI ( Blue ) merged images was presented in each of the image panels . DOI: http://dx . doi . org/10 . 7554/eLife . 08647 . 01410 . 7554/eLife . 08647 . 015Figure 4—figure supplement 4 . KPNA isoforms are not critical for cellular rhythmicity . Knockdown of various KPNA isoforms does not significantly affect circadian promotor activity rhythm ( Blue = Negative control [Control siRNA] , Green = Positive control [CRY2 siRNA] , Red = Data [Target KPNA siRNA] ) ( A ) KPNA1 ( importin α5 ) , ( B ) KPNA3 ( importin α4 ) , ( C ) KPNA4 ( importin α3 ) , ( D ) KPNA5 ( importin α6 ) , ( E ) KPNA6 ( importin α7 ) , ( F ) KPNA7 ( importin α8 ) . Each of data set ( A–F ) was retrieved from BioGPS website; http://biogps . org ( Wu et al . , 2009; Zhang et al . , 2009 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08647 . 015 The majority of core clock components are conserved between flies ( cycle , clk , per , cry ) and humans ( BMAL1/BMAL2 , CLOCK/NPAS2 , PER1/PER2/PER3 , CRY1/CRY2 ) , across 600 million years of evolution . Indeed , the kinases that regulate PER/CRY translocation to the nucleus are also conserved ( dbt , sg; CKIε/CKIδ , GSK3β ) . Because KPNB1 was required for rhythmicity in human cells , we investigated the role of the Drosophila ortholog , Fs ( 2 ) Ket , ketel , in the fly clock by RNAi followed by locomotor activity monitoring ( Mason and Goldfarb , 2009 ) . We used the UAS/Gal4 system to target UAS-ketel RNAi to clock cells in Drosophila ( Brand and Perrimon , 1993 ) . dicer was also co-expressed to increase the RNA interference effect ( Dietzl et al . , 2007 ) . Behavioral analysis of flies expressing UAS ketel RNAi in central clock cells , targeted by the relatively weak Pigment dispersing factor ( Pdf ) -Gal4 driver , revealed that 62% of flies exhibited no rhythms in constant darkness ( DD ) . Although 38% of flies were rhythmic , their rhythm strength determined by fast Fourier transform ( FFT ) value was low ( Figure 5A–C , Supplementary file 1 ) . With a strong tim-UAS-Gal4 driver that is expressed in all clock cells , 100% of flies were arrhythmic ( Figure 5A , B , Supplementary file 1 ) . To exclude possible developmental effects on behavioral phenotypes in ketel knockdown flies , we restricted the ketel dsRNA expression to adult PDF+ neurons using the pdf Gene Switch ( pdf-GS ) . We crossed ketel RNAi transgenic flies with pdf-GS and then drove ketel dsRNAs expression with a drug ( RU486 ) during adulthood ( McGuire et al . , 2004 ) . After 6–7 days in constant darkness , this conditional knockdown of ketel also resulted in weak to arrhythmic locomotor activity ( Figure 5D , Supplementary file 1 ) . 81% of flies were arrhythmic and the rest showed weak rhythmicity with low FFT values ( Figure 5E , Supplementary file 1 ) . These results show that importin β in PDF+ neurons is required for normal locomotor rhythms in adult flies . 10 . 7554/eLife . 08647 . 016Figure 5 . Inhibition of conserved importin β ( ketel ) abolishes circadian behavior in flies . ( A ) Flies in which ketel was downregulated in PDF+ cells and in all clock neurons were assayed for rest/activity rhythms in constant dark ( DD ) . ( B ) Summary of the circadian rhythms of various flies under DD . Quantification shows percentage of rhythmic ( R , orange ) and arrhythmic ( AR , gray ) flies ( n = 16–26 ) . ( C ) Some rhythmic Pdf-Gal4/+; dicer/+ flies showed weak rhythmicity with lower fast Fourier transform value compared to both ketel RNAi/+; dicer/+ and Pdf-GAL4/+; dicer/+ control flies ( p < 0 . 001 , p < 0 . 0001 , by Student's t-test ) ( See also Supplementary file 1 ) . ( D ) ketel knockdown in PDF+ cells during adulthoods leads to a long period , which eventually degenerates into arrhythmia in DD . They were fed either 500 mM RU486 or ethanol ( EtOH , vehicle control ) from the time of entrainment . ( E ) Rhythm strength of rhythmic RU486-treated flies was lower than that of EtOH-treated control flies after 7 days in DD ( p < 0 . 05 , by Student's t-test ) ( See also Supplementary file 1 ) . ( F , G ) PDF expression in ketel knockdown flies was only detected in large lateral ventral neurons ( l-LNvs ) , not in the small lateral ventral neurons ( s-LNvs ) . Downregulation of KETEL in PDF+ cells impairs nuclear translocation of PER in l-LNvs . In RU486-treated flies , PER expression in l-LNvs was detected in cytoplasm at ZT1 , a time when PER is nuclear in control flies ( ***p < 0 . 0001 , by Student's t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08647 . 016 To investigate whether this behavioral phenotype is related to nuclear translocation of clock proteins , we determined the cellular localization of PER in RU486 treated pdf-GS > ketel RNAi flies at ZT1 , when PER is nuclear in wild type flies . PDF is expressed in the large and small ventral lateral neurons ( LNvs ) , so we focused on PER expression in these cells . In ketel knockdown flies , PER levels were low and primarily cytoplasmic in large LNvs , while PER staining was intensely nuclear in control flies ( Figure 5F , G ) . The small LNvs were undetectable in ketel RNAi transgenic flies , as is also the case in other arrhythmic fly mutants such as Clk and cyc ( Figure 5F , right panel images ) ( Park et al . , 2000 ) . Taken together , these data suggest that like mammalian importin β , Ketel mediates the nuclear localization of PER , and consequently , is required for functioning of the core circadian clock . Timely nuclear translocation of the PER/CRY ( or Per/Tim ) repressor complex is a critical and conserved step in maintaining rhythmicity in metazoans . However , the functional carrier molecules responsible for nuclear localization of the feedback regulators have yet to be determined . Our study provides the first evidence that KPNB1 plays a key role as a nuclear transporter of the PER/CRY complex in negative feedback repression in both mammalian and fly clocks ( Figure 6 ) . 10 . 7554/eLife . 08647 . 017Figure 6 . Proposed model for KPNB1 function in nuclear translocation of PERs/CRYs controlling negative feedback regulation of the molecular clock . KPNB1 directly associates with PERs/CRYs and guides their nuclear entry thus facilitating negative feedback repression of CLOCK/BMAL1-mediated transcription of the repressor genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08647 . 017 Many mammalian and Drosophila clock studies have shown that phosphorylation and nuclear translocation of PERs and CRYs are closely coupled ( Gallego and Virshup , 2007 ) . We noted that CKIε , GSK3β , and DYRK2 , clock kinases in both mammals and flies , are also constituents of the NRON complex , which regulates cytoplasmic to nuclear translocation of NFAT and consequently its function in the immune systems ( Willingham et al . , 2005 ) . The NRON complex also contains several molecules involved in protein stability and , critical for this study , nuclear translocation ( KPNB1 , TNPO1 , CSEL1 ) . Subsequent study showed that KPNB1 is the major transport receptor for NFAT nuclear entry and activity upon its dephosphorylation ( Sharma et al . , 2011 ) . Thus , we investigated the role of KPNB1 on nuclear localization of PER/CRY complex ( Figure 1 , Figure 1—figure supplement 1 ) . We found that KPNB1 knockdown trapped PER and CRY proteins at the nuclear membrane , preventing their nuclear accumulation . Consistent with this , in our transcriptional study , we found KPNB1 knockdown significantly relieved PER1 transcriptional repression by the PER/CRY complex . Depletion of the other NRON complex translocation factors , TNPO1 and CSE1L , did not ( Figure 3 , Figure 3—figure supplement 1 ) , nor did they interact with PER or CRY proteins ( data not shown ) . These data suggest a specific role of KPNB1 as a nuclear transport receptor directly involved in negative feedback regulation of the core clock , in addition to its previously described role in NFAT signaling . Despite not playing a role in PER/CRY function , both TNPO1 and CSE1L displayed circadian gene expression in various mouse tissues , and also affected circadian period length ( TNPO1 ) or were required for cellular rhythms ( CSE1L ) when silenced ( [Pizarro et al . , 2013] , data not shown ) . Interestingly , the recent in situ hybridization analysis showed that TNPO1 exhibits circadian expression profiles in its mRNA in SCN ( Sato et al . , 2011 ) . Thus , these proteins may also play a role in mammalian clock function independent from PER/CRY , perhaps regulating import and export of other clock related components . Notably , in our biochemical analyses , KPNB1 strongly interacted with PER proteins and exhibited rhythmic nucleocytoplasmic localization and concurrent nuclear accumulation with PER/CRY in transcriptional repression phase ( Figure 2 ) . These data suggest KPNB1 drives circadian nuclear entry of PER/CRY . This raises the question: what causes phase-dependent KPNB1 nuclear entry ? Most recently , it was reported that intracellular calcium levels regulate importin β-dependent nuclear import ( Kaur et al . , 2014 ) . Further , a previous model suggests that KPNB1 nuclear-translocates NFAT immediately after stimulation dependent intracellular calcium influx ( Sharma et al . , 2011 ) . In this regard , several studies have demonstrated the important function of Ca2+ dynamics in circadian clock regulation . For example , periodic Ca2+ influx in the cytosol , resulting from circadian variation in membrane potential or by the release of Ca ( 2+ ) from ryanodine-sensitive stores in SCN neurons , is a critical process for circadian pacemaker function ( Ikeda et al . , 2003; Lundkvist et al . , 2005 ) . Based on this evidence , we postulate that cytosolic Ca2+ rhythms contribute to driving timely nuclear localization of KPNB1 and its PER/CRY cargo as a critical step in negative feedback regulation of the clock . In early models , KPNB1 mediates docking of the importin α/substrate complex in the nuclear pore . More recent evidence shows that KPNB1 can function as nuclear transport receptor independently of importin α by directly recognizing substrates such as CREB , AP-1 transcription factors , and parathyroid hormone-related protein ( Lam et al . , 1999; Forwood et al . , 2001; Ghildyal et al . , 2005 ) . Interestingly , a previous report showed that importin α isoforms ( RCH , QIP-1 , NPI-2 ) mediate nuclear localization of CRY2 through direct interactions but did not investigate the functional consequence of this interaction ( Sakakida et al . , 2005 ) . In this regard , our data suggest that in mammals KPNB1 mediates nuclear localization of the PER/CRY complex independently of importin α , as overexpression of a KPNB1 mutant lacking an importin α interaction domain had a similar phenotype ( Figure 1 , Figure 3—figure supplement 2 ) . Consistent with this , both KPNA2 and KPNA1 , the well-known KPNB1 interacting importin alpha proteins ( importin α1 , importin α5 ) , when depleted in cells , did not affect circadian rhythmicity or subcellular localization of PER2 or CRY2 protein ( Figure 4—figure supplement 1 , Figure 4—figure supplement 2 , Figure 4—figure supplement 3 ) . In a recent report , cytoplasmic retention of PER proteins was accelerated by KPNA2 overexpression in mouse stem cells ( Umemura et al . , 2014 ) . However , intriguingly we found that KPNA2 overexpression substantially reduced cells expressing PER2-Venus protein in our imaging studies ( Figure 4—figure supplement 1E ) . Moreover , both the endogenous and overexpressed KPNA2 protein localized predominantly in the nucleus , which was confirmed by nuclear cytoplasmic fraction Western analysis ( Figure 4—figure supplement 1C , E , Figure 4—figure supplement 2A ) . This suggests that KPNA2 might regulate PER2 protein level through an unknown mechanism in U2 OS cells . Thus , we speculate that these different results are likely due to the different cell types and species used . On the other hand , knockdown of KPNB1 had strong ( PER1 , PER2 ) , modest ( CRY1 ) , or no effect ( CRY2 ) on single tagged PER and CRY proteins , while it had strong effects on all combinations of PER/CRY complexes ( Figure 1A–C ) . Similarly , knockdown of KPNB1 increased cytoplasmic retention of endogenous PER1 , PER2 , CRY1 , and CRY2 to varying degrees , probably through complex interactions ( Figure 1D , Figure1—figure supplement 2 ) . The immunoprecipitation analysis showed that KPNB1 interacted strongly with PERs compared with CRYs ( Figure 2—figure supplement 1A ) . These data suggest that the PERs mediate interaction with KPNB1 and gate nuclear translocation of PER/CRY complex . This is reminiscent of a previous report indicating that the nuclear entry of CRY is mainly dependent on PER , in vivo ( Lee et al . , 2001 ) . Finally , knockdown of KPNB1 or its fly ortholog ketel resulted in arrhythmic U2 OS cells or fly locomotor activity rhythms , while knockdown of most of KPNA isoforms had no significant effect on the rhythm ( Figure 4 , Figure 5 , Figure 4—figure supplement 4 ) . These data suggest a model in which KPNB1 plays a necessary role in mediating nuclear entry of the PER/CRY complex in core clock regulation ( Figure 6 ) . Recently , importin β emerged as a potential molecular target for cancer prevention ( van der Watt et al . , 2013 ) . So far , several novel small molecule regulators of KPNB1 have been identified , though further investigation is needed to improve their specificity and efficacy ( Hintersteiner et al . , 2010; Soderholm et al . , 2011 ) . In this respect , our study provides a rationale for KPNB1 as molecular target for regulating clock physiology . U2 OS or HEK293T cells were cultured in Dulbecco's modified Eagle's medium supplemented with 10% FBS , 1% L-Glutamine , and 1% penicillin-streptomycin ( Invitrogen , Grand Island , NY , United States ) at 37°C under 5% CO2 . The cells were transfected with siRNAs using Lipofectamine RNAiMAX ( Invitrogen ) and DNA plasmids using Fugene HD reagents ( Promega , Madison , WI , United States ) . The combined transfection of DNA plasmids with siRNA into the cells were performed using Lipofectamine 2000 ( Invitrogen ) . For overexpression of NRON in cells and generation of its transgenic cell lines , full-length genomic DNA fragments of NRON from U2 OS cells was obtained by PCR with primers containing the flanking restriction sites ( NotI , XhoI ) and inserted into pcDNA 3 . 1 mammalian expression vector ( Invitrogen ) . Plasmids encoding several flag-tagged clock proteins ( Flag-PER1 , Flag-PER2 , Flag-CRY1 , Flag-CRY2 , Flag-CLOCK , Flag-REVERBα , Flag-CHRONO ) were generated as described previously ( Anafi et al . , 2014 ) . For plasmids expressing Venus-tagged PER1 , PER2 , CRY1 , CRY2 ( PER1-Venus , PER2-Venus , CRY1-Venus , CRY2-Venus ) or BiFC plasmids expressing Venus N terminal ( VN ) or C-terminal ( VC ) -tagged PER1 , PER2 , CRY1 , and CRY2 ( PER1-VN , PER2-VN , CRY1-VC , CRY2-VC ) , full-length DNA fragment of the each of genes was subcloned into pCMV-Venus , pCMV-VN , or pCMV-VC using a restriction-free cloning method as described previously ( Shyu et al . , 2006; Bond and Naus , 2012 ) . For luciferase assays , pCMV Sport6 plasmids encoding full length cDNAs of CLOCK , BMAL1 , PER1 , PER2 , CRY1 , CRY2 , KPNB1 was obtained from the Mammalian Gene Collection ( Thermo Fisher Scientific , Grand Island , NY , United States ) . For generation of plasmids encoding mutant KPNB1 , DNA fragments of the KPNB1 gene with an importin-α binding domain deletion encoding a truncated protein ( 1–771 ) were sub-cloned into pCMV Sport6 expression vector ( Invitrogen ) using NotI and SalI restriction enzymes . HEK293T cells transfected with plasmids encoding proteins as indicated in the figures were assayed for luciferase reporter activity using DualGlo luciferin reagent ( Promega ) 24 hr post-transfection according to the manufacturer's protocol . Anti-PER2 used for immunoblot analysis of immunoprecipitated samples from mouse liver extracts were kindly provided by Dr Cheng Chi Lee's lab ( University of Texas Health Science Center at Houston ) . Anti-KPNB1 ( A300-482A ) for immunoblotting analysis was obtained from Bethyl Labratories ( Montgomery , TX , United States ) , and Anti-KPNB1 ( ab2811 ) for immunoprecipiation with mouse liver extracts or immunostaining of U2 OS cells was purchased from Abcam ( Cambridge , MA , United States ) . For ChIP or immunostaining experiments , Anti-PER1 ( AB2201 , Millipore , Darmstadt , German ) and Anti-PER2 ( NB100-125 , Norvus Biologicals , Littleton , CO , United States ) were used . Anti-Flag ( F7425 , Sigma , St . Louis , MO ) , Anti-GFP ( G1544 , Sigma ) , Anti-PER1 ( AB2201 , Millipore ) , Anti-PER2 ( 20359-1-AP , Proteintech , Chicago , IL , United States ) , Anti-CRY1 ( sc-33177 , Santa Cruz Biotech , Dallas , Texas , United States ) , Anti-CRY2 ( 13997-1-AP , Proteintech ) , were used for immunopreciptation , immunoblotting , and immunostaining experiments as indicated in figures . Normal mouse IgG ( NI03 ) was purchased from Calbiochem . Anti-Tubulin ( Ab7291 , Abcam ) , anti-TBP ( 22006-1-AP , Proteintech ) , anti-Lamin ( #2032 , Cell Signaling , Danvers , MA , United States ) . and anti-GAPDH ( sc25778 , Santa Cruz Biotech ) were used as loading control antibodies for cytoplasmic , nuclear , and total protein extracts respectively . U2 OS cells were co-transfected with each of the following DNA plasmids encoding Venus ( V ) -tagged PER1 , PER2 , CRY1 , or CRY2 ( PER1-V , PER2-V , CRY1-V , CRY2-V ) and siRNA against KPNB1 . Cells were collected 48 hr post-transfection by briefly trypsinizing and washing twice in ice cold PBS . Nuclear–cytoplasmic fractionation of the cells was performed using the NE-PER Nuclear and Cytoplasmic Extraction Reagents kit ( 78835 , Thermo Fisher Scientific ) according to the manufacturer's protocol . The same kit protocol was used for preparation of cytoplasmic and nuclear extracts from mouse liver tissues collected at 4 hr interval for 24 hr in constant darkness . At 48 hr post-transfection of plasmids encoding PER2-Venus into U2 OS cells , the cell lysates were harvested in radioimmunoprecipitation assay ( RIPA ) buffer ( 50 mM HEPES [pH 7 . 4] , 150 mM NaCl , 1% NP-40 , 1 mM EDTA , 1 mM EGTA , 1 mM phenylmethylsulfonyl fluoride , 0 . 5% sodium deoxycholate , 1 mM NaF , 1 mM Na3VO4 , and protease inhibitor cocktail [Roche , Indianapolis , IN , United States] ) . Protein-G coated magnetic beads ( 10004D , Life Technologies , Grand Island , NY , United States ) were pre-incubated with 3 µg of anti-GFP ( G1544 , Sigma ) antibody at 4°C for 6 hr . The antibody conjugated beads were incubated with equal amounts of total protein at 4°C overnight . The final immune complexes were analyzed by immunoblot assay using antibodies as described . For tissue immunoprecipitation , liver tissue was homogenized in 1 vol of RIPA buffer 1 ( 50 mM Tris-HCl [pH 8 . 0] , 450 mM NaCl , 1% Triton X-100 , 1 mM EDTA , 1 mM EGTA , 1 mM phenylmethylsulfonyl fluoride , 0 . 5% sodium deoxycholate , 1 mM NaF , 1 mM Na3VO4 , and 1 protease inhibitor cocktail [Roche] ) . Homogenates were cleared by dilution with 2 vol of RIPA buffer 2 ( RIPA buffer 1 without NaCl ) . Further procedures were as described above . Immunoblot analyses were performed on 7 . 5% sodium dodecyl sulfatepolyacrylamide gels and transferred to polyvinylidene difluoride membranes ( Immobilon P; Millipore ) . Target proteins were detected with antibodies as indicated . The immune complexes were visualized with an ECL detection kit ( Pierce , Grand Island , NY , United States ) . For ChIP analysis , cell lysates were collected from U2 OS cells and the following ‘Materials and methods’ were performed as described previously ( Schmidt et al . , 2009 ) . Briefly , the pre-cleared chromatin was immunoprecipitated overnight at 4°C by agitating with 5 μg of PER1 or PER2 antibody as described . The cell extracts incubated in the absence of antibody were used for input controls . Immune complexes were collected by incubation with protein-G coated magnetic beads ( 10004D , Life Technologies ) and the final eluted DNA was extracted by phenol–chloroform–isoamyl alcohol ( 25:24:1 ) and ethanol precipitation . The primer sets used for ChIP Quantitative PCR ( qPCR ) analysis of human Per1 promoter region spanning canonical ( CACGTG ) were as follows: Forward primer: 5′-TCTCCCTCTCTCCTCCCTTCC-3′ , Reverse primer: 5′-GCCTGATTGGCTAGTGGTCTT-3′ , Probe: GTGTGTGACACAGCCCTGACC . At 24 hr post-transfection of expression vectors encoding Venus-tagged clock proteins ( PER1-Venus , PER2-Venus , CRY1-Venus , CRY2-Venus ) or BiFC plasmids expressing ( PER1-VN , PER2-VN , CRY1-VC , CRY2-VC ) as indicated , the cells were fixed with 4% paraformaldehyde in PBS and visualized using GFP filter set in fluorescence microscopy . For IF analysis , U2 OS cells was incubated with various antibodies as indicated by secondary antibodies conjugated to Alexa Fluor 488 and/or 568 ( Invitrogen ) . Cells were visualized using FITC/TRITC/DAPI filter set in fluorescence microscopy . Real-time bioluminescence of pPer2 or pBmal1 dLuc reporter U2 OS cells after synchronization with 1 µM Dex ( Sigma ) were monitored using a LumiCycle luminometer ( Actimetrics , Wilmette , IL , United States ) as previously described ( Zhang et al . , 2009 ) . The period of the luminescence data was determined through the WaveClock algorithm ( Price et al . , 2008 ) . cDNAs were generated with the High Capacity cDNA Archive Kit using the manufacturer's protocol ( Applied Biosystems , Grand Island , NY , United States ) . qPCR reactions were conducted using iTaq PCR mastermix ( BioRad , Hercules , CA , United States ) in combination with gene expression assays ( Applied Biosystems ) on a 7800HT Taqman machine ( Applied Biosystems ) . GAPDH was used as an endogenous control for all experiments . 3–5 days old flies were collected at indicated ZT on the fourth day of LD entrainment . Fly heads of each genotype were opened and brains were immediately fixed with 4% paraformaldehyde ( in 1× PBS ) for 15–20 min , followed by dissection in 1× PBST ( 0 . 3% Triton X-100 in PBS ) at room temperature . After 30min-wash with 1× PBST at room temperature , brains were blocked with 5% normal donkey serum ( NSD ) for 20 min , and then incubated overnight at 4°C with primary antibody: rat anti-PER ( UPR34 , 1:1000 ) and mouse anti-PDF ( C7 , Developmental Studies Hybridoma Bank , University of Iowa , 1:500 ) . After 30-min wash in 1× PBST at room temperature , brains were incubated with secondary antibodies ( Jackson ImmunoResearch Laboratories , West Grove , PA , United States , 1:500 ) for 2 hr in NDS at room temperature , followed by extensive 30 min-wash . Samples were imaged using Leica TCS SP5 confocal microscope . 8–10 brains were examined at each time point . Targeted expressions of ketel RNAi in PDF+ cells and tim-expressing clock neurons were performed using the UAS/GAL4 system . We tested two independent UAS-ketel RNAi lines ( v22348 , v107622; Vienna Drosophila RNAi center ) and isogenic w1118 ( iso31 ) strain was used as a wild-type control . Male flies of the UAS-ketel RNAi lines or iso31 ( control ) were crossed to female flies from Gal4 drivers . Male progenies of the crosses were entrained to a 12 hr:12 hr LD cycle at 25°C at least for 3 days . Individual flies were loaded in the glass locomotor tubes containing 5% sucrose and 2% agar . Locomotor activity was measured for 7–15 days in DD using the Drosophila Activity Monitoring System as previously described ( Williams et al . , 2001 ) . Power of rhythmicity was determined by FFT value . Flies with FFT value >0 . 01 were counted as a rhythmic . This activity records were analyzed using ClockLab software ( Actimetrics ) .	Most organisms have an internal clock—known as the circadian clock—that regulates many aspects of their biology and behavior in roughly 24-hr long cycles . In animals , the core of the circadian clock is made of two ‘activator’ proteins and two ‘repressor’ proteins that inhibit the activators so that the levels of all four proteins in cells fluctuate over the cycle . The activator proteins switch on the genes that encode the repressor proteins . This increases the production of the repressor proteins in an area of the cell called the cytoplasm . The repressor proteins then bind to each other and then move into the nucleus of the cell to inactivate the activator proteins . However , it was not clear how the repressor proteins move into the nucleus . Lee et al . used a technique called ‘RNA interference’ to study the circadian clock in human cells and fruit flies . The experiments show that a protein called importin β enables the repressor proteins to move into the nucleus . Importin β directly interacted with only one of the repressor proteins ( called PER ) . Previous studies have shown that importin β is able to interact with another protein called importin α , but Lee et al . 's results show that this interaction is not important for importin β's role in the movement of the repressor proteins . Blocking importin β activity resulted in the loss of circadian rhythms in both human cells and fruit flies , which suggests that importin β performs the same role in many different animals . The circadian clock is disrupted in many cancers , so Lee et al . 's findings may also help to lead us to new treatments to fight these diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "genetics", "and", "genomics" ]	2015	KPNB1 mediates PER/CRY nuclear translocation and circadian clock function
Dephosphorylation of eukaryotic translation initiation factor 2a ( eIF2a ) restores protein synthesis at the waning of stress responses and requires a PP1 catalytic subunit and a regulatory subunit , PPP1R15A/GADD34 or PPP1R15B/CReP . Surprisingly , PPP1R15-PP1 binary complexes reconstituted in vitro lacked substrate selectivity . However , selectivity was restored by crude cell lysate or purified G-actin , which joined PPP1R15-PP1 to form a stable ternary complex . In crystal structures of the non-selective PPP1R15B-PP1G complex , the functional core of PPP1R15 made multiple surface contacts with PP1G , but at a distance from the active site , whereas in the substrate-selective ternary complex , actin contributes to one face of a platform encompassing the active site . Computational docking of the N-terminal lobe of eIF2a at this platform placed phosphorylated serine 51 near the active site . Mutagenesis of predicted surface-contacting residues enfeebled dephosphorylation , suggesting that avidity for the substrate plays an important role in imparting specificity on the PPP1R15B-PP1G-actin ternary complex . Reversible phosphorylation of the alpha subunit of translation initiation factor 2 ( eIF2a ) is pivotal to control of global rates of protein synthesis and to modulating mRNA-specific translation in eukaryotes ( Sonenberg and Hinnebusch , 2009 ) . Phosphorylated eIF2 inhibits its guanine nucleotide exchange factor , eIF2B , attenuating the translation of most mRNA , whilst the translation of a small subset of mRNAs , with special 5′ untranslated regions , is increased ( Hinnebusch , 2005 ) . As the latter encode potent transcription factors , such as GCN4 in yeasts and ATF4 in animals , eIF2a phosphorylation activates gene expression programs with broad physiological ramifications: the integrated stress response ( ISR ) in mammals and its yeast counterpart , the general control response ( Harding et al . , 2003 ) . Four kinases are known to couple diverse upstream signals to eIF2a phosphorylation ( Ron and Harding , 2007 ) . PERK restrains protein synthesis in response to unfolded proteins in the endoplasmic reticulum . HRI accomplishes the same in response to heme restriction in developing erythroid precursors , whereas PKR is activated by double-stranded RNA to curtail viral protein synthesis in infected cells . The oldest eIF2a kinase , GCN2 , is activated by uncharged tRNAs to restore amino acid balance by ISR activation . In animal cells , eIF2a phosphorylation is reversed by cellular phosphatase complexes consisting of a protein phosphatase 1 catalytic subunit ( PP1 ) and a substrate-specific regulatory subunit . Two such regulatory subunits have been identified in mammals: PPP1R15A ( known as GADD34 ) is encoded by an ISR-inducible gene ( Novoa et al . , 2001; Ma and Hendershot , 2003 ) , whereas PPP1R15B ( known as CReP ) is constitutively present ( Jousse et al . , 2003 ) . Cells lacking PPP1R15A are impaired in recovery of protein synthesis during resolution of the stress response ( Novoa et al . , 2003; Marciniak et al . , 2004 ) , whereas elimination of PPP1R15B results in developmental impairment and perinatal lethality of mice . Importantly , inactivation of both PPP1R15 isoforms is lethal to cells ( Tsaytler et al . , 2011 ) , but can be rescued by converting serine 51 of eIF2a to the non-phosphorylatable alanine , providing genetic proof for the narrow in vivo substrate specificity of the PPP1R15 family of regulatory subunits ( Harding et al . , 2009 ) . In mammals , PPP1R15 genes encode proteins of over 600 amino acids , but conspicuous conservation of sequence between the A and B isoform is confined to a stretch of ∼70 residues at their C-termini . This is also the only sequence in eIF2a-specific regulatory subunits that is conserved broadly across phyla and is encoded by a separate exon; a conserved feature of the genomic organization of the two isoforms of PPP1R15 . Viruses have co-opted PPP1R15; presumably to neutralize the effects of host PKR ( He et al . , 1996 , 1997 ) and conservation amongst these viral proteins is also confined to the same stretch of 70 residues . The preeminence of the PPP1R15 C-terminus is also supported by functional experiments , as over-expression of this portion was the basis of their identification as ISR inhibitors in the first place ( Novoa et al . , 2001; Jousse et al . , 2003 ) , whereas the extended N-terminal region of PPP1R15 participates in membrane binding , regulating subcellular localization and turnover ( Brush et al . , 2003; Brush and Shenolikar , 2008; Zhou et al . , 2011 ) . Binding to the PP1 catalytic subunit has been mapped to the C-terminal functional core of the PPP1R15 family ( Connor et al . , 2001; Novoa et al . , 2001 ) and is shared by the pared-down viral proteins ( He et al . , 1998 ) . An RVxF motif , common to many PP1 binding proteins , is present in the conserved functional core of the PPP1R15 family members and their viral counterparts , and plays an important role in assembly of the active holophosphatase complex ( He et al . , 1998; Connor et al . , 2001; Novoa et al . , 2001 ) . However , beyond these important insights the basis for substrate-specific dephosphorylation remains largely mysterious . Here , we report on the in vitro reconstitution of an active eIF2a-directed phosphatase from bacterially-expressed PP1 and PPP1R15 functional core . Our studies reveal the essential role of a third cellular factor , G-actin , in endowing the complex with substrate specificity and provide the first glimpse at the structure of a ternary holophosphatase complex possessing such substrate specificity . Alignment of regulatory subunits mediating eIF2aP dephosphorylation reveals that sequence conservation in the PPP1R15 family is limited to a stretch of ∼70 residues in their C-termini ( Figure 1A , B and Figure 1—figure supplement 1 ) , which is also the only conserved region of these proteins predicted to be ordered by DISOPRED ( Ward et al . , 2004 ) ( Figure 1—figure supplement 2A , B ) . 10 . 7554/eLife . 04871 . 003Figure 1 . A core functional domain in the C-terminus of PPP1R15 family members inhibits the integrated stress response ( ISR ) . ( A ) Cartoon depicting the domain organization of mammalian and viral PPP1R15 proteins . The N-terminal membrane-interaction domain is depicted by a broad line , the repeats , found in the PPP1R15A/GADD34 isoform , are shown in orange and the C-terminal , conserved functional domain that binds the catalytic subunit ( PP1 ) is shown in green . ( B ) Alignment of the conserved 70 residue portion of representative PPP1R15 family members: human PPP1R15B/CReP ( hsR15B; UNIPROT: Q5SWA1 ) , mouse PPP1R15A/GADD34 ( muR15A; UNIPROT: P17564 ) , frog PPP1R15 ( xlR15; UNIPROT: Q9W1E4 ) , fruit-fly PPP1R15 ( dmR15; UNIPROT: Q9W1E4 ) , human herpes simplex virus PPP1R15 ( HSV1; UNIPROT: P36313 ) , insect virus PPP1R15 ( coGV-gp036; UNIPROT: Q1A4R0 ) . Residues conserved in members of the PPP1R15 family across phyla are noted above the alignment . ( C ) Two-dimensional plots of fluorescent intensity of individual CHO cells containing a stably-integrated ISR reporter , CHOP::GFP ( Novoa et al . , 2001 ) , following transfection with plasmids encoding mCherry and fusions of PPP1R15 with mCherry . Where indicated the cells were exposed to the ER stress-inducing toxin tunicamycin ( TUN ) . GFP fluorescent intensity , reporting on the activity of the UPR ( X-axis ) was detected at 530 nm following excitation at 488 nm , whereas mCherry fluorescent intensity , reporting on the level of PPP1R15-mCherry fusion proteins ( Y-axis ) was detected at 610 nm following excitation at 561 nm . Suppression of the ISR by PPP1R15 is reflected by the accumulation of CHOP::GFPweak/mCherrystrong cells in quadrant 1 ( Q1 ) and is conspicuous in cells transduced with either the full-length PPP1R15 expression vectors ( R15A-mCherry and R15B-mCherry ) or their C-terminal core functional fragments [R15A ( 539–614 ) -mCherry and R15B ( 631–700 ) -mCherry] . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 00310 . 7554/eLife . 04871 . 004Figure 1—figure supplement 1 . An extended alignment of PPP1R15 regulatory subunit family members reveals that the conservation of amino acid sequence is confined to a ∼70 residue segment . Alignment of C-terminal sequences of PPP1R15B from mouse ( Q8BFW3 ) , rat ( D3ZP67 ) , human ( Q5SWA1 ) and cow ( E1BKX2 ) , PPP1R15A from frog ( Q9W1E4 ) , mouse ( P17564 ) , rat ( Q6IN02 ) , cow ( Q2KI51 ) , human ( O75807 ) , fruit-fly ( Q9W1E4 ) and their hypothetical viral analogues from African swine fever virus ( VF71_ASFV7 , Q65212 ) , human herpes simplex virus 1 ( ICP34_HHV11 , P36313 ) , Amsacta moorei entomopoxvirus ( NP_064975 ) , Glossina pallidipes salivary gland hypertrophy virus ( SGHV144 , YP_001687092 ) , Choristoneura occidentalis granulovirus ( CoGV_orf36 , YP_654457 ) and Trichoplusia niascovirus 2c ( YP_803309 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 00410 . 7554/eLife . 04871 . 005Figure 1—figure supplement 2 . Disordered and ordered regions predicted in PPP1R15 regulatory subunits . Disorder prediction ( based on DISOPRED [Ward et al . , 2004] ) along the primary sequence of mouse PPP1R15A ( UNIPROT P17564 , ‘A’ ) and human PPP1R15B ( UNIPROT Q5SWA1 , ‘B’ ) , superimposed on a schema of their functional domains ( from Figure 1 ) . Note that the conserved potentially ordered region is confined to the C-terminal 70 residues implicated functionally in eIF2a dephosphorylation . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 005 The ability to promote eIF2aP dephosphorylation can be assayed by the repression of ISR target genes whose induction in stressed cells is dependent on levels of eIF2 phosphorylation . Dual-channel flow cytometry analysis of Chinese Hamster Ovary ( CHO ) cells stably expressing a CHOP::GFP ISR reporter gene ( Novoa et al . , 2001 ) revealed that introduction of an mCherry fusion of PPP1R15A or PPP1R15B resulted in strong repression of the ISR in cells exposed to tunicamycin , a toxin that promotes protein misfolding in the endoplasmic reticulum to activate the eIF2a kinase PERK ( Harding et al . , 1999 ) . C-terminal fragments limited to the conserved region of PPP1R15 family members [mouse PPP1R15A ( 539–614 ) and human PPP1R15B ( 631–700 ) ] were sufficient for ISR attenuation ( Figure 1C ) , confirming that the information necessary to promote eIF2aP dephosphorylation is conveyed by its conserved C-terminal minimal functional core . To study the PP1-PPP1R15 binary complex in isolation from other cellular factors , the two proteins were co-expressed in a heterologous Escherichia coli system . The C-terminal minimal functional core of PPP1R15 was fused to a cleavable glutathione S-transferase ( GST ) tag and the co-expressed catalytic PP1 subunit was left untagged . PPP1R15 engaged PP1 in a tight complex , reflected in their co-purification on a glutathione-sepharose affinity resin . Following GST tag removal by TEV proteolysis , the two proteins co-eluted from a size exclusion column at a position predicted of a dimer containing one catalytic and one regulatory subunit ( Figure 2A ) . 10 . 7554/eLife . 04871 . 006Figure 2 . PPP1R15 engages the PP1 catalytic subunit through multiple contacts . ( A ) UV protein absorbance trace of a PPP1R15B ( 630–701 ) -PP1G ( 7–300 ) binary complex expressed in bacteria and resolved by size-exclusion chromatography . The indicated fractions from the chromatogram are presented in the Coomassie-stained SDS-PAGE below . The position of PP1 and the PPP1R15 peptide is indicated . ( B ) Cartoon representation of the indicated segments of human PPP1R15B ( in colored ribbon diagram ) in complex with mouse PP1G ( in cyan surface representation ) shown from three perspectives . The position of the hydrophobic ( H ) , C-terminal ( C ) and acidic ( A ) groves of PP1 are provided for orientation and the metal ions at the enzymes active site are colored pink . ( C ) Cartoon representation of PPP1R15B residues K639 to R653 in stick diagram and the corresponding density map , contoured to 1 rmsd . The backside of PP1G is shown in grey surface presentation . ( D ) Structure of the PP1-binding portion of PPP1R15B from the complex shown in ‘B’ aligned to the PP1-binding portions of the regulatory subunits PNUTS ( PDB: 4MP0 ) and spinophilin ( PDB: 3EGG ) . Side-chains of PPP1R15B residues that make important contacts with PP1G and their counterparts in PNUTS and spinophilin are shown in stick diagram . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 00610 . 7554/eLife . 04871 . 007Figure 2—figure supplement 1 . Variable occupancy of the M1 metal binding site in the PP1G-PPP1R15B binary complexes . ( A ) Detail of the PP1G active site in the binary complexes constituted with different PPP1R15B chain lengths: 631–660 ( magenta ) , 631–669 ( cyan ) , 631–684 ( green ) . Note an identically-positioned metal atom in the M2 position of all three structures but occupancy of the M1 site that is observed only in the crystal forms derived from the 631–669 construct . ( B ) Grey contours show the final 2nFo-DFc electron density and green contours show the SAD log-likelihood gradient map ( McCoy and Read , 2010 ) that detects anomalous scatterers , for the PP1G-PPP1R15B ( 631–669 ) crystal form in which both metal sites are occupied . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 007 PPP1R15A and PPP1R15B both formed stable complexes with PP1 ( both the PP1A and PP1G isoforms were tested ) but crystals suited for structural determination only arose from complexes between human PPP1R15B and mouse PP1G . Furthermore , binary complexes containing the entire C-terminal functional core of PPP1R15 ( residues 639–701 in the B isoform and 547–614 in the A isoform ) were relatively insoluble ( more on this point below ) but X-Ray diffracting crystals were obtained for complexes between PP1G 7–300 and human PPP1R15B 631–660 ( PDB: 4V0V ) , PPP1R15B 631–669 ( PDB: 4V0W ) or PPP1R15B 631–684 ( PDB: 4V0X ) and their structures were solved by molecular replacement ( Table 1 ) . 10 . 7554/eLife . 04871 . 008Table 1 . Structure parametersDOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 008PP1G-PPP1R15B ( 631–660 ) PP1G-PPP1R15B ( 631–669 ) PP1G-PPP1R15B ( 631–684 ) PP1G-PPP1R15B ( 631–701 ) -actinData collection Synchrotron beamlineDiamond I02Diamond I02Diamond I03Diamond I04-1 Space groupP21212P21212P41212C121 Cell dimensions a , b , c; Å66 . 8 , 67 . 89 , 156 . 3867 . 01 , 67 . 86 , 156 . 7567 . 54 , 67 . 54 , 158 . 01103 . 9 , 149 . 9 , 318 . 7 α , β , γ; ⁰90 , 90 , 9090 , 90 , 9090 , 90 , 9090 , 91 . 03 , 90 Resolution , Å51 . 26-1 . 61 ( 1 . 65-1 . 61 ) 33 . 94-1 . 55 ( 1 . 58-1 . 55 ) 51 . 34-1 . 85 ( 1 . 89-1 . 85 ) 82 . 79-7 . 88 ( 8 . 08-7 . 88 ) Rmerge0 . 084 ( 0 . 807 ) 0 . 097 ( 0 . 737 ) 0 . 094 ( 0 . 982 ) 0 . 142 ( 0 . 680 ) Rmeas0 . 101 ( 0 . 956 ) 0 . 12 ( 0 . 927 ) 0 . 107 ( 1 . 118 ) 0 . 199 ( 0 . 953 ) <I/σ ( I ) >12 . 5 ( 3 . 0 ) 7 . 4 ( 1 . 8 ) 9 . 7 ( 1 . 5 ) 7 . 2 ( 1 . 6 ) CC1/20 . 997 ( 0 . 746 ) 0 . 995 ( 0 . 685 ) 0 . 998 ( 0 . 914 ) 0 . 980 ( 0 . 627 ) Completeness , %99 . 8 ( 100 ) 92 . 6 ( 99 . 9 ) 100 ( 100 ) 98 ( 99 . 2 ) Redundancy6 . 3 ( 6 . 7 ) 5 . 1 ( 4 . 9 ) 7 . 8 ( 8 . 2 ) 3 . 4 ( 3 . 5 ) Refinement Rwork0 . 1760 . 1720 . 1760 . 370 Rfree0 . 2030 . 2030 . 2220 . 400 No . of reflections9258496347320785111 No . of atoms54935701266228185 Average B-factors24 . 425 . 245 . 2334 Metal ion occupanciesChain A: M2 0 . 95Chain A: M2 0 . 79Chain A: M2 0 . 76n/aM1 0 . 25Chain C: M2 0 . 99Chain C: M2 0 . 90M1 0 . 22 rms deviations Bond lengths ( Å ) 0 . 0060 . 0060 . 0110 . 0097 Bond angles ( ⁰ ) 1 . 0441 . 0541 . 2211 . 271 Ramachandran favoured region , %96 . 696 . 496 . 897 . 2 Ramachandran outliers , %0000 MolProbity score ( percentile ) 1 . 23 ( 98% ) 1 . 2 ( 98% ) 1 . 22 ( 98% ) 1 . 58 ( 100% ) PDB code4V0V4V0W4V0X4V0U The PPP1R15B peptide could be traced from K639 , immediately N-terminal of the RVxF motif ( K640VtF in human PPP1R15B ) , through to W662 ( Figure 2B–D ) . PPP1R15B followed a trajectory along the surface of PP1G similar to that traced by the regulatory subunits spinophilin ( Ragusa et al . , 2010 ) and PNUTS/PPP1R10 ( Choy et al . , 2014 ) . As expected , the side chains of V641 and F643 ( of the RVxF motif ) engage hydrophobic crevices on the backside of the catalytic subunit , an interaction common to PP1 regulatory subunits ( Egloff et al . , 1997 ) . From there PPP1R15B winds its way through the C-terminal groove to the front of the catalytic subunit . Along this path PPP1R15B Y650 occupies a position similar to spinophilin F459 and PNUTS F411 whereas PPP1R15B I652 occupies a position similar to spinophilin T461 and PNUTS F413 , whose side chains engage the ϕϕ motif binding-site of PP1 ( Choy et al . , 2014 ) . The side chain of PPP1R15B R658 engages a deep pocket on the PP1 surface , occupying the same position as spinophilin R469 and PNUTS R420 and forming a conserved salt bridge with PP1G D71 in the ‘Arg site’ , as predicted by Choy and colleagues ( Choy et al . , 2014 ) ( Figure 2D ) . The interactions between PPP1R15B and PP1G were thus confined to residues conserved in all PP1 isoforms , explaining the lack of preference of PPP1R15 for binding different PP1 isoforms . Despite their presence in the crystallized proteins , none of the residues C-terminal to W662 could be assigned in the density maps , suggesting that the C-terminal portion of PPP1R15's functional core is disordered in the binary complex . Interestingly both spinophilin and PNUTS also appear to disengage from the surface of PP1 at a similar position ( D475 of spinophilin and N424 of PNUTS ) , suggesting that the C-terminal extension of these regulatory subunits evolved to engage component ( s ) missing from the binary complex . PP1G in these crystals conformed to previous structural determinations of PP1 isoforms ( RMS: 0 . 19–0 . 29 Å over more than 240 Cα atoms ) , indicating that the binding of PPP1R15 did not change PP1's structure . Previous structures of PP1 revealed two metal ions , M1 and M2 ( likely Mn2+ ) in the shallow groove implicated in catalysis . Two metal ions were also detected in the complex of PP1G 7–300 with PPP1R15B 631–669 ( PDB: 4V0W ) , occupying the sites observed in other crystal forms of PP1 ( Figure 2—figure supplement 1A , B ) . However , in the crystal structure of the binary complex of PP1G with the shorter , PPP1R15B 631–660 ( PDB: 4V0V ) , and the longer PPP1R15B 631–684 ( PDB: 4V0X ) , only a single metal ion , occupying an identical M2 position , was observed ( Figure 2—figure supplement 1A ) . An unoccupied M1 site was previously observed in the crystal structure of PP1G-Inhibitor-2 complex ( PDB: 2O8G ) and the okadaic acid-bound PP1G ( PDB: 1JK7 ) , suggesting that the metal ion at M1 site is more labile than that at the M2 site and that the structures determined here provide different snapshots of metal ion occupancy of the catalytic subunit . The structure of the binary complex of PP1G and PPP1R15B 631–684 ( PDB: 4V0X ) , with ∼2800 Å2 of buried surface at the interface of its two protomers , is consistent with the high affinity of PPP1R15 for PP1 , a property it shares with other regulatory subunits . However , given their similar trajectory on the surface of the catalytic subunit , these structures do not provide a ready answer for the basis of substrate specificity of the different holophosphatases . Therefore , we tried to measure the specificity of binary complexes formed by the complete functional core of PPP1R15 631–701 and PP1 . Appending E . coli maltose binding protein ( MalE ) to the functional core of PPP1R15B ( 631–701 ) was found to stabilize the binary complex in solution and enabled its recovery in amounts and purity suited for biochemical studies ( Figure 3A ) . 10 . 7554/eLife . 04871 . 009Figure 3 . The PPP1R15B-PP1 binary complex lacks substrate specificity . ( A ) UV protein absorbance trace of a PPP1R15B-PP1G complex co-expressed in bacteria as a GST-PPP1R15B ( 631–701 ) -MBP fusion protein alongside untagged PP1G ( 7–300 ) , purified by glutathione affinity chromatography and , after cleavage of the GST tag , resolved by size-exclusion chromatography . The indicated fractions from the chromatogram are presented in the Coomassie-stained SDS-PAGE below . The position of PP1G and the PPP1R15B peptide ( PPP1R15B-MBP , a fusion with E . coli maltose binding protein , MBP , that maintains its stability ) are indicated . ( B ) Images of Coomassie-stained Phos-Tag SDS-PAGE in which the non specific substrate and product ( GSTP and GST0 , upper panel ) and the specific substrate and product ( eIF2aP and eIF2a0 , lower panel ) have been resolved . Escalating concentrations of the binary complex ( the enzyme ) shown in ‘A’ ( ranging from ∼6 . 5 nM to 25 nM ) have been applied identically to the two substrates ( GSTP at 1 . 7 µM and eIF2aP at 2 . 8 µM ) for 3 , 5 , 12 , or 25 min . The fraction of the substrate dephosphorylated is indicated under each experimental point . The phosphorylated and unphosphorylated substrates were loaded onto lanes 1 and 2 to serve as a reference for their mobility in the absence of enzyme . Shown is a representative experiment reproduced 4 times . ( C ) As in ‘B’ . Dephosphorylation reactions were conducted with a fixed concentration of the binary complex shown in ‘A’ [PPP1R15B-MBP and PP1G , at 26 nM] and varying concentration of substrate over a time frame of 3 , 5 , 12 , and 25 min . ( D ) Plot of the initial velocity of the reaction ( calculated at the 5 min time point , when excess substrate remains ) as a function of substrate concentration derived from the data shown in ‘C’ . Note that across a concentration range of substrate over which both reaction velocities are substrate-dependent , the binary complex is more effective at dephosphorylating the non-specific substrate , GSTP , than the specific substrate , eIF2aP . Shown is a representative experiment reproduced twice . ( E ) Plot of the logarithm of the time-dependent change in ratio of substrate concentration at t = 0 to the substrate concentration at t [log ( S0/St ) ] from the six dephosphorylation time courses experiments shown in ‘C’ . The slope ( mean ± SD ) , indicative of the relative enzyme velocity , and the linear regression coefficient of the different reactions initiated at the indicated substrate concentration [S0] are noted . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 00910 . 7554/eLife . 04871 . 010Figure 3—figure supplement 1 . The PPP1R15A-PP1 binary complex also lacks selectivity towards the specific ( eIF2aP ) substrate over the non-specific ( GSTP ) substrate . Images of Coomassie-stained Phos-Tag SDS-PAGE in which phosphorylated and de-phosphorylated non-specific substrate and product ( GSTP and GST0 , upper panel ) and the specific substrate and product ( eIF2aP and eIF20 , lower panel ) have been resolved . Escalating concentrations of the PPP1R15A ( 539–614 ) -PP1G ( 7–300 ) ( ranging from 7 nM to 60 nM ) have been applied in triplicate , identically to the two substrates ( at 2 . 7 µM ) for 30 min . Note that exposure to enzyme that results in nearly complete dephosphorylation of the non-specific substrate ( GSTP ) leave much of the specific substrate ( eIF2aP ) in its phosphorylated state . The phosphorylated and unphosphorylated substrates were loaded onto lanes 1 and 2 to serve as a reference for their mobility in the absence of enzyme . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 01010 . 7554/eLife . 04871 . 011Figure 3—figure supplement 2 . Phosphatase activity of the purified bacterially-expressed complexes is dependent on the presence of both the regulatory ( PPP1R15A ) and catalytic ( PP1G ) subunits . ( A ) Image of Coomassie-stained Phos-Tag SDS-PAGE in which phosphorylated ( GSTP ) and non-phosphorylated ( GST0 ) GST have been resolved . Escalating concentrations of the indicated complexes purified from bacteria by glutathione affinity chromatography were applied to the substrate ( GSTP , ∼2 µM ) for 20 min . The complex used in lanes 1–3 was purified from E . coli expressing GST alone ( no catalytic subunit ) ( and applied at concentrations of 0 . 2–1 . 6 µM ) . The complex used in lanes 4–6 was purified from E . coli expressing GST and untagged PP1G catalytic subunit ( and applied at concentrations 0 . 1–0 . 8 µM ) . The complex used in lanes 7–9 was purified from E . coli expressing GST-PPP1R15A ( 539–614 ) and untagged PP1G catalytic subunit ( and applied at concentrations of 0 . 08–0 . 64 µM ) . The complex used in lanes 10–12 was purified from E . coli expressing GST fusion to a mutant PPP1R15A ( 539–614 ) V549E that binds weakly to PP1 , and untagged PP1G catalytic subunit ( and applied at concentrations of 0 . 08–0 . 64 µM ) . ( B ) The lower panel is a Coomassie-stained Phos-Tag SDS-PAGE of the same reaction shown in panel A , but in the absence of substrate . The migration of the GST and GST-PPP1R15A ( 539–614 ) is indicated . The position of the substrate ( GSTP ) , product ( GST0 ) , and the conspicuous GST signal emanating from reactions in which high concentrations of GST were introduced as a control enzyme ( lanes 1 and 4 ) generating a signal that overlaps with the GST0 signal is indicated . Note that dephosphorylation of the substrate is restricted to lanes 7–9 in which both PP1 and PPP1R15 have been expressed in the E . coli . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 011 Escalating concentrations of binary complex were assayed for their phosphatase activity directed towards two different substrates ( presented at similar concentrations ) : GSTP , consisting of globular glutathione S-transferase with a short unstructured C-terminal peptide derived from the regulatory portion of the transcription factor CREB that is stoichiometrically phosphorylated on a single serine by protein kinase A ( Ron and Dressler , 1992 ) , served as the reference , non-specific substrate , whereas the soluble N-terminal lobe of human eIF2a ( 1–185 ) ( Ito et al . , 2004 ) , phosphorylated on serine 51 by bacterially-expressed PERK ( Marciniak et al . , 2006 ) , served as the specific substrate . Surprisingly , across a range of enzyme concentrations over which the reaction velocity was enzyme-limited ( 6 . 5–25 nM ) , the non-specific GSTP substrate ( at 1 . 7 µM ) was more rapidly dephosphorylated by the PPP1R15B-PP1G binary complex than the specific eIF2aP substrate ( despite the latter being present at a higher concentration , 2 . 8 µM ) ( Figure 3B ) . A similar hierarchy was observed with reactions in which the concentration of binary complex was fixed ( at 26 nM ) and the concentration of substrate varied: across a range of substrate concentrations over which the reaction velocity was substrate-limited ( 0 . 5–3 µM ) the non-specific GSTP substrate was threefold ( 95% confidence limits 3 . 6 to 2 . 6 ) more rapidly dephosphorylated by the PPP1R15B-PP1G binary complex than the specific eIF2aP substrate ( Figure 3C , D ) . These differences in velocity were observed at early time points of the reaction ( Figure 3D ) and persisted into later time points ( with considerable substrate depletion ) , obeying first order kinetics with consistently more rapid dephosphorylation of GSTP compared with eIF2aP ( Figure 3E ) . A binary complex produced with PP1G and the homologous region of mouse PPP1R15A ( 539–614 ) similarly proved more effective at dephosphorylating the non-specific GSTP substrate ( Figure 3—figure supplement 1 ) . Phosphatase activity was strictly dependent on the presence of both the regulatory and catalytic subunits in the E . coli expression system ( Figure 3—figure supplement 2 ) , leading to the conclusion that binary complexes of PP1 and the in vivo-defined functional core of PPP1R15 possess no measureable selectivity toward eIF2aP over GSTP in vitro . Given that the functional core of PPP1R15 promotes eIF2aP dephosphorylation in cells , these observations suggested that something might have been missing from the complex constituted of bacterially-expressed proteins . Therefore , we measured the ability of tissue lysates to complement the activity of the binary complex and endow it with substrate-specificity . Mouse pancreas ( a tissue in which the ISR plays an important role ) was homogenized to obtain a cytoplasmic lysate , which was added in escalating amounts to binary complex ( 1 . 5–7 . 5 nM ) and substrate ( ∼2 µM ) . At the concentrations used ( up to 100 ng/µl of protein ) the lysate itself had minimal GSTP or eIF2aP-directed phosphatase activity ( compare lanes 1 and 4 in Figure 4A ) , however , when added to a binary complex of PPP1R15B and PP1 the lysate selectively increased the eIF2aP-directed phosphatase activity without affecting the dephosphorylation of GSTP ( Figure 4A ) . These experiments suggest that the lysate is able to provide ingredient ( s ) missing from the binary complex that endow it with specificity towards eIF2aP . 10 . 7554/eLife . 04871 . 012Figure 4 . An activity in tissue lysate that endows the PPP1R15B-PP1G binary complex with specificity towards eIF2aP can be mimicked by pure G-actin . ( A ) Images of Coomassie-stained Phos-Tag SDS-PAGE in which a phosphorylated non-specific substrate and de-phosphorylated product ( GSTP and GST0 , upper panel ) and the specific substrate and product ( eIF2aP and eIF20 , lower panel ) have been resolved . Escalating concentrations of tissue lysate ( ranging from 12–100 ng/µl of protein ) were incubated with low ( 1 . 5 nM ) and high concentration ( 7 . 5 nM ) of the PPP1R15B-PP1G complex ( shown in ‘Figure 3A’ ) and applied identically to the two substrates ( at ∼2 µM ) for 20 min after which the substrate was purified away from other proteins and resolved on the gel shown . The fraction of the substrate dephosphorylated is indicated under each experimental point . ( B ) As in ‘A’ with the G-actin-binding compound cytochalasin D added to a final concentration ranging from 12 . 5 to 100 µM . ( C ) As in ‘A’ with the F-actin-stabilizing compound jasplakinolide added at a final concentration of 10 µM . ( D ) As in ‘A’ but with pure , latrunclulin B-blocked G-actin ( ∼1 µM ) in place of tissue lysate . As the entire content of the reaction was loaded ( without further purification of the substrate ) the actin is visible in this gel , whilst the PPP1R15B-PP1 binary complex , present in quantities below the detection limit of the stained gel , is invisible . ( E ) Plot of the substrate concentration-dependence of the velocity of dephosphorylation of eIF2aP or phosphorylase A ( PYGMP ) by apo-PP1G , the binary PP1G-PPP1R15B ( BC ) or the ternary PP1G-PPP1R15B-actin complex ( TC ) , assayed over a physiologically relevant concentration range of the eIF2aP substrate ( well below the enzyme's Km for that substrate , see Figure 4—figure supplement 1 ) . Shown are the mean ± SD of measurements performed in triplicate . The regression coefficient ( R2 ) and slope ( ±SD ) , of the linear regression of each of the six enzyme-substrate pairs are indicated . Note: at substrate concentrations below the Km , the slope of the linear regression reflects the relative velocity of the reaction ( proportional to the Kcat/Km of the particular enzyme-substrate pair ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 01210 . 7554/eLife . 04871 . 013Figure 4—figure supplement 1 . Substrate concentration dependence of the velocity of eIF2aP dephosphorylation by the PPP1R15B-PP1G-Actin ternary complex . ( A ) Image of Coomassie stained Phos-Tag gels of eIF2aP dephosphorylation reactions by a ternary complex constituted of PPP1R15B-PP1G ( 5 nM ) and G-actin ( 1 µM ) . Time points of 3 , 5 , 12 , and 25 min were sampled and the concentration of substrate was varied over a range from 0 . 5 to 15 µM . To ensure resolution of the Phos-Tag gel , the volume of the sample loaded from the reactions containing high concentration of substrate ( lanes 1–12 ) was adjusted to ensure equality in the total mass of substrate plus product ( total eIF2a ) in each lane . The entire reaction containing lower concentrations of substrate was loaded onto the gel ( lanes 13–24 ) . Shown is an experiment reproduced twice with similar results . ( B ) Plot of the relation between the initial velocity of the reaction ( calculated at the 5 min time point ) and concentration of substrate at t = 0 . Note that the reaction velocity increased linearly over this concentration range and that the enzyme could not be saturated by substrate over the 30× concentration range accessible to testing . ( C ) Plot of the logarithm of the time-dependent change in ratio of substrate concentration at t = 0 to the substrate concentration at t [log ( S0/St ) ] from the six dephosphorylation time courses experiments shown in ‘A’ . The slope ( mean ± SD ) , indicative of the relative enzyme velocity , and the linear regression coefficient of the reactions initiated at different substrate concentration [S0] are indicated . Note the similar slope and strong fit to a linear regression of reactions initiated at different substrate concentration and allowed to progress through to extensive substrate depletion . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 01310 . 7554/eLife . 04871 . 014Figure 4—figure supplement 2 . Estimation of the concentration of eIF2a by quantitative immunoblotting of HEK293T cell lysates . ( A ) Plot of the relationship between fluorescent intensity of the eIF2a signal in immunoblot to the mass of recombinant eIF2a ( in pico moles ) applied to the blot . An image of the immunoblot is provided in the inset . ( B ) Immunoblot of ascending volumes of HEK293T cell lysates from biological duplicate samples applied to the same membrane used to derive the calibration curve shown in ‘A’ , above . The intensity of the fluorescent signal , the corresponding mass of eIF2a ( derived from the standard curve shown in ‘A’ ) the packed cell volume and the corresponding estimated cytosolic volume of the lysate applied to each lane are indicated , as is the derivative estimated concentration of eIF2a in that sample . The latter was calculated from the first two loadings of each sample ( lanes 1 , 2 , 5 , and 6 , where the fluorescent signal is not saturated ) and is based on the assumption that 60% of the packed cell volume is intracellular , of which ½ is comprised of cytosol ( the rest being nucleus and organelles ) . Yielding an estimated eIF2a concentration of 1 . 13 ± 0 . 15 µM ( mean ± SD , n = 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 01410 . 7554/eLife . 04871 . 015Figure 4—figure supplement 3 . Incorporation of G-actin inhibits PPP1R15B-PP1G phosphatase activity directed against non-specific substrates . ( A ) Images of Coomassie stained Phos-Tag gels of a time course ( 2 . 5–40 min ) of dephosphorylation reactions of phosphorylase A ( PYGMP , upper panel ) or GSTP ( lower panel ) ( both at 1 µM ) by a complex constituted of PPP1R15B-PP1G ( 15 nM ) in the presence or absence of G-actin ( 1 µM ) . Shown is a representative sample of an experiment reproduced three times . ( B ) Plot of the logarithm of the time-dependent change in ratio of substrate concentration at t = 0 to the substrate concentration at t [log ( S0/St ) ] from the four dephosphorylation time courses experiments shown in ‘A’; phosphorylase A ( PYGMP ) or GSTP , by a binary complex of PPP1R15B-PP1G ( BC , without Actin ) or a ternary complex of PPP1R15B-PP1G-Actin ( TC , with Actin ) . The slope ( mean ± SD ) , indicative of the relative enzyme velocity , and the linear regression coefficient of the different reactions are noted . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 015 A clue to the identity of the missing ingredient was provided by experiments showing that G-actin readily joins PPP1R15 and PP1 to form a ternary complex whose abundance and activity respond to changes in actin dynamics in cells; described in detail in the accompanying manuscript ( Chambers et al . , 2015 ) . In keeping with this clue , we found that cytochalasin D , a low molecular weight natural compound that binds in the cleft between lobes I and III of actin and thus disrupts the interactions of G-actin with many of its binding partners ( Nair et al . , 2008 ) , also reversed the stimulatory activity of lysate on eIF2aP dephosphorylation by binary complexes ( Figure 4B ) . Furthermore , as the cell-based experiments in the accompanying manuscript suggested ( Chambers et al . , 2015 ) , jasplakinolide , a marine toxin that promotes actin polymerization ( Holzinger , 2009 ) , modestly but consistently antagonized the stimulatory activity of lysate on eIF2aP dephosphorylation by binary complexes ( Figure 4C , compare lanes 5 , 7 , 9 and 11 with 6 , 8 , 10 and 12 ) but had no effect on the non-specific phosphatase activity of the binary complex in the absence of lysate ( lanes 1 and 3 vs 2 and 4 ) . Addition of pure G-actin ( whose polymerization was blocked by latrunculin B ) selectively stimulated the eIF2aP-directed phosphatase activity of the PPP1R15B-PP1 binary complex ( Figure 4D , lanes 1–6 ) , but had no positive effect on the dephosphorylation of GSTP ( Figure 4D , lanes 7–12 ) . The enzymatic activity of the actin-stimulated PPP1R15B-PP1 binary complex was strongly substrate-concentration dependent and could not be saturated with substrate over the concentration range accessible to testing with the available methodology , precluding the extraction of a Km for the substrate , or a Vmax ( Figure 4—figure supplement 1A , B ) . This is typical of PP1-holophosphatases in that often they cannot be saturated in vitro by their substrate ( MacKintosh , 1993 ) . The strong linear relationship between the log change of substrate concentration with reaction time indicated that the dephosphorylation of eIF2aP by the PPP1R15B-PP1G-actin ternary complex proceeded as a first order process , obeying Michaelis–Menten kinetics for reactions well below the substrate Km ( Figure 4—figure supplement 1C ) . Comparison of the relative velocity of the binary and ternary complex suggested that G-actin stimulated dephosphorylation of eIF2aP by 13 . 7-fold ( 95% confidence limits 11 . 3–16 . 9 , compare Figure 3D with Figure 4—figure supplement 1B ) . To obtain a better physiological perspective on the aforementioned observations pointing to actin's role in stimulating the dephosphorylation of eIF2aP , we measured the substrate concentration in cells by quantitative immunoblotting ( using known quantities of purified bacterially-expressed eIF2a to calibrate the assay , Figure 4—figure supplement 2 ) . Our estimate of 1 . 13 ± 0 . 15 µM eIF2a in the cytosol of HEK 293T cells agrees well with an estimate derived from the number of molecules of the yeast homolog , Sui2p , in a haploid yeast cells ( 1 . 71 × 104 molecules/cell , [Ghaemmaghami et al . , 2003] , which , assuming a yeast cell volume of 60 fl , of which half is cytosol , yields a concentration of 0 . 95 µM ) . This information enabled a detailed quantitative comparison of the initial velocity of eIF2aP dephosphorylation over a range of physiologically relevant substrate concentrations by three different purified enzymes: apo-PP1G , a binary complex of PPP1R15B-PP1G and a ternary complex of actin-PPP1R15B-PP1G . At substrate concentrations in the physiological range ( 0 . 5–4 µM ) , a 16 . 5-fold acceleration ( 95% confidence limits 21 . 2–13 . 2-fold ) of eIF2aP dephosphorylation was effected by G-actin joining the PP1G-PPP1R15B binary complex to form a ternary complex ( Figure 4E ) . Whereas in the absence of actin , PPP1R15B failed to accelerate eIF2aP dephosphorylation over that observed by apo-PP1 ( Figure 4E ) . In the same assay , actin inhibited the dephosphorylation of a reference substrate , phosphorylase A ( PYGMP ) , by the PPP1R15B-PP1G binary complex 3 . 1-fold ( 95% confidence limits 6 . 8–1 . 4-fold ) ( Figure 4E ) ; the inhibitory effect of actin on an unstructured substrate , GSTP , was less conspicuous ( Figure 4—figure supplement 3 ) . Together , these observations fit well with prevailing concepts on the basis of substrate-specific dephosphorylation by PP1-holophosphatase complexes , in that their regulatory component ( s ) , PPP1R15 and G-actin in this case , endow the enzyme with specificity towards its physiological substrate ( eIF2aP ) whilst inhibiting the dephosphorylation of an irrelevant structured substrate ( phosphorylase A ) ( Peti et al . , 2013 ) . Addition of G-actin stimulated the phosphatase activity of both PPP1R15A and PPP1R15B-containing binary complexes . The concentration of G-actin required to elicit a half maximal stimulatory effect was similar for both complexes , <100 nM ( Figure 5A ) , whilst an irrelevant protein , bovine serum albumin had no stimulatory effect on eIF2aP dephosphorylation ( Figure 5—figure supplement 1 ) . Cytochalasin D and jasplakinolide , which antagonized the stimulatory activity of tissue lysate , also reversed stimulation of the eIF2aP-directed phosphatase activity of the binary complex by purified G-actin ( Figure 5B , C ) . The diminishing stimulatory effect of G-actin at the highest concentrations ( Figure 5C , lanes 9–12 ) is consistent with high monomer concentrations accelerating F-actin formation during the dephosphorylation reaction ( which is conducted at physiological salt concentrations that favor actin polymerization ) , a feature that is more conspicuous in the jasplakinolide-treated sample . 10 . 7554/eLife . 04871 . 016Figure 5 . G-actin activates the eIF2aP-directed phosphatase activity of both PPP1R15A and PPP1R15B-containing binary complexes with an EC50 in the submicromolar range . ( A ) Trace of the velocity of eIF2aP dephosphorylation by 16 nM PPP1R15A-PP1G ( in red ) and 68 nM PPP1R15B-PP1G ( in blue ) binary complexes in the presence of the indicated concentrations of G-actin ( fitted to a non-linear regression ) . An image of the gel of a representative experiment is presented above the trace . ( B ) Image of a Coomassie-stained gel of dephosphorylation reactions by a specific [PPP1R15B-PP1G ( 25 nM ) and actin ( 1 µM ) ] ternary complex ( lanes 2–8 ) or a non-specific [PPP1R15B-PP1G ( 125 nM ) ] binary complex ( lanes 9–14 ) in the presence of escalating concentrations of cytochalasin D ( 1 . 2–100 µM ) . ( C ) As in ‘B’ but with escalating concentrations of G-actin in the absence and presence of jasplakinolide , 1 µM . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 01610 . 7554/eLife . 04871 . 017Figure 5—figure supplement 1 . An irrelevant protein , bovine serum albumin , has no effect on the dephosphorylation of eIF2a by the PPP1R15B-PP1G binary complex . Trace of the velocity of eIF2aP dephosphorylation by 68 nM PPP1R15B-PP1G binary complexes in the presence of the indicated concentrations of G-actin ( grey ) or bovine serum albumin ( BSA , red ) both fitted to a non-linear regression . An image of the gel of a representative experiment reproduced twice is presented above the trace . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 017 To further explore the role of G-actin in the holophosphatase complex , we measured the effect of mutations in residues conserved between PPP1R15 proteins on the ability of actin to stimulate eIF2aP-directed phosphatase activity of binary complexes . To compensate for any enfeebling effect the mutations might have on the association of PPP1R15 with PP1 during purification from the bacteria—an important consideration , given the in vivo evidence for cooperativity of actin and PP1 binding ( Chambers et al . , 2015 ) —we adjusted the concentration of binary complex in the reaction to provide comparable baseline levels of eIF2aP dephosphorylation ( in the absence of added actin ) . Thus , the concentration of binary complex in the reactions varied from 15–45 nM . Replacing the conserved residues R658 ( PPP1R15B ) or R571 ( PPP1R15A ) , seen to engage the arginine pocket of PP1 with alanine , markedly enfeebled the ability of actin to serve as an activator of substrate specific dephosphorylation ( Figure 6 ) . Attenuated response to actin was also observed in mutations affecting other conserved residues , W662 , F672 , and I676 of PPP1R15B ( Figure 6A ) and their counterparts in PPP1R15A ( Figure 6B ) . The more severe R571A and W575A mutations also markedly attenuated PPP1R15A's ability to inhibit activation of an ISR target gene in vivo whilst the weaker PPP1R15A F585A and I589A mutations ( that have retained a measure of responsiveness to G-actin in vitro , Figure 6B ) were indistinguishable from the wildtype in their ability to reverse the ISR ( Figure 6—figure supplement 1 ) , further supporting the functional significance of the interactions observed in the crystal structures of the binary complex . As the response to actin of the mutant binary complexes was normalized for the recovery of functional catalytic subunit in vitro ( reflected in the dephosphorylation activity in the absence of actin ) , these experiments imply that actin's ability to serve as a substrate-specific activator is dependent on structural features of the ternary complex formed and not merely on anchoring the catalytic subunit to the regulatory one and are in keeping with prevailing concepts on the mechanism of specificity of PP1-containing holophosphatase complexes ( Peti et al . , 2013 ) . 10 . 7554/eLife . 04871 . 018Figure 6 . Mutations in conserved residues of the PPP1R15 core functional domain enfeeble its activation by actin . ( A ) Traces of the velocity of eIF2aP dephosphorylation by wildtype and the indicated mutant PPP1R15B-PP1G complexes in the presence of the indicated concentrations of actin . ( B ) As in ‘A’ but with wildtype and mutant PPP1R15A-PP1G complexes . Note that the concentration of binary complex varied from 15–45 nM in the assays shown . It was purposely titrated to attain a velocity of dephosphorylation ( in the absence of actin ) comparable to ∼1/5 of that achieved by the wildtype enzyme in the presence of saturating concentration of actin . This ensures comparable activity of the wildtype and mutant enzymes in the absence of actin . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 01810 . 7554/eLife . 04871 . 019Figure 6—figure supplement 1 . Mutations in conserved residues of PPP1R15A compromise function in vivo . ( A ) Dual channel flow cytometric analysis of CHOP::GFP cells transfected with mCherry , fusion of wildtype PPP1R15A to mCherry or fusion of the indicated mutant derivatives of PPP1R15A to mCherry . Where indicated the cells were treated with the ER stress-inducing agent tunicamycin ( TUN ) to activate the ISR marker gene CHOP::GFP . ( B ) The bar diagram quantifies CHOP::GFP activation ( the ISR marker ) by tunicamycin in the mCherry marked cells ( shown is the median ± InterQuartile Range of the GFP intensity of cells in quadrants 1 and 2 , from ‘A’ above ) . It is suppressed in vivo by the wildtype and the weak F585A and I589A mutations , but the strong R571A and W575A mutations reversed the ability of PPP1R15A to suppress the ISR . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 019 G-actin readily joined binary complexes of GST-PPP1R15 and PP1 . Stable ternary complexes were formed by incubating G-actin ( purified from rabbit muscle ) with bacterially-expressed [GST-PPP1R15 and PP1G] binary complexes immobilized on a glutathione sepharose resin . Cleavage of the GST tag released a complex of the three components in a stoichiometry of 1:1:1 , which eluted from a size exclusion column at a position predicted of a trimer ( Figure 7A , Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 04871 . 020Figure 7 . G-Actin joins PPP1R15B-PP1G binary complexes to extend its active site-facing surface . ( A ) UV protein absorbance trace of a PPP1R15B ( 631–701 ) -PP1G ( 7–323 ) -G-actin complex assembled from the bacterially-expressed binary complex and rabbit muscle G-actin and resolved by size-exclusion chromatography . The indicated fractions from the chromatogram are presented in the Coomassie-stained SDS-PAGE below . The position of G-actin , PP1 , and the PPP1R15B peptide ( R15B ) is indicated , as is a degradation product of the PPP1R15B peptide that elutes in later fractions ( * ) . ( B ) Cartoon representation of the PP1G ( purple ) , PPP1R15B ( cyan ) , G-actin ( yellow ) holophosphatase complex . The peptide backbone is in colored ribbon diagram , the metal ions in the PP1G active site are shown as pink spheres whereas latrunculin B and ATP in the actin nucleotide binding pocket are shown in stick diagram ( blue and green respectively ) . Actin's four lobes ( 1–4 ) are marked for reference . The image on the left provides a view of the holoenzyme's active site , whereas the view to the right is of the back side . Note that density attributable to PPP1R15B is only traceable through residue W662 . ( C ) Close-up of actin's barbed end in the holophosphatase complex . Actin is shown in gray surface representation , PP1 in cyan , and PPP1R15B in magenta . Difference electron density after averaging in coot ( likely representing PPP1R15B ) is shown as a green mesh . The actin-binding helical peptide of drosophila Ciboulot , a prototypical G-actin binding protein ( PDB: 1SQK ) ( Hertzog et al . , 2004 ) , in pink ribbon presentation and cytochalasin D , a small molecule inhibitor ( PDB: 3EKS ) ( Nair et al . , 2008 ) , in orange stick diagram , are provided as landmarks . ( D ) Alignment of the C-terminal most residues of the conserved functional core of human PPP1R15B ( R15B ) and mouse PPP1R15A ( R15A ) with amphipathic helices of G-actin binding proteins seen to engage the cleft between domains I and III in the indicated PDB files ( shaded grey ) . The residues observed ( or predicted , in the case of PPP1R15 ) to constitute the hydrophobic face of the amphipathic helix are highlighted in teal . ( E ) Traces of the velocity of eIF2aP dephosphorylation by wildtype ( grey ) and the C-terminal truncation mutant PPP1R15B lacking F696-Q700 ( blue ) in the presence of the indicated concentrations of actin . ( F ) As in ‘D’ but comparing wildtype ( grey ) and C-terminal truncation mutant PPP1R15A lacking W609-R613 ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 02010 . 7554/eLife . 04871 . 021Figure 7—figure supplement 1 . G-Actin also joins PPP1R15 A-PP1G binary complexes to form a stable ternary complex . UV protein absorbance trace of a PPP1R15A ( 539–614 ) -PP1G ( 7–323 ) -G-actin complex assembled from the bacterially-expressed binary complex and rabbit muscle G-actin and resolved by size-exclusion chromatography on tandem S200 and S75 30/300 columns . The indicated fractions from the chromatogram are presented in the Coomassie-stained SDS-PAGE below . The positions of G-actin , PP1 , the GST remnant , the PPP1R15A peptide and its degradation product are indicated ( the latter by an asterisk ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 02110 . 7554/eLife . 04871 . 022Figure 7—figure supplement 2 . A ternary complex of DNase I , G-actin , PP1G and PPP1R15A retains its eIF2aP-directed phosphatase activity . ( A ) UV protein absorbance trace of a PPP1R15A ( 539–614 ) -PP1G ( 7–323 ) -G-actin and DNase I complex assembled from the bacterially-expressed binary complex , rabbit muscle G-actin and bovine pancreatic DNase I , resolved by size-exclusion chromatography . The indicated fractions from the chromatogram are presented in the Coomassie-stained SDS-PAGE below . The positions of G-actin , PP1 , DNase I , and the PPP1R15A peptide are indicated . ( B ) Cartoon representation of a model of the PPP1R15B , PP1G , and G-actin ternary complex with DNase I placed by superimposing the actin and DNase I complex ( PDB: 2A41 ) ( Chereau et al . , 2005 ) onto the PPP1R15B , PP1G and G-actin ternary complex ( PDB: 4V0U ) . Note that DNase I is bound to the backside of the ternary complex , facing away from the PP1 active site ( arrow ) . ( C ) Images of Coomassie-stained Phos-Tag SDS-PAGE in which phosphorylated and dephosphorylated eIF2a ( eIF2aP and eIF2a0 ) have been resolved . Escalating amounts of G-actin or a complex of G-actin and DNase I ( final concentration , 10 nM–1 µM ) were added to a reaction containing 25 nM PPP1R15B-MBP and PP1G complex ( as in Figure 3A ) and 2 µM eIF2aP substrate for 20 min . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 022 Crystals of ternary complexes containing either PPP1R15A or PPP1R15B were obtained , but only the PPP1R15B-containing crystals diffracted X-rays and that to a resolution of only 7 . 9 Å . Nonetheless , the structure could be solved by molecular replacement , placing five copies of the PPP1R15B-PP1G binary complex ( PDB: 4V0U ) and G-actin ( PDB: 4BIY ) ( Mouilleron et al . , 2008 ) in the crystal unit cell , providing a model of the active holophosphatase ( Figure 7B , and Table 1 ) . PP1 and actin assemble to form an elongated flat object . The C-terminal structured portion of PPP1R15B faces lobe IV of actin , suggesting that the portion of the regulatory subunit C-terminal of W662 that is unstructured in the binary complex is poised to interact with actin and bridge the gap between actin and PP1 . Whilst the resolution of the ternary complex is insufficient to trace the regulatory subunit's trajectory beyond that defined by the higher resolution binary complexes ( i . e . , C-terminal to W662 ) , the most significant feature in the averaged difference density map is observed in the cleft between domains I and III in actin's barbed end ( Figure 7C ) ; suggesting that PPP1R15B extends to engage this site and providing a plausible explanation for the inhibitory effect of cytochalasin D on eIF2aP dephosphorylation by the ternary complex ( Figure 5B ) . The C terminal-most residues of the PPP1R15 functional core ( F696-Q700 in human PPP1R15B and W609-R613 in mouse PPP1R15A ) are good candidates for mediating this interaction . This portion of PPP1R15 proteins can be modeled to form an amphipathic helix , which is found in other G-actin barbed end-binding proteins , exemplified by the drosophila Ciboulot helix ( Dominguez and Holmes , 2011 ) ( Figure 7C , D ) . And deletion of this portion abrogates actin-mediated acceleration of eIF2aP dephosphorylation in vitro ( Figure 7E , F ) . Actin lobe II faces away from PP1 and its D-loop is predicted to be free to engage other ligands . Consistent with this prediction , we found that DNase I , which has high affinity for actin's D-loop ( Mannherz et al . , 1980 ) , readily joins actin-PPP1R15-PP1 to form a quaternary complex ( Figure 7—figure supplement 2A ) . DNase I binds to the backside of the ternary complex and thus would not be expected to disrupt access to the PP1 active site ( Figure 7—figure supplement 2B ) . Consistent with this prediction , we found that a quaternary complex of PPP1R15A-PP1G-Actin-DNaseI retained its ability to de-phosphorylate eIF2aP ( Figure 7—figure supplement 2C ) . High Ambiguity Driven protein docking with HADDOCK ( Dominguez et al . , 2003 ) revealed that the platform formed between lobe IV of actin and the catalytic face of PP1 can readily accommodate the N-terminal regulatory lobe of eIF2a ( 1–185 ) with phospho-S51 inserted deep into the enzyme's active site in proximity to the catalytic metal ions . This mode of binding predicts polar contacts between yeast eIF2a side chains K66 and K86 ( R66 and K86 in human eIF2a ) and PP1 D220 , eIF2a E92 and PP1 K211 and a web of hydrogen bonding involving eIF2a R74 and D83 and actin D222 and N225 ( Figure 8A ) . 10 . 7554/eLife . 04871 . 023Figure 8 . Mutation of residues predicted to affect substrate-enzyme binding enfeeble dephosphorylation by the selective ternary complex . ( A ) High ambiguity driven protein docking by HADDOCK ( Dominguez et al . , 2003 ) model of yeast eIF2a's regulatory N-terminus ( PDB: 1Q46 ) with substrate phospho-S51 docked at the active site of the PPP1R15B-PP1-actin ternary complex ( PDB: 4V0U ) . The left close-up view shows the web of predicted hydrogen bonding between eIF2a E92 and PP1 K211 and eIF2a K66 ( R66 in the mammalian eIF2a ) and K86 and PP1 D220 . The right close-up view is of the eIF2a residues R74 and D83 and actin residues D222 and N225 . ( B ) Images of Coomassie-stained Phos-Tag SDS-PAGE in which phosphorylated and de-phosphorylated wildtype and the indicated eIF2a mutant forms have been resolved . The upper panel ( ‘+G-actin’ ) is of reactions with substrate at ∼2 µM exposed to 8 . 6 nM of the [PPP1R15B-MBP and PP1] binary complex with 1 µM ( saturating ) concentration of G-actin for 5 , 10 , 15 , and 30 min . In the lower panel ( ‘−G-actin’ ) , the concentration of binary complex was higher ( 0 . 26 µM ) . The specificity factor ( SF ) of the mutant substrates , based on the relative velocity of dephosphorylation by the PPP1R15B-MBP-PP1-actin ternary complex ( TC ) compared to the PPP1R15B-MBP-PP1 binary complex ( BC ) and normalized to the wildtype substrate ( see Figure 8—figure supplement 2 ) , is noted below each construct ( mean and with 95% confidence limits ) :SF=[VTC ( mut ) VBC ( mut ) ]÷[VTC ( wt ) VBC ( wt ) ] . Shown is a typical experiment reproduced four times . ( C ) As above; dephosphorylation reactions performed with escalating concentrations of binary complexes constituted with wildtype or the D220K PP1G mutant catalytic subunit and allowed to progress for 30 min . The substrate in the upper panel was wildtype eIF2aP ( ∼4 µM ) and G-actin was added at a saturating concentration . The substrate in the lower panel is the non-specific GSTP ( ∼4 µM ) . ( D ) Plot of the relationship between dephosphorylation velocity and enzyme concentration for the enzyme-substrate pairs shown in panel ‘C’ above . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 02310 . 7554/eLife . 04871 . 024Figure 8—figure supplement 1 . The wildtype and mutant eIF2aP substrates have indistinguishable retention profiles on size exclusion chromatography . Shown are plots of UV protein absorbance traces of the indicated eIF2aP substrates discussed in the results section and utilized in the experiment shown in Figure 8 . The bacterially-expressed proteins were purified by affinity tag ( His 6X ) chromatography , phosphorylated to completion by the kinase PERK , in vitro and resolved by size exclusion chromatography on a Superdex 75 30/300 mm column . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 02410 . 7554/eLife . 04871 . 025Figure 8—figure supplement 2 . Kinetic analysis of dephosphorylation reactions shown in Figure 8B . ( A ) Plot of time-dependent dephosphorylation of the indicated substrates by the PPP1R15B-MBP-PP1-actin ternary complex ( TC ) and the PPP1R15B-MBP-PP1 binary complex ( BC ) . ( B ) Plot of the logarithm of the time-dependent change in ratio of substrate concentration at t = 0 to the substrate concentration at t [log ( S0/St ) ] from dephosphorylation of the indicated substrates by the PPP1R15B-MBP-PP1-actin ternary complex ( TC ) . The slope ( mean ± SD ) , indicative of the relative enzyme velocity , and the linear regression coefficient of the different reactions are noted . ( C ) Dephosphorylation of the indicated substrates by PPP1R15B-MBP-PP1- binary complex ( BC ) , analyzed as in ‘B’ above . DOI: http://dx . doi . org/10 . 7554/eLife . 04871 . 025 To test the importance of these predicted contacts to the dephosphorylation of eIF2a we systematically introduced charge substitutions into the aforementioned residues in eIF2a and compared the ability of the wildtype and mutant proteins to serve as substrates for the non-specific PPP1R15B-PP1G binary complex and actin-containing substrate-specific holophosphatase ternary complex . Mutations R66E , R74E , K86E , and E92K and the R66E; K86E double mutant were well expressed and monomeric ( Figure 8—figure supplement 1 ) and lent themselves to stoichiometric phosphorylation by active PERK . The D83K mutation compromised solubility and could not be studied further . The R66E , K86E , and E92K mutations had a modest negative effect on dephosphorylation rates ( data not shown ) . Consistent with contacts made with PP1 , the R66E; K86E double mutant was compromised ∼twofold as a substrate of the binary complex but more than 10-fold as a substrate of the ternary complex ( Figure 8B and Figure 8—figure supplement 2 ) . By contrast the R74E mutation , affecting a predicted contact to G-actin , was selectively compromised in its ability to serve as a substrate of the ternary complex . The extent of this compromise was quantified by a calibrated specificity factor ( SF ) : as the ratio of the dephosphorylation rate of a given substrate at physiological concentration ( 2 µM ) by the ternary complex compared to the binary complex and set to 1 for wildtype eIF2a:SF=[ViTC ( mut ) ViBC ( mut ) ]÷[ViTC ( wt ) ViBC ( wt ) ] . The specificity score was lowest for R74E , the mutation that was predicted to affect the interaction with actin , whereas the R66E; K86E double mutant , which is also compromised in its ability to serve as a substrate of the binary complex , had a higher specificity score ( Figure 8B , and Figure 8—figure supplement 2 ) . The importance of contacts between eIF2a residues R66 and K86 and PP1 D220 is also supported by the reciprocal D220K mutation in PP1G , which has no measurable effect on the dephosphorylation of the non-specific substrate , GSTP , but reproducibly weakens dephosphorylation of eIF2aP ( Figure 8C , D ) . Together , these observations support the plausibility of a binding mode predicted by the computational docking exercise . Genetics has taught us that the non-redundant , essential , function of the PPP1R15 family is to enable dephosphorylation of eIF2a . It was surprising therefore to learn that the PPP1R15-PP1 complex is as active in dephosphorylating a non-specific substrate as in dephosphorylating eIF2aP . However , a factor found in cell lysates , G-actin , can provide the necessary specificity . The critical role of G-actin in constituting an eIF2aP-specific holophosphatase is consistent with its emergence as a conserved binding partner of PPP1R15 across phyla and with evidence that manipulations of the cellular cytoskeleton that diminish the pool of free G-actin promote higher levels of phosphorylated eIF2a ( Chambers et al . , 2015 ) . The latter observations point to a role for cytoskeletal dynamics in regulating rates of eIF2aP dephosphorylation and are consistent with the finding that most of the stimulatory activity of cell lysate is subject to inhibition by the actin-binding drug cytochalasin D . Nonetheless , our observations are not incompatible with the existence of other cellular proteins contributing to alternative PPP1R15-containing eIF2aP-directed holophosphatases . Like other PP1 regulators ( e . g . , spinophilin , inhibitor 2 ) , the functional core of PPP1R15 is natively unstructured ( Yu et al . , 2004 ) and attains its structure by wrapping around the catalytic subunit . In so doing it follows closely the path of other regulatory subunits , spinophilin ( Ragusa et al . , 2010 ) and PNUTS ( Choy et al . , 2014 ) . However , these well-resolved interactions ( involving residues R639-W662 ) account for slightly less than half the length of the PPP1R15B functional core . Residues C-terminal to W662 , which are not conserved in other regulatory subunits , are also not resolved in crystal structures of the PPP1R15B-PP1G binary complex . Furthermore , binary complexes containing the entire functional core of PPP1R15B require the presence of actin for stability and solubility . Unfortunately , the resolution of the ternary complex is too low to place the portion of PPP1R15B C-terminal of W662 in the density maps , but mutagenesis suggests that it plays an important role in actin-mediated substrate-specific dephosphorylation . The gap between actin and PP1 is potentially suited to accommodate PPP1R15B residues C-terminal to W662 whose density is not resolved in the crystal . Furthermore , the presence of density in the cleft between actin domains I and III , which is known to accommodate a short amphipathic helix of other G-actin binding partners ( Dominguez and Holmes , 2011 ) , is consistent with the engagement of the C-terminal five residues of the functional core of PPP1R15 . Deletion of this predicted short amphipathic helical region of PPP1R15 markedly impairs activation by actin in vitro and de-stabilizes actin's association with PPP1R15 in vivo ( Chambers et al . , 2015 ) . Together , the in vivo and in vitro approaches suggest two complementary roles for actin: stabilizing the PPP1R15-PP1 complex and endowing it with substrate specificity . The two functions can be separated in vitro by mutations that retain sufficient PP1 binding to enable reliable measurement of the non-specific phosphatase activity of the binary PPP1R15-PP1 complex and all but eliminate actin's ability to promote substrate specificity . In the ternary complex of PPP1R15-PP1-actin , the active site ( of the catalytic subunit ) is positioned at the bottom of a shallow platform constituted by actin and PP1 . The N-terminal lobe , containing the phosphorylated residue of eIF2a , can be computationally docked into this composite surface , bringing phospho-S51 in contact with the catalytic site . This model of the enzyme-bound substrate identifies surface residues in PP1 and actin that could engage the substrate in ionic interactions . In support of this model , we note that mutations in several of these residues enfeeble dephosphorylation of eIF2aP by the specific ternary complex , but less so by the binary complex . This feature is noted not only for the eIF2a R74E mutation , predicted to affect the binding to actin , but also in mutations affecting residues predicted to interact with charged surface residues of PP1 ( eIF2a residues R66E and K86E ) , suggesting that such substrate-PP1 contacts are especially important in the setting of the actin-containing ternary complex . Together these observations support a simple model whereby the composite surface of actin and PP1 , held together by PPP1R15 , cooperatively engages its substrate to provide specificity through enhanced affinity of binding . G-actin reproducibly enfeebled the phosphatase activity of the PP1-PPP1R15 complex towards an irrelevant structured substrate ( phosphorylase A ) . The enfeebling effect of actin is modest by comparison to its stimulatory effect on dephosphorylation of the specific substrate , eIF2aP . This last observation is consistent with the shallowness of the composite surface , which would likely have a modest negative effect on accessibility of a structured substrate to the active site and even less so to inhibit an unstructured phosphorylated peptide ( such as GSTP ) . However , as bacterially-expressed PP1 has more promiscuous phosphatase activity than the enzyme purified from natural sources ( Alessi et al . , 1993; MacKintosh et al . , 1996 ) , we may be underestimating the extent to which G-actin biases against the dephosphorylation of other PP1 substrates , such as phosphorylase A . These considerations may also account for the apparent difference in the extent of the bias against phosphorylase A noted in this study and an earlier one which used PP1 derived from rabbit muscle ( Connor et al . , 2001 ) . The functional importance of an inhibitory facet of regulatory subunit action is influenced by stoichiometric considerations . As the pool of free PP1 in cells is believed to be low , the dramatic increase in PPP1R15A levels in stressed cells may have consequences not only in terms of enhanced dephosphorylation of eIF2aP but also in terms of diminished dephosphorylation of other substrates . Actin is an abundant protein . In vitro , the EC50 of G-actin for activation of eIF2aP dephosphorylation by PP1-PPP1R15 complexes is <100 nM ( for both isoforms of PPP1R15 ) . It is therefore possible that under most circumstances G-actin is not limiting to holophosphatase formation . The nearly wildtype ISR-antagonizing activity of the weaker PPP1R15A F585A and I589A mutants ( Figure 6—figure supplement 1 ) observed in the face of a nearly log-order higher EC50 for actin activation in vitro ( Figure 6B ) , could be explained by high G-actin concentration in CHO cells . However , an informative precedent exists for regulation of cellular processes by variation in the abundance of G-actin: physiological changes in the ratio of monomeric G-actin to filamentous F-actin regulate the activity of serum response factor ( SRF ) through G-actin's ability to engage SRF's activation partner , MAL/MRTF-A , with affinities in the low micromolar range ( Mouilleron et al . , 2008 ) . Given the abundance of cellular G-actin-binding proteins , many of which engage the cleft between domains I and III ( Dominguez and Holmes , 2011 ) and are thus predicted to compete with PPP1R15 , it is possible that in some circumstances G-actin availability may also be limiting to holophosphatase formation and that actin dynamics may thus be coupled to eIF2aP dephosphorylation . The eIF2a phosphorylation-dependent ISR strongly affects memory formation ( Costa-Mattioli et al . , 2005; Sidrauski et al . , 2013 ) and actin dynamics are important in synaptogenesis ( Dillon and Goda , 2005 ) . Therefore , it is tempting to speculate on localized changes in G-actin levels modulating levels of phosphorylated eIF2 and protein synthesis through localized changes in holophosphatase activity in the dynamic synapse . The ISR also plays an important role in protein folding homeostasis , especially in the endoplasmic reticulum , where PERK-mediated eIF2a phosphorylation defends the stressed organelle by limiting the influx of newly synthesized proteins . The phenotype of combined deletion of PPP1R15A and PPP1R15B tells us that the capacity to reverse this process , by eIF2aP dephosphorylation , is essential to homeostasis and cell survival ( Harding et al . , 2009 ) . However , under conditions of severe stress , the normal induction of the PPP1R15A gene overshoots its mark , such that genetic ( Marciniak et al . , 2004 ) or pharmacological ( Boyce et al . , 2005; Tsaytler et al . , 2011 ) attenuation of PPP1R15-mediated phosphatase activity is protective . Accordingly , small molecules that engage the composite actin-PP1 surface might serve as specific inhibitors of eIF2aP dephosphorylation to reverse the aforementioned failure of homeostasis , without affecting the many other PP1-containing holophosphatases in the cell . Supplementary file 1 lists the plasmids used , their lab names , description and notes their first appearance in the figures and their corresponding label , and provides a published reference , where available . A combination of PCR-based manipulations , restriction digests , and site-directed mutagenesis procedures was used to mobilize the coding sequence and produces in-frame fusions with the affinity tags ( GST , His X6 or FLAG epitope ) or mCherry fluorescent tag , and to create the deletions and the point mutations indicated in the text . Actin was purified from rabbit muscle as described ( Pardee and Spudich , 1982 ) , dialyzed against G buffer ( 2 mM Tris pH 8 , 0 . 2 mM ATP , 0 . 5 mM DTT , 0 . 1 mM CaCl2 , 1 mM NaN3 ) , blocked from further polymerization by incubation with a fivefold molar excess of Latrunculin B ( #428020 , Calbiochem ) , and used in biochemical and structural studies . Chromatographically purified bovine pancreatic DNase I ( >2000 units/mg ) was purchased from Worthington Biochemical Corporation ( #LS002006 , Lakewood NJ ) and constituted into a complex with actin as previously described ( Mannherz et al . , 1980 ) . Binary complexes of mouse PP1G and PPP1R15 with diverse endpoints ( Supplementary file 1 ) were created by co-transforming BL21 T7 Express lysY/Iq E . coli ( #C3013 , New England Biolabs , Ipswich , MA ) with ampicillinr marked GST-tagged PPP1R15 expression plasmids and kanamycinr marked untagged PP1G expression plasmids . Colonies bearing both resistance markers were selected on LB plates with 50 µg/ml kanamycin and 100 µg/ml ampicillin . Both plasmids were stable under this dual selection regime . Cultures of 0 . 5–6 litres of LB media with 1 mM MnCl2 , were inoculated with 1/100 volume of a saturated over-night culture ( both under dual selection ) , allowed to progress to an OD600 of 0 . 6–0 . 8 at 37°C , at which point they were switched to 18°C and induced with 1 mM IPTG , and cultured further for 20 hr until harvest . Bacterial pellets were chilled on ice , suspended in 4–8 pellet volumes of ice-cold lysis buffer ( 50 mM Tris pH 7 . 4 , 500 mM NaCl , 1 mM MnCl2 , 1 mM MgCl2 , 1 mM TCEP , 100 µM PMSF , 20 mTIU/ml aprotonin , 2 µM leupeptin , and 2 µg/ml pepstatin in 10% glycerol ) , and lysed with an EmulsiFlex-C3 homogenizer ( Avestin , Inc , Ottawa , Ontario ) at 4°C . Lysates were clarified in a JA-25 . 50 rotor ( Beckman Coulter ) at 33 , 000×g for 30 min , loaded onto a suspension of glutathione sepharose 4B beads and allowed to bind at 4°C for 1–2 hr . The beads were batch-washed with 45 bed volumes of lysis buffer , transferred to a 10 ml column and further washed with 30 bed volumes of lysis buffer , and eluted in 50 mM Tris pH 7 . 4 , 100 mM NaCl , 40 mM GSH , 0 . 5 mM MnCl2 , 0 . 5 mM TCEP , 10% glycerol . Tobacco Etch Virus protease ( TEV ) cleavage ( 12 . 5 µg TEV protease/mg protein ) was performed overnight ( at 4°C ) and the clarified mixture of free GST , PPP1R15-containing complex and residual uncleaved precursor proteins was loaded onto a tandem array of two 10/300 mm columns , Superdex 75 and Superdex 200 , with a 1 ml GSTrap 4B column at the outflow ( all from GE Healthcare , Buckinghamshire , UK ) and developed in gel filtration buffer ( 20 mM HEPES , 100 mM NaCl , 0 . 1 mM MgCl2 , 0 . 5 mM MnCl2 , 0 . 1 mM ADP , 0 . 2 mM TCEP , protease inhibitors ) . In this configuration , the binary complex elutes first and the free GST and any uncleaved binary complexes are retained in the GSTrap 4B column , eluting later with the glutathione rich ‘salt’ peak . Ternary complexes of PP1G-PPP1R15 and actin were assembled by combining stoichiometric amounts of the binary complex ( assembled on the glutathione-sepharose resin ) with latrunculin B-blocked G-actin , incubated for 90–120 min at 4°C , and eluted with the elution buffer noted above , cleaved and fractionated by the tandem size exclusion chromatography setup described above . Binary complexes used in biochemical studies only , were purified as described above , with the following modifications: the lysis buffer contained 0 . 1% Triton X-100 to reduce non-specific binding; the elution buffer contained 20 mM Tris pH 7 . 4 , 100 mM NaCl , 40 mM GSH , 1 mM MnCl2 , protease inhibitors , 1 mM TCEP , 10% glycerol; and the TEV cleavage step was omitted in some instances . PERK kinase domain , N-terminal lobe of eIF2a ( with and without an EGFP tag ) and GSTag were expressed from plasmids PerkKD-pGEX4T-1 , GST_TEV_eIF2a-NM_EGFP , eIF2a-NM_pET30a ( and mutant variants ) , and pGSTag ( Supplementary file 1 ) in bacteria , and purified by glutathione sepharose or nickel affinity chromatography accordingly . GST-PERK , assembled on glutathione-sepharose beads was extensively washed in 20 mM Tris pH 7 . 4 , 150 mM NaCl , 4 mM DTT , 0 . 01% Triton X-100 , adjusted to 60% ( vol/vol ) glycerol . In this configuration , kinase activity is retained for months at −20°C . His X6 tagged proteins were recovered in lysis buffer: 50 mM Tris pH 7 . 4 , 500 mM NaCl , 20 mM imidazole , 1 mM MgCl2 , protease inhibitors , 1 mM TCEP , 0 . 2% Triton X-100 in 10% glycerol , washed in the same and eluted in: 50 mM Tris pH 7 . 4 , 100 mM NaCl , 500 mM imidazole , 1 mM TCEP in 10% glycerol and further purified by size exclusion chromatography ( Superdex 75 or 200 10/300 , GE Healthcare ) in gel filtration buffer: 25 mM Tris pH 7 . 4 , 100 mM NaCl , 0 . 1 mM EDTA , 1 mM TCEP in 10% glycerol , snap frozen in liquid nitrogen and stored in small aliquots at −80°C . Mouse pancreas lysate ( cytosolic fraction ) was produced by homogenization of fresh tissue in a Teflon-glass homogenizer in 4 vol of homogenization buffer ( 250 mM sucrose , 50 mM Tris–HCl pH 7 . 4 , 5 mM MgCl2 , 1 mM DTT , and protease inhibitors ) followed by a series of clarification steps: 800×g , 15 min , twice; 6000×g , 15 min , twice , and 100 , 000×g in a TLA-100 rotor ( Beckman Coulter ) , for 1 hr , all at 4°C . The clarified lysate had a protein concentration of 5–10 mg/ml , based on Bradford's method . Molar concentrations of solutions of pure proteins were estimated from the UV absorbance spectrum and the extinction coefficient , predicted by the ProtParam tool of ExPasy http://web . expasy . org/protparam/ . Fractions of PP1G-PPP1R15B binary complex and PP1G-PPP1R15-Actin ternary complex after gel filtration chromatography were pooled for protein concentration to 6–10 mg/ml by centrifugal filter units with molecular weight cutoff of 10 kDa ( Millipore , Amico Ultra ) . Then protein ( 200 nl ) and screen plate reservoir buffer ( 100 nl ) for each drop in 96-well crystallization plate were set up as sitting drops for crystal growth in incubators at either 20°C or 14°C . Crystals of PP1G ( 7–300 ) -PPP1R15B ( 631–660 ) grew at 20°C in a solution of 2 . 4 M sodium malonate , pH7 . 0 . Crystals of PP1G ( 7–300 ) -PPP1R15B ( 631–669 ) grew in 3 M NaCl , 0 . 1 M Hepes , pH7 . 5 . Crystals of PP1G ( 7–300 ) -PPP1R15B ( 631–684 ) grew in 2 . 0 M sodium malonate , pH6 . 0 . Crystals of PP1G ( 1–323 ) -PPP1R15B ( 631–701–H6 ) -Actin grew over a period of 21 days in solution containing 0 . 2 M CaCl2 , 0 . 1 M Hepes , pH7 . 0 , 20% PEG6000 . Crystals were harvested with perfluoropolyether cryo oil as cryoprotective agent and diffraction data were collected at 100 K at the Diamond Light Source ( see Table 1 for details of beamlines used ) . The data were integrated with either XDS ( Kabsch , 2010 ) ; PP1G ( 7–300 ) -PPP1R15B ( 631–660 ) and PP1G ( 1–323 ) -PPP1R15B ( 631–701–H6 ) -actin or iMosflm ( Leslie and Powell , 2007 ) ; PP1G ( 7–300 ) -PPP1R15B ( 631–669 ) and PP1G ( 7–300 ) -PPP1R15B ( 631–684 ) , then scaled with aimless ( Evans , 2006 ) in the CCP4 program suite ( Murshudov et al . , 2011 ) . All structures were solved by molecular replacement in Phaser ( McCoy et al . , 2007 ) . For PP1G ( 7–300 ) :PPP1R15B ( 631–660 ) the structure of PP1G from PDB: 1FJM was used as a model , then after refinement this structure was used as a model for the remaining structures . The actin component of the ternary complex was obtained from PDB: 4B1Y . The three binary complexes were refined using phenix . refine ( Afonine et al . , 2012 ) . The ternary complex was refined with Refmac5 ( Murshudov et al . , 2011 ) ; to cope with the low resolution of the data , the refinement used jelly-body and NCS restraints , TLS parameterisation to describe thermal motion , and the actin structure in PDB: 1IJJ was used as an external reference model . Data collection and structure refinement statistics are reported in Table 1 . Ribbon , surface and density map representations were prepared in the PyMOL Molecular Graphics System ( Version 1 . 5 . 0 . 4 , Schrödinger , LLC ) . The quaternary complex of the PP1G-PPP1R15B-actin ternary complex with its substrate , the N-terminal lobe of eIF2a was modeled by the HADDOCK server ( Dominguez et al . , 2003 ) . Chains B , F , and G of PDB: 4V0U were used as inputs for the ternary complex and the N-terminal lobe of yeast eIF2a ( chain A of PDB: 1Q46 ) as the input for the substrate . Distance restrains between phosphoserine 51 of yeast eIF2a N-terminal domain and Asn124 , His125 , Asp208 , and the two metal ions in the active site of PP1G from the ternary complex were applied . Chinese Hamster Ovary ( CHO ) cells were cultured and electroporated using the Neon transfection system ( Life Technologies ) as described ( Tsunoda et al . , 2014 ) . The effect of PPP1R15A and PPP1RR15B on the activity of the unfolded protein response was studied by transient transfection of stable CHOP::GFP reporter cells ( Novoa et al . , 2001 ) . 12 hr after transfection cells were exposed to 2 . 5 μg/ml tunicamycin ( Calbiochem ) for an additional 12 hr to activate CHOP::GFP and analyzed by flow cytometry . The data were statistically analyzed using FlowJo . Differences in the median reporter values between the wildtype and mutant variants of PPP1R15A were evaluated by One-way ANOVA , by the medians Kruskal–Wallis test ( cells not expressing mCherry and with an intensity above 103 were discarded ) using GraphPad Prism version 6 . 0e ( GraphPad Software ) . HEK293T cells were resuspended in phosphate buffered saline and pelleted gently ( 800×g at 4°C ) . The supernatant was decanted , the cell mass measured by weight and the corresponding packed cell volume derived by the assumption of a buoyant density of 1 . 005 gm/ml ( Loken and Kubitschek , 1984 ) . Cells were lysed in homogenization buffer ( 20 mM HEPES pH = 7 . 5 , 150 mM NaCl , 1% Triton X-100 , 1 mM EDTA , 1 mM DTT , 10% glycerol , and protease inhibitors ) , and the clarified lysate resolved by SDS-PAGE alongside samples with a known mass of recombinant eIF2a ( purified from bacteria as described above ) , immunoblotted with a primary rabbit serum directed to the N-terminus of eIF2a ( residues 1–185 ) ( lab name NY1308 ) and an IRDye fluorescently labeled secondary anti-rabbit IgG ( LiCor ) . The fluorescence signals were detected with an Odyssey near-infrared imager ( LiCor ) and quantified by ImageJ ( NIH ) .	For a cell to build a protein , it must first copy the instructions contained within a gene . A complex molecular machine called a ribosome then reads these instructions and translates them into a protein . This translation process involves a number of steps . Proteins called eukaryotic translation initiation factors ( or eIFs for short ) coordinate the first step in the process , which is known as ‘initiation’ . The eIFs also provide the cell with ways to control how quickly it makes proteins . For example , when a cell is stressed , either by starvation or toxins , it adds a phosphate group onto part of an eIF protein , called eIF2α . This modification makes this eIF protein less able to initiate translation , and so the cell builds fewer proteins and conserves more of its resources during times of stress . Once the stressful conditions are over , the phosphate group is removed from eIF2α by an enzyme called a phosphatase . This phosphatase contains two subunits: one that recognizes eIF2α and another that removes the phosphate group . However , experiments that attempted to recreate this phosphatase activity using just these two subunits in a test tube failed to generate a working enzyme that specifically targeted the phosphate group of eIF2α . This suggests that in cells this enzyme contains an additional unknown subunit . Now , Chen et al . ( and Chambers , Dalton et al . ) report the identity of a ‘missing’ third subunit as a protein known as globular-actin or G-actin . First , Chen et al . looked at the three-dimensional structure of a two-subunit complex formed from the previously known subunits of the phosphatase enzyme , and confirmed that it could remove phosphate groups from a range of proteins and not just eIF2α . However , when a mixture of other proteins taken from mouse cells was added to this two-subunit complex , the complex could specifically remove the phosphate group on the eIF2α protein . Further experiments revealed that G-actin was the protein in the mixture that , when added to the two-subunit complex , made it specifically target the eIF2α protein . Chen et al . then used a combination of biochemical and structural biology techniques to investigate the phosphatase activity of the three-subunit complex . These findings suggest a plausible molecular mechanism by which the three-subunit complex becomes selective for its target , but further refinements to the structural work will be needed to critically test these suggestions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2015	G-actin provides substrate-specificity to eukaryotic initiation factor 2α holophosphatases
Centrosome structure , function , and number are finely regulated at the cellular level to ensure normal mammalian development . Here , we characterize PPP1R35 as a novel bona fide centrosomal protein and demonstrate that it is critical for centriole elongation . Using quantitative super-resolution microscopy mapping and live-cell imaging we show that PPP1R35 is a resident centrosomal protein located in the proximal lumen above the cartwheel , a region of the centriole that has eluded detailed characterization . Loss of PPP1R35 function results in decreased centrosome number and shortened centrioles that lack centriolar distal and microtubule wall associated proteins required for centriole elongation . We further demonstrate that PPP1R35 acts downstream of , and forms a complex with , RTTN , a microcephaly protein required for distal centriole elongation . Altogether , our study identifies a novel step in the centriole elongation pathway centered on PPP1R35 and elucidates downstream partners of the microcephaly protein RTTN . The centrosome is a membrane-less organelle whose major role is to organize , orient , and regulate the site of microtubule formation . In somatic dividing cells , the centrosome is critical for ensuring faithful and timely chromosome segregation and establishment of the correct cell division axis , whereas in non-dividing and differentiated cells , it is critical for cellular polarization and cilia formation ( Conduit et al . , 2015; Vertii et al . , 2016 ) . Centrosomes are essential for normal human development and health ( Nigg and Holland , 2018 ) . Loss of function mutations in centrosomal proteins , including components of the centriolar cartwheel , elongation machinery , appendages , and pericentriolar material , are responsible for developmental defects such as primary recessive microcephaly ( Barbelanne and Tsang , 2014 ) , primordial dwarfism ( Care4Rare Canada Consortium et al . , 2015; Rauch et al . , 2008; Zheng et al . , 2016b ) , and ciliopathies ( Reiter and Leroux , 2017 ) . Defects in centrosome number and structure are a major hallmark of tumorigenesis ( Gönczy , 2015; de Cárcer and Malumbres , 2014; Nigg and Holland , 2018 ) . Recently , studies in mouse models indicated that centrosome over-duplication concomitant with mutations in p53 drives tumor formation in the epidermis ( Serçin et al . , 2016 ) and can drive tumor formation in certain other tissues , even in the absence of concurrent p53-/- mutations ( Levine et al . , 2017 ) . Therefore , it is essential to characterize the critical set of proteins required for centrosome assembly to understand the molecular mechanism of disease and identify therapeutic targets ( Nigg and Holland , 2018 ) . Due to its important role in cell and tissue homeostasis , the centrosome is built in a highly-regulated , stepwise manner through the assembly of a multiplicity of protein complexes ( Conduit et al . , 2015; Mennella et al . , 2014 ) . Significant progress has been made in understanding how centrosome duplication begins in most somatic cells—at the G1/S phase boundary—with the assembly of the cartwheel , a nine-fold symmetrical scaffold made of SAS6 , STIL , and CEP135 . While SAS6 molecules can undergo remarkable self-assembly in vitro , the kinase Plk4 promotes cartwheel formation and centriole duplication by phosphorylating STIL to favor its interaction with SAS6 ( Vulprecht et al . , 2012; Lin et al . , 2013b; Dzhindzhev et al . , 2014; Arquint and Nigg , 2016 ) . The initial binding of Plk4 to the centriole is governed by CEP63 ( Brown et al . , 2013 ) , CEP152 ( Brown et al . , 2013; Kim et al . , 2013; Sonnen et al . , 2013; Dzhindzhev et al . , 2010; Hatch et al . , 2010; Cizmecioglu et al . , 2010 ) , and CEP192 ( Kim et al . , 2013; Sonnen et al . , 2013 ) . After cartwheel formation , CPAP , recruited by STIL ( Tang et al . , 2011 ) , aids in the formation of the centriole microtubule wall ( Pelletier et al . , 2006; Schmidt et al . , 2009 ) by regulating centriolar microtubule plus-end dynamics ( Basten and Giles , 2013; Zheng et al . , 2016a ) . CEP135 facilitates the stabilization of the centriole structure ( Ohta et al . , 2002; Basten and Giles , 2013 ) but may also play a more direct role in initial cartwheel formation as recombinant Drosophila SAS6 and Bld10 ( Drosophila CEP135 homolog ) can self-organize into a nine-fold symmetrical cartwheel structure ( Guichard et al . , 2017 ) . Once the initial steps of procentriole formation occur , centriole elongation can proceed . However , we have a limited understanding of the essential components required for centriole elongation , which happens between S and G2 phases , and how they are assembled in a stepwise manner . CPAP has been shown to interact with CEP120 ( Lin et al . , 2013a ) and SPICE ( Comartin et al . , 2013 ) in a complex that regulates centriole elongation at the centriolar microtubule wall ( Archinti et al . , 2010; Lin et al . , 2013b ) . Centrobin has also been implicated in directly regulating centriolar microtubule elongation ( Lee et al . , 2010; Zou et al . , 2005 ) and stability by binding to α-Tubulin ( Gudi et al . , 2011 ) and by regulating CPAP levels ( Gudi et al . , 2015; Gudi et al . , 2014 ) . Centrobin is further required to recruit CP110 , a protein forming a cap-like structure on the distal end of the centriole that suppresses centriole elongation ( Schmidt et al . , 2009 ) . Proximal to CP110 , several proteins localized to the distal luminal end of centrioles such as POC5 ( Azimzadeh et al . , 2009 ) , POC1B ( Venoux et al . , 2013 ) , and OFD1 ( Singla et al . , 2010 ) have been implicated in promoting the elongation of the centriole’s distal region . More recently , additional proteins have been identified , namely CEP295 ( Chang et al . , 2016 ) and RTTN ( Chen et al . , 2017 ) , which have been proposed to play a scaffolding role in the elongation process by connecting the centriole wall to the luminal centriolar region . However , it remains unclear if there are components in the lumen of the centriole that stabilize interactions with centriolar wall proteins . RTTN ( rotatin ) was originally identified as a protein critical for axial rotation and left-right symmetry specification in mice ( Faisst et al . , 2002 ) . Subsequently , mutations in human RTTN have been shown to cause primary microcephaly and primordial dwarfism ( Kheradmand Kia et al . , 2012; Grandone et al . , 2016; Care4Rare Canada Consortium et al . , 2015 ) . Recent reports have shed light on the cellular function of RTTN . The Drosophila RTTN homolog , Ana3 , was demonstrated to be a centrosomal component critical for maintaining the structural integrity of centrioles ( Stevens et al . , 2009 ) , whereas human RTTN , localized near the centriolar cartwheel , has been shown to be dispensable for initial centriole assembly , but critical for formation of a full-length centriole ( Chen et al . , 2017 ) . It remains unclear what factors are downstream of RTTN and how they promote the elongation and stabilization of the centriole once the cartwheel is formed . Here , we characterize human PPP1R35 , the product of the gene C7orf47 , which was previously identified in fractions co-purifying with centrosomes in a high-throughput mass spectrometry study ( Jakobsen et al . , 2011 ) . Our study demonstrates that PPP1R35 is a centrosomal component located in the proximal centriolar lumen above the cartwheel . We further demonstrate that PPP1R35 is not important for early centriole assembly but is critical for centriole elongation by impacting the recruitment of the microtubule-binding elongation machinery . In addition , we show that PPP1R35 is downstream of RTTN in the elongation pathway and that they form a protein complex . Altogether , we describe a novel centriolar component essential for centriole formation and identify a new mechanistic step downstream of RTTN in the pathway to reach a fully elongated centriole and functional centrosome . To examine if PPP1R35 is a bona fide centrosomal protein , we generated a U2OS cell line constitutively expressing GFP-PPP1R35 under the control of a low copy protein expression promoter ( Kim et al . , 2011 ) , integrated into the genome through the Flp-In system . GFP-PPP1R35 showed two main protein populations: one enriched in a diffraction limited spot located in the middle of the cell adjacent to the nucleus , consistent with centrosomal localization , and a cytoplasmic pool ( Figure 1 ) . To verify that the observed PPP1R35 was located at the centrosomes , we transfected the GFP-PPP1R35 U2OS Flp-In cell line with a vector expressing Centrin 1-mCherry and observed co-localization of Centrin 1-mCherry with GFP-PPP1R35 ( Figure 1a ) . To examine the dynamics of PPP1R35 during the cell cycle , we conducted long-term live-cell imaging by spinning disc confocal fluorescence microscopy ( Figure 1b and Video 1 ) . PPP1R35 was found on two centrosomes ( grandmother and mother ) throughout the entire cell cycle ( Figure 1b ) . We observed some cells that have four GFP-PPP1R35 spots prior to mitosis ( Video 1 ) and noted that a second GFP-PPP1R35 spot was always present after mitosis , suggesting that PPP1R35 is recruited on daughter centrioles prior to mitosis . To verify that both the mother and daughter centrioles have recruited GFP-PPP1R35 , we leveraged the ~1 . 5 x resolution increase of sub-diffraction live cell imaging . In all G2 cells examined , prior to centrosome separation , two GFP-PPP1R35 spots are resolvable on each of the centrioles ( grandmother and mother ) , confirming that PPP1R35 resides on both the mother and daughter centrioles and must be recruited early in the duplication cycle , in S or early G2 phase ( Figure 1c ) . To confirm that GFP-PPP1R35 localization is consistent with the endogenous protein , we imaged U2OS cells labeled with antibodies against PPP1R35 and γ-tubulin by confocal microscopy ( Figure 1d ) and cells labeled with antibodies against PPP1R35 and CETN1 by 3D structured illumination microscopy ( 3DSIM ) , and observed co-localization ( Figure 1e ) . Since the anti-PPP1R35 antibody showed high background staining we used GFP-PPP1R35 to conduct further studies . To ensure that the GFP tag did not impact the localization of the protein , we examined the morphology of the centrosome by 3DSIM and did not observe a difference between WT and GFP-PPP1R35-expressing U2OS cells ( Figure 1—figure supplement 1a ) . Furthermore , we verified that the GFP-PPP1R35 construct did not alter centrosome biogenesis by measuring the total number of CEP152-labeled centrosomes in WT and GFP-PPP1R35-expressing U2OS cells ( Figure 1—figure supplement 1b ) . In addition , the GFP-PPP1R35 construct rescues the centriole duplication phenotype when PPP1R35 levels are knocked down by siRNA targeting the 3’ untranslated region ( 3’UTR , see below Figure 3c ) . Next , to determine whether PPP1R35 was continuously recruited or was stably associated to the centrosome , we performed Fluorescence Recovery After Photobleaching ( FRAP ) experiments . Comparison of the fluorescence recovery curves of the cytoplasmic versus centrosomal PPP1R35 pools revealed that centrosomal PPP1R35 did not fully recover to pre-bleach levels after photobleaching , therefore indicating that the protein has low turnover and is stably associated at the centrosome ( Figure 1f , g ) . This observation is consistent with a previous analysis that identified PPP1R35 as co-purifying with centrosomal components and observed only a 22% turnover in centrosomal PPP1R35 as measured by stable isotope labeling of amino acids in cell culture ( SILAC ) mass spectrometry ( Jakobsen et al . , 2011 ) . Altogether , our imaging experiments demonstrate that PPP1R35 is a resident protein of the centrosome and is recruited to the nascent daughter centriole early in the duplication cycle . To further dissect the role of PPP1R35 at the centrosome , we used super-resolution microscopy to precisely map the position of PPP1R35 relative to several reference markers , whose position at the centrosome has been previously characterized by EM and fluorescence imaging ( Figure 2 ) . To perform these experiments , we used linear 3DSIM , a technique that provides a 2-fold resolution increase over standard confocal/widefield fluorescence microscopy , which is sufficient to resolve the relative distribution of many centrosomal proteins and allows for straightforward multicolor imaging ( Sydor et al . , 2015; Mennella et al . , 2012 ) . 3DSIM imaging showed that GFP-PPP1R35 is located in the centrosomal lumen , as suggested by the position of PPP1R35 in the middle of the ring structure formed by CEP152 ( Hatch et al . , 2010; Cizmecioglu et al . , 2010 ) ( Figure 2a ) . Next , we used several proximal ( SAS6 , CEP135 , CPAP , CEP250 ) and distal ( CETN1 , POC1B , POC5 ) proteins to locate the position of PPP1R35 along the centrosomal longitudinal axis . Qualitative assessment of the 3DSIM micrographs showed that the position of PPP1R35 is biased toward proximal markers such as CPAP and CEP135 more than either the utmost proximal ( i . e . CEP250 ) or distal ( i . e . CETN1 ) ends of the centriole ( Figure 2b ) . To precisely map PPP1R35 , we performed a quantitative analysis of the distance between PPP1R35 and many centriolar reference markers . We collected hundreds of 3DSIM images and analyzed micrographs with centriole side views where PPP1R35 was in the same z-plane of the protein of interest for measurement to avoid distortions due to anisotropic resolution ( Figure 2c ) . 3DSIM molecular mapping shows that PPP1R35 is located furthest from the distal end proteins ( Centrin-1: 230 ± 50 nm; POC5: 160 ± 50 nm; POC1B: 140 ± 60 nm ) , but closer to proximal end markers such as CEP135 ( 90 ± 40 nm ) and CPAP ( 60 ± 30 nm ) , yet not as proximal as SAS6 ( 120 ± 40 nm ) or CEP250 ( 170 ± 50 nm ) ( Figure 2d ) . Together , we conclude that PPP1R35 localizes to the proximal centriolar lumen just above the cartwheel ( Figure 2e ) . Since PPP1R35 is recruited early in the centrosome duplication pathway , we hypothesized that it might play a role in regulating centrosome biogenesis . To test this possibility , we depleted PPP1R35 protein levels in U2OS cells by targeting the mRNA with two non-overlapping siRNA strands , one designed to be complementary to an exon in the conserved C-terminal region and the second to the 3’ UTR of the PPP1R35 mRNA ( Figure 3a , b ) . The specificity of the siRNA strands toward PPP1R35 was validated by western blotting of cells expressing GFP-PPP1R35 ( Figure 3—figure supplement 1 ) and RT-qPCR ( Figure 3—figure supplement 2 ) . We opted to deplete cells of PPP1R35 via siRNA rather than CRISPR/Cas9 gene editing since previous studies demonstrated cell lethality in the absence of PPP1R35 ( Hart et al . , 2015; Neumann et al . , 2010 ) . Cells were treated with siRNA for 72 hr , thereby allowing cells to progress through multiple cell cycles and accumulate any centriolar defects ( Figure 3b and Figure 3—figure supplement 3 ) . With both siRNA strands , a significant decrease was observed in centrosomal staining of CEP152 , a protein recruited in the last stages of daughter centriole formation ( Fu et al . , 2016 ) ( Figure 3c ) . This phenotype is rescued by exogenously expressing GFP-PPP1R35 , demonstrating the specificity of the siRNA and the resultant phenotype of PPP1R35 loss ( Figure 3c ) . We next sought to narrow down the stage of centrosome duplication at which PPP1R35 plays a role by labeling PPP1R35-depleted cells with several centrosomal proteins sequentially recruited during its assembly ( Conduit et al . , 2015; Fu et al . , 2015; Loncarek and Bettencourt-Dias , 2018 ) . This analysis revealed that centriolar components recruited early in the pathway such as SAS6 ( Dzhindzhev et al . , 2014 ) , CEP135 ( Loncarek and Bettencourt-Dias , 2018; Fu et al . , 2015 ) and Centrin 1 ( Middendorp et al . , 1997 ) , are modestly affected at centrioles in the absence of PPP1R35 , as opposed to proteins recruited in later stages , such as CEP295 ( Chang et al . , 2016 ) , POC1B ( Venoux et al . , 2013 ) , and CEP152 ( Loncarek and Bettencourt-Dias , 2018; Fu et al . , 2015 ) that are drastically reduced ( Figure 3e and Figure 3—figure supplement 4 ) . To assay for centrosome function , we examined whether centrosomes could recruit the pericentriolar material or efficiently nucleate microtubules in the absence of PPP1R35 by staining for Cdk5rap2 and γ-tubulin . In both cases , we observed a significant reduction upon PPP1R35 knockdown ( Figure 3—figure supplement 5 ) . A more significant impact on centriolar components recruited later in the pathway is more noticeable in a time-course experiment , in which cells are assayed at 24 hr intervals after siRNA treatment ( Figure 3f and Figure 3—figure supplement 3 ) . In these assays , there is little change in the recruitment of early-centriolar components such as SAS6 , CETN1 , and CEP135 up to the 72 hr time point . In contrast , defective recruitment of other components , such as CPAP and CEP152 , is present around the 48 hr time point . When cells treated with PPP1R35 siRNA were stained for CEP152 and SAS6 , the proportion of engaged centrosomes with cartwheels was not significantly different ( Figure 3g ) , further suggesting that PPP1R35 loss does not influence the early stages of centriole biogenesis . It is also noteworthy that at longer timepoints ( >72hr ) , CETN1 levels drastically decrease suggesting that overall centriole formation is being impacted . Altogether , these results demonstrate that PPP1R35 loss of function results in decreased centrosome number and suggest that PPP1R35 is critical for the recruitment of centriolar components after cartwheel formation . To better understand the mechanistic role of PPP1R35 in centriole duplication , we conducted biotinylation-dependent proximity mapping ( BioID ) ( Roux et al . , 2012 ) experiments using stable cell lines expressing protein fusions with a FLAG-BirA ( R118G ) ( henceforth referred to as BirA* ) tag on either the N- or C-terminus of PPP1R35 . BioID analysis revealed a proximity map with several high-confidence hits ( FDR score <1% ) . As expected , the proximity interactome of PPP1R35 shows several centrosomal proteins , including AZI1 ( CEP131 ) , CEP85 , and KIAA0753 ( Moonraker ) . One of the most robust hits , as evidenced by the high numbers of peptides identified by both N- and C-terminal BirA* tags , is RTTN , a recently characterized protein ( H . -Y . Chen et al . , 2017 ) whose mutations in patients cause microcephaly ( Care4Rare Canada Consortium et al . , 2015; Grandone et al . , 2016 ) , dwarfism , and polymicrogyria ( Kheradmand Kia et al . , 2012 ) ( Figure 4a; Supplementary file 1 ) . Stable HEK293T T-REX Flp-In cell lines showed normal centriolar localization as determined via confocal imaging with the marker CEP152 ( Figure 5f ) . We reasoned that if PPP1R35 and RTTN are in close proximity to each other and are both located at the centrosome near the cartwheel ( H . -Y . Chen et al . , 2017 ) , they might form a bona fide protein complex . To test this hypothesis , we performed FLAG immunoprecipitation ( IP ) of the N- and C-terminal tagged PPP1R35 constructs ( Figure 4b and Supplementary file 1 ) . Notably , RTTN was found to form a complex with the N-terminally tagged BirA-PPP1R35 . RTTN was the only protein identified with high confidence by BioID that also co-immunoprecipitated with PPP1R35 , suggesting a strong link between this microcephaly protein and PPP1R35 function . Interestingly , IP-mass spectrometry data using the N-terminal FLAG-tagged PPP1R35 , but not the C-terminal FLAG-tag construct , detected a high-confidence interaction with RTTN . IP with the PPP1R35 C-terminal FLAG-tag construct still identified RTTN peptide counts above that of the controls , but below our confidence level cut-off ( Supplementary file 1 ) , indicating that the interaction between the two proteins has been severely impaired but not completely abolished . Altogether , this suggests that the binding site might reside within the conserved C-terminal region of PPP1R35 . RTTN is a 298 kDa protein predicted to have an elongated , solenoid conformation ( Fournier et al . , 2013 ) that has been recently reported to localize to basal bodies ( Stevens et al . , 2009; Kheradmand Kia et al . , 2012 ) and the centrosome ( H . -Y . Chen et al . , 2017; Stevens et al . , 2009; Care4Rare Canada Consortium et al . , 2015 ) . Specifically , RTTN has been shown to localize to the proximal lumen of centrioles near CEP135 and the cartwheel ( H . -Y . Chen et al . , 2017 ) . To further characterize the structural and functional relationship of PPP1R35 and the microcephaly protein RTTN , we mapped the position of RTTN relative to PPP1R35 by 3DSIM imaging and quantitative analysis ( Figure 4c ) . To detect RTTN we used both an N-terminal mCherry-RTTN construct and an antibody recognizing residues 50–150 of RTTN . Our data show that RTTN localizes to the proximal centriole and it is located in close proximity to PPP1R35 , consistent with our BioID findings ( PPP1R35 distance from mCherry-RTTN , 80 ± 50 nm; anti-RTTN , 110 ± 50 nm; Figure 4d ) . To further explore the functional relationship between PPP1R35 and RTTN , we depleted PPP1R35 from U2OS cells by siRNA and examined RTTN recruitment . The presence of RTTN at the centrosome is moderately , yet significantly , diminished upon PPP1R35 depletion ( Figure 4e ) . When the reciprocal recruitment was explored by RTTN depletion , a major reduction in centrosomal PPP1R35 was observed ( Figure 4f ) . This phenotype appears to be unrelated to the decrease in centriole number expected as a result of RTTN knockdown ( Chen et al . , 2017 ) , because the number of cells with at least 2 centrin spots remains unchanged ( Figure 4—figure supplement 1 ) , yet the number of cells lacking GFP-PPP1R35 is significantly reduced . These results show that the two proteins co-localize at the centriole and that both proteins are mutually required for each other’s recruitment to the centriole , with RTTN exerting a more significant impact on PPP1R35 recruitment to the centriole . Altogether , our data suggest that RTTN and PPP1R35 form a complex and that RTTN acts upstream of PPP1R35 . PPP1R35 is a highly conserved protein whose homologs are found across a wide range of eukaryotic species , ranging from the simple multicellular organism Trichoplax adhaerens to Homo sapiens ( Figure 5—figure supplement 1 ) . Interestingly , PPP1R35 homologues are found only in Holozoa species , correlating well with species presenting centrosomes , with the exception of Caenorhabditis elegans ( Hodges et al . , 2010 ) . PPP1R35 can be divided into two major domains based on amino acid sequence homology: the highly divergent N-terminal domain and the more conserved C-terminal domain ( Figure 5—figure supplement 2 ) . Despite its variability across evolution , the N-terminus contains several highly conserved residues in mammalian species including three serine residues ( S45 , S47 , S52 in Homo sapiens PPP1R35 ) previously found phosphorylated in large scale phospho-proteomic studies in both human and mouse cells ( Olsen et al . , 2010; Dephoure et al . , 2008; Chi et al . , 2008 ) ( Figure 5a and Figure 5—figure supplement 2 ) . In particular , S47 and S52 have been reported to be Cdk phosphorylation sites ( Chi et al . , 2008 ) ( Figure 5a ) . As such , we hypothesized that these residues could be candidates for regulating PPP1R35 activity during centrosome duplication , as Cdk2 ensures that centrosome duplication takes place concomitantly with DNA synthesis in S-phase ( Fu et al . , 2015 ) . To probe the importance of these residues in the interaction with RTTN , we mutated all three serine residues to either non-phosphorylatable alanines ( S45A , S47A , S52A ) or to phospho-mimetic aspartic acids ( S45D , S47D , S52D ) and generated inducible HEK293 T-Rex Flp-In cell lines expressing the mutant N-terminal BirA-PPP1R35 constructs . Neither the triple alanine nor the triple aspartic acid mutant significantly impacted the presence of PPP1R35 at the centrosome , nor its proximity to RTTN ( Figure 5b , g ) . Furthermore , co-IP demonstrated that neither phospho-mutant impacted the interaction between PPP1R35 and RTTN ( Figure 5—figure supplement 3 ) . We further evaluated the role of PPP1R35 phosphorylation on centriole duplication by examining whether the above phospho-mutants are able to rescue our centriole defect phenotype . When cells were depleted of endogenous PPP1R35 by the 3’ UTR-targeting siRNA and expressed the triple aspartic acid mutant ( S45D , S47D , S52D ) GFP-PPP1R35 in trans , we did not observe any reduction in centrosome number ( Figure 5c ) . Despite multiple attempts , we were unable to generate a triple alanine ( S45A , S47A , S52A ) mutant cell line in U2OS cells , therefore we examined both triple mutant cell lines in HEK293 cells ( Figure 5—figure supplement 4 ) . Analysis of individual alanine mutants ( S45A and S47A ) in U2OS cells is also consistent with the notion that that despite their conservation , these resides are not playing a critical for PPP1R35’s function in centriole biogenesis ( Figure 5c ) . PPP1R35 is predicted to contain a canonical RVxF PP1-binding site ( Peti et al . , 2013 ) , encompassing residues 77–81 ( Hendrickx et al . , 2009 ) in the N-terminal domain . This site is conserved only among Chordata species ( Figure 5—figure supplement 1 and Figure 5—figure supplement 2 ) . Interestingly , disruption of the predicted PP1 binding site by mutating two conserved residues , V79 and F81 , to alanine ( Peti et al . , 2013 ) leads to proper targeting to the centriole ( Figure 5g ) and does not disrupt PPP1R35’s proximity to , or interaction with , RTTN ( Figure 5d , Figure 5—figure supplement 3 ) . Furthermore , the V79A , F81A mutant nearly completely rescued our centriole duplication phenotype ( Figure 5e ) , suggesting that it is not critical for centriole biogenesis . Since PPP1R35 forms a complex with the microcephaly protein RTTN and this protein has been previously linked to centriole elongation , where its loss resulted in shortened centrioles ( Chen et al . , 2017 ) , we investigated whether PPP1R35 knockdown results in diminished centriole length . To this effect , we used 3DSIM to measure the distance between the proximal end of centrioles labeled with acetylated tubulin to CP110 , which localizes to the centriole’s distal end ( Figure 6a ) . Acetylated tubulin has been suggested to be an early tubulin modification during centriole duplication ( Balashova et al . , 2009 ) and it has been previously used for conducting centriole length measurements ( Chen et al . , 2017 ) . Furthermore , we verified that tubulin acetylation is unaffected when PPP1R35 is knocked down by siRNA ( Figure 6—figure supplement 1 ) . To ensure that we were examining mature centrioles , we focused our analysis only on mother centrioles in G2 phase cells in which a clear mother and daughter centriole were present . The length of mother centrioles was determined to be 356 ± 65 nm in control-RNAi treated cells , in agreement with previous reports ( Thauvin-Robinet et al . , 2014 ) . When PPP1R35 levels are knocked down by siRNA , we see a significant reduction in centriole length to 263 ± 69 and 246 ± 83 nm for the exon and 3’ UTR siRNA , respectively ( Figure 6b ) . On the contrary , overexpression of PPP1R35 does not significantly change centriole length ( 382 ± 89 nm ) . Due to the small effect observed on RTTN recruitment when PPP1R35 levels are reduced , we hypothesized that the shorter centriole length may be due to the inability of nascent centrioles to recruit proteins involved in elongation . We then examined cells treated with PPP1R35 siRNA for recruitment of proteins involved in either microtubule stabilization/recruitment such as CPAP and SPICE ( Archinti et al . , 2010; Zheng et al . , 2016a; Tang et al . , 2009; Comartin et al . , 2013; Lin et al . , 2013b ) or the elongation of the distal portion of the centriole such as POC5 ( Azimzadeh et al . , 2009 ) and both proteins were significantly reduced in the absence of PPP1R35 . Consistently , CP110 , a negative regulator of centriole elongation recruited early in the elongation pathway , was not significantly changed relative to control RNAi-treated cells ( Figure 6c ) . Altogether , this demonstrates that PPP1R35 is a critical factor for centriole assembly by promoting recruitment of centriole elongation proteins . PPP1R35 was initially suggested to be a centrosomal protein by mass spectrometry studies that identified PPP1R35 as co-purifying with isolated centrosomes ( Jakobsen et al . , 2011 ) . Here we show that the uncharacterized protein PPP1R35 is stably associated at the centrosome throughout the cell cycle where it plays a critical role in its elongation . Our loss of function analysis places PPP1R35 relatively early in the centriole duplication pathway , after cartwheel formation and before complete centriole elongation . In addition , we demonstrate that the ultimate downstream effect of PPP1R35 loss is shortened centrioles , suggesting that PPP1R35 directly controls the elongation pathway . It is interesting to note that this diminished centriole duplication defect takes several days to become very pronounced suggesting that either the short centrioles are still competent to duplicate but to a diminished degree , or that only low levels of PPP1R35 are needed for proper biological activity . Phenotypic analysis places PPP1R35 upstream of CPAP , CEP295 , SPICE and POC5 , which are all proteins involved in centriole elongation ( Tang et al . , 2009; Lin et al . , 2013b; Comartin et al . , 2013; Chang et al . , 2016 ) , but downstream of RTTN , which also affects the recruitment of POC1B , POC5 , and CEP295 similarly to PPP1R35 ( Chen et al . , 2017 ) . Furthermore , BioID , IP , and 3DSIM data show that PPP1R35 and RTTN form a protein complex and that this complex is likely a result of a direct interaction as RTTN is the only protein identified in both BioID and IP experiments . Whereas several distal luminal proteins have been reported to date ( Azimzadeh et al . , 2009; Venoux et al . , 2013; Paoletti et al . , 1996 ) , PPP1R35 is the first to be mapped to the proximal luminal region , an uncharacterized ‘outpost’ right above the cartwheel . Since most proteins involved in centriole elongation are localized along the microtubule wall or the distal end of the centriole ( Comartin et al . , 2013; Hatzopoulos et al . , 2013; Azimzadeh et al . , 2009; Schmidt et al . , 2009 ) , the interesting distribution of PPP1R35 suggests that it likely acts directly through RTTN , which our imaging show luminal localization for the N-terminus of the protein , in close proximity to PPP1R35 . Our phylogenetic analysis demonstrates that PPP1R35 is conserved in a wide range of species , including Drosophila , parasitic worms , and mammals ( Figure 5—figure supplement 1 ) . PPP1R35 is likely the human homolog of the Drosophila protein Reduction of Cnn dots 4 ( Rcd4 ) , a protein identified in a large RNAi screen aimed at discovering novel proteins impacting centrosome formation and PCM assembly ( Dobbelaere et al . , 2008 ) . Alignment of Homo sapiens PPP1R35 and Drosophila Rcd4 results in an overall similarity of 24% , with the greatest homology in the conserved C-terminal domain ( Figure 5—figure supplement 2 ) . In all identified PPP1R35 homologs , the N-terminus exhibits a large degree of variability , hinting at an organism-specific specialization for this domain . We probed several conserved residues in PPP1R35 including several conserved serines ( S45 , S47 , S52 ) , but none of these mutants appeared to drastically impact centriole biogenesis . Intriguingly , the aforementioned serines and putative PP1-binding site , which are well conserved in Chordata , are not conserved in Rcd4 . Furthermore , changes to the phosphorylation state did alter the overall BioID proximity map of PPP1R35 , including altering the proximity to CEP85 , AZI1 , and OFD1 , the latter two shown to have important roles in ciliogenesis ( Hall et al . , 2013; Ma and Jarman , 2011; Wilkinson et al . , 2009; Romio et al . , 2004; Ferrante et al . , 2006 ) . The centrosomal duplication cycle is closely linked to the cell cycle and tightly controlled by a host of kinases and phosphatases ( Pihan , 2013; Fujita et al . , 2016 ) . While kinases inherently possess temporal and spatial specificity , protein phosphatases require a regulatory component to properly function ( Heroes et al . , 2013; Peti et al . , 2013; Korrodi-Gregório et al . , 2014 ) . To date , only a handful of centrosomal PP1 regulatory components have been identified ( Katayama et al . , 2001; Meraldi and Nigg , 2001; Mi et al . , 2007; Helps et al . , 2000; DeVaul et al . , 2013; Huang et al . , 2005 ) and overall knowledge of their interaction and role with PP1 is limited in scope . PPP1R35 is annotated to be a PP1-regulatory protein and contains a canonical PP1-binding site . Despite previous reports that demonstrated PP1-binding and inhibition ( Hendrickx et al . , 2009; Fardilha et al . , 2011 ) , we were unable to identify any PP1 isoform in our BioID or IP screens using HEK293 cycling cells . This is not completely surprising as previous studies have encountered similar difficulties in identifying interactions between protein phosphatases and their interactors , frequently due to the transient nature of binding ( St-Denis et al . , 2016 ) . We tested the role of this interaction in regard to centriole duplication by mutating the canonical PP1-binding site but found that this interaction with PP1 does not appear to be critical for PPP1R35’s role at the centrosome . Overall , this suggests that despite PPP1R35’s annotation and previous demonstration as a PP1 regulator , this activity may not be related to centriole biogenesis . However , we cannot yet rule out the possibility that a second , non-canonical PP1-binding site is involved or that the PP1-regulating activity of PPP1R35 is required only in specific cellular functions not investigated here , such as ciliogenesis . Nonetheless , the large number of robust , non-centrosomal BioID hits suggests that PPP1R35 may serve other functions in the cell apart from centriole duplication and perhaps these other functions require PPP1R35’s PP1-regulation activity . Despite the importance of centriole elongation to numerous human diseases , the exact mechanisms through which elongation takes place is still poorly understood ( Loncarek and Bettencourt-Dias , 2018 ) . To date , two major pathways governing centriole elongation have been described , one positive-growth mechanism acting on assisting microtubule elongation ( CPAP/CEP120/SPICE ) ( Kohlmaier et al . , 2009; Schmidt et al . , 2009; Tang et al . , 2009; Lin et al . , 2013b; Comartin et al . , 2013 ) and a second negative-growth mechanism involving CP110 and CEP97 , which form a cap-like structure on the centriole to restrict microtubule growth ( Schmidt et al . , 2009; Spektor et al . , 2007; Franz et al . , 2013; Chen et al . , 2002 ) . However , even with the discovery of additional proteins such as POC5 ( Azimzadeh et al . , 2009 ) , CEP295 ( Chang et al . , 2016 ) , and Centrobin ( Gudi et al . , 2015 ) , all of which impact centriole elongation , we apparently have yet to acquire a complete picture of this process . Here , we have identified a novel key player of this process , PPP1R35 . Our data suggest that PPP1R35 primarily impacts the CPAP/CEP120/SPICE and RTTN/CEP295 pathways of centriole elongation ( Figure 6d , e ) . Previously , RTTN was proposed to be critical for stabilizing the early procentriole containing STIL , CPAP , and SAS6 and for recruiting CEP295 , which in turn can recruit POC5 and POC1B ( Chen et al . , 2017 ) . Our data are consistent with a model where the impact on centriole elongation occurs primarily through PPP1R35’s interaction with RTTN . Such a role is consistent with the localization of both PPP1R35 and RTTN , which are uniquely positioned just above the cartwheel in the region where elongation following initial centriole formation would occur . PPP1R35 could be involved in either modulating RTTN’s turnover or interactions with other proteins . This reasoning is supported by the observation that RTTN loss has been shown to cause lack of proper CEP295 , POC1B , and POC5 recruitment and an interdependency with CPAP and CEP135 ( Chen et al . , 2017 ) , consistent with our observed phenotype when PPP1R35 is knocked down . Mutations in numerous centriolar proteins have been linked to microcephaly ( Barbelanne and Tsang , 2014; Kaindl et al . , 2010 ) , including components involved in centriole elongation such as CPAP ( Leal et al . , 2003 ) and RTTN ( Care4Rare Canada Consortium et al . , 2015; Grandone et al . , 2016 ) . The discovery of additional microcephaly proteins will aid in our understanding of the disease and assist in the development of future therapies . Our data show that that PPP1R35 impacts the process of centriole elongation through a close relationship with a known microcephaly protein , RTTN , therefore suggesting that PPP1R35 may be one such candidate microcephaly gene . Future DNA sequencing of microcephaly patients and animal model studies are needed to address this issue . Tables detailing primers ( Supplementary file 2 ) and siRNA strands ( Supplementary file 3 ) used in this study are available in the Supplemental Material section . The construct pIRES Centrin1 mCherry was a gift from Matthieu Piel ( Addgene plasmid # 64338 ) . For cloning experiments , PCR products were amplified from plasmid cDNA ( PPP1R35 cDNA , MGC clone Image ID 4773899; RTTN cDNA , GE Dharmacon ORFeome cDNA 25914 ) , verified for specificity of amplification on an agarose gel and purified using the PureLink PCR Purification kit ( Thermo Fisher Scientific ) or gel-extracted ( Qiagen DNA Gel Extraction Kit ) when necessary . All cloning experiments were conducted using Gibson Assembly ( New England Biolabs ) according to the manufacturer’s instructions . Site-directed mutagenesis for the single serine mutants was conducted using QuikChangeII XL Lightning site-directed mutagenesis kit ( Agilent ) . Gene synthesis ( Invitrogen GeneArt ) was used to generate mutant PPP1R35 containing the triple Ala , Asp , and V79A , F81A mutations and were subsequently cloned into the GFP-containing FRT vector via Gibson Assembly as described above . Newly generated plasmid constructs were verified using Sanger Sequencing ( ACGT Corp . , Toronto ) . U2OS and U2OS Flp-In cells were cultured in McCoy’s 5A medium ( Gibco ) supplemented with 10% fetal bovine serum ( Wisent ) and 1 X antibiotic/antimycotic ( Gibco , 100 units/ml penicillin , 100 µg/ml streptomycin , and 0 . 25 µg/ml amphotericin B ) . HEK 293T TREX cells were maintained in Dulbecco’s Modified Eagle’s Medium ( Gibco ) supplemented with 10% tetracycline-free fetal bovine serum ( Wisent ) and 1 X antibiotic/antimycotic . All cells were cultured at 37°C with 5% CO2 and routinely passed . The cells were tested routinely for mycoplasma contamination using Invitrogen’s Mycoplasm Detection Kit . For cellular transfection of DNA plasmids , JetPrime ( Polyplus ) was used according to the manufacturer’s instructions . Cells were subsequently selected in the appropriate antibiotic ( puromycin for constructs in U2OS Flp-In cells and hygromycin for constructs in HEK293 Flp-In TREX cells ) to generate cell lines with stably integrated transgenes . For siRNA transfections , Lipofectamine RNAiMax ( Invitrogen ) was used according to the manufacturer’s instructions . All siRNA strands were transfected at a final concentration of 40 nM and cells were assayed 72 hr post transfection . Scrambled siRNA and siRNA targeting GAPDH were used as negative and positive controls , respectively . GAPDH control knockdown efficacy was monitored by immunofluorescence and by western blot . The number of labeled centrosomes per cell ( classified into categories as either 0 , 1 , 2 or >2 labeled centrosome; see Figure 3d for example of centrosome spots imaged in cells and Figure 3—figure supplement 4 for example of categories of centrosomal counts ) were manually counted and all siRNA experiments were conducted in at least triplicate except for the HEK293 siRNA experiments that were conducted in at least duplicate . For live-cell imaging , cells were seeded on KOH-washed coverglass ( Electron Microscope Sciences ) to reduce background fluorescence and subsequently left overnight to adhere . The standard culture media was replaced with DMEM medium lacking phenol red ( Gibco ) supplemented with 10% fetal bovine serum and 1 X antibiotic/antimycotic . Cells were imaged on a Zeiss Axio Observer Spinning-disc microscope equipped with Yogokawa spinning disk head , Phototronics EM CCD camera , and a 63x objective ( NA = 1 . 4 ) . The samples were maintained at 37°C with 5% CO2 during imaging in an incubation chamber . Automated acquisition of a 30 µm z-stack with a 0 . 75 µm step size every 40–45 min was obtained using the Zeiss Zen Blue software . During acquisition the lowest possible minimal laser power was used to avoid phototoxicity , resulting in movies of an average length of 14 hr . The cells were imaged on a Leica SP8 Confocal DMI6000 microscope equipped with a HyD detector and a 63x ( NA = 1 . 4 ) oil objective . The samples were maintained at 37°C with 5% CO2 during imaging in an incubation chamber . Samples were bleached with a white light laser for approximately 1 . 5 s and the subsequent recovery monitored for an additional 26 . 5 s . The resultant plots were analyzed and fit to a single exponential curve using the build-in FRAP analysis function in the Leica Analysis Suite X software . A table detailing all antibodies used in this study , including concentrations and suppliers , is available in the key resources table . Cells were plated onto coverslips ( Electron Microscope Sciences; previously cleaned with KOH ) and left overnight to adhere . The cells were treated with 0 . 02% w/v digitonin in PBS for 5 min at RT to remove the cytoplasmic population of PPP1R35 followed by fixation with −20°C methanol for 20 min . The cells were blocked for 1 hr using 5% FBS in PBS supplemented with 0 . 5% Tween-20 . The cells were incubated with primary and secondary antibodies for 40 min each at RT . To detect specific primary antibodies , Alexa 488- , Alexa 568- , or Alexa 647-conjugated IgGs were used as secondary antibodies at a dilution of 1:1000 ( Invitrogen ) . Cell nuclei were stained with Hoescht 33342 ( Thermo Fisher ) . Cells were mounted with 0 . 5% n-propyl gallate in 80% glycerol mounting media . 3DSIM data were collected using an ELYRA PS . 1 ( Carl Zeiss Microscopy ) with a Plan-Apochromat 63x or 100x/1 . 4 Oil immersion objective lens with an additional 1 . 6x optovar . An Andor iXon 885 EMCCD camera was used to acquire images with 101 nm/slice z-stack intervals over a 5–10 µm thickness . The fluorophores were excited with 405 , 488 , 555 and 647 nm wavelengths and band-pass 420–480 , 495–550 , 570–620 , long-pass 655 and 750 nm filters were used to collect the emission wavelengths . Laser powers at the objective focal plane of 52 . 6 mW in the 2–12% range , exposure time between 50–250 ms and EMCCD camera gain values between 5 and 50 were used during image acquisition . For each image field , grid excitation patterns were collected for five phases and three rotation angles ( −75o; −15o , +45o ) . The raw data were reconstructed using the SIM module of ZEN Black Software ( version 8 . 1 ) with noise filter values between −6 and −3 . Channel alignment was conducted using calibrated file generated from super-resolution Tetraspec beads ( Carl Zeiss Microscopy ) . If appropriate , whole-volume images or maximum intensity projections were exported as tiff files to be further analyzed in ImageJ/Fiji ( NIH ) . To measure the position of PPP1R35 relative to various centriolar markers , only 3DSIM images in which both the fluorescence maxima of PPP1R35 and the corresponding reference protein were on the same z-slice were analyzed . The distance between the peak maxima for the two markers were determined using the caliper function built in to the Zeiss Zen Black software ( see Figure 2c for an example ) . The centriole length measurements were conducted in an identical manner using CP110 as a distal end marker and the acetylated tubulin signal as a proximal end marker ( see Figure 6a for an example ) . BioID was conducted as previously described ( Firat-Karalar and Stearns , 2015; Gupta et al . , 2015 ) . To generate stable cell lines expressing recombinant BirA fusion proteins for BioID experiments , HEK293 Flp-In T-Rex cells were co-transfected with the pcDNA5/FRT/TO PPP1R35-FLAG-BirA* or pcDNA5/FRT/TO FLAG-BirA-PPP1R35 plasmid and Flp Recombinase Expression plasmid pOG44 in a 1:20 ratio , and then selected for multiple passages with increasing antibiotics concentrations to reach final concentrations of 400 µg/ml Hygromycin B ( Invitrogen ) and 15 µg/ml Blasticidin ( Gibco , Thermo Fisher Scientific ) . HEK293 TREX Flp-In cells expressing the appropriate transgene were cultured until 90–100% confluency and treated for 24 hr with 1 µg/ml tetracycline to induce BirA expression and 50 µM biotin to allow biotinylation of proteins . HEK293T TREX Flp-In cells transfected with a vector containing either the N- or C-terminal FLAG-BirA but no PPP1R35 , were processed in parallel as controls . Cells were collected , pelleted , and washed three times with PBS prior to freezing . Cell pellets were processed for Bio-ID and FLAG ImmunoPrecipitation ( IP ) experiments as described previously ( Coyaud et al . , 2015 ) . Interactor classification: bona fide interactors were defined as high confidence protein identifications ( ProteinProphet p>0 . 85 ) with a SAINT score ≥0 . 75 , based on 4 independent MS runs . Histone hits were eliminated . Fold-change was calculated as described previously ( Coyaud et al . , 2015 ) . Total cell lysates were collected using RIPA lysis buffer ( Pierce ) supplemented with mammalian protease inhibitor ( BioBasic; 100 mM PMSF , 1 mM Bestatin , 1 . 5 mM Pepstatin A , 1 . 4 mM E-64 , 0 . 08 mM Aprotinin , 1 mM Leupeptin ) and cell debris pelleted by spinning for 30 min at 12 , 000 rpm . Protein concentrations were determined using a BCA protein assay kit ( Pierce ) . Protein lysate containing ~15–30 µg of total protein was loaded onto well of 4–12% Bis-Tris gels ( Invitrogen ) . Proteins were transferred to nitrocellulose membrane for 2 hr on an Invitrogen Bolt Minigel Apparatus at 10 V and blocked with 5% skim milk for 1 hr . Membranes were subsequently incubated with specific antibodies overnight at 4°C . Secondaries conjugated with HRP ( Cell Signalling ) were used at a 1:2000 dilution . Blots were developed using the ECL Chemiluminescent Substrate Kit ( Invitrogen ) . RNA was extracted from cells using the GeneJet RNA Purification kit ( Thermo Scientific ) and subsequently treated with the RapidOut DNA Removal kit ( Thermo Scientific ) . Purified RNA was quantitated and only RNA with an A260/A280 ratio greater than 1 . 8 was used for reverse transcription with the BioRad iScript cDNA Synthesis kit with 1 µg of RNA as the template . All quantitative PCR was performed using a CFX Connect Real-Time System ( BioRad ) with SsoAdvanced Universal SYBR Green Supermix ( BioRad ) and 500 nM combined primer concentration per well . The relative expression of the target genes were normalized to RNA polymerase II and TATA binding protein transcript levels for each condition and then relative to expression in the scrambled siRNA-treated sample . Primer sequences can be found in Supplementary file 2 . No-template and no-reverse transcriptase controls were run for each primer pair to confirm the lack of primer–dimer formation/DNA contamination and genomic DNA contamination , respectively . At least three biological replicates were run per condition . Data were analyzed using the CFX Maestro software ( BioRad ) . All kits were conducted as per the manufacturer’s protocol . All siRNA experiments were analyzed as 2 × 2 contingency tables in which all cells for a given population ( i . e . cells with >1 CEP152 spots ) were pooled for all replicates . To determine the p-values compared to the scrambled siRNA control for each dataset , Barnard’s Test was used in R with unpooled variances ( package by Kamil Erguler; available at https://github . com/kerguler/Barnard ) ( Erguler , 2015 ) . A summary of all statistics for the siRNA experiments can be found in Supplementary file 4 . For all other statistical tests , the Student’s T-Test was used . Error bars represent the standard deviation for all replicates . For all figures , the following conventions were used: ns ( p>0 . 05 ) , * ( p≤0 . 05 ) , ( p≤0 . 01 ) , * ( p≤0 . 001 ) , **** ( p≤0 . 0001 ) . The NCBI protein database was queried with the search term ‘PPP1R35’ and all resultant hits were downloaded . For species with multiple annotated isoforms , the longest was selected . Any entries that were also annotated as a protein of known function ( i . e . transposase , helicase , etc ) were removed . Furthermore , only one organism per genus was selected to ensure broad coverage yet avoiding artifacts caused by over-sampled genera . All entries were from the Holozoa group of Eukaryotes . In order to ensure that no sequences from other major eukaryotic groups were missed , Delta Blastp searches using both the Homo sapien and Drosophila melanogaster PPP1R35 sequences were used to search for homologs in representative genera from the remaining eukaryotic groups ( exact genera probed are those found in Figure 1 of Ref . [Hodges et al . , 2010] ) . No additional homologs were identified outside of the Holozoa . Multiple sequence alignments were performed using Clustal Omega ( Sievers et al . , 2011 ) with the default settings . The phylogeny was inferred using the Bayesian method implemented with MrBayes v . 3 . 2 . 6 ( mixed amino acid rate mode ) and run for 2 . 5 million generations until the standard deviation of split frequencies was 0 . 199 . Drosophila melanogaster Sds22 , a PP1 regulator protein identified to have diverged early from homologous PP1 regulators ( Ceulemans et al . , 2002 ) , was used as the outgroup . Trees were drawn using FigTree v . 1 . 4 . 3 .	Most animal cells contain a structure called the centrosome , which plays a vital role in helping cells to divide for producing new cells . Early in the cell division process , cells make a copy of their centrosome . Each centrosome includes two cylindrical structures called centrioles encased in a complex web of other proteins . The centrioles must get longer for the duplication process to work correctly , but it is not clear which proteins help the centrioles to elongate . Previous work suggested that a protein called PPP1R35 might be a centrosome protein . To investigate its role , Sydor et al . performed experiments that reduced the amount of PPP1R35 in cells grown in the laboratory . Cells that contained fewer PPP1R35 proteins also contained fewer centrioles; these centrioles were also shorter and lacked some of the proteins that can elongate them . Super-resolution microscopy found PPP1R35 in the centre of the centrioles , in a region involved in the early stages of elongation . Sydor et al . also found that PPP1R35 interacts with a protein called RTTN , which is linked to centriole elongation . RTTN contributes to a condition called microcephaly , which prevents the brain from developing properly and results in individuals having a small head . Future work that builds on the findings presented by Sydor et al . could therefore help researchers to understand the causes of microcephaly in patients .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2018	PPP1R35 is a novel centrosomal protein that regulates centriole length in concert with the microcephaly protein RTTN
We present in vivo single-cell FRET measurements in the Escherichia coli chemotaxis system that reveal pervasive signaling variability , both across cells in isogenic populations and within individual cells over time . We quantify cell-to-cell variability of adaptation , ligand response , as well as steady-state output level , and analyze the role of network design in shaping this diversity from gene expression noise . In the absence of changes in gene expression , we find that single cells demonstrate strong temporal fluctuations . We provide evidence that such signaling noise can arise from at least two sources: ( i ) stochastic activities of adaptation enzymes , and ( ii ) receptor-kinase dynamics in the absence of adaptation . We demonstrate that under certain conditions , ( ii ) can generate giant fluctuations that drive signaling activity of the entire cell into a stochastic two-state switching regime . Our findings underscore the importance of molecular noise , arising not only in gene expression but also in protein networks . Cellular physiology is deeply shaped by molecular fluctuations , resulting in phenotypic diversity and temporal variability that can be both detrimental and beneficial ( Rao et al . , 2002; Kussell and Leibler , 2005; Lestas et al . , 2010; Hilfinger et al . , 2016 ) . One of the most important and well-studied sources of intracellular fluctuations is stochastic gene expression ( Elowitz et al . , 2002; Eldar and Elowitz , 2010; Raj and van Oudenaarden , 2008 ) , which can generate substantial cell-to-cell variability in protein levels within isogenic populations under invariant environmental conditions . Such heterogeneity in protein counts are readily measurable by fluorescent-protein reporters ( Elowitz et al . , 2002; Ozbudak et al . , 2002 ) , but mechanistically tracing the consequences of such molecular noise to the level of complex cellular phenotypes such as signaling and motility remains a significant challenge , in part due to the multitude of interactions between gene products , but also because each of those interactions can , in principle , become an additional source of noise . In this paper , we study how multiple sources of molecular noise , arising in both gene expression and protein-protein interactions , affect performance of the E . coli chemotaxis network , a canonical signaling pathway . In bacteria , gene-expression noise tends to manifest itself as stable cell-to-cell differences in phenotypes that persist over the cell’s generation time , because typical protein lifetimes are longer than the cell cycle ( Li et al . , 2014 ) . The architecture of signaling networks can have a profound influence on their sensitivity to such noise-induced differences in protein levels , and it has been shown that the design of the E . coli chemotaxis network confers robustness of a number of signaling parameters , such as precision of adaptation , against variability in gene expression ( Barkai and Leibler , 1997; Kollmann et al . , 2005 ) . On the other hand , cell-to-cell differences in behavior can also be advantageous for isogenic populations under uncertain and/or time-varying environments , and it has been argued that the manner in which the chemotaxis network filters gene expression noise to shape phenotype distributions could be under selective pressure ( Frankel et al . , 2014; Waite et al . , 2016 ) . In principle , molecular noise arising in processes other than gene expression , such as protein-protein interactions within signaling pathways , can also contribute to cellular variability . However , such noise sources tend to be harder to study experimentally because , in contrast to gene-expression noise , which can be characterized by measuring fluorescent reporter levels ( Elowitz et al . , 2002; Raser et al . , 2004 ) , requirements for in vivo measurements of protein-protein interactions tend to be more demanding and no generically applicable strategies exist . The E . coli chemotaxis system provides a compelling experimental paradigm for addressing protein-signaling noise , because a powerful technique for in vivo measurements of protein signaling , based on Förster resonance energy transfer ( FRET ) , has been successfully developed ( Sourjik and Berg , 2002a; Sourjik et al . , 2007 ) . The chemotaxis network controls the motile behavior of E . coli , a run-and-tumble random walk that is biased by the signaling network to achieve net migrations toward favorable directions . The molecular mechanisms underlying this pathway have been studied extensively ( for recent reviews , see refs . ( Wadhams and Armitage , 2004; Tu , 2013; Parkinson et al . , 2015 ) ) . In brief , transmembrane chemoreceptors bind to ligand molecules , inhibiting the autophosphorylation of a central kinase , CheA . When active , CheA transfers its phosphate to CheY to form CheY-P . Meanwhile , the phosphatase CheZ dephosphorylates CheY-P to limit the signal lifetime . CheY-P binds to a flagellar motor , which in turn increases the chance of the motor to turn clockwise , leading to a tumble . An adaptation module consisting of the enzymes CheR and CheB implements negative integral feedback by tuning the sensitivity of the chemoreceptors via reversible covalent modifications that restore the kinase activity ( and CheY-P level ) . Despite its relative simplicity , this pathway exhibits many interesting network-level functionalities , such as cooperative signal amplification ( Segall et al . , 1986; Sourjik and Berg , 2002a; Bray et al . , 1998 ) , sensory adaptation ( Barkai and Leibler , 1997; Alon et al . , 1999 ) , and Weber’s law and fold-change detection ( Mesibov et al . , 1973; Lazova et al . , 2011; Clausznitzer et al . , 2014 ) , and FRET microscopy has proven extremely powerful in characterizing such signal processing of the chemotaxis pathway , especially in E . coli ( Sourjik and Berg , 2002a; Sourjik and Berg , 2004; Shimizu et al . , 2010; Oleksiuk et al . , 2011 ) , but also in Salmonella ( Lazova et al . , 2012; Rosier and Lazova , 2016 ) and B . subtilis ( Yang et al . , 2015 ) . It has been implemented in various ways ( Sourjik and Berg , 2002a; Sourjik and Berg , 2002b; Shimizu et al . , 2006; Kentner and Sourjik , 2009; Neumann et al . , 2012 ) , but most commonly by using CFP and YFP as the FRET donor-acceptor pair , fused to CheY and CheZ , respectively . To date , however , nearly all applications of FRET in the bacterial chemotaxis system have been population-level measurements in which signals from hundreds to thousands of cells are integrated to achieve a high signal-to-noise ratio . A pioneering study applied FRET at the single-cell level to study spatial heterogeneities in CheY-CheZ interactions ( Vaknin and Berg , 2004 ) , but those measurements were limited to relatively short times due to phototoxicity and bleaching . By exploring a range of fluorescent proteins as FRET pairs , and improving measurement protocols , we have developed a robust method for single-cell FRET measurements of chemotactic signaling dynamics in single bacteria over extended times . The data reveal extensive cell-to-cell variability , as well as temporal fluctuations that are masked in population-level FRET measurements . In contrast to previous single-cell experiments that relied on measurements of motor output or swimming behavior ( Berg and Brown , 1972; Spudich and Koshland , 1976; Segall et al . , 1986; Korobkova et al . , 2004; Park et al . , 2010; Masson et al . , 2012 ) , FRET alleviates the need to make indirect inferences about intracellular molecular interactions through the highly noisy 2-state switching of the flagellar motor , whose response function can vary over time due to adaptive remodeling ( Yuan et al . , 2012 ) . In a typical experiment , we are able to obtain dozens of ( up to ∼100 ) single-cell FRET time series simultaneously , to efficiently collect statistics of phenotypic diversity and temporal variability . To measure variability in intracellular signaling , we adapted a FRET assay for chemotaxis widely used for population-level measurements with fluorescent fusions to CheY and its phosphatase CheZ ( Sourjik and Berg , 2002a ) . On timescales longer than the relaxation of CheY’s phosphorylation/dephosphorylation cycle , the FRET level reflects the phosphorylation rate of CheY by the CheA kinase , thus providing an efficient in vivo measurement of the network activity ( Figure 1—figure supplement 1 ) . Instead of the conventional CFP/YFP FRET pair we used the fluorophores YFP and mRFP1 to avoid excitation with blue light , which induces considerably stronger photoxicity and also perturbs the chemotaxis system as a repellent stimulus ( Taylor and Koshland , 1975; Taylor et al . , 1979; Wright et al . , 2006 ) . Fusions of these fluorophores to CheZ and CheY still yield a fully functional phenotype ( Wolfe and Berg , 1989 ) , when observing chemotaxis on soft agar ( see Figure 1—figure supplement 1d ) . A field of E . coli cells expressing this FRET pair were immobilized on a glass surface imaged in two fluorescence channels , and segmented offline to obtain fluorescence intensities of donor and acceptor . From the fluorescence ratio , FRET time series for each cell in the field of view ( see Materials and methods ) can be computed , after dividing out the decay ( Figure 1—figure supplement 1 ) in each channel due to bleaching . Ratiometric FRET provides an anti-parallel response signature and confers robustness to parallel fluctuations that affect both fluorescent channels , such as differences in absolute fluorescence intensity due to inhomogeneous illumination and differences in cell size . For wildtype cells ( Figure 1a ) we found that the ensemble mean of single-cell FRET responses , ⟨FRET⟩⁢ ( t ) , agrees well with previous population-level measurements ( Sourjik and Berg , 2002a ) . Upon prolonged stimulation with a saturating dose of attractant α-methylaspartate ( MeAsp ) , ⟨FRET⟩⁢ ( t ) rapidly fell to zero before gradually returning to the pre-stimulus level due to adaptation . Upon removal of attractant , ⟨FRET⟩⁢ ( t ) rapidly increased to a maximum before returning to the pre-stimulus baseline . Single-cell FRET time series , FRETi⁢ ( t ) , had qualitatively similar profiles , but the kinetics of adaptation and response amplitudes demonstrate differences from cell to cell . For each cell , FRETi⁢ ( t ) is limited by the autophosphorylation rate of CheA and hence is proportional to ai[CheA]T , i ( provided [CheY] and [CheZ] are sufficiently high , see Materials and methods ) , in which ai is the activity per kinase ( 0 ≤ ai ≤ 1 ) and [CheA]T , i the total concentration of receptor-kinase complex of the i-th cell . The FRET level of each cell is thus bounded at a value FRETi , max which occurs when its kinases are fully active ( ai=1 ) , and can be measured by the removal of a sufficiently large stimulus after adaptation ( as in the experiment of Figure 1 ) . Hence from FRETi⁢ ( t ) the activity per kinase ai⁢ ( t ) can be readily determined by normalizing each FRET time series by its maximum response ai⁢ ( t ) =FRETi⁢ ( t ) /FRETi , max ( Figure 1b ) . The steady-state activity a0 , i , defined as the time-average of ai ( t ) before the addition of attractant , was found to vary from cell-to-cell with a coefficient of variation CV ( a0 ) =0 . 23 ( Figure 1c ) . The network activity controls the flagellar motor rotation , and hence this is consistent with the observation that cells in an isogenic population exhibit a broad range of steady-state tumble frequencies ( Spudich and Koshland , 1976; Bai et al . , 2013; Dufour et al . , 2016 ) . The adaptation precision is defined as its post-adaptational activity level divided by the pre-stimulus level ( Π=aadapted , i/a0 , i ) , hence a precision of 1 refers to perfect adaptation . The adaptation kinetics are quantified by the recovery time τrecovery , the time required for each cell to recover to 50% of its post-adaptation activity level ( aadapted , i ) . When observing the distributions of these parameters we noted that the cell-to-cell variability is high in the precision Π ( Figure 1d , CV=0 . 40 ) but the average precision ( 0 . 79 ) agrees well with population measurements ( Neumann et al . , 2014 ) . The variation is also substantial in τrecovery ( Figure 1e , CV=0 . 20 ) considering that the underlying kinetics of receptor methylation ( catalyzed by CheR ) involve thousands of events per cell , but falls within the range of ~20-50% from previous reports in which single-cell recovery times were estimated from motor-rotation or swimming-behavior measurements ( Berg and Tedesco , 1975; Spudich and Koshland , 1976; Min et al . , 2012 ) . The time required to recover from a saturating amount of attractant is determined not only by the stimulus size , but also the methylation rate of receptor modification sites catalyzed by CheR and the number of such sites that need to be methylated . Variability in the recovery time is thus likely to reflect cell-to-cell variability in the ratio between the expression level of CheR and that of the chemoreceptor species responding to ligand ( Tar for the experiment in Figure 1 ) . The diversity we observed here in adaptation precision , recovery time and steady-state activity was not explained by variation in salient experimental parameters ( Figure 1—figure supplement 2a–f ) , are reproducible across experimental days ( Figure 1—figure supplement 2g ) , and , on average , agree well with previous population-level FRET experiments and single-cell flagellar-based experiments . We thus conclude that single-cell FRET allows efficient measurement of signaling dynamics within individual bacteria to reveal variability in a wide variety of signaling parameters . The chemoreceptor clusters in E . coli are the central processing units and are responsible for signal integration and amplification . The sensory output of the cluster , the activity of the kinase CheA , is activated by a mixture of chemoreceptors . Cooperative interactions within the receptor-kinase complex leads to amplifications of small input stimuli and weighting different input signals . It has been shown that the composition of the receptor-kinase complexes can affect both the amplification as well as the weighting of different input signals ( Ames et al . , 2002; Sourjik and Berg , 2004; Kalinin et al . , 2010 ) , but how the amplification and integration varies across a population has not been characterized . To bridge the gap between collective behavior and its underlying single-cell motility it is essential to determine the variability of these important signaling parameters , as well as the origin of the variability . Also , current estimates of the apparent gain in the response ( defined as the fractional change in output divided by fractional change in input ) are based on population-averaged measurements which may or may not reflect single-cell cooperativity levels . In population averaged measurements , the largest gain is observed in adaptation-deficient ( CheRB- ) cells ( Sourjik and Berg , 2004 ) , in which the receptor population is homogeneous with respect to their adaptational modification state and hence in these cells variability in ligand sensing can be studied separately from variability induced by the adaptation enzymes . We probed the ligand sensitivity of CheRB- cells ( TSS58 ) at the single-cell level by FRET dose-response measurements in which step stimuli of successively larger amplitudes were applied over time ( Figure 2 ) . Considerable variability in the response to the attractant L-serine were observed across the population of immobilized cells simultaneously experiencing the same stimulus , with response magnitudes often ranging from virtually zero to full response ( Figure 2a ) . The resulting dose-response data could be well described by a Hill curve of the form [1+ ( [L]/K ) H]-1 , where the parameters ( 1/K ) and H are defined as the sensitivity and steepness , respectively , of the response of each cell . The family of dose response curves constructed from this ensemble of fit parameters reveals considerable variability from cell to cell in the shape of the response curve ( Figure 2b ) . What could be the cause of the diversity in ligand response in the absence of adaptation-induced heterogeneity ? We reasoned that expression-level variability of the five chemoreceptor species of E . coli , which are known to form mixed clusters with cooperative interactions ( Ames et al . , 2002; Sourjik and Berg , 2004 ) , could endow isogenic populations with sensory diversity . In line with this idea , CheRB- cells expressing only a single chemoreceptor species ( Tsr ) demonstrated not only higher cooperativity , but also attenuated variability in the dose-response profile from cell to cell ( Figure 2b–c ) , showing that the composition of the receptor population is important not only to tune the average ligand response of a population , but also in generating a wide range of sensory phenotypes within an isogenic population . It has been shown that expression level of chemoreceptors changes during growth of E . coli batch cultures: concomitant with the slowing of growth upon the transition from the exponential phase towards early stationary phase , the relative expression level ratio Tar/Tsr , the two most abundant chemoreceptors , increases from majority Tsr ( Tar/Tsr<1 ) to majority Tar ( Tar/Tsr>1 ) ( Salman and Libchaber , 2007; Kalinin et al . , 2010 ) . To probe the consequence of such changes for ligand-sensing diversity , we measured single-cell dose response curves in populations harvested at different cell densities during batch growth ( Figure 2d ) . The resulting population-averaged responses show a dependence of dose-response parameters on the optical density ( OD ) of the culture , shifting from highly sensitive ( low K ) and highly cooperative ( high H ) at low cell densities ( OD ≈ 0 . 3 ) to less sensitive ( high K ) and less cooperative ( low H ) at increased cell densities ( OD ≈ 0 . 45 , and OD ≈ 0 . 6 ) ( Figure 2d , open triangles , and Figure 2—figure supplement 2 ) . This trend is also visible at the level of single cells , but we found the responses to be highly variable under each condition ( Figure 2d , filled points ) . Remarkably , both K and H varied by over an order of magnitude , far exceeding the uncertainty in parameter estimates due to experimental noise ( Figure 2—figure supplement 3 ) . To further test the idea that ligand-response diversity is governed by differences in receptor expression levels , we considered the pattern of covariation between the fitted sensitivity K and cooperativity H in single cells ( Figure 2b , blue ) . In contrast to cells expressing Tsr as the only chemoreceptor , in which the variability in K is only ~0 . 15-20% ( Figure 2c ) , single cells expressing a wildtype complement of chemoreceptors demonstrated strong variation in K . This variation was negatively correlated with the cooperativity H ( Figure 2d ) . Noting that this overall pattern of covariation agrees well with dose response parameters obtained from population-level FRET experiments in which the Tar/Tsr ratio was experimentally manipulated via plasmid-based expression control ( Figure 2d , red circles; data from ( Sourjik and Berg , 2004 ) ) , we proceeded to quantitatively estimate the diversity in the Tar/Tsr ratio via fits of a multi-species MWC model ( Mello and Tu , 2005; Keymer et al . , 2006 ) to single-cell FRET data ( see Materials and methods ) . The resulting distribution of single-cell Tar/Tsr estimates ( Figure 2e ) was dominated by Tsr in cells harvested early ( OD ≈ 0 . 3 ) but the relative contribution of Tar increased in cells harvested at later stages of growth ( OD ≈ 0 . 45 ) and OD ≈ 0 . 6 ) . Interestingly , in addition to this increase in the mean of the Tar/Tsr distribution during batch growth , which confirms previous reports that found increased Tar/Tsr ratios at the population level ( Salman and Libchaber , 2007; Kalinin et al . , 2010 ) , we find that the breadth of the distribution also increases at later stages of growth . Thus , modulation of receptor expression during growth provides a means of tuning not only response sensitivity and cooperativity , but also single-cell diversity in the response of cell populations experiencing identical changes in their common environment . The large variability in the Tar/Tsr ratio ( CV≈0 . 5 at OD=0 . 45 ) is somewhat surprising given that the mean expression level of both receptors are known to be high and of order 103-104 copies per cell ( Li and Hazelbauer , 2004 ) . At such high expression , intrinsic noise in expression levels ( i . e . due to the production and degradation process of proteins , expected to scale as the square root of the mean ) could be as low as a few percent of the mean , and gene-expression fluctuations are expected to be dominated by extrinsic noise components ( i . e . those affecting regulation of gene expression , which do not scale with the mean ) . Quantitative measurements of gene expression reported in previous studies indicate a high degree of covariation among the expression level of chemotaxis genes , both at the population level under changes in growth conditions ( Li and Hazelbauer , 2004 ) and at the single-cell level across isogenic cells sampled from the same growth culture ( Kollmann et al . , 2005 ) . Correlated expression-level variation is also expected given the architecture of the flagellar regulon , in which all chemotaxis genes are under the control of a common master regulator ( Chilcott and Hughes , 2000 ) . These results indicate that the extrinsic ( correlated ) component of variation is greater than the intrinsic ( uncorrelated ) variability . Interestingly , however , a recent study ( Yoney and Salman , 2015 ) found using single-cell flow-cytometry a high degree of variability in the ratio of Tar/Tsr promotor activities ( CV≈0 . 45 at OD=0 . 51 ) comparable to the range of ratios extracted from our analysis of dose response data . Given that cell-to-cell variability in the Tar/Tsr ratio is much greater than achievable lower bounds of gene-expression noise in bacteria , it would be interesting to investigate the mechanistic sources of this variability , such as operon organization , promotor stochasticity , and translation-level regulatory structures ( Frankel et al . , 2014 ) . Variability in receptor expression could also explain the distribution of adaptation precision we observed in wildtype cells ( Figure 1d ) . In a previous population-level study , it has been shown that adaptation precision depends strongly on the expression-level ratio between the multiple chemoreceptor species , with the highest adaptation precision being achieved when the ligand-binding receptor is a minority within the total receptor population ( Neumann et al . , 2014 ) . Thus , the substantial heterogeneity in adaptation precision we observed ( CV=0 . 40 ) upon a saturating MeAsp stimulus is consistent with strong variability in the Tar/Tsr ratio . While bacteria can exploit molecular noise for beneficial diversification , variability can also limit reliable information transfer and degrade sensory performance . In the framework of E . coli’s run-and-tumble navigation strategy , chemotactic response to gradients requires that cells maintain a finite tumble bias , the fraction of time a bacterium spends tumbling , and avoids extreme values zero and one . The latter cases would correspond to unresponsive phenotypes that fail to switch between run and tumble states in response to the environmental inputs . One important mechanism that ensures responsiveness to stimuli over a broad range of input levels is sensory adaptation mediated by the methyltransferase/methylesterase pair CheR/CheB . These receptor-modifying enzymes provide negative feedback through the dependence of their catalytic activity on the receptor’s signaling state: the rate of methylation ( demethylation ) by CheR ( CheB ) is a decreasing ( increasing ) function of receptor-kinase activity ( Borczuk et al . , 1986; Amin and Hazelbauer , 2010 ) . This dependence of enzyme activity on the substrate conformation provides negative integral feedback that ensures precise adaptation ( Barkai and Leibler , 1997 ) toward the pre-stimulus steady-state activity a0 . Interestingly , one of the two adaptation enzymes , CheB , can be phosphorylated by CheA , the kinase whose activity CheB controls through its catalytic ( demethylation ) activity on receptors . Effectively , this adds an additional negative feedback loop to the network , but the role of this phosphorylation-dependent feedback has remained elusive since it has been shown to be dispensable for precise adaptation ( Alon et al . , 1999 ) . Through theoretical analysis , it has been conjectured that this secondary feedback loop might play a role in attenuating effects of gene-expression noise ( Kollmann et al . , 2005 ) , but experimental verification has been lacking . We therefore sought to investigate the influence of perturbations to this network topology on the variability of chemotactic signaling activity . CheB consists of two domains connected by a flexible linker ( Figure 3a ) . A regulatory domain , with structural similarity to CheY , can be phosphorylated at residue Asp56 ( Djordjevic et al . , 1998; Stewart et al . , 1990 ) . A catalytic domain mediates binding to specific residues on chemoreceptor cytoplasmic domains and removes a methyl group added by the counterbalancing activity of CheR . Phosphorylation induces a conformational change and activates CheB ( CheB* ) ( Djordjevic et al . , 1998; Lupas and Stock , 1989 ) . Several mutants of CheB lack phosphorylation feedback while retaining catalytic activity . Here , we focus on two specific mutants: CheBD56E , which bears a point mutation at the phosphorylation site , and CheBc , which expresses only the catalytic domain of CheB ( Stewart et al . , 1990; Alon et al . , 1999 ) . Cells expressing these mutants have an altered network topology ( Figure 3b ) which lacks CheB phosphorylation feedback . To study the influence of network topology on cell-to-cell variability , we expressed different forms of CheB ( CheBWT , CheBD56E , CheBc ) from an inducible promoter in a ΔcheB strain and measured the response to a saturating amount of attractant ( 500 μM MeAsp ) . The expression levels of each mutant are tuned such that they approximate the wildtype steady state activity level . The response of CheBWT was qualitatively very similar to cells in which CheB is expressed from its native chromosomal position ( compare Figure 3—figure supplement 1a and Figure 1a ) despite the fact that plasmid expression breaks the translational coupling with CheR ( Løvdok et al . , 2009 ) . By contrast , cells expressing either of the two CheB mutants defective in phosphorylation demonstrated increased cell-to-cell variability in the steady-state activity compared to cells expressing CheBWT . The increased variability of the CheB phosphorylation-deficient mutants ( CheBD56E and CheBc ) was manifested not only in a higher coefficient of variation in a0 ( 1 . 07 and 1 . 10 , respectively , and WT 0 . 7 ) , but also a qualitatively different shape of the distribution of a0 across the population ( Figure 3c ) . Whereas the distribution demonstrated a single peak in CheBWT cells with phosphorylation feedback , the distribution for the phosphorylation-feedback mutants demonstrated a bimodal shape with peaks close to the extreme values a0={0 , 1} . We tested whether these strong differences in cell-to-cell variability might be the result of gene expression noise , by comparing expression-level distributions of the CheB mutants . We constructed fluorescent fusions of each cheB allele to the yellow fluorescent protein mVenus and quantified the distribution of single-cell fluorescence levels under the same induction conditions as in the FRET experiments ( Figure 3—figure supplement 1 ) . The ratio between the measured expression-levels ( CheBc:WT:D56E≈0 . 7:1:2 . 5 ) was compatible with expectations from the hierarchy of reported in vitro catalytic rates of CheB ( kbD56E<kbW⁢T<kbc ) ( Anand and Stock , 2002; Simms et al . , 1985; Stewart , 1993 ) , and expression-level variability was very similar between the three strains ( CV’s of 0 . 87 , 0 . 90 and 0 . 82; we note that these rather high CV values likely include contributions from plasmid copy number variability ) . These findings suggest that the differences in cell-to-cell variability observed in FRET are not due to differences between the expression-level distributions of the three cheB alleles , but rather to the differences they impose on the signaling network topology . What feature of the signaling network could generate such broad ( and even bimodal ) distributions of a0 ? A general paradigm for models of adaptation that exhibit precise adaptation is activity-dependent ( integral ) feedback ( Barkai and Leibler , 1997; Yi et al . , 2000 ) , which in bacterial chemotaxis can be implemented by the activity of the feedback enzymes CheR and CheB being dependent of the conformational state ( i . e . activity ) of their substrate chemoreceptors . This results in a steady-state activity a0 that only depends on the [R]/[B] expression-level ratio and not on their absolute abundance . We can view this mapping as a transfer function ƒ between the ratio [R]/[B] and the steady-state activity , a0=f⁢ ( [R]/[B] ) Depending on the function ƒ , the input variance PRB ( [R]/[B] ) may lead to high or low variance in the distribution P ( a0 ) . This is because the manner in which the transfer function ƒ filters the [R]/[B] distribution , P⁢ ( a0 ) =PR⁢B⁢ ( f-1⁢ ( a0 ) ) \|f′⁢ ( f-1⁢ ( a0 ) ) \| . Hence a steep function ƒ can impose bimodality in the methylation level , and thereby also in the activity of steady-state CheA activity , a0 , even at quite modest input variances for distributions of the ratio [R]/[B] . Thus , even if expression-level noise for both CheR and CheB are modest , a sensitive transfer function ƒ can effectively amplify the variation in [R]/[B] , and if the distribution of the latter ratio , PRB ( [R]/[B] ) extends below and above the narrow region over which ƒ is steep , the decreased slope of ƒ ( i . e . lower ƒ' ( [R]/[B] ) in those flanking regions will tend to increase the weight on both sides of the broad P ( a0 ) distribution to produce a bimodal profile . On the other hand , if the network topology effectively reduces the steepness of ƒ , the resulting P ( a0 ) will have a reduced variance for the same input PRB ( [R]/[B] ) ( Figure 3d ) . Our results suggest that ƒ is much steeper in the absence of phosphorylation feedback than in its presence . We find that models with linear or supra-linear dependence of the methylation rate on activity generate a function ƒ that is very shallow ( Figure 3—figure supplement 3 ) , making them unsuitable for explaining the observed bimodal behavior . However , if we assume CheR and CheB follow Michaelis-Menten kinetics in which the dependence of the methylation rates on receptor activity is sub-linear , the dependence of ƒ on [R]/[B] can become very steep . It has been conjectured ( Barkai and Leibler , 1997; Emonet and Cluzel , 2008 ) that in vivo the enzymes CheR and CheB operate at or near saturation , an idea supported by population-level FRET measurements of adaptation kinetics ( Shimizu et al . , 2010 ) . An important consequence of enzyme saturation in such reversible modification cycles is that the steady-state activity of the substrate can become highly sensitive to the expression level ratio of the two enzymes , a phenomenon known as zero-order ultrasensitivity ( ( Goldbeter and Koshland , 1981 ) ; see Materials and methods ) . Within the chemotaxis system , saturation of both CheR and CheB can thus render the receptor modification level , and in turn , the CheA activity a0 , ultrasensitive to the [R]/[B] concentration ratio ( Emonet and Cluzel , 2008 ) . Could the known biochemical differences between the three forms of CheB ( CheBWT , CheBD56E , CheBc ) explain the contrasting patterns of a0 variability observed in our single-cell FRET experiments ? In the absence of any feedback , the steepness of ƒ' ( [R]/[B] ) is solely determined by the low Michaelis-Menten constants KB , R , which corresponds to saturated kinetics of the enzymatic activity of CheRB and hence ultransensitivity of the steady-state substrate activity . The expression ratio of CheR/CheB which determines the crossover point ( a0=0 . 5 ) is set by the ratio of catalytic rates of CheR and CheB ( kr , b ) . Hence the phosphorylation deficient mutants CheBD56E and CheBc both have steep curves but are shifted along the R/B axis due to very different catalytic rates . However , in the case of phosphorylation feedback , CheBWT , the same enzyme can be in two states , each with equal Kr , b but one low and one high kr . Whether CheB is in the one state or the other is determined by the activity-dependent phosphorylation feedback . As a result , the curve of CheBWT is activity dependent ( ƒ ( a , [R]/[B] ) ) and changes with activity by shifting between the two curves corresponding to the extremes of all phosphorylated or all unphosphorylated . Effectively , this makes the resulting curve ƒ less steep ( Emonet and Cluzel , 2008 ) . The mean of the distributions PRB are tuned such to get the same mean activity level ( ⟨a0⟩ ) , but the same variance in PRB leads to very wide P ( a0 ) distributions in absence of phosphorylation , while phosphorylation feedback ensures a much smaller , single-peaked distribution . It has also been conjectured that the CheB phosphorylation feedback is responsible for the highly nonlinear kinetics of recovery from repellent ( or attractant removal ) responses ( Shimizu et al . , 2010; Clausznitzer et al . , 2010 ) . Indeed , in cells expressing CheBc , the kinetics of recovery from the response to removal of 500 μμM MeAsp after adaptation appeared qualitatively different from that in cells expressing wildtype CheB , lacking the characteristic rapid recovery and instead appearing more symmetric with the CheR-mediated recovery upon addition of a saturating dose of attractant ( Figure 3—figure supplement 4 ) . By contrast , CheBD56E was found to still possesses a fast component , despite being defective in phosphorylation , albeit also with somewhat slower kinetics than wt . In summary , the clearest difference between wildtype and phosphorylation-defective CheB mutants is found in the variability of the steady-state signal output ( i . e . kinase activity ) . The bimodal distribution in kinase activity we observed in the phosphorylation-deficient CheB mutants implies that a large fraction of cells have a CheY-P concentration far below or far above the motor’s response threhold and hence will impair chemotactic responses to environmental gradients . Consistent with this idea , in motility-plate experiments ( Supplementary Figure 3—figure supplement 5 ) we found that chemotactic migration on soft-agar plates was severely compromised for both CheBD56E and CheBc compared to CheBWT , indicating that the phosphorylation feedback is important for efficient collective motility . The slow kinetics of the adaptation enzymes CheR and CheB have been hypothesized to play a role not only in determining the steady-state kinase activity a0 , but also in generating temporal fluctuations of the intracellular signal ( Korobkova et al . , 2004; Emonet and Cluzel , 2008; Park et al . , 2010; Celani and Vergassola , 2012 ) . We found substantial differences between wildtype ( CheRB+ ) and adaptation-deficient ( CheRB- ) cells in the variability of their FRET signals across time ( Figure 4 ) . The effect is clearly visible upon comparing long ( ∼1 hr ) FRET time series obtained from cells of these two genotypes ( Figure 4a ) . The FRET signal in wildtype cells demonstrated transient excursions from the mean level that were far greater in amplitude than those in CheRB- cells . To distinguish between variability across cells in a population ( which we discuss in terms of coefficients of variation , CV ) and that over time within a single cell , we denote the temporal noise amplitude as η≡σa/a0 . This amplitude was quantified by computing the variance of each single-cell time series , low-pass filtered with a moving average filter of 10 s , and shows that the fluctuation amplitudes are much larger in wildtype cells compared to adaptation-deficient cells ( ⟨η⟩ = 0 . 44 and 0 . 09 respectively , Figure 4b ) . Importantly , these experiments were carried out under conditions in which no protein synthesis can occur due to auxotrophic limiation ( see Materials and methods ) , thus ruling out gene-expression processes as the source of these fluctuations . Power spectral density ( PSD ) estimates computed from such time series confirm a nearly flat noise spectrum for CheRB- cells , whereas CheRB+ cells demonstrated elevated noise at low frequencies ( Figure 4c ) . The amplitude of these low-frequency noise components do clearly vary from cell to cell , as can be gleaned in the diversity of single-cell power spectra . To quantify this protein-level noise due to CheR/CheB activity , we describe the fluctuating signal as an Ornstein-Uhlenbeck ( O-U ) process of the single variable a , with relaxation timescale τ and diffusion constant c , which can be interpreted as a linear-noise approximation ( Van Kampen , 1981; Elf and Ehrenberg , 2004 ) to the multivariate stochastic kinetics of the underlying chemical network controlling the mean kinase activity a ( Tu and Grinstein , 2005; Emonet and Cluzel , 2008 ) : ( 1 ) d⁢ad⁢t=-1τm⁢a⁢ ( t ) +c⁢Γ⁢ ( t ) where Γ⁢ ( t ) is a Gaussian white noise process . The parameters τmτm and c for each cell are readily extracted via the power-spectrum solution of the O-U process: ( 2 ) Sa⁢ ( ω ) =2⁢c⁢τ21+ ( 2⁢π⁢ω⁢τm ) 2+Ewhere we have added to the standard Lorentzian solution ( Gillespie , 1996 ) a white-noise term E that may vary from cell to cell to account for experimental shot noise in the photon-limited FRET signal . Single-cell PSD data were well fit by Equation 2 ( Figure 4d ) , and the average of extracted single-cell fluctuation timescales ( ⟨τm⟩=12 . 6s ) ( Figure 4e ) are in good agreement with previously reported correlation times of flagellar motor switching ( Park et al . , 2010; Korobkova et al . , 2004 ) , as well as the kinetics of CheRB-mediated changes in receptor modification from in vivo measurements using radioactively labeled methyl groups ( Lupas and Stock , 1989; Terwilliger et al . , 1986 ) . The variance of the fluctuations obtained from the fits of the PSD , σa=c⁢τm/2 yielded very similar noise amplitudes ηO⁢U≡σa , O⁢U/a0 as calculated from the time series ( ⟨ηO⁢U⟩=0 . 42 , Figure 4—figure supplement 3 ) . We note that these noise levels are larger than expected - in a considerable fraction of cells , the standard deviation of fluctuations is comparable to the mean level of activity , and the steady-state fluctuations span the full range of kinase activity ( see e . g . that represented by the red curve in Figure 4a ) . Previous studies had predicted a value of ~10-20% , based either on reported fluctuation amplitudes of motor switching ( Korobkova et al . , 2004; Tu and Grinstein , 2005 ) or biochemical parameters of the intracellular signaling network ( Emonet and Cluzel , 2008; Shimizu et al . , 2010 ) . The noise amplitudes are also highly variable ( CV=0 . 55 , ση=0 . 24 ) from cell to cell . In summary , we confirmed the presence of strong temporal fluctuations in single-cell chemotaxis signaling attributable to the stochastic kinetics of the adaptation enzymes CheR/CheB , and further found that the amplitude of these fluctuations vary considerably across cells in an isogenic population . The fluctuation amplitude η in CheRB+ cells ( Figure 4b ) is much greater than previous estimates from pathway-based models that considered sublinear kinetics in the enzymatic activities of CheR and CheB ( Emonet and Cluzel , 2008 ) and receptor cooperativity ( Shimizu et al . , 2010 ) as possible mechanisms that amplify noise originating in the stochastic kinetics of receptor methylation/demethylation . A possible explanation for this discrepency is the presence of one or more additional noise source ( s ) independent of methylation/demethylation dynamics . Although we found that the noise amplitude η was much lower than wildtype in unstimulated CheRB- cells ( Figure 4 ) , it is possible that the strong activity bias of these cells in the absence of chemoeffectors ( a0≈1 ) masks noise contributions that would be observable if receptors were tuned to the more responsive regime of intermediate activity ( e . g . as in wt cells , where a0≈1/3 ) . We reasoned that in CheRB- cells , tuning the activity to an intermediate level by adding and sustaining a sub-saturating dose of attractant could reveal additional noise sources . Hence we measured the temporal variability of CheRB- cells during prolonged stimulation with 50 μM L-serine , which elicits a half-maximal population-level response ( Figure 5b ) . Although no large fluctuations were be observed in the population-averaged time series ( Figure 5b ) , averaging the power spectra computed from all single-cell time series revealed a somewhat elevated noise level at low frequencies , compared to the case without ligand ( Figure 5a ) , indicating the possibility of a noise source independent of receptor methylation . To further test whether and how these methylation-independent fluctuations are affected by the composition of the chemoreceptor arrays , we also measured the response of CheRB- cells expressing Tsr as the sole chemoreceptor during a sustained stimulus of magnitude close to the population-level K ( Figure 5c ) . Surprisingly , the averaged single-cell power spectra ( Figure 5a ) indicated the presence of very large fluctuations , even surpassing the fluctuation magnitude in CheRB+ cells . The time series of single-cell responses demonstrated strong deviations from the population average ( Figure 5d and - Video Supplement ) . Whereas all cells responded identically to the saturating dose of attractant , the behavior during the sub-saturating step was highly diverse . Some cells ( 11/141 ) showed no apparent response in kinase activity , whereas in others ( 32/141 ) complete inhibition was observed ( Figure 5d , yellow curves ) . The majority of cells ( 98/141 ) , however , had an intermediate level of activity when averaged over time , but demonstrated strong temporal fluctuations , often with magnitudes exceeding those observed in wildtype cells . We further noted that within this subset of cells with large temporal fluctuations , a large fraction ( 54/98 ) demonstrated fluctuations that resemble rapid step-like transitions between discrete levels of relatively stable activity that could be identified as peaks in the distribution of activity values across time ( Figure 5d , marginal histograms ) . Among these ‘stepper’ cells , the majority ( 37/54 ) appeared to transition between three or more discrete activity levels ( Figure 5d , brown curve ) , whereas the remaining sizable minority of steppers ( 17/54 ) demonstrated binary switching between two discrete levels corresponding to the maximum ( a≈1 ) and minimum ( a≈0 ) receptor-kinase activity states ( Figure 5d , red curve ) . The remaining fraction of cells ( 44/98 ) demonstrated fluctuations that were also often large but in which discrete levels could not be unambiguously assigned ( Figure 5d , black curve ) . The numbers of cells corresponding to each of the categories described above are summarized in Figure 5e . The observation of cells that demonstrate spontaneous two-level switching is particularly surprising , given the large number of molecules involved in receptor-kinase signaling . The expression level of each protein component of the chemoreceptor-CheW-CheA signaling complex in our background strain ( RP437 ) and growth medium ( TB ) has been estimated ( by quantitative Western Blots ) to be of order 104 copies/cell ( Li and Hazelbauer , 2004 ) . Considering that the core unit of signaling has a stoichiometric composition of receptor:W:A = 12:2:2 ( monomers ) ( Li and Hazelbauer , 2011 ) , the number of core units is likely limited by the number of receptors , leading to an estimate 104/12~103 core units for a typical wildtype cell . This estimate does not apply directly to the experiments of Figure 5 because receptors are expressed from a plasmid in a strain deleted for all receptors . But the FRET response amplitudes of these cells were similar to those of cells with a wildtype complement of receptors , and we thus expect the number of active core units per cell in the experiments of Figure 5 to be similar to or greater than that in wildtype cells . We analyzed further the temporal statistics of the discrete transitions in the subset of cells exhibiting two-level switching ( Figure 5g–h ) . We first quantified the duration of such transitions by fitting segments of the activity time series over which these switches occured ( Figure 5d ) by a symmetrized exponential decay function ( see Materials and methods ) to obtain switch durations τ+ and τ- for upward and downward transitions , respectively . The fitted values for τ+ and τ- correspond to the duration over which the activity trajectory traverses a fraction 1-e-1 of the transition’s full extent , and were found to be similar between switches in both directions: ⟨τ+⟩±στ+= 4 . 2 ± 2 . 2 s and ⟨τ-⟩±στ-= 3 . 5 ± 3 . 2 s ( Figure 5e ) . We note that these transition times are significantly greater than , but close to , the data acquisition interval ( 1 s ) , and so the shape of the fitted function should be considered a first approximation to the true rise and decay dynamics . We then considered the duration of time between switching events . We defined Δtup , k and Δtdown , k as the duration of the k-th time interval between transitions with high- and low-activities , respectively ( Figure 5d ) , and computed the average over all k of Δtup/down , k for each individual cell to estimate its residence timescales τup/down for states of high/low activity , respectively . From each cell’s set of intervals {Δtup/down , k} we also computed a parameter a1/2 , defined as the fraction of time the cell spent in the high activity level , as a measure of its time-averaged activity during the sub-saturating ( 20μM ) L-serine stimulus that yielded a population-averaged response ⟨a⟩≈1/2 ( see Materials and methods ) . We found that the logarithms of the mean residence times τup and τdown scale approximately linearly with ln⁡[a1/2/ ( 1-a1/2 ) ] ( Figure 5f ) . The latter can be considered a free-energy difference ( −ΔG ) =Gdown−Gup between the inactive and active states of an equilibrium two-state switching process in which the time-averaged activity a1/2 is given by the probability of being in the active state , a1/2=p⁢ ( active ) = ( 1+eΔ⁢G ) -1 . The residence time in each state can then be described by an Arrhenius-type relation with characteristic time for barrier crossing τr and the height of the energy barrier dependent on Δ⁢G , ( 3 ) τdown=τrexp⁡[−γdownΔG/kBT]τup=τrexp⁡[−γupΔG/kBT]where the ( dimensionless ) constants γdown and γup describe how the barrier heights of the down and up states , respectively , depend on the free-energy difference Δ⁢G=kB⁢T⁢ln⁡[ ( 1-a1/2 ) /a1/2] . We find γdown=−0 . 4±0 . 1 , γup=0 . 6±0 . 1 , and the characteristic timescale τr , defined here as equivalent to τup=τdown when Δ⁢G=0 ( and hence a1/2=0 . 5 ) , was found to be 110 ± 10 s . The fact that the mean residence times ( τup , τdown ) scale exponentially with the apparent free energy difference ( Δ⁢G ) indicates that receptor-kinase switching can , to a first approximation , be treated as a barrier-crossing process . In summary , these data demonstrate the existence of a signaling noise source that is independent of the adaptation enzymes CheR/CheB . The fluctuations they generate can be very strong in cells expressing Tsr as the sole chemoreceptor , leading to two-level switching in a subset of cells . The latter observation suggests that cooperativity among signaling units in homogeneous chemoreceptor arrays can reach extremely high values , with up to ∼103 units switching in a cooperative fashion . The temporal statistics of these two-level switches are consistent with a barrier-crossing model in which the residence time of both states depend on the activity bias ln⁡[a1/2/ ( 1-a1/2 ) ] in a nearly symmetric manner with opposing signs . A key feature of bacterial chemotaxis as an experimental system is that one can study in vivo signaling and behavior in a manner that is decoupled from gene expression and growth . Being an entirely protein-based signaling network , chemotaxis signaling responses do not require changes in gene expression , and the relatively short timescales of signaling reactions ( subsecond to minutes ) are well separated from those of changes in protein counts due to gene expression noise ( minutes to hours ) . The ensemble of single-cell FRET time series measured in each of our experiments thus provide a snapshot of cell-to-cell variability due to stochastic gene expression in a variety of signaling parameters . Our data revealed high variability in important signaling parameters connected to the adaptation system ( Figure 1 ) . In the case of the variability in recovery times ( CV=0 . 20 ) , this is likely due to variability in the CheR/receptor ratio from cell to cell . What consequences might such variability have on chemotactic behavior ? A recent theoretical study has established that long ( short ) adaptation times are better suited for maximizing chemotactic migration rates in shallow ( steep ) gradients ( Frankel et al . , 2014 ) . Thus , variability in adaptation times could partition the population into cells that will be more efficient in running up steep gradients , and others better suited to climbing shallow ones . Interestingly , it was also found that optimal performance at each gradient involves tuning not only the adaptation time , but also other parameters such as swimming speed or tumble bias , leading to the prediction that selective pressures act not only on the distribution of individual parameters , but also on the pattern of covariation among them ( Frankel et al . , 2014; Waite et al . , 2016 ) . Exploring such correlated variation of signaling parameters , both under changes in environmental conditions such as nutrient levels ( Khursigara et al . , 2011 ) and within identically grown populations , would be a fruitful avenue for future single-cell FRET studies . In the ligand response of the network , we observed large cell-to-cell variability in the sensitivity ( 1/K ) and steepness ( H ) of dose-response relations , for cells with a wildtype receptor population ( Figure 2 ) . Using a mixed-species MWC model ( Mello and Tu , 2005 ) , we were able to estimate the Tar/Tsr ratio in single cells , which spans a broad range from nearly zero to more than two . This strong variability in the receptor-cluster composition has the potential to dramatically impact behavior . In their natural habitats , cells likely experience a variety chemoeffector gradients simultaneously , each associated with an unknown fitness payoff for chemotactic pursuit . Generating diversity in the chemoreceptor ratio , which has been shown to determine which gradient to climb when challenged with such conflicting possibilities ( Kalinin et al . , 2010 ) , could allow the isogenic population to hedge its bets to maximize net fitness gains . The Tar/Tsr ratio has also been shown to play an important role in setting the preferred temperature for thermotaxis ( Salman and Libchaber , 2007; Yoney and Salman , 2015; Paulick et al . , 2017 ) . Variability in Tar/Tsr would allow diversification of the preferred temperature across cells in the population , which will promote spreading of bacteria in environments with temperature gradients . Finally , when chemotactic bacteria colonize an initially nutrient-rich environment , they are known to successively exploit resources by emitting multiple traveling waves of chemotactic bacteria , each of which consumes and chases by chemotaxis a different nutrient component outward from the colony origin ( Adler , 1966 ) . Our observation that the population diversity in receptor ratios , and hence chemotactic preference , varies concomitantly with population growth could provide a means to tune the population fractions that engage in such excursions into virgin territory , and those that remain for subsequent exploitation of remaining resources . Thus , the diversity in ligand response and preference generated by variability in the Tar/Tsr ratio could have nontrivial consequences in a variety of behavioral contexts encountered by isogenic chemotactic ( and thermotactic ) populations . The role of phosphorylation feedback has been a long standing open question in the field of bacterial chemotaxis signaling , ever since its presumed role in providing precise adaptation was decisively ruled out by ( Alon et al . , 1999 ) . In the ensuing years , a diverse set of hypotheses have been proposed to explain its purpose . Apart from precise adaptation , CheB phosphorylation has been suggested as possibly responsible for the non-linear response of CheB activity to changes in CheA kinase activity ( Shimizu et al . , 2010; Clausznitzer et al . , 2010 ) , ligand sensitivity of wildtype cells ( Barkai et al . , 2001 ) , and has been implicated theoretically as a possible mechanism to buffer gene-expression noise to suppress detrimental variability in the steady-state kinase activity ( Kollmann et al . , 2005; Emonet and Cluzel , 2008; Pontius et al . , 2013 ) . Here , we tested the latter hypothesis , by severing the phosphorylation feedback loop as a possible noise-reduction mechanism . Our single-cell FRET data revealed that , not only does CheB phosphorylation feedback strongly attenuate the magnitude of variability in the steady-state kinase activity a0 , it also qualitatively changes the shape of the distribution P⁢ ( a0 ) across cells to convert an otherwise bimodal distribution into a unimodal one ( Figure 3d ) . The highly polarized bimodal distribution of steady-state activities in CheB phosphorylation mutants are likely detrimental , as they could drive a0 of a large fraction of the population too far from the flagellar motor’s steep response threshold ( Cluzel et al . , 2000; Yuan and Berg , 2013 ) to effectively control swimming . By analyzing simplified models of adaptation kinetics , we found that a bimodal P⁢ ( a0 ) could occur in the absence of phosphorylation feedback if the enzyme kinetics of CheR and CheB depend sublinearly on the activity a of their receptor substrates . As a limiting case , when both enzymes work at or near saturation , this model leads to zero-order ultrasensitivity ( Goldbeter and Koshland , 1981; Emonet and Cluzel , 2008 ) , which could act as a strongly non-linear transfer function f ( [R]/[B] ) that converts a unimodal distribution P ( [R]/[B] ) into a bimodal P⁢ ( a0 ) . We note that ultrasensitivity due to sublinear ( Michaelis-Menten ) enzyme kinetics is by no means the only possible explanation for the observed bimodality in P⁢ ( a0 ) . Any mechanism that renders f ( [R]/[B] ) a strongly nonlinear ( sigmoidal ) function could lead to the same effect . The merit of the sublinear kinetic ( ultrasensitivity ) model is in its simplicity , but it is worth noting that reality is likely to be more complex due to , for example , effects of spatial organization . It is known that both CheR and CheB interact with chemoreceptors not only at their substrate modification residues , but also with a second binding site on a flexible tether at the receptor C-terminus . Such bivalent interactions with the receptor array could affect the movement of these enzymes across the receptor lattice ( Levin et al . , 2002 ) , and such movements could shift the balance between processivity and distributivity of enzyme activity on their substrate receptors ( Pontius et al . , 2013 ) , which could in turn attenuate or enhance the nonlinearity in the relationship f⁢ ( [R]/[B] ) between the enzyme expression ratio [R]/[B] and the steady-state activity a0 of their substrate receptors ( Takahashi et al . , 2010 ) . In addition to cell-to-cell variability in signaling parameters , single-cell FRET allowed us to resolve temporal fluctuations in signaling about the steady-state output within individual cells . In wildtype cells , we found that the steady-state activity fluctuates slowly ( Figure 4 , correlation time τ≈ 10s ) with a large amplitude ( η=σa/⟨a⟩≈ 40% ) , but this amplitude also varies significantly from cell to cell ( CV ≈ 0 . 6 ) . Fluctuations on this timescale were absent in CheRB- cells defective in receptor methylation/demethylation , indicating that these fluctuations are generated by stochastic processes in the activity of the adaptation enzymes CheR and CheB . Whereas the fluctuation correlation time τ in our FRET experiments was in close agreement with those from previously reported flagellar motor switching experiments ( Korobkova et al . , 2004; Park et al . , 2010 ) , the fluctuation amplitude ⟨η⟩≈40% was surprisingly large . Theoretical analysis of the motor-based noise measurements indicated that , in the frequency range of our experiments , stochastic methylation kinetics are indeed the dominant source of noise ( Clausznitzer and Endres , 2011 ) . Another theoretical study of the motor noise ( Tu and Grinstein , 2005 ) , had predicted a modest noise level of intracellular noise , with a lower bound of 20% of the mean . The discrepancy is likely due , at least in part , to the recently discovered adaptation at the level of the flagellar motor ( Yuan et al . , 2012 ) , which must effectively act as a high-pass filter that attenuates frequencies near or below a cutoff frequency determined by its own characteristic timescale for adaptation . The fluctuation amplitude η was also much greater than previous estimates from pathway-based models and we have shown that there is an additional noise source , independent from methylation , which contributes to the total noise amplitude in wildtype cells and not considered in previous modeling efforts . The large temporal noise we observed in wildtype ( CheRB+ ) cells may seem counterintuitive , given that the chemotaxis pathway is a transduction path for sensory information , and noise generally reduces information transmission capacity of communication channels ( Shannon , 1949 ) . However , the chemotaxis signaling pathway is not only a sensory system but also a control circuit for motile behavior , and recent studies have highlighted the importance of considering the behavioral context in understanding the design of this signaling pathway ( Dufour et al . , 2016; Wong-Ng et al . , 2016; Long et al . , 2017 ) . The temporal noise we observed could have profound implications for E . coli’s random-walk motility strategy , because slow fluctuations in the intracellular signal can enhance the likelihood of long run events and stretch the tail of the run-length distribution to yield power-law-like switching-time distributions over a range of time scales ( Korobkova et al . , 2004; Tu and Grinstein , 2005 ) . Such non-exponential statistics are known to yield superior foraging performance in environments where resource distribution is sparse ( Viswanathan et al . , 1999 ) , and temporal fluctuations in run-tumble behavior has also been shown theoretically to enhance climbing of shallow gradients by generating runs that are long enough to integrate over the faint gradient a detectable difference in ligand input ( Flores et al . , 2012; Sneddon et al . , 2012 ) . Hence , the noise generated by the adaptation system can be advantageous in resource-poor environments ( deserts ) in which efficient exploration of space for sparsely distributed sources ( oases ) is of utmost importance . By contrast , strong temporal noise clearly degrades response fidelity in rich environments where the gradient signal is strong enough for detection with short runs , and might also complicate coordination of cells in collective behaviors such as the aforementioned traveling-wave exploitation of nutrients . Our finding that the noise amplitude varies strongly from cell to cell thus suggests that isogenic populations might be hedging their bets by partitioning themselves between specialists for local exploitation of identified resource patches and those for long-range exploration in search for new ones . We found the most dramatic temporal fluctuations in adaptation-deficient ( CheRB- ) cells expressing Tsr as the sole chemoreceptor species ( Figure 5 ) . When brought close to their dose-response transition point ( K ) by attractant stimulation , these cells demonstrated strong temporal fluctuations , revealing that there exist sources of signal fluctuations that are independent of CheR and CheB activity . The origin of these adaptation-independent fluctuations remain unknown , but in broad terms , one can envisage that they are due to either intrinsic sources ( i . e . fluctuations arising within the components of the receptor-kinase complex ) , extrinsic sources ( i . e . fluctuations in other cellular processes/environmental variables ) , or both . Possible intrinsic sources include coupled fluctuations in protein conformations ( Duke and Bray , 1999; Shimizu et al . , 2003; Mello et al . , 2004; Skoge et al . , 2011 ) , the slow-timescale changes in receptor ‘packing’ that have been observed by fluorescence anisotropy measurements ( Frank and Vaknin , 2013; Vaknin , 2014 ) , and the stochastic assembly dynamics of receptor clusters ( Greenfield et al . , 2009 ) . Possible extrinsic sources include fluctuations in metabolism , membrane potential , or active transport/consumption of ligand . Many of these possibilities could be tested by experiments of the type presented here with appropriate mutant strains and environmental controls , and present promising directions for future research . The adaptation-independent fluctuations we observed were not only large in amplitude but often ( though not always ) took the form of discrete steps in activity , in some cases between only two levels . Two-state descriptions of receptor signaling are a common feature of nearly all mechanistic models of bacterial chemotaxis signaling addressing both cooperativity ( Duke and Bray , 1999; Shimizu et al . , 2003; Mello et al . , 2004; Mello and Tu , 2005; Keymer et al . , 2006 ) and adaptation ( Asakura and Honda , 1984; Barkai and Leibler , 1997; Morton-Firth et al . , 1999; Endres and Wingreen , 2006; Emonet and Cluzel , 2008; Tu et al . , 2008 ) , yet direct evidence for two-state switching by receptor-kinase complexes has been lacking . Although as noted above , it is yet possible that the two-level switching we observed ( Figure 5d ) is due to extrinsic noise sources ( e . g . metabolism or transport ) , the temporal statistics ( Figure 5e–h ) are compatible with a simple model in which two stable signaling states are separated by an energy barrier sensitive to both environmental stimuli and internal cell variables . Regarding cells that exhibited step-like transitions among more than two stable states , a plausible interpretation is that the underlying transitions are actually two-level , but the majority of the receptor-kinase population is partitioned into two or more disjoint signaling arrays which fluctuate independently . While two-state switching has been observed in small oligomers such as ion channels ( Keller et al . , 1986 ) and larger protein assemblies such as the bacterial flagellar motor ( Silverman and Simon , 1974; Bai et al . , 2010 ) , controlled by up to a few dozen units , our findings suggest ( as discussed in results ) that at least many hundreds , if not thousands of receptor-kinase units can switch in a concerted fashion . The rather long timescale associated with intervals between switches ( ≈102 s ) is clearly distinct from the methylation-dependent fluctuation timescale ( ≈101 s ) observed in CheRB+ cells , and might reflect the large size of the cooperatively switching signaling array . The switching duration ( ≈4 s ) , is also much slower than the sub-second response to attractant stimuli ( Segall et al . , 1982; Sourjik and Berg , 2002b ) . These fluctuations of surprisingly large magnitude indicate the possibility that cooperativity between arrayed chemoreceptors are much stronger than suggested by previous population-averaged measurements , and represent a promising direction for future investigations . We described a new single-cell FRET technique capable of resolving intracellular signaling dynamics in live bacteria over extended times . Our results highlight how a protein-based signaling network can either generate or attenuate variability , by amplifying or filtering molecular noise of different molecular origins . Gene expression noise is harnessed , on the one hand , to generate diversity in the ligand response of isogenic populations , or attenuated , on the other the hand , in the control of steady-state signal output . In addition , we showed that signaling noise generated at the level of interacting gene products can have a profound impact . Stochastic protein-protein interactions within the signaling network , as well as other ‘extrinsic’ fluctuations , can be amplified by the signaling network to generate strong temporal fluctuations in the network activity . All strains used are descendants of E . coli K-12 HCB33 ( RP437 ) . Growth conditions were kept uniform by transforming all strains with two plasmids . All strains and plasmids are shown in Tables 2 and 3 . The FRET acceptor-donor pair ( CheY-mRFP and CheZ-YFP ) is expressed in tandem from a IPTG inducible pTrc99A plasmid , pSJAB12 or pSJAB106 , with respective induction levels of 100 and 50 μM IPTG . The differences between pSJAB12 and pSJAB106 are ( i ) the presence of a noncoding spacer in pSJAB106 to modify the ribosome binding site of CheZ ( Salis et al . , 2009 ) , such that CheZ is expressed approximately three fold less , and ( ii ) a A206K mutation in YFP to enforce monomerity . We also used pVS52 ( CheZ-YFP ) and pVS149 ( CheY-mRFP1 ) to express the fusions from separate plasmids with induction levels of 50 μM IPTG and 0 . 01 % arabinose , respectively . We transformed the FRET plasmids in an adaptation-proficient strain ( VS104 ) to yield CheRB+ and an adaptation-deficient strain ( VS149 ) to get CheRB- . For attachment with sticky flagella from pZR1 we used the equivalent strains in fliC background ( VS115 and TSS58 ) . Experiments with Tsr as the sole chemoreceptor were performed in UU2567 or TSS1964 , in which the native FliC gene is changed to sticky FliC ( FliC* ) . Tsr is expressed from pPA114 Tsr , a pKG116 derivative , at with an induction of 0 . 6 μμM NaSal . For the experiments with the CheB mutants , pSJAB12 was transformed into VS124 together with plasmids expressing CheBW⁢T , CheBD⁢56⁢E and truncated mutant CheBc ( plasmids pVS91 , pVS97 and pVS112 , respectively , with induction levels of 1 . 5E-4 , 6E-4 and 3E-4 % arabinose . Föster Resonance Energy Transfer [FRET] microscopy was performed as previously reported ( Sourjik et al . , 2007; Vaknin and Berg , 2004 ) . Cells were grown to OD = 0 . 45–0 . 5 in Tryptone Broth ( TB ) medium from a saturated overnight culture in TB , both with 100 μg/mL ampicillin and 34 μg/mL chloramphenicol and appropriate inducers in the day culture . For the FRET experiments we used Motility Media ( MotM , ( Shimizu et al . , 2006 ) ) , in which cells do not grow and protein expression is absent . Cells were washed in 50 mL MotM , and then stored 0 . 5–6 hr before experiment . In the dose-response curve experiments and the temporal fluctuation measurements , cells were stored up to three hours at room temperature to allow for further red fluorescence maturation . A biological replicate or independent FRET experiment was defined as a measurement from separately grown cultures , each grown on a separate day . Cells were attached by expressing sticky FliC ( FliC* ) from a pKG116 plasmid or the chromosome ( TSS1964 ) , induced with 2μμM Sodium Salicylate ( NaSal ) , or with Poly-L-Lysine ( Sigma ) , or with anti-FliC antibodies column purified ( Using Protein A sepharose beads , Amersham Biosciences ) from rabbit blood serum and pre-absorbed to FliC- cells ( HCB137 , gifts from Howard Berg ) . We found FRET experiments with sticky FliC to have the highest signal-to-noise ratio . Fluorescent images of the cells were obtained with a magnification of 40-100x ( Nikon instruments ) . For excitation of YFP , we either used 514 nm laser excitation set to 30 mW for 2 ms or an LED system ( CoolLED , UK ) with an approximate exposure time of 40 ms to approximate the same illumination intensity per frame . The sample was illuminated stroboscopically with a frequency between 1 and 0 . 2 Hz . RPF excitation was performed by 2 ms exposure of 60 mW 568 nm laser or equivalent with LED to measure acceptor levels independently from FRET . Excitation light was sent through a 519 nm dichroic mirror ( Semrock , USA ) . Epifluorescent emission was led into an Optosplit ( Cairn Research , UK ) with a second dichroic mirror 580 nm and two emission filters ( 527/42 nm and 641/75 nm , Semrock , USA ) to project the RFP and YFP emission side by side on an EM-CCD ( Princeton Instruments , USA ) with multiplication gain 100 . Images were loaded and analyzed by means of in-house written scripts ( Image segmentation script FRETimaging . py available online ) in MATLAB and Python . For ratiometric FRET experiments , we segmented single cells using the donor emission with appropriate filter steps to remove clusters of cells or cells improperly attached to the coverslip . At the position of each cell a rectangular ROI is defined in which all fluorescence intensity is integrated . For FRET experiments in which the concentration of donor molecules may influence the FRET signal , the experiments on the CheB mutants , segmentation was done separately for each frame to determine the cell shape and then linking these segmented images with a tracking algorithm ( Crocker and Grier , 1996 ) , afterwards , fluorescence intensities are normalized for the cell size ( mask surface area ) in segmentation , intensities were corrected for inhomogeneous illumination , and cells with low acceptor intensities were excluded from the analysis . The ROI for the donor intensity were subsequently used to obtain the acceptor intensity per cell , both in photon-count per pixel . Fluorescence intensities were corrected for long-time drift ( primarily due to bleaching , but also including contributions from fluorophore maturation and/or recovery from long-lived dark states ) by fitting a linear , single exponential or double exponential function to the fluorescence decay , separately for both donor and acceptor channels . The net decay in the FRET signal was dominated by photobleaching of the donor ( YFP ) intensity ( on average 25% over the course of a 30 min experiment; Figure 1—figure supplement 2 ) . Red fluorescent proteins tend to have long maturation times , which under our experimental conditions ( in which gene expression is halted upon harvesting via auxotrophic limitation ) could result in a residual increase in red fluorescence intensity during experiments . In control experiments , we determined mRFP1’s maturation half time under our conditions to be ~2–3 hr , with a maximum increase in the FRET signal of ~25% at ~5–6 hr . Cells in which the intensity decay could not accurately be corrected were excluded from the analysis . In non-ratiometic fluorescence experiments ( CheB-mVenus ) the fluorescence intensities obtained after segmentation were corrected for inhomogeneous illumination and divided by cell area . The FRET signal is calculated from fluorescent time series . We observe changes in the ratio R=A/D , in which A and D are the fluorescence intensities of the acceptor and donor . In previous population-averaged FRET experiments the FRET per donor molecule ( Δ⁢D/D0 ) is calculated as ( Sourjik and Berg , 2002a; Sourjik et al . , 2007 ) : ( 4 ) Δ⁢DD0=Δ⁢Rα+R0+Δ⁢Rin which R0 is the ratio in absence of FRET , α=\|Δ⁢A/Δ⁢D\| is a constant that depends on the experimental system ( in our case α = 0 . 30 ) and the change in ratio as a result of energy transfer ΔR and R0 are obtained through observing the ratio just after adding and removing saturated attractant stimuli . This expression is convenient for population FRET since is invariant to attachment densities of a population . However , in single-cell FRET this expression may generate additional variability in FRET due to variable donor levels from cell to cell . Hence it is more convenient to define the FRET levels in terms of the absolute change in donor level Δ⁢D , since this reflects the number of resonance energy transfer pairs ( 5 ) FRET⁢ ( t ) =Δ⁢D=D0⁢Δ⁢Rα+R0+Δ⁢R Since FRET occurs only when CheY-P and CheZ interact , the FRET level is proportional to the concentration of complex [Yp-Z] . If we assume the CheY-P dephosphorylation by CheZ follows Michaelis-Menten kinetics we can describe the [Yp-Z] concentration in terms of the activity of the kinase CheA . For this , we assume the system is in steady-state for timescales much larger than CheY phosphorylation-dephosphorylation cycle ( ≈100 ms ) . In that case , the destruction rate should equal the rate of CheA phosphorylation and hence the FRET signal is proportional to the activity per kinase a and the amount of CheA in the receptor-kinase complex ( Sourjik and Berg , 2002a; Oleksiuk et al . , 2011 ) : ( 6 ) FRET∝[Yp-Z]=a⁢kAkZ⁢[CheA]≈a⁢kAkZ⁢[CheA]T This last step is only valid if we further assume CheA autophosphorylation being the rate-limiting step . This is the case only if sufficient amounts of CheZ and CheY present in the cell . We have found that the FRET level initially increases with donor ( CheZ ) levels , but then saturates and remains constant for CheY and CheZ ( see Figure 3—figure supplement 2 ) . In many cases the most relevant parameter is the normalized FRET response . The FRET level reaches maximum if all kinases are active ( a≈1 ) . In case of CheRB+cells , this is the case when removing a saturating amount of attractant after adaptation ( Sourjik and Berg , 2002a ) . For CheRB- cells the baseline activity is ( Sourjik and Berg , 2002a; Shimizu et al . , 2010 ) close to 1 . Hence the normalized FRET FRET⁢ ( t ) /FRETmax represents the activity per kinase a⁢ ( t ) and is the relevant parameter for many quantitative models for chemoreceptor activity ( Tu , 2013 ) . ( 7 ) a⁢ ( t ) =FRET ( t ) FRETm⁢a⁢xand from a⁢ ( t ) the steady-state activity a0 can be determined by averaging a⁢ ( t ) over baseline values before adding attractant stimuli . From FRET time series of length T and acquisition frequency f we calculated Power Spectral Density ( PSD ) estimates as ( 8 ) SFRET⁢ ( ω ) =1T⁢\|ℱ⁢ ( ω ) \|2where ℱ⁢ ( ω ) is the ( discrete-time ) Fourier transform of the FRET time series FRET⁢ ( t ) . We only consider positive frequencies and multiply by two to conserve power . To study the influence of experimental noise and the effect of estimating τ and c from a finite time window , we generated O-U time series using the update formula ( Gillespie , 1996 ) ( 9 ) X⁢ ( t+Δ⁢t ) =X⁢ ( t ) -τ-1⁢Δ⁢t+c1/2⁢n⁢ ( Δ⁢ ( t ) ) 1/2in which n denotes a sample value from a normally distributed random variable ( μ=0 , σ=1 ) . To the generated time series Gaussian white noise was added to simulate experimental noise . The experimental noise amplitude was obtained from the average power at high frequencies . Since the amplitude of two-state switches is much greater than the noise , switching events times t0 could be easily read off by eye . We obtained switching durations by fitting the function ( 10 ) a ( t ) =12±12t−t0\|t−t0\| ( 1−e−2\|t−t0\|/τ± ) to the normalized FRET time series in a 30 s time window , approximately ±15 s from t0 . The residence times Δtup , i , k and Δtdown , i , k of event k in cell i were defined by the time between transitions or the beginning/end of the 25 μM stimulus time window . The steady-state activity during activity was then calculated as ( 11 ) a1/2 , i=∑kΔtup , i , k∑kΔtdown , i , k+∑kΔtup , i , kand for the residence times we take the mean over k to get τdown and τup . If we treat the system as an equilibrium process we can use the Arrhenius equations that describe the residence times as a function of the distance to the energy barrier ( 12 ) τdown=τrexp⁡[γdownln⁡[a1/2/ ( 1−a1/2 ) ]/kBT]τup=τrexp⁡[γupln⁡[a1/2/ ( 1−a1/2 ) ]/kBT]in which γdown and γup are constants corresponding to the slopes of ln⁡τdown and ln⁡τup against ln⁡[a1/2/ ( 1-a1/2 ) ] , respectively . The fit parameters and standard error are obtained with the robustfit function in Matlab ( statistics toolbox ) . Normalized FRET responses to different levels of ligand are fit to a hill curve of the form ( 13 ) a=[L]H[L]H+[K]H This can be connected to an MWC-type model ( Monod et al . , 1965 ) of receptor cluster activity ( Tu et al . , 2008 ) in the regime KI≪[L]≪KA , resulting in the correspondence keyH=NK=KIefm ( m ) which relates the Hill slope directly to the cluster size N , and sensitivity K to the methylation energy of the receptor . We plot K on a logarithmic scale to scale linearly with energy . The parameter estimate uncertainties of K and H are defined by the covariance matrix for each cell i ( 14 ) COVi=[σK⁢KσH⁢KσK⁢HσH⁢H]iin which σ denotes the standard error from the fit . For each covariance matrix the corresponding eigenvectors and eigenvalues are determined . The eigenvalues and vectors constitute an ellipsoid which represent error basins in K-H space . To obtain expression level estimates of different receptor species we use a different MWC model . Following ( Mello and Tu , 2005 ) , we use as an expression for the normalized response of cells to ligand [L] serine ( 15 ) a=ϵ0⁢ϵSNS⁢ϵANA⁢ ( 1+C⁢[L]/K~ ) Ns ( 1+[L]/K~ ) Ns+ϵ0⁢ϵSNS⁢ϵANA⁢ ( 1+C⁢[L]/K~ ) Nsin which NA is the number of Tar receptors in the cluster and NS is the number of Tsr receptors . Parameters ϵA , ϵS , ϵ0 are the energies corresponding to binding of ligand to Tar , Tsr and the other three receptors and are the same for each cell , like C and K~ which describe the disassociation constant for the active state as KA=K~/C , while NA and NT may vary from cell to cell . This yields the minimization problem for all 128 cells ( 16 ) min⁢∑iNc⁢e⁢l⁢l⁢s∑jNs⁢t⁢i⁢m ( mi , j-ai , j ) 2in which mi , j the measured FRET response normalized to the response amplitude of cell i to stimulus Lj . This function was minimized using the matlab function fmincon ( optimization toolbox ) . The total number NT=NA+NS is limited to 32 . When fitting the model used the energy parameters ϵ from reference ( Mello and Tu , 2005 ) where used as initial guess with a maximum of ±5% deviation . This yielded an estimate of NA and NS for each cell . Under the assumption that receptor clusters are well-mixed , this yields a Tar/Tsr ratio of NA/NS . For our model , we consider CheR and CheB to perform opposite operations on the same substrate . For simplicity , we do not explicitly describe the methylation and demethylation of the receptors explicitly but instead assume that CheR ( R ) activates the receptor-kinase complex directly ( A* ) , and that CheB ( B ) deactivates it ( A ) In general , the corresponding reaction equation is a function of the methylation of inactive kinases by CheR , and demethylation of active kinases by CheB , described by two functions g and h ( 17 ) dadt=g ( vr , a ) −h ( vb , a ) with vr and vb being the rates for CheR and CheB , respectively . We now assume that these reactions follow Michaelis-Menten kinetics , following ( Goldbeter and Koshland , 1981 ) and ( Emonet and Cluzel , 2008 ) , and the total amount of kinase complexes is constant ( AT=A+A ) . Hence the change in activity a=A/AT has a sublinear dependence on a: ( 18 ) d⁢ad⁢t=vr⁢1-aKr+1-a-vb⁢aa+Kb The Michaelis-Menten constants Kb and Kr are in units of AT and are therefore dimensionless numbers . We are interested in the steady-state level a0 and its dependence on the kinetic parameters in equation 18 . This is described by the Goldbeter-Koshland function ( Tyson et al . , 2003 ) , an exact solution to the system in case [R] and [B] are much smaller than [A]T . ( 19 ) a0⁢[vr , vb , Kr , Kb]=2⁢vr⁢Kr ( vb-vr+vbKb+vrKr+ ( vb-vr+vb⁢Kb+vr⁢Kr ) 2-4⁢ ( vb-vr ) ⁢vr⁢Kr The shape of this curve is sigmoidal if the Michaelis-Menten constants Kr and Kb are much smaller than one . For CheB phosphorylation , we assume the phosphorylation rate depends linearly on active CheA and write ( 20 ) d⁢[Bp]d⁢t=kp⁢[B]⁢a⁢ ( vr , vb , Kb , Kr ) -kd⁢p⁢[Bp]with the corresponding conservation law BT = BP + B . For the case for wild-type CheB , with phosphorylation feedback , the rates can be described in terms of catalytic rate times the enzyme ( subspecies ) concentration ( 21 ) vb=kb ( [BT]−[Bp] ) +Mkb[Bp]vr=kr[R]in which M stands for the ratio of demethylation rates of unphosphorylated and phosphorylated CheB . The fraction of the phosphorylated CheB , [Bp]/[B]Tthen determines the effective activity of CheB . Equation 20 is solved numerically using Mathematica ( Mathematica model source code available online ) for [Bp] and the result is substituted in equation 19 . In the absence of feedback , the activity can be directly calculated from equation 19 with the rates being simplyvb=kb⁢[B]vr=kr⁢[R] We only need to consider the ratio of rate constants kr and kb which determines at which expression ratio [CheR]/[CheB] the activity equals 1/2 . We assume kr=kb for simplicity , since the shape of the curve from Equation 19 is not affected by the values of kr and kb , changing their ratio only shifts the curve along the horizontal axis . Similarly , we only consider the ratio of phosphorylation and dephosphorylation rates . This leaves the system of equations above only has a few parameters: Kb , r and M; and the ratios kr/bb and kp/kd⁢p . In Table 5 , the parameters used for the calculations are listed . We first fixed the phosphorylation rates kp=1/2kdp . This means that the steady-state phosphorylated level of CheB [Bp]/[BT] at activity ≈1/3 is around 15 % . This parameter is not constrained by any direct observation , but it is clear the system benefits from a relatively low fraction of phosphorylation , to be able to up and down regulate the levels effectively upon changes in activity . Generally , we assume CheB-D56E to behave like unphosphorylated CheB . The gain in catalytic rate of activated CheB is estimated to be nearly a 100 fold , but this does not agree with the expression level differences between the different CheB mutants so we made a conservative estimate of 15 ( the attenuating effect increases with the gain ) . CheBc behaves approximately like phosphorylated CheB ( albeit with increase of only seven compared to D56E ) , qualitatively consistent with measured in vitro rates for CheBc and phosphorylated intact CheB ( Anand and Stock , 2002 ) . The difference between predicted rates and might be due to the fact that the rate experiments were performed in vitro . Michaelis Menten constants used in the model are lower than 1 , but how low is not well constrained by data , and estimations do not take into account the possible attenuating effect of phosphorylation . Our experimental data on the distribution of a0 implies the sigmodial curve is steep in the absence of phosphorylation and hence that Kb and Kr are quite small . The variability in a0 for CheBc is lower than D56E , implying that the curve is less steep and hence we have chosen are Kr which is not quite as low as D56E . To simulate gene expression noise , we simulated [CheR]/[CheB] log-normal distributions with σ=0 . 18 for all three strains . The mean of the distribution was chosen to yield an average steady-state network activity ( a0 ) of 0 . 4 . The resulting distribution of a0 was calculated using the corresponding Goldbeter-Koshland function for each genotype . Instead of assuming a sub-linear ( Michaelis-Menten ) dependence of CheR- and CheB-catalyzed rates on the receptor-kinase activity a , one may also assume linear , quadratic or cubic dependence of the methylation rates on the activity , as was for example done in ( Clausznitzer et al . , 2010 ) . Here , CheR feedback is assumed to be linear ( g=kr⁢[R]⁢ ( 1-a ) ) , while CheB feedback can be linear ( h=kb[B]a ) , quadratic ( h=kb⁢[B]⁢a2 ) or cubic ( h=kb⁢[B]⁢a3 ) in the receptor-kinase activity a . The supralinear ( quadratic and cubic ) forms of dependence are intended to model the case with CheB phosphorylation , and the linear form the case without CheB phosphorylation . The steady-state activity a0 can be found by solving g⁢ ( vr , a ) =h⁢ ( vb , a ) and the dependence of a0 to [R]/[B] ( a0=f ( [R]/[B] ) ) for these linear and supralinear cases are shown in Figure 3—figure supplement 3 .	Many sophisticated computer programs use random number generators to help solve challenging problems . These problems range from achieving secure communication across the Internet to deciding how best to invest in the stock market . Much research in recent years has found that randomness is also widespread in living cells , where it is often called “noise” . For example , the activity of some genes is so unpredictable to the extent that it appears random . Yet , relatively little is known about how such gene-expression noise propagates up to change how the cell behaves . Many open questions also remain about how cells might exploit these or other fluctuations to achieve complex tasks , like people use random number generators . Bacteria perform a number of complex tasks . Some bacteria will swim toward chemicals that suggest a potential reward , such as food . Yet they swim away from chemicals that could lead them to harm . This ability is called chemotaxis and it relies on a network of interacting enzymes and other proteins that coordinates a bacterium’s movements with the input from its senses . Keegstra et al . set out to find sources of noise that might act as random number generators and help the bacterium E . coli to best perform chemotaxis . An improved version of a technique called in vivo Förster resonance energy transfer ( or in vivo FRET for short ) was used to give a detectable signal when two proteins involved in the chemotaxis network interacted inside a single bacterium . The experiments showed that this protein network amplifies gene-expression noise for some genes while lessening it for others . In addition , the interactions between proteins encoded by genes acted as an extra source of noise , even when gene-expression noise was eliminated . Keegstra et al . found that the amount of signaling within the chemotaxis network , as measured by in vivo FRET , varied wildly over time . This revealed two sources of noise at the level of protein signaling . One was due to randomness in the activity of the enzymes involved in tuning the cell’s sensitivity to changes in its environment . The other was due to protein interactions within a large complex that acts as the cell’s sensor . Unexpectedly , this second source of noise under some conditions could be so strong that it flipped the output of the cell’s signaling network back and forth between just two states: “on” and “off” . Together these findings uncover how signaling networks can not only amplify or lessen gene-expression noise , but can themselves become a source of random events . The new knowledge of how such random events interact with a complex trait in a living cell – namely chemotaxis – could aid future antimicrobial strategies , because many bacteria use chemotaxis to help them establish infections . More generally , the new insights about noise in protein networks could help engineers seeking to build synthetic biochemical networks or produce useful compounds in living cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology" ]	2017	Phenotypic diversity and temporal variability in a bacterial signaling network revealed by single-cell FRET
Recent phylogenetic analyses indicate that RNA virus populations carry a significant deleterious mutation load . This mutation load has the potential to shape patterns of adaptive evolution via genetic linkage to beneficial mutations . Here , we examine the effect of deleterious mutations on patterns of influenza A subtype H3N2's antigenic evolution in humans . By first analyzing simple models of influenza that incorporate a mutation load , we show that deleterious mutations , as expected , act to slow the virus's rate of antigenic evolution , while making it more punctuated in nature . These models further predict three distinct molecular pathways by which antigenic cluster transitions occur , and we find phylogenetic patterns consistent with each of these pathways in influenza virus sequences . Simulations of a more complex phylodynamic model further indicate that antigenic mutations act in concert with deleterious mutations to reproduce influenza's spindly hemagglutinin phylogeny , co-circulation of antigenic variants , and high annual attack rates . Seasonal influenza viruses infect up to 15% of the world's human population annually , with the majority of flu cases attributable to influenza type A subtype H3N2 ( A/H3N2 ) ( World Health Organization , 2014 ) . This substantial disease burden stems from the virus's rapid antigenic evolution , which enables it to infect hosts within several years of a previous infection . A large body of research has therefore focused on understanding the process by which influenza evolves antigenically , particularly how point mutations in the virus's hemagglutinin ( HA ) protein allow for immune escape ( Wiley et al . , 1981; Wilson and Cox , 1990; Koel et al . , 2013 ) and how virus strains interact immunologically to shape this subtype's evolutionary patterns in the long term ( Ferguson et al . , 2003; Tria et al . , 2005; Koelle et al . , 2006; Recker et al . , 2007; Bedford et al . , 2012; Zinder et al . , 2013 ) . Distinct from these efforts , several phylogenetic analyses have indicated that influenza A/H3N2 in humans carries a deleterious mutation load ( Fitch et al . , 1997; Pybus et al . , 2007; Strelkowa and Lässig , 2012 ) . Specifically , early work by Fitch et al . ( 1997 ) found the number of nonsynonymous changes on tip branches of the HA phylogeny to be higher than expected , indicative of either strain selection bias or the presence of transiently circulating deleterious mutations in the influenza viral population . In more recent work , Pybus et al . ( 2007 ) performed a comprehensive phylogenetic analysis of over 140 viruses , including influenza A/H3N2 . For H3N2's M1 protein , as well as for the majority of the other viral proteins examined in the study , they found heightened ratios of non-synonymous-to-synonymous substitutions on external tree branches relative to those found internally . This finding again points towards transiently circulating deleterious mutations in influenza and , more generally , across RNA virus populations . Other recent work on predicting the short-term evolution of influenza has highlighted the necessity of accounting for fitness costs associated with sublethal deleterious mutations when projecting the frequencies of influenza clades into the next season ( Łuksza and Lässig , 2014 ) . Together , these results indicate that purifying selection is not sufficiently strong to immediately eliminate deleterious mutations from the influenza A/H3N2 virus population that circulates among humans . As a result of genetic linkage within genes and , to a lesser extent , across genes , these deleterious mutations have the potential to interact with beneficial mutations in determining which viral lineages will persist and which ones will ultimately be lost . Indeed , a recent statistical analysis of HA sequences from influenza A/H3N2 has suggested that interference effects largely determine the fates of viral mutants , rather than their inherent selective effects ( Illingworth and Mustonen , 2012 ) . These interference effects are possible because of an extensive genetic linkage across influenza's HA ( Strelkowa and Lässig , 2012 ) and arise from variation in the background fitness of viral strains and from variation in the fitness effects of subsequent mutations . Taken together , this body of work indicates that sublethal deleterious mutations commonly arise and circulate for sufficiently long periods of time to be able to impact the trajectories of influenza A/H3N2 strains . However , the impact that these deleterious mutations have on the population dynamics and long-term evolutionary patterns of this subtype has to date not been explored . Here , we address this question with a set of increasingly complex population genetic and population dynamic models , under the common assumption that influenza's adaptive evolution is driven by antigenic changes that allow for escape from herd immunity . We start by extending classic population genetic models into an epidemiological context . As expected from previous analyses of these types of models ( Fisher , 1930; Birky and Walsh , 1988; Peck , 1994; Barton , 1995; Orr , 2000 ) , we find that circulating sublethal deleterious mutations in influenza A/H3N2's viral population reduce the rate of adaptive evolution and increase the average size of the beneficial mutants that fix . Extending this analysis to models explicitly incorporating epidemiological dynamics , we further show that the accumulation of deleterious mutations can contribute to explaining the invasion dynamics of new antigenic clusters , defined as sets of viral strains that are antigenically similar to one another ( Smith et al . , 2004 ) . This model further predicts three distinct molecular pathways by which antigenic cluster transitions can occur , and we find empirical patterns consistent with each of these pathways in sequence data from 1992 to 2004 . Gaining intuition from these simple models , we then present more extensive phylodynamic simulations that incorporate the occurrence of both antigenic and non-antigenic ( largely deleterious ) mutations . This extension is critical given that antigenic mutations acquire their selective advantage through immune escape , which reduces competition for susceptible hosts and therefore in principle could allow for long-term coexistence of virus strains through niche partitioning of the host population ( Cobey , 2014 ) . Indeed , in the absence of other contributing processes , it has been shown that reduced between-strain competition resulting from antigenic evolution leads to explosive genetic and antigenic diversity ( Ferguson et al . , 2003 ) , a pattern that is inconsistent with the long-term evolutionary dynamics of influenza A/H3N2 in humans . Intriguingly , the phylodynamic model we present robustly reproduces the spindly phylogeny of influenza A/H3N2's HA protein ( Fitch et al . , 1997 ) under parameterizations relevant to A/H3N2 in humans . It further reproduces the recently described patterns of co-circulation of minor antigenic variants ( Strelkowa and Lässig , 2012 ) , as well as high annual attack rates ( World Health Organization , 2014 ) . In the discussion , we situate these findings in the context of previously published models used to explain the characteristic evolutionary dynamics of influenza A/H3N2 in humans , speculate on the applicability of these findings to other influenza-host systems , and comment on the consequences of these findings for the predictability of influenza evolution . To develop an understanding for how circulating deleterious mutations will impact patterns of influenza's antigenic evolution , we first extend existing population genetic models ( Haigh , 1978; Peck , 1994 ) to acute infectious diseases undergoing immune escape . As is common in many population genetic models , we assume an infinite population and consider this population subject to frequent sublethal deleterious mutations that act independently from one another in reducing fitness . In an explicit susceptible-infected-recovered ( SIR ) epidemiological context that does not yet incorporate immune escape , these assumptions lead to a deleterious mutation-selection balance in the virus population ( ‘Materials and methods’ , Figure 1A ) . This balance is given by a Poisson distribution with mean λ/sd , where λ is per-genome deleterious mutation rate and sd is the transmission fitness cost of sublethal deleterious mutations . The virus population at epidemiological equilibrium will have an overall net reproductive rate ( i . e . , mean absolute fitness ) of R = 1 , with more transmissible viruses that carry fewer deleterious mutations having reproductive rates above one and less transmissible viruses that carry more deleterious mutations having reproductive rates below one ( Figure 1A , inset ) . This within-population variation in viral transmission rates is also reflected in the distribution of infected individuals' basic reproductive rates ( R0 values ) ( Figure 1A inset ) . 10 . 7554/eLife . 07361 . 003Figure 1 . The effect of a deleterious mutation load on the fate of an antigenic mutant . ( A ) A resident antigenic strain at its deleterious mutation-selection balance . The histogram shows pk , the proportion of infected individuals carrying a virus with k deleterious mutations at its endemic equilibrium . The viral class carrying the fewest number of deleterious mutations is defined as mutation class k = 0 . Inset: variation in the basic reproductive rate of infected individuals ( gray histogram ) and variation in the net reproductive rate R of infected individuals ( brown histogram ) resulting from variation in the number of deleterious mutations carried by circulating viruses . Model parameters: λ = 0 . 10 , sd = 0 . 008 , μ = 1/30 years−1 , γ = 1/4 days−1 , R0 , k = 0 = 2 . 25 . ( B ) Simulations showing the three dynamical fates that an antigenic mutant can experience: rapid loss ( blue , with expanded inset ) , transient circulation ( green ) , and successful establishment ( red ) . The blue antigenic mutant ( with σ = 0 . 008 ) arises in a genetic background with k = 16 deleterious mutations . The green antigenic mutant ( with σ = 0 . 04 ) arises in a background of k = 10 deleterious mutations . The red antigenic mutant ( with σ = 0 . 06 ) arises in a background of k = 5 deleterious mutations . All other parameters are as in ( A ) . ( C ) The proportion of antigenic mutants that result in each of the three dynamical fates shown in ( B ) as a function of their antigenic size σ . All parameters as in ( A ) . ( D ) The average antigenic size of successfully establishing antigenic mutants under different deleterious mutation rates λ ( x-axis ) and for three transmission fitness costs ( see ( E ) for legend ) . The size of arising antigenic mutations σ are assumed to be gamma distributed with mean of 0 . 012 and a shape parameter of 2 . All other parameters are as in ( A ) . ( E ) The relative rate of antigenic evolution under different deleterious mutation rates λ ( x-axis ) and for the three transmission fitness costs shown in ( D ) . Other model parameters are as in ( D ) . The relative rate of antigenic evolution is given by the fraction of arising antigenic mutants that establish under the deleterious mutation load relative to the fraction of arising antigenic mutants that would establish under a no-load scenario . See ‘Materials and methods’ for choice of model parameters . DOI: http://dx . doi . org/10 . 7554/eLife . 07361 . 003 Following Peck ( 1994 ) , we first examine the fate of a single advantageous mutant arising in such a population . This advantageous mutant will necessarily arise in a genetic background with a certain number of deleterious mutations and , in our case , carry an immune-escape mutation that is beneficial to its spread . The genetic background in which the antigenic mutation arises and the size of the antigenic change jointly determine the mutant's initial reproductive rate Rm ( t = 0 ) in the virus population ( ‘Materials and methods’ ) . If this initial reproductive rate is less than one , the antigenic mutant is likely to be rapidly lost from the virus population . If the mutant's initial reproductive rate is instead greater than one , the mutant will invade the virus population if it is not initially stochastically lost . In the case of invasion , the mean reproductive rate of the antigenic mutant lineage will necessarily decline because it will accumulate its own set of deleterious mutations . ( The mutant lineage's mean reproductive rate will also necessarily decline because the size of its susceptible host pool will decline over time , a factor we for now ignore but return to in later models . ) For this invading antigenic mutant lineage , we can calculate a final reproductive rate Rm ( t = ∞ ) , which is the mean reproductive rate of this lineage once it has reached its own mutation-selection balance ( ‘Materials and methods’ ) . If this final reproductive rate falls below one , an invading antigenic mutant will therefore only transiently circulate before deterministically declining as a result of deleterious mutation accumulation . If this final reproductive rate exceeds one , however , the invading antigenic mutant lineage , under the assumptions of this model , will successfully establish . Antigenic mutants therefore experience one of three possible fates ( Figure 1B ) : rapid loss , transient circulation , or successful establishment . Which of these three fates awaits an antigenic mutant depends in part on the number of deleterious mutations carried by the strain in which the antigenic mutation arises: the lower the number of background deleterious mutations , the higher the antigenic mutant's chances are of at least transient establishment , consistent with the background fitness interference effects found in Illingworth and Mustonen ( 2012 ) . Which fate occurs also depends on the extent to which the antigenic mutant escapes immunity ( Figure 1C ) . The vast majority of small-sized antigenic mutants are rapidly lost; the remaining ones only circulate transiently before the accumulation of deleterious mutations results in their ultimate loss . Antigenic mutants that significantly escape immunity are less subject to rapid loss and also less likely to circulate only transiently . Given a specified size distribution for antigenic mutations , the average size of antigenic mutations that will successfully establish can be calculated . Figure 1D shows that this average antigenic size increases with increases in the deleterious mutation rate λ . The magnitude of the transmission fitness cost sd also affects the average size of antigenic mutants that will successfully establish . For any given deleterious mutation rate λ , as sd decreases the average size of antigenic mutants that fix increases ( Figure 1D ) . These results can be interpreted in the context of findings from the population genetics literature: increases in λ and decreases in sd similarly increase the virus population's fitness variance ( as quantified by the variance in net reproductive rate R , Figure 1A inset , ‘Materials and methods’ ) . Increases in the fitness variance of asexual populations makes fixation of a beneficial mutant increasingly dependent on genetic background; only beneficial mutants that exceed a characteristic large size will have a high probability of fixing in populations with substantial fitness variance ( Peck , 1994; Barton , 1995; Schiffels et al . , 2011; Good et al . , 2012 ) . In addition to their effect on the sizes of successfully establishing antigenic mutants , circulating deleterious mutations will act to slow the tempo of antigenic evolution ( Figure 1E ) ; that is , they will reduce the number of antigenic mutants that go to fixation in a given amount of time . This particular effect has previously been remarked upon in the context of a population genetics model for influenza's HA protein ( Strelkowa and Lässig , 2012 ) . Again , increases in the fitness variance of the viral population is the culprit: increases in the deleterious mutation rate λ and decreases in the fitness cost of deleterious mutations sd similarly act to increase fitness variance; with increased fitness variance in the population , the genetic background in which an antigenic mutant needs to arise in to have a chance at fixation will be increasingly restrictive . This leads to a reduced tempo of antigenic change , consistent with a reduction in the rate of adaptation that is known from the population genetics literature ( Peck , 1994; Barton , 1995 ) . Our model's findings can now be situated in the context of influenza A/H3N2's characterized evolutionary dynamics in humans . In particular , detailed antigenic analyses have demonstrated that this virus undergoes punctuated antigenic evolution , with predominantly single amino acid changes of large antigenic effect being responsible for the occurrence of antigenic cluster transitions ( Smith et al . , 2004; Koel et al . , 2013 ) . This dynamic is consistent with our results that antigenic evolution—as traced by lineages that ultimately fix—should occur via mutations of characteristically large size . These same analyses , as well as others ( Plotkin et al . , 2002 ) , have further indicated that antigenic cluster transitions occur only every 2 to 6 years , an incredibly slow pace given the virus's high mutation rate and the need for only a single amino acid to substantially alter antigenicity . Again , this dynamic is consistent with our results that the tempo of antigenic evolution should be severely reduced by circulating deleterious mutations . Thus , our model posits that the punctuated and surprisingly slow nature of influenza A/H3N2's antigenic evolution are related features of this largely asexual , adapting population subject to a deleterious mutation load: adaptive evolution requires not only that large antigenic mutants occur , but also that they occur in good genetic backgrounds . The above model contains a number of simplifying assumptions . Among these are that the susceptible host pool is negligibly affected over the time period of the antigenic mutant's establishment and that the successful establishment of an antigenic mutant will lead to the fixation of the mutant lineage in the virus population . Although these assumptions may be reasonable under some scenarios , they may not always be in an epidemiological context . For example , an antigenic mutant might invade sufficiently slowly to erode its frequency-dependent advantage prior to the exclusion of the existing antigenic lineage . Due to only partial cross-immunity between these lineages , this would lead to long-term coexistence of the variants . Another scenario is one of an antigenic mutant with a particularly large selective advantage: this mutant might burn through its susceptible host population so rapidly that its net reproductive rate drops significantly below one , leading to the possibility of its own extinction along with that of the previously circulating variant ( Ballesteros et al . , 2009 ) . This scenario underscores the importance of considering the possibility of a variable virus population size; population genetic models that assume a constant population size may not be appropriate under certain epidemiological conditions . To relax both the assumption of a time-invariant selective advantage and the assumption that successful establishment of a mutant leads to fixation of the mutant lineage ( including replacement of the resident lineage ) , we now consider a more complex epidemiological model that explicitly incorporates the dynamics of susceptible hosts ( ‘Materials and methods’ ) . Figure 2 shows the dynamics of the resident strain and the antigenic mutant under four distinct scenarios: a scenario in which a small antigenic mutant arises in a low-load ( ‘good’ ) genetic background ( Figure 2A ) ; a scenario in which a small antigenic mutant arises in an average-load genetic background ( Figure 2B ) ; a scenario in which a large antigenic mutant arises in a good genetic background ( Figure 2C ) ; and a scenario in which a large antigenic mutant arises in an average genetic background ( Figure 2D ) . These simulations indicate that the three fates predicted in the simpler model still play out in the explicit context of epidemiological dynamics when parameterized for influenza A/H3N2 in humans . Rapid loss is expected to occur when a small antigenic mutant arises in an average genetic background ( Figure 2B ) . As this combination occurs commonly , rapid loss is the most frequent fate experienced by antigenic mutants . Transient circulation occurs when a small antigenic mutant arises in a good genetic background ( Figure 2A ) , provided that it survived genetic drift . Transient circulation also occurs when a large antigenic mutant arises in an average genetic background ( Figure 2D ) , again provided that it survived genetic drift . Successful establishment can only occur when a large antigenic mutation arises in a good genetic background ( Figure 2C ) , a ‘jackpot’ combination . In this case , the resident strain is competitively excluded as a result of strain cross-immunity . Intriguingly , the presence of deleterious mutations can also affect the invasion dynamics of antigenic mutants having this ‘jackpot’ combination: because offspring of antigenic mutants accumulate deleterious mutations , and these deleterious mutations reduce viral transmissibility , the invasion dynamics of particularly large antigenic mutants are considerably less explosive than would be expected in the absence of deleterious mutation accumulation ( Figure 3 ) . Consequently , large antigenic mutants do not readily burn themselves out during attempted establishment , as might in theory be expected ( Ballesteros et al . , 2009 ) . 10 . 7554/eLife . 07361 . 004Figure 2 . Epidemiological dynamics following the emergence of an antigenic mutant . First row: antigenic mutants arising in low-load genetic backgrounds ( k = 3 deleterious mutations ) . Second row: antigenic mutants arising in average-load genetic backgrounds ( k = 13 deleterious mutations ) . Left column: antigenic mutants that are of a small size ( σ = 0 . 004 ) . Right column: antigenic mutants that are of a large size ( σ = 0 . 045 ) . ( A ) Small antigenic mutants arising in good genetic backgrounds transiently circulate . ( B ) Small antigenic mutants arising in average genetic backgrounds are rapidly lost . ( C ) Large antigenic mutants arising in good genetic backgrounds can successfully establish and exclude the resident antigenic strain , resulting in an antigenic cluster transition . ( D ) Large antigenic mutants arising in average genetic backgrounds transiently circulate . All simulations assume a host population size of N = 4 billion and start with a single antigenic strain at evolutionary and epidemiological equilibrium . The remaining parameters are as in Figure 1A . DOI: http://dx . doi . org/10 . 7554/eLife . 07361 . 00410 . 7554/eLife . 07361 . 005Figure 3 . Explosiveness in cluster invasion dynamics in the presence and absence of deleterious mutation accumulation . ( A ) A representative example of the population dynamics of an antigenic mutant in the presence of deleterious mutation accumulation . The new antigenic strain invades and successfully establishes , while excluding the resident strain , characteristic of a successful antigenic cluster transition . Model parameters are N = 4 billion , μ = 1/30 years−1 , γ = 1/4 days−1 , λ = 0 . 10 , sd = 0 . 008 , R0 , k = 0 = 2 . 25 , and σ = 0 . 05 . In this simulation , the antigenic mutant arises in a genetic background with k = 4 deleterious mutations . ( B ) A representative example of the population dynamics of an antigenic mutant in the absence of deleterious mutations . The new antigenic strain invades explosively , leading to its own burn-out along with exclusion of the resident strain . Model parameters are N = 4 billion , μ = 1/30 years−1 , γ = 1/4 days−1 , R0 = 2 . 04 , and σ = 0 . 121 . The value of R0 was chosen such that , prior to the invasion of the antigenic mutant , the fraction of the host population susceptible to infection and the number of infected hosts was the same across the two simulations . In ( B ) , the value of σ was chosen such that Rm ( t = 0 ) was the same across the two simulations ( at a value of 1 . 13 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07361 . 005 Figure 2C shows that an antigenic mutant can exclude a resident antigenic strain , provided that it arises in a good genetic background and carries an antigenic mutation of large effect . This is consistent with detailed molecular studies of influenza A/H3N2 that have shown that single amino acid changes of large antigenic effect can precipitate an antigenic cluster transition ( Smith et al . , 2004; Koel et al . , 2013 ) , although the importance of the genetic background in which the antigenic mutation arises has not been discussed in the context of this work . The BE92-to-WU95 cluster transition , precipitated by an amino acid change from N to K at site 145 , is a good example of a cluster transition occurring via a single mutational step ( Koel et al . , 2013 ) ( Figure 4A ) . Of note , several clades witnessed the 145NK amino acid substitution before it occurred in the clade that founded the WU95 viral lineage . Our models above indicate that the failure of these early 145K clades to establish could in principle be a consequence of this N-to-K amino acid change occurring in insufficiently good genetic backgrounds , as also suggested in Neher et al . ( 2014 ) . Other explanations for the failure of early 145K clades to establish are of course possible . One such explanation is that these clades might have circulated in spatial locations that are not sufficiently well-connected globally . Recent work has further emphasized the important role that spatial ecology plays in the global establishment of antigenic variants ( Russell et al . , 2008; Lemey et al . , 2014; Bedford et al . , 2015 ) , with findings suggesting that Asia plays a dominant role in sourcing these new variants . Thus , if the early 145K clades were not geographically well-situated , the spatial context ( rather than the genetic context ) of these clades may have led to their failure in establishing . We thus examined the spatial locations of the sequences from the three largest 145K clades that failed to establish: all three clades were geographically widespread , spanning at least two continents . Furthermore , two of the three clades contained sequences from Asia , despite Asia being undersampled during this time period . It is thus unlikely that these early 145K clades were geographically restricted to ‘sink’ populations . An alternative explanation for the failure of these early 145K clades to establish is that herd immunity levels against BE92 may not yet have been high enough to result in a sufficiently large selective advantage for these WU95-like lineages . Given these alternative explanations , an in vitro experimental study that quantifies relative viral fitness of these 145K lineages would thus be necessary to confirm our genetic background hypothesis . 10 . 7554/eLife . 07361 . 006Figure 4 . Influenza phylogenies consistent with the three distinct molecular pathways by which antigenic cluster transitions may occur . ( A ) Maximum clade credibility ( MCC ) phylogeny showing the BE92-WU95 antigenic cluster transition , reconstructed from sequences spanning years 1993–1997 . The phylogeny shows evolutionary dynamics that are consistent with a ‘jackpot’ combination of a large antigenic mutation arising in a rare low-load genetic background . Hemagglutination inhibition assays experimentally indicated that only a single amino acid change of large antigenic effect ( 145NK ) was necessary to precipitate the cluster transition ( Koel et al . , 2013 ) . Nodes are colored by the amino acid present at this site ( 145N = blue; 145K = red; other = black ) . ( B ) MCC phylogeny showing the WU95-SY97 antigenic cluster transition , reconstructed from sequences spanning years 1995–1999 . The phylogeny shows evolutionary dynamics consistent with the two-step antigenic change molecular pathway leading to antigenic cluster transitions , as depicted in Figure 5A . Hemagglutination inhibition assays experimentally indicated that two amino acid changes ( 156KQ and 158EK ) were necessary to precipitate the cluster transition ( Koel et al . , 2013 ) . Nodes are colored by the amino acids present at these sites ( 156K158E = red; 156Q158K = blue; other = black ) . Note that both amino acid changes occur on the same short internal branch , such that this apparently rapid transition is unlikely to be an artifact of sparse sampling . ( C ) MCC phylogeny showing the SY97-FU02 and FU02-CA04 cluster transitions , reconstructed from sequences spanning years 2001–2005 . The phylogeny shows evolutionary dynamics consistent with the two-step reassortant molecular pathway leading to antigenic cluster transitions , as depicted in Figure 5B . Hemagglutination inhibition assays experimentally indicated that only a single amino acid change ( 156QH ) antigenically defined the SY97-FU02 cluster transition ( Koel et al . , 2013 ) . Nodes are colored by the amino acid present at this site ( 156Q = blue; 156H = red ) . Vertical bar shows the FU02 reassortant clade . The genetically distant non-hemagglutinin ( HA ) parent lineage of this reassortant clade is also shown . Phylogenies in ( A–C ) were inferred using BEAST ( Drummond et al . , 2012 ) from full HA1 sequences with specified sampling dates . DOI: http://dx . doi . org/10 . 7554/eLife . 07361 . 006 While hitting the ‘jackpot’ combination is a viable molecular pathway for precipitating an antigenic cluster transition , the combination of a large antigenic mutation arising in a low-load genetic background is expected to occur infrequently . We can therefore consider whether there might be alternative molecular pathways open to antigenic mutants that would similarly yield a successful cluster transition . One possibility is for a more common small- to medium-sized antigenic mutation to first occur in a good genetic background . This would result in a transient rise of this mutant ( Figure 2A ) , thereby increasing the number of individuals infected with the virus carrying few deleterious mutations . A less common , large-sized antigenic mutation could then occur , effectively piggy-backing on the good genetic background that the smaller antigenic mutant inflated . Because smaller antigenic mutations can only circulate transiently , and accumulate deleterious mutations during their circulation , the large antigenic mutation must not only follow , but also rapidly follow , the rise of the smaller antigenic mutation for this molecular pathway to yield an antigenic cluster transition . This scenario is depicted in Figure 5A , and phylogenetically would result in a sudden appearance of a viral lineage carrying two antigenic mutations . A phylogenetic analysis of the WU95-SY97 antigenic cluster transition provides an empirical example that is consistent with this molecular pathway of antigenic turnover , with a seemingly simultaneous accumulation of two antigenic amino acid changes ( 156KQ and 158EK ) occurring on a short branch of the reconstructed phylogeny ( Koel et al . , 2013 ) ( Figure 4B ) . 10 . 7554/eLife . 07361 . 007Figure 5 . Two-step approaches to antigenic cluster transitions . ( A ) A cluster transition arising from two consecutive antigenic mutations . A small antigenic mutation ( σ = 0 . 003 , dashed line ) first arises in a good genetic background ( deleterious mutation load k = 2 ) of the resident strain ( dotted line ) . Shortly after , a second and larger-sized antigenic mutation ( σ = 0 . 045 , solid line ) occurs in an individual infected with the single antigenic mutant . This sequence of events precipitates an antigenic cluster transition , with the double mutant replacing the resident strain and the low-frequency single mutant . We assume that the degree of immune escape is additive , such that σ between the resident strain and the double mutant is σ = 0 . 048 . ( B ) A cluster transition arising from intrasubtypic viral reassortment . A large-sized antigenic mutation ( σ = 0 . 06 , dashed line ) first arises in an average genetic background ( k = 10 ) of the resident strain ( dotted line ) . After 2 . 5 years of circulation , a coinfection that leads to the generation of low-load mutant ( k = 4 ) occurs . This low-load mutant ( solid line ) replaces the resident strain and the average-load carrying antigenic mutant , ultimately precipitating an antigenic cluster transition . Other model parameters in ( A ) and ( B ) are N = 4 billion , μ = 1/30 years−1 , γ = 1/4 days−1 , R0 , k = 0 = 2 . 25 , λ = 0 . 10 , and sd = 0 . 008 . DOI: http://dx . doi . org/10 . 7554/eLife . 07361 . 007 A third molecular pathway that could in principle precipitate an antigenic cluster transition is for a large antigenic mutation to first arise in an average genetic background and , during its transient circulation ( Figure 2D ) , to purge itself of a large number of deleterious mutations . This purging could arise through within-subtype viral reassortment taking place in an individual coinfected with a strain belonging to the resident cluster and a strain belonging to the transiently circulating antigenic mutant . Even though both of the strains infecting this individual would likely carry an average deleterious mutation load , it is highly unlikely that they will carry the same set of deleterious mutations if phylogenetically sufficiently far apart . Reassortment of the eight gene segments within the coinfected host could therefore significantly lower the number of deleterious mutations carried by viral progeny characterized as belonging to the new antigenic cluster . Once the deleterious mutation load has largely been shed , the reassortant virus would quickly rise and cause an antigenic cluster transition ( Figure 5B ) . Indeed , many historical instances of intrasubtypic reassortment contributing to antigenic turnover have been documented ( Morens et al . , 2009 ) ; these instances have been associated with high incidence levels as would be expected by purging of deleterious mutations . A phylogenetic analysis of the SY97-FU02 cluster transition provides an especially compelling example that is consistent with this molecular pathway of antigenic turnover . In this cluster transition , a virus antigenically characterized as FU02 reassorted with a genetically distant virus antigenically characterized as SY97 ( Barr et al . , 2005; Holmes et al . , 2005 ) ( Figure 4C ) . This reassortant viral lineage circulated extensively in Australia and New Zealand in 2003 and subsequently in the US in the 2003–2004 influenza season , causing substantial morbidity and mortality ( Barr et al . , 2005 ) . Although this viral lineage may ultimately have led to the replacement of the non-reassortant FU02 viral lineage , both of these FU02 lineages were excluded by the subsequent CA04 antigenic cluster , which appears to have originated from the non-reassortant FU02 lineage ( Figure 4C ) . Intriguingly , the CA04 lineage carried with it not only an HA mutation of large antigenic size , but also two amino acid changes in its polymerase gene segment that enhanced replicative fitness ( Memoli et al . , 2009 ) . Our above analyses have relied on simple epidemiological models to gain intuition for how circulating sublethal deleterious mutations would impact patterns of influenza A/H3N2's antigenic evolution . These analyses indicate that deleterious mutation loads should lower the rate of antigenic evolution ( Figure 1E ) . Based on previous modeling work ( Koelle et al . , 2006 , 2009 , 2010; Zinder et al . , 2013 ) , a lower rate of antigenic evolution is known to constrain genetic and antigenic diversity and thus might lead to a spindly HA phylogeny . Our analyses also indicate that deleterious mutation loads increase the average size of antigenic variants that establish in the long run ( Figure 1D ) ; observed patterns of punctuated antigenic evolution ( Smith et al . , 2004 ) may thus be better reproduced with a model that integrates sublethal deleterious mutations than one that ignores these mutations . Furthermore , our above analyses indicate that antigenic variants can reach appreciable numbers even when their ultimate fate is one of only transient circulation . This pattern of transient circulation points towards the possibility of co-circulation of a substantial number of antigenic variants , as argued for in statistical analyses of influenza's HA sequences ( Strelkowa and Lässig , 2012 ) . In our model , these antigenic variants need not necessarily strongly compete with one another for susceptible hosts for only a single lineage to persist . In the absence of exceptionally strong competition for susceptible hosts , the co-circulation of these antigenically distinct variants may thus be capable of reproducing empirically observed high annual attack rates ( World Health Organization , 2014 ) . To determine whether spindly HA phylogenies , co-circulation of antigenic variants , and high annual attack rates indeed come out of a model that incorporates deleterious mutations , we implemented a phylodynamic model that simulates the occurrence of both non-antigenic ( largely deleterious ) mutations and antigenic mutations ( ‘Materials and methods’ ) . When simulated under parameters appropriate for influenza A/H3N2 in humans , this model yields a spindly HA phylogeny with low-load viruses populating the trunk and higher load viruses populating the tips of the phylogeny ( Figure 6A ) . This distribution of deleterious mutation loads on the simulated phylogeny is consistent with the excess number of non-synonymous substitutions on external tree branches that was found for human influenza A/H3N2 ( Fitch et al . , 1997; Pybus et al . , 2007 ) and arises because deleterious mutations contribute a substantial fraction of the fitness variance in the viral population ( Figure 7A ) . 10 . 7554/eLife . 07361 . 008Figure 6 . Viral phylogenies from a simulation of the phylodynamic model incorporating antigenic and non-antigenic mutations . ( A ) Simulated phylogeny reproducing H3N2's spindly HA phylogeny , with low levels of genetic and antigenic diversity over the long run . Lineages are colored according to their deleterious mutation loads . ( B ) Simulated phylogeny shown in ( A ) with lineages colored according to their antigenic type . Similarly colored lineages that are genetically distinct are antigenically distinct ( colors were re-used due to their limited number ) . ( C ) Simulated phylogeny shown in ( A ) with lineages colored according to the fraction of the host population susceptible to infection with that lineage ( Seff/N ) . ( D ) Simulated phylogeny shown in ( A ) with lineages colored according to their net reproductive rate R . DOI: http://dx . doi . org/10 . 7554/eLife . 07361 . 00810 . 7554/eLife . 07361 . 009Figure 7 . Fitness variance dynamics , epidemiological dynamics , and times to most recent common ancestor for a simulation of the phylodynamic model incorporating antigenic and non-antigenic mutations . ( A ) The fraction of ( log ) viral fitness variation explained by deleterious mutations over time for the simulation whose phylogenies are plotted in Figure 6 . The fraction not explained by mutation load is due to antigenic variation in the population . ( B ) Simulated epidemiological dynamics showing co-circulation of multiple antigenic variants and sustained prevalence levels over time for this same simulation . ( C ) Times to the most recent common ancestor ( tMRCAs ) , computed over time from the phylogenies shown in Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 07361 . 009 The phylodynamic model simulation further reproduces antigenic variant co-circulation ( Figures 6B , 7B ) , consistent with findings that lineages that are lost nevertheless undergo appreciable antigenic evolution ( Strelkowa and Lässig , 2012 ) and that antigenic diversity levels within clusters can exceed antigenic distances between clusters ( Smith et al . , 2004 ) . The simulation yields prevalence levels of 10–180 cases per 100 , 000 individuals ( Figure 7B ) and annual attack rates of approximately 2–10% , consistent with empirical estimates of influenza incidence ( World Health Organization , 2014 ) . Despite extensive co-circulation of antigenic variants , the overall phylogeny remains ladder-like due to the selective sweeps initiated by rare large-sized antigenic mutations that arise in good genetic backgrounds . This dynamic is evident by jointly considering Figure 6A and Figure 6C , with Figure 6C showing , for each lineage , the fraction of the host population that is susceptible to infection with that lineage . From these figures , it is clear that the trunk of the phylogeny carries low-load viruses ( Figure 6A ) that have a high number of susceptible hosts ( Figure 6C ) . This combination together yields high-fitness viruses , as epidemiologically given by their net reproductive rates R ( Figure 6D ) . It is these viruses that establish and thus form the trunk of the tree . Neither a low mutation load alone nor a high number of susceptible hosts alone suffice in generating a sufficiently high-fitness viral lineage that will ensure its long-term evolutionary success . Inspection of Figure 6C also shows that trunk lineages abruptly gain susceptible hosts . These abrupt gains are a result of single large antigenic mutations that , when occurring in good genetic backgrounds , initiate the selective sweeps that ultimately limit influenza's genetic diversity . How these selective sweeps affect levels of standing genetic diversity in the viral population is shown in Figure 7C , where we use the time to the most recent common ancestor ( tMRCA ) of all circulating lineages as a proxy for total genetic diversity . This tMRCA plot reproduces quantitative features of influenza A/H3N2's tMRCA plot presented in Bedford et al . ( 2011 ) , including its interannual variation and the observed major drops in tMRCA following the emergence of new and antigenically very distinct clusters . The importance of deleterious mutations in constraining the genetic and antigenic diversity of influenza can be further illustrated by simulating the phylodynamic model under the assumption of their absence . These simulations very rapidly yield explosive genetic and antigenic diversity ( Figure 8A ) and , as a consequence , prevalence levels that generally increase over time ( Figure 8B ) . Reassuringly , this finding is consistent with predictions from previous influenza modeling work incorporating only strain-specific immunity ( Ferguson et al . , 2003 ) . 10 . 7554/eLife . 07361 . 010Figure 8 . Simulations of the phylodynamic model in the absence of a deleterious mutation load . ( A ) Simulated phylogeny showing explosive genetic and antigenic diversity over a 20-year period . Lineages are colored according to their antigenic type , with similarly colored lineages that are genetically distinct being antigenically distinct ( colors were re-used due to their limited number ) . ( B ) Simulated epidemiological dynamics showing prevalence levels generally increasing over time . DOI: http://dx . doi . org/10 . 7554/eLife . 07361 . 010 Given that purifying selection alone is known to reduce genetic diversity ( Charlesworth et al . , 1993; Walczak et al . , 2012 ) , we further simulated the phylodynamic model under the assumption of no antigenic mutations . In these simulations , we phenomenologically incorporated antigenic drift by simulating susceptible-infected-recovered-susceptible ( SIRS ) dynamics such that prevalence levels were similar to those shown in Figure 7B . Compared with the results shown in Figure 6 , these simulations gave rise to significantly higher levels of genetic diversity ( Figure 9A ) and longer tMRCAs ( Figure 9B ) . This indicates that purifying selection alone does not account for the spindly phylogenies shown in Figure 6 . Rather , it is antigenic evolution in the context of fitness variation generated by deleterious mutations that constrains the viral phylogeny . 10 . 7554/eLife . 07361 . 011Figure 9 . Simulations of the phylodynamic model in the absence of antigenic mutations . ( A ) Simulated phylogeny under a parameterization with no antigenic mutations , showing genetic diversity generally increasing over time . ( B ) Times to the most recent common ancestor ( tMRCAs ) computed over time from the phylogeny shown in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07361 . 011 While the above simulations demonstrate that a model incorporating both antigenic and deleterious mutations can reproduce influenza A/H3N2's spindly phylogeny , its antigenic co-circulation patterns , and its high annual attack rates , a number of the parameters that required specification are empirically not well characterized . Most notably , these are the evolutionary parameters of the model: the deleterious mutation rate λ , the fitness cost of deleterious mutations sd , and the antigenic mutation rate λantigenic . In Figure 10 , we show how the evolutionary and epidemiological dynamics of the model simulations depend on these three parameters . Specifically , we vary one parameter while keeping the remaining two parameters fixed . For each parameter set considered , we perform 20 simulations to determine the range of dynamics predicted under a single parameterization . For each simulation , we quantify the virus's evolutionary dynamics by plotting the minimum and maximum tMRCA calculated over a continuous 10-year period ( years 15–25 of the simulation ) as a measurement of the extent of genetic diversity in the viral population . To quantify the virus's epidemiological dynamics , we plot the minimum and maximum annual attack rates over this same time period . Figure 10A shows that in the absence of deleterious mutations ( λ = 0 ) the model simulations yield explosive viral diversity , with the maximum tMRCA ranging between 5 and 25 years . The observed increases in viral genetic and antigenic diversity result in unrealistically high maximum annual attack rates , in the range of 15–45% ( Figure 10B ) . Simulated under this parameterization , the results shown in Figure 8A , B are representative of these results . With increasing deleterious mutation rates , the maximum tMRCAs decline as do maximum annual attack rates ( Figure 10A , B ) . For λ values of 0 . 10 and higher , empirically documented annual attack rates can be consistently reproduced . Maximum and minimum tMRCAs are best reproduced for λ values between 0 . 10 and 0 . 15 . 10 . 7554/eLife . 07361 . 012Figure 10 . Sensitivity of evolutionary and epidemiological dynamics to parameters of the phylodynamic model . Subplots ( A , B ) show model sensitivity to the deleterious mutation rate λ . Subplots ( C , D ) show model sensitivity to the fitness cost of deleterious mutations sd . Subplots ( E , F ) show model sensitivity to the antigenic mutation rate λantigenic . The top row shows maximum ( red dots ) and minimum ( black dots ) times to the most recent common ancestor ( tMRCAs ) for 20 independent simulations . The red dashed line indicates the maximum tMRCA inferred from a phylogenetic analysis of influenza A/H3N2's HA ( Bedford et al . , 2011 ) ; the black dashed line indicates the minimum tMRCA inferred from this same analysis . The bottom row shows maximum ( red dots ) and minimum ( black dots ) annual attack rates for the same 20 simulations . The red dashed line indicates an estimate of the maximum annual attack rate for influenza A/H3N2; the black dashed line indicates an estimate of the minimum annual attack rate for influenza A/H3N2 . These values are based on annual attack rate estimates in adults of 5–10% , such that the maximum annual attack rate is on the order of 10% , and the minimum annual attack rate is shown at 1% ( which would correspond to years of negligible circulation of this influenza subtype ) . Each simulation was run for 28 years , and minimum and maximum tMRCAs and attack rates were computed from years 15–25 of the simulation . In subplots ( A ) and ( B ) , λ is varied , sd = 0 . 008 and λantigenic = 0 . 00075 . In subplots ( C ) and ( D ) , λ = 0 . 10 , sd is varied , and λantigenic = 0 . 00075 . In subplots ( E ) and ( F ) , λ = 0 . 10 , sd = 0 . 008 , and λantigenic is varied . All other parameter values are as listed in Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 07361 . 012 Figure 10C , D shows the evolutionary and epidemiological effects of the fitness cost of deleterious mutations , sd . It is clear from these plots that neither the mean maximum nor the mean minimum tMRCA across the simulations depends strongly on sd ( Figure 10C ) . However , the range of maximum tMRCA values is considerably higher at lower sd values ( Figure 10C ) . This may be because selective sweeps are expected to occur more rarely at lower sd values ( Figure 1E ) , such that the viral population is homogenized less frequently at these lower values , leading to higher tMRCA values . This explanation is consistent with the slightly lower annual attack rates at lower sd values ( Figure 10D ) : when the rate of antigenic evolution is slower , individuals cannot be reinfected as rapidly and thus annual attack rates would be lower . Despite the dependency of evolutionary and epidemiological dynamics on sd , Figure 10C , D shows that a broad range of sd values yields dynamics that are consistent with influenza A/H3N2 dynamics in humans . Finally , Figure 10E , F shows the sensitivity of the model to the antigenic mutation rate λantigenic . In the absence of antigenic evolution ( λantigenic = 0 ) , maximum tMRCAs are significantly higher than empirically documented ( Figure 10E ) and maximum annual attack rates do not exceed 3 . 8% ( Figure 10F ) . These findings are consistent with , Figure 9 , which indicates that a spindly viral phylogeny cannot be reproduced under purifying selection alone in a model that is parameterized for influenza . ( Figure 9's model differs slightly from the model parameterized with λantigenic = 0 , whose simulations are shown in Figure 10E , F: Figure 9 shows simulation results from an SIRS model that yields annual attack rates consistent with those empirically observed for flu; the results shown in Figure 10E , F under λantigenic = 0 use the same parameterization as in Figure 6 , with the exception of λantigenic , and thus simulate a simple SIR model . In either case , purifying selection alone cannot reproduce a spindly phylogeny . ) At increasing λantigenic values , tMRCAs decrease and annual attack rates increase to empirically documented values . These plots further indicate that influenza's evolutionary and epidemiological dynamics can be reproduced over a wide range of antigenic mutation rates as long as the rate exceeds a certain minimum value . Finally , taken together , Figure 10A , B , E , F again indicates that it is the interaction between deleterious and advantageous immune escape mutations that consistently yields a spindly phylogeny and high annual attack rates . Neither deleterious mutations nor immune escape mutations alone succeed in reproducing these dynamic features of influenza . Here , we have shown that population genetic and population dynamic models incorporating sublethal deleterious mutations can reproduce the characteristic features of influenza A/H3N2's evolutionary dynamics in humans . These include the virus's rare punctuated antigenic evolution and the low genetic diversity of its hemagglutinin protein . The low genetic diversity of the virus's HA , reflected in the spindliness of its phylogeny , has been a particular evolutionary characteristic that influenza modelers have sought to reproduce . To date , three other models exist that can explain this evolutionary characteristic of influenza ( Ferguson et al . , 2003; Koelle et al . , 2006; Bedford et al . , 2012 ) . All three of these models , however , are subject to criticism . Influenza's epochal evolution model ( Koelle et al . , 2006 ) assumes that neutral or nearly neutral mutations accumulate at HA epitopes and that these changes enable a previously neutral mutation to exact a large antigenic effect and thereby to precipitate an antigenic cluster transition . Criticisms of this model are several . First , recent work indicates that amino acid changes that are responsible for cluster transitions have large antigenic effects in consensus sequences ( Koel et al . , 2013 ) , such that genetic context is unlikely to be of utmost importance in determining the antigenic effect of mutations . Second , a molecular evolutionary analysis following the publication of the epochal evolution model has indicated that positive selection acts not only between antigenic clusters but also within antigenic clusters ( Suzuki , 2008 ) . In support of this finding , a more recent statistical analysis has shown that multiple antigenic variants co-circulate ( Strelkowa and Lässig , 2012 ) . Both of these empirical analyses are inconsistent with the assumptions of the epochal evolution model and indicate that the evolution of influenza is unlikely to be limited by the occurrence of antigenic mutations . The two other existing models that can reproduce influenza's spindly HA phylogeny are the generalized cross-immunity model put forward by Ferguson et al . ( 2003 ) and the canalization model put forward by Bedford et al . ( 2012 ) . When simulated , both of these models yield antigenic variant co-circulation and are thus consistent with the analyses detailed above that have shown that influenza evolution is not antigenic mutation limited . In Ferguson et al . ( 2003 ) , generalized ( strain-transcending ) cross-immunity lasting on the order of 6 months was invoked to limit the genetic and antigenic diversity of influenza A/H3N2 in humans . The major criticism of this model is lack of empirical support for this duration and form of generalized cross-immunity: while there is evidence for long-lasting cross-immunity between more genetically distant strains ( including heterologous influenza A subtypes ) , it appears to reduce pathology and possibly accelerate viral clearance rather than prevent infection ( Grebe et al . , 2008 ) . A recent experimental study in ferrets indicates that generalized cross-immunity that prevents infection appears to exist , but that it lasts for less than a week ( Laurie et al . , 2015 ) , which is insufficiently long to limit genetic and antigenic diversity in the model of Ferguson et al . The canalization model by Bedford et al . does not invoke generalized cross-immunity but instead assumes that antigenic mutations move viral strains in a two-dimensional ( or higher ) antigenic space . While these antigenic spaces or maps have been used to visualize the trajectories of flu's antigenic evolution ( Smith et al . , 2004; de Jong et al . , 2007 ) , models that start with the assumption of these maps considerably inflate the degree of competition between antigenic variants . This inflation of competition results from antigenic distances between daughter variants ( variants that are produced from the same parent ) necessarily being subadditive in this space . Inflated competition for susceptible hosts between antigenic variants is expected to lead to a more spindly phylogeny due to increased ‘niche overlap’ and therefore more frequent occurrences of between-strain competitive exclusion . In light of our findings presented here and existing knowledge on the theory of asexual evolution , we can reflect on why the deleterious mutation and canalization models can reproduce limited genetic and antigenic diversity in simulations of influenza A/H3N2 in humans , whereas Ferguson et al . ( 2003 ) initially found that strain-specific immunity did not suffice in reproducing these patterns . From the population genetics literature on asexual populations in which interference effects are at play , it is well known that an increase in the fitness variance of a population acts to slow adaptive evolution and make the characteristic size of adaptive mutations that fix larger . While here we have invoked deleterious mutations in the generation of fitness variation , beneficial mutations can similarly contribute to inherited variation in fitness ( Birky and Walsh , 1988; Barton , 1995 ) . Thus , in population genetic models , interference between beneficial mutations ( or a combination of beneficial and deleterious mutations ) similarly slows down the rate of adaptive evolution and increases the size of adaptive mutants that fix ( Birky and Walsh , 1988; Barton , 1995; Gerrish and Lenski , 1998; Rouzine et al . , 2003 , 2008; Park et al . , 2010; Sniegowski and Gerrish , 2010; Schiffels et al . , 2011; Good et al . , 2012 ) . However , all of these population genetic models assume a constant population size and therefore full resource competition between individuals . In the context of influenza , however , it is a change in antigenicity that is thought to provide the selective advantage . Because antigenic changes allow for escape from herd immunity , these changes by definition reduce competition for susceptible hosts ( the virus resource ) . As such , antigenic mutants create a partially new niche and so do not necessarily lead to competitive exclusion . The establishment of antigenic mutants can therefore instead lead to long-term coexistence and , as a result of only partial cross-immunity , a larger infected population size . This is why explosive genetic and antigenic diversity is expected in the presence of only strain-specific immunity ( Ferguson et al . , 2003 ) . In this context , generalized cross-immunity acts to considerably increase the competition between strains for susceptible hosts ( as well as to reduce the overall infected population size ) . As such , the fitness variance generated by beneficial antigenic mutations in this model results in a decrease in the rate of antigenic evolution , an increase in the size of antigenic mutations that establish , and limited diversity in the long run , as in population genetic models that assume full competition between beneficial mutations by considering populations of constant size ( Good et al . , 2012 ) . Similarly , the canalization model by Bedford et al . likely reproduces punctuated antigenic evolution and long-term limited genetic and antigenic diversity as a result of interference effects generated by inflated competition between antigenic mutations . The generalized cross-immunity model , the canalization model , and the deleterious mutations model presented here therefore share fundamental similarities: they can reproduce the characteristic features of influenza evolution in humans by generating enough fitness variation among competing strains in the viral population . However , the first two of these models create fitness variation by beneficial antigenic mutations alone; because these mutations obtain their selective advantages by reducing competition for susceptible hosts , these models need other components ( generalized cross-immunity or mutations that are necessarily subadditive in effect ) to augment strain competition . In contrast , the deleterious mutation model we present here does not need to invoke processes to augment competition between antigenic strains because a large proportion of fitness variation in the virus population arises from differences in deleterious mutation loads ( Figure 7A ) . Letting fitness variance be generated by deleterious mutations allows for limited diversity in the long run despite the co-circulation of antigenically very distinct variants that do not necessarily compete strongly for susceptible hosts . As such , our model can reproduce empirically observed high annual attack rates on the order of 10–15% ( Figure 10B for λ = 0 . 10 ) . Because deleterious mutations are known to circulate in the influenza A/H3N2 virus population ( Fitch et al . , 1997; Pybus et al . , 2007; Strelkowa and Lässig , 2012 ) , and in light of existing criticisms of the other two models , we therefore argue that the model presented here provides a more plausible mechanistic explanation for influenza's characteristic evolutionary features . Our finding that specifically non-antigenic fitness variation is an important contributing driver in shaping the characteristic features of influenza's evolutionary dynamics in humans sheds light on other recent virological findings . One such finding is that cellular receptor binding avidity is an important phenotype that impacts viral fitness ( Hensley et al . , 2009 ) . In a naive host , influenza viruses with low receptor binding avidities have a selective advantage , whereas , in a more immune host , influenza viruses with high receptor binding avidities have a selective advantage ( Hensley et al . , 2009 ) . Which receptor binding avidity phenotype can be considered the ‘deleterious’ variant is therefore subject to which individual the virus finds itself in , as well as the overall degree of herd immunity in the host population ( Yuan and Koelle , 2013 ) . Because changes in receptor binding avidity frequently also alter antigenicity ( Hensley et al . , 2009 ) , it is therefore also easily conceivable that certain of these changes increase virus fitness via two distinct mechanisms . As such , even though the genetic change is one and the same , the change in binding avidity that results can be thought of as increasing the background fitness of a virus strain , whereas the change in antigenicity that results can be considered as in this paper . Successful cluster transitions via the ‘jackpot’ strategy , as depicted in Figure 2C , may therefore preferentially involve mutations that simultaneously affect binding avidity and antigenicity . Indeed , the majority of cluster-transition mutations that have been recently characterized fall near the receptor binding site of influenza's HA ( Koel et al . , 2013 ) , suggesting that changes in the receptor binding avidity phenotype that occur with these antigenic mutations may improve virus fitness . In addition to cellular receptor binding avidity , glycosylation of influenza's HA is known to impact virus fitness . Whether glycosylation sites accumulate over evolutionary time ( as in the case of H3N2 in humans [Blackburne et al . , 2008] ) or do not ( as in the case of H1N1 in humans [Das et al . , 2011] ) therefore will depend on how these sites impact the background fitness of virus strains . In the case of H1N1 , for example , glycosylation has been shown to significantly reduce receptor binding avidity and therewith to lower overall virus fitness , despite the beneficial effect of glycosylation on escape from antibody-mediated neutralization ( Das et al . , 2011 ) . Compensatory mutations are therefore needed to restore virus fitness following the addition of a glycosylation site ( Das et al . , 2011 ) . Other phenotypes that impact virus fitness include those that influence protein stability ( Bloom and Glassman , 2009 ) and those that confer resistance to antivirals ( Herlocher et al . , 2004 ) . In the case of permissive mutations that influence protein stability ( Bloom et al . , 2010; Gong et al . , 2013 ) , these mutations may not directly influence virus fitness; their occurrence , however , may impact the viability of other mutations that have fitness consequences and thus may alter the size distribution of viable beneficial mutations , including those that allow for immune escape . Epistatic interactions such as these can therefore be accommodated within this general framework of fitness variation generated by phenotypes other than antigenicity in driving patterns of influenza's antigenic evolution . While we here simply model this fitness variation as arising from circulating deleterious mutations , any of these non-antigenic phenotypes can similarly contribute to this fitness variation . Indeed , deleterious mutations are necessarily deleterious as a consequence of some phenotype , whether it is susceptibility to antivirals , a suboptimal receptor binding avidity , a protein with reduced stability , or simply another phenotype that reduces viral replication . The complementary phenotypes to these can conversely be considered as beneficial mutations that contribute to fitness variation . As such , there is not always a clear source of fitness variation in the influenza virus population . What is critical , however , is that a significant proportion of the virus's fitness variation has to arise from non-antigenic phenotypes that do not reduce competition between virus strains , as antigenic mutants alone result in explosive genetic and antigenic diversity as a result of effective niche partitioning . Our choice of modeling simply deleterious mutation accumulation , instead of specific non-antigenic phenotypes , stems from the finding that virus populations , and more specifically influenza virus populations , carry substantial deleterious mutation loads ( Fitch et al . , 1997; Pybus et al . , 2007 ) . Returning to our model , given that an antigenic mutation arises in a strain carrying some deleterious mutation load , one might expect the virus population's deleterious mutation load to increase in the long run , causing a long-term decline in influenza A/H3N2 fitness . Two processes exist , however , that can keep this long-term accumulation of deleterious mutations from occurring . First , as the virus population becomes less fit the proportion of non-antigenic mutations that are beneficial is likely to increase , such that the mutation-selection balance becomes an evolutionary attractor ( Goyal et al . , 2012 ) . Second , within-subtype viral reassortment could occur sufficiently frequently to keep any long-term decline in viral fitness at bay by combining segments with low mutational load onto the same genetic background . The possibility that influenza A/H3N2 has been declining in fitness over time may , however , also be entertained: a recent virological analysis of human H3N2 viruses points towards a long-term decrease ( since 1968 ) in the propensity of the virus to bind human sialic acid receptors ( Lin et al . , 2012 ) . This decrease has been invoked to explain the reduction in this virus's disease impact over the last 10 years ( Lin et al . , 2012 ) . Whether this finding can be interpreted as evidence for a ‘weakening’ of the virus over time is unclear , especially because ‘weakening’ by any epidemiological measure has not been empirically demonstrated . Clearly , more virological studies are needed to determine the fitness trajectory of influenza A/H3N2 in humans over the past decades . While we focused on the role of deleterious mutations in shaping influenza A/H3N2's antigenic evolution , the presence of circulating deleterious mutations should also impact the adaptive evolutionary dynamics of other flu types/subtypes in humans . For example , influenza B is known to have a lower mutation rate than influenza A/H3N2 ( Nobusawa and Sato , 2006; Sanjuán et al . , 2010 ) . Given that the deleterious mutation rate λ is likely to decrease with the overall mutation rate , we would expect influenza B to carry a lower mean mutation load and further to have lower fitness variance . As such , we would expect smaller antigenic mutants to be able to successfully establish ( Figure 1D ) . Our model would therefore predict that influenza B evolves antigenically in a less punctuated manner than influenza A/H3N2—a pattern that has been recently documented ( Bedford et al . , 2014 ) . All else equal , we would also expect influenza B to have a faster rate of antigenic evolution ( Figure 1E ) . However , a lower mutation rate would also surely reduce the rate at which antigenic mutations occur . The slower rate of antigenic evolution observed for influenza B ( Bedford et al . , 2014 ) is therefore not inconsistent with our model . Relative to these two influenza viruses , human influenza A/H1N1 shows a similar pattern to H3N2 in terms of punctuated antigenic evolution ( Bedford et al . , 2014 ) . Despite this similarity , H1N1's rate of antigenic evolution is considerably slower than H3N2's , although it is faster than influenza B's rate of antigenic evolution ( Bedford et al . , 2014 , 2015 ) . The observed difference in rate of antigenic evolution between H3N2 and H1N1 is unlikely to reflect differences in these virus's mutation rates , since both are influenza A subtypes . Instead , these differences may stem from differences in their basic reproductive rates and , therefore , differences in selection pressures . This explanation is consistent with the finding that H1N1 appears to experience weaker antigenic selection than H3N2 in humans ( Bhatt et al . , 2011 ) . Alternatively , the lower rate of antigenic evolution in H1N1 relative to H3N2 may be a consequence of differences in these viruses' global circulation patterns , which have only recently been described ( Bedford et al . , 2015 ) . Differences in the evolutionary dynamics of influenza viruses also exist across host species . For example , the same H3N2 virus that is circulating in humans emerged in pigs in the early 1970s . This swine influenza A/H3N2 evolves genetically at a rate that is similar to that of human influenza A/H3N2 ( de Jong et al . , 2007 ) . However , its rate of antigenic evolution is approximately six times slower than the same virus's in humans ( de Jong et al . , 2007 ) . This difference in the rate of antigenic evolution likely stems more from the stark ecological differences between the two hosts , rather than from differences in their deleterious mutation loads , with escape from humoral immunity being a less important evolutionary driver of HA in short-lived hosts than in humans . Beyond the influenza viruses , circulation of deleterious mutations has been established in a wide range of RNA viruses ( Pybus et al . , 2007 ) . The evolutionary dynamics of many of these viruses are characterized by spindly phylogenies and punctuated phenotypic changes . Whether deleterious mutations can account for these apparently similar patterns remains an open question . One especially intriguing case is that of norovirus , for which punctuated antigenic evolution has been documented ( Lindesmith et al . , 2008 , 2011 ) and for which deleterious mutations along with differential binding to histoblood group antigens may contribute to fitness variation ( Donaldson et al . , 2008; Lindesmith et al . , 2008 ) . Similar to the case of influenza , an interplay between antigenic and non-antigenic/deleterious mutations may alone be sufficient to explain this viral evolutionary pattern , obviating the need to invoke the mutation-limited process of epochal evolution ( Siebenga et al . , 2007; Lindesmith et al . , 2008 ) . In addition to helping us understand patterns of viral adaptive evolution , acknowledging the ‘rubbish around the ruby’—to paraphrase ( Peck , 1994 ) —may also help us predict the course of adaptive evolution . Indeed , two recent publications have made great strides in predicting the genetic evolution of influenza A/H3N2 in humans over the short term ( Neher et al . , 2014; Łuksza and Lässig , 2014 ) . Łuksza and Lässig's approach incorporated knowledge of antigenic and non-antigenic sites in developing a fitness model for this virus . With the assumption that mutations at non-antigenic sites were weakly deleterious , the authors were able to predict which influenza clades would grow and which ones would decline with an accuracy of 93% and 76% , respectively ( Koelle and Rasmussen , 2014; Łuksza and Lässig , 2014 ) . Instead of using knowledge specific to influenza A/H3N2 , Neher et al . ’s approach relied on branching patterns present in the HA phylogeny to predict strain evolution , with a success rate comparable to that of Łuksza and Lässig . The ability of both of these models to predict influenza evolution rests on the presence of substantial fitness variation in the influenza A/H3N2 virus population . The work here contributes to this understanding by reconciling the presence of fitness variation in the virus population with the observation of limited genetic and antigenic diversity of influenza A/H3N2's HA in the long run: cluster transitions will only succeed when large antigenic mutations find themselves in good genetic backgrounds , which must be largely determined by non-antigenic phenotypes . Successfully predicting cluster transitions will therefore require not only better characterizing the antigenic effects of mutations but also characterizing relative virus fitness , as mediated by differential deleterious mutation loads and contributions of non-antigenic phenotypes , in influenza's HA and other gene segments . An integrative understanding of these non-antigenic components of viral fitness will require the continued work of virologists and modelers alike , and ideally their interaction , for predicting viral evolution . The deleterious mutation-selection balance was classically derived in the context of a discrete generation , constant-size Wright–Fisher population ( Haigh , 1978 ) . We here briefly re-derive the deleterious mutation-selection balance for a virus population in an explicit epidemiological context . To do so we consider a virus population subject exclusively to deleterious mutations . Because we are not considering antigenic variation at this point , the epidemiological dynamics are governed by a basic SIR model . In this model , the number of susceptible hosts increases only through births into the host population and decreases through background mortality and infection of susceptible hosts . The number of infected hosts increases through infection of susceptible hosts and decreases through recovery from infection and through natural mortality of infected hosts . The number of recovered hosts increases through recovery of infected individuals and decreases through natural mortality of recovered hosts . To extend this basic SIR model to allow for a virus population undergoing deleterious mutations , we classify infected individuals according to the number of deleterious mutations harbored by the virus they carry . This classification makes the implicit assumption that an infected host harbors a genetically homogeneous virus population; that is , in this case , that there is no variation in the number of deleterious mutations carried by distinct virions that make up the intrahost viral population . Although the assumption of a genetically homogeneous within-host virus population is clearly a simplifying assumption , it is an assumption that is commonly made in population-level models of influenza evolution ( Andreasen et al . , 1997; Gog and Grenfell , 2002; Ferguson et al . , 2003; Koelle et al . , 2006; Bedford et al . , 2012 ) . As in Haigh ( 1978 ) , we assume that mutations occur at birth , which , in the context of a virus population , are disease transmission events . We choose this approach to introduce new deleterious mutations over an approach that assumes that infected individuals can ‘mutate’ from being infected with a virus having a specified number of deleterious mutations to being infected with a virus having a higher number of deleterious mutations . This is because it is unlikely that a deleterious mutation that arises within a host could rapidly sweep to fixation in that host once the intrahost viral population is large . Introducing deleterious mutations at disease transmission events is therefore a more biologically reasonable assumption . Again as in Haigh ( 1978 ) , we further let the number of deleterious mutations that arise at ‘birth’ be Poisson distributed with mean λ , where λ is the per-genome per-transmission deleterious mutation rate . With these assumptions , the rate of change in the number of infected individuals carrying a virus with k deleterious mutations is given by: ( 1 ) dIkdt=SN∑j=0k ( βk−je−λλjj ! Ik−j ) − ( μ+γ ) Ik , where the first term in this equation captures the increase in the number of individuals infected with virus in mutation class k arising from the transmission process and the second term captures the decrease in the number of infected individuals resulting from background mortality ( at per capita rate μ ) and recovery from infection ( at per capita rate γ ) . In the first term , S is the number of susceptible hosts , N is the ( constant ) host population size , and βi is the transmission rate of a virus carrying i deleterious mutations . The Poisson term e−λλjj ! provides the probability that j deleterious mutations occur at transmission . As in Haigh ( 1978 ) , we assume multiplicative fitness effects of deleterious mutations , with each deleterious mutation exacting a fitness cost of size sd . We can thus write βi in terms of the transmission rate of a virus in the highest fitness class ( i . e . , the lowest deleterious mutation class ) : ( 2 ) βi=β0 ( 1−sd ) i . The dynamics of susceptible hosts are given by: ( 3 ) dSdt=μN−μS−SN∑k=0∞βkIk , where the first term captures births into the host population ( at a per capita rate μ , equal to the background mortality rate ) , the second term captures background mortality of susceptible hosts , and the third term captures depletion of susceptible hosts through the transmission process . The dynamics of recovered individuals are simply given by dRdt=γ∑k=0∞Ik−μR , where the first term captures the increase in the number of recovered individuals through recovery of infected individuals and the second term captures background mortality . For this set of equations , we can define the basic reproductive rate R0 , k as the expected number of secondary infections ( belonging to any mutation class ) produced by a single infected individual carrying a virus with k deleterious mutations in a completely susceptible population . This mutation class-specific basic reproductive rate is given by R0 , k=βk ( μ+γ ) . To solve for epidemiological equilibrium , we can now consider the dynamics of the first infected mutation class k = 0: dI0dt=β0e−λSNI0− ( μ+γ ) I0 . Setting this equation to zero , the equilibrium fraction of the population susceptible to infection can be solved for: ( 4 ) S^N= ( μ+γ ) β0e−λ , S^N is greater than 1R0 , k=0 by a factor of 1e−λ . That the fraction of susceptible hosts exceeds the inverse of the basic reproductive number of the highest-fitness virus class makes sense as the highest-fitness virus class occupies only a fraction of the total virus population . The equilibrium number of infected individuals carrying virus with k deleterious mutations , I^k , can now be solved . Substituting Equations 2 , 4 into Equation 1 and simplifying yields:dIkdt= ( μ+γ ) Itot∑j=0k ( ( 1−sd ) k−jλjj ! pk−j ) − ( μ+γ ) Ik , where the total number of infected individuals is given by Itot=∑i=0∞Ii and pi is the proportion of infected individuals carrying virus with i deleterious mutations , pi=IiItot . Setting this equation to 0 and solving for pk yields:pk=∑j=0k ( ( 1−sd ) k−jλjj ! pk−j ) . This equation mirrors Equation 4 in reference ( Haigh , 1978 ) , with solution: ( 5 ) pk=e−θθkk ! , where θ = λ/sd . The total number of infected individuals at equilibrium , I^tot , can now be solved by substituting Equations 2 , 4 into Equation 3 , and replacing Ik with pkItot . Setting equal to 0 and solving for Itot yields:I^tot=e−λμN ( 1−μ+γβ0e−λ ) ( μ+γ ) ∑k=0∞ ( 1−sd ) kpk . Substituting Equation 5 into the above expression and simplifying yields: ( 6 ) I^tot=μN ( μ+γ ) ( 1−μ+γβ0e−λ ) . The equilibrium number of mutation class-specific infected individuals can then be calculated using I^k=pkI^tot for any deleterious mutation class k . Absolute fitness in epidemiological models is provided by the net reproductive rate R . With a population experiencing deleterious mutations , each mutation class k will have its own net reproductive rate Rk at equilibrium . Given the above definition of the basic reproductive rate for viruses carrying k deleterious mutations , the fitness cost associated with deleterious mutations ( Equation 2 ) , and the equilibrium fraction of susceptible hosts ( Equation 4 ) , the mutation class k net reproductive rate is given by: ( 7 ) Rk=R0 , k ( S^N ) =eλ ( 1−sd ) k . As expected , the mean net reproductive rate of the virus population ( ∑k=0∞ ( Rkpk ) ) is 1 when the population is at epidemiological equilibrium , with some viruses having a net reproductive rate above 1 and other viruses having a net reproductive rate below 1 . The variance in the net reproductive rate is given by ∑k=0∞ ( Rk2pk ) −1 , which increases with increases in λ and increases with decreases in sd . To compute the antigenic mutant's initial and final net reproductive rates , we first use an epidemiological model to mathematically describe the interaction between the resident antigenic strain and the antigenic mutant . We denote the resident strain with super- and subscripts r and the antigenic mutant with super- and subscripts m , and use a history-based model ( Andreasen et al . , 1997 ) to specify the immunological interaction between these two strains . Note here that we again make the implicit assumption that the intrahost viral population is genetically homogeneous with respect to both antigenicity and the number of deleterious mutations carried . With deleterious mutations accumulating at transmission , we have:dS0dt=μN−μS0−S0N∑k=0∞ ( βk ( I0 , kr+IB , kr+I0 , km+IA , km ) ) , dI0 , krdt=S0N∑j=0k ( βk−je−λλjj ! ( I0 , k−jr+Im , k−jr ) ) − ( μ+γ ) I0 , kr , dI0 , kmdt=S0N∑j=0k ( βk−je−λλjj ! ( I0 , k−jm+Ir , k−jm ) ) − ( μ+γ ) I0 , km , ( 8 ) dSrdt=γ∑k=0∞I0 , kr−μSr−σSrN∑k=0∞ ( βk ( I0 , km+Ir , km ) ) , dSmdt=γ∑k=0∞I0 , km−μSm−σSmN∑k=0∞ ( βk ( I0 , kr+Im , kr ) ) , dIm , krdt=σSmN∑j=0k ( βk−je−λλjj ! ( I0 , k−jr+Im , k−jr ) ) − ( μ+γ ) Im , kr , dIr , kmdt=σSrN∑j=0k ( βk−je−λλjj ! ( I0 , k−jm+Ir , k−jm ) ) − ( μ+γ ) Ir , km , dS{r , m}dt=γ∑k=0∞ ( Ir , km+Im , kr ) −μSr , m , where Sx is the number of uninfected individuals who have previously been infected with strain ( s ) x , and Ix , ky is the number of individuals previously infected with strain x who are currently infected with a virus of strain y carrying k deleterious mutations . The parameter σ quantifies the extent to which susceptibility to infection with one strain is affected if the individual has previously been infected with the other strain . With σ = 1 a previous infection does not reduce susceptibility to infection with the second strain , whereas with σ = 0 a previous infection results in complete protection from reinfection with a second infection . A higher value of σ therefore corresponds to a greater degree of immune escape ( i . e . , a larger antigenic change ) . As expected , at the time immediately prior to the antigenic mutant's emergence , Equation 8 reduce to Equations 1 , 3 . Given Equation 8 , an antigenic mutant that arises in a background with i deleterious mutations initially has a net reproductive rate of: ( 9 ) Rm ( t=0 ) = ( βiμ+γ ) ( S0N+σSrN ) , where we have defined time t in terms of the time since the antigenic mutant's emergence . With S^0N= ( μ+γ ) β0e−λ ( Equation 4 ) and with S^rN≈1−S^0N , the mutant's initial net reproductive rate is given by: ( 10 ) Rm ( t=0 ) =eλ ( 1−sd ) i ( 1+σ ( β0e−λμ+γ−1 ) ) . This expression would be exact if we allowed for coinfection , treating individuals currently infected with the resident strain similarly to individuals previously infected with the resident strain . Equation 10 consists of three components: ( i ) eλ is the net reproductive rate of resident strain viruses that are in mutational class k = 0; ( ii ) ( 1 − sd ) i quantifies the extent to which the antigenic mutant's initial reproductive rate is reduced as a result of the deleterious mutations it carries; and ( iii ) the term ( 1+σ ( β0e−λμ+γ−1 ) ) quantifies the extent to which the antigenic mutant's initial reproductive rate is increased as a result of immune escape . To make the link stronger between this model and traditional population genetic models , we can define the selective advantage of an antigenic mutant at the time of its emergence as sb=σ ( β0e−λμ+γ−1 ) . The reproductive rate of a strain carrying an antigenic mutation necessarily decreases as it establishes through its own accumulation of deleterious mutations . Neglecting any changes in the host immune landscape , the final mean reproductive rate of the antigenic mutant lineage is: ( 11 ) Rm ( t=∞ ) =eλ ( ∑j=i∞pj−i ( 1−sd ) j ) ( 1+σ ( β0e−λμ+γ−1 ) ) , where the second term of the product quantifies the extent to which the antigenic mutant lineage's reproductive rate is reduced as a result of the deleterious mutations it carries once it has reached its own deleterious mutation-selection balance . Because this term ignores the possibility of stochastic loss of mutation class i during the strain's establishment , Equation 11 is an upper estimate for Rm ( t = ∞ ) . Changes in the host immune landscape would change the third term of the product . Under this simple model , successful establishment can only occur when Rm ( t = ∞ ) > 1 . From Equation 11 , it is therefore clear that the probability of an antigenic mutant's establishment is higher the lower its original mutation load i . This is because , in the absence of any compensatory or back-mutations , i provides the lower limit to the deleterious mutation load carried by the new antigenic strain . It is also clear from this equation that the probability of an antigenic mutant's establishment is higher the greater the ability of the mutant strain to reinfect previously infected individuals , reflected in a higher σ . Two evolutionary parameters determine influenza's deleterious mutation load: the per-genome per-transmission deleterious mutation rate λ and the transmission fitness cost of a deleterious mutation sd . Neither of these parameters have been estimated specifically for influenza . We therefore do our best with estimating them using existing data from other viruses or via rough estimates that incorporate existing knowledge about influenza . The parameter λ was roughly computed by first multiplying the genome size of influenza's major coding regions ( 12 , 741 nucleotides [Holmes et al . , 2005] ) with the empirical estimate of 2 . 3 × 10−5 for the number of substitutions that occur per nucleotide per cell infection in influenza A viruses ( Sanjuán et al . , 2010 ) . This yielded a per-genome , per-transmission substitution rate of 0 . 293 . With roughly 2/3 of mutations being non-synonymous and roughly 50% of non-lethal , non-synonymous mutations being deleterious ( Sanjuán et al . , 2004 ) , we calculated the per-genome , per-transmission deleterious mutation rate λ as the product: 0 . 293×23×0 . 5≈0 . 10 . Given the uncertainty in λ , we look across a range of λ values from λ = 0 to λ = 0 . 20 in Figure 1D , E . To get a rough estimate of the transmission fitness cost of deleterious mutations for influenza , we start with empirical estimates of mean fitness cost values for the five viruses studied in Sanjuán ( 2010 ) : 0 . 103 , 0 . 107 , 0 . 112 , 0 . 126 , and 0 . 132 . These fitness costs are in vitro fitness costs associated with viral growth in cells . These costs therefore differ from transmission fitness costs at the population level , which is how sd is defined in the models we present . To obtain a rough estimate for transmission fitness cost from these within-host viral growth fitness costs , we used a published within-host model of influenza dynamics ( Baccam et al . , 2006 ) to simulate viral load dynamics under two scenarios: a ‘wild-type’ parameterization and a ‘deleterious mutant’ parameterization , where the mutant carries a specified fitness cost . The ‘wild-type’ scenario used parameter values that were estimated in Baccam et al . ( 2006 ) using viral load data from six individuals who were experimentally infected with influenza . To parameterize the ‘deleterious mutant’ scenario , we assumed that the in vitro fitness cost manifested itself through a reduction in the within-host viral production rate . We therefore set all of the within-host parameters of the ‘deleterious mutant’ scenario to be equal to the ones used in the ‘wild-type’ scenario , with the exception of the viral production rate . This parameter we set to the product of the ‘wild-type’ scenario's viral production rate and the quantity 1 minus the in vitro fitness cost . For the six patients in Baccam et al . ( 2006 ) , we simulated the ‘wild-type’ and the ‘deleterious mutant’ scenarios for each of the five in vitro fitness costs listed above . We then mapped these simulated viral load dynamics to between-host transmission rates by assuming that viral transmission rates are proportional to the log of viral load: β ( τ ) ∝log10 ( V ( τ ) ) . This assumption is commonly used ( see Handel et al . , 2013 ) and has some empirical support , particularly from studies of HIV ( Quinn et al . , 2000; Fraser et al . , 2007 ) . The population-level transmission fitness costs sd for the five empirical within-host fitness cost estimates from Sanjuán ( 2010 ) could thus be inferred for each of the patients studied in Baccam et al . ( 2006 ) . Our simulations indicate , first and foremost , that relatively large fitness costs within a host translate into much smaller fitness effects at the population level . This is in part unsurprising given that we assume that transmission rates scale with the log of the viral load . We further found that in vitro fitness costs in the last three patients studied in Baccam et al . ( 2006 ) would translate into fitness benefits in terms of transmission . Rather than trusting this to be the case biologically , this paradoxical result is likely to be a result of the rough mapping that we ( and others ) use for translating between viral load and transmissibility . We therefore restricted ourselves to the first three patients studied in Baccam et al . ( 2006 ) . For them , we found that within-host fitness costs of 0 . 103–0 . 132 yielded transmission fitness costs between 0 . 002 and 0 . 018 , with a mean transmission fitness cost sd of 0 . 008 , which we use in Figure 1A–C . Given the uncertainties present in our estimate of sd , we instead show three different values for sd in Figure 1D , E . These indicate that the qualitative effects of circulating deleterious mutations ( larger average antigenic sizes of mutants that establish and slower rates of antigenic evolution ) are robust to specific values for sd . In addition to these two evolutionary parameters , the models we present contain three epidemiological parameters: the birth/death rate μ , the recovery rate γ , and the basic reproductive rate of a virus in the k = 0 mutation class ( R0 , k = 0 ) . We use a birth/death rate of μ = 1/30 years−1 , based on crude birth rate estimates of approximately 33 per 1000 individuals per year ( United Nations , 2011 ) . This value for μ is also consistent with a recent flu modeling study ( Bedford et al . , 2012 ) . The recovery rate γ was chosen based on an in-depth meta-analysis of 56 human challenge studies ( Carrat and Flahault , 2007 ) . In this study , the authors found an average duration of viral shedding of 4 . 8 days for influenza viruses ( regardless of subtype ) . They also calculated an average generation time of 3 . 1 days for influenza A/H3N2 , where generation time is defined as the time between an individual becoming infected and transmitting the virus . Because , in an epidemiological model with a constant transmission rate and a constant recovery rate , the duration of infectiousness ( taken to be the duration of viral shedding ) and the generation time take the same value when a virus is endemically circulating , we chose an intermediate value between these estimates , letting the recovery rate γ = 1/4 days−1 . We chose an R0 , k = 0 of 2 . 25 . This value falls within the range of recent R0 estimates for influenza A/H3N2 ( range = 1 . 21–3 . 58 for influenza A/H3N2's second pandemic wave [Jackson et al . , 2010] ) . Figure 1D , E require specification of a distribution for the degree of immune escape σ . We assume that σ is gamma distributed with mean of 0 . 012 and a shape parameter of 2 . The choice of a gamma distribution is based on a virological study showing that the distribution of beneficial mutation effects appear to be gamma distributed ( Sanjuán et al . , 2004 ) . Our choice of mean σ value is not based on an independent empirical estimate as relevant data to parameterize this value to our knowledge do not exist . Assuming an exponential distribution with a mean of 0 . 012 for σ instead of a gamma distribution did not change the qualitative findings that circulating deleterious mutations act to slow antigenic evolution and make it more punctuated in nature . Figures 2 , 3 in the main text show stochastic simulations of the epidemiological model given by Equation 8 . To simulate the model stochastically , we used Gillespie's τ-leap method with a time step of 1 hr . These stochastic simulations required further specification of a host population size N . We used a population size of N = 4 billion hosts , corresponding to the human population size around 1980 , which can also be roughly considered today's tropical population size . To explore the full evolutionary dynamics of our model with non-antigenic and antigenic mutations , we implemented an individual-based model that allowed for an arbitrary number of new strains to enter the population and co-circulate . Individuals in this model were categorized as either currently infected or currently uninfected . Using the same notation as in history-based multi-strain models ( Andreasen et al . , 1997 ) , each currently uninfected individual carries a list of antigenic types with which he has previously been infected . Each infected individual similarly carries this list of previously experienced antigenic types . In addition to this list , each infected individual carries a current viral infection that is characterized by the infecting virus's deleterious mutation load and its antigenic type . Again , we assume for the sake of model simplicity that the intrahost viral population is genetically homogeneous such that a single deleterious mutation load and a single antigenic type suffice in characterizing a viral infection . Upon recovery , infected individuals add to their strain history list the antigenic type of the viral infection from which they are recovering . Viral antigenic types are defined by their relationship to one another in terms of their pairwise cross-immunity values σ . The minimum antigenic distance between the challenging strain and the host's repertoire of previously encountered strains ( as provided by the host's antigenic type list ) determines the probability of a currently uninfected host becoming infected by the challenging strain . Specifically , the probability of infection given contact with a host infected with strain i was given by pij = min ( 1 . 0 , σij ) , where strain j is the strain in the host's repertoire that is antigenically closest to strain i . This formulation leads to complete immunity to reinfection with antigenic types that a host has previously experienced and complete susceptibility to infection for naive hosts . To compute any σij value , the model uses tracked parent–offspring relationships of distinct antigenic variants , with mother–daughter variants differing by one antigenic mutation having a degree of cross-immunity that is drawn from the specified antigenic size distribution ( see below ) . The degree of cross-immunity σij between two strains i and j is assumed to be additive; for example , if strain i gives rise to strain l and strain l gives rise to strain j , then σij = σil + σlj . When they occurred , both antigenic and non-antigenic mutations occurred during transmission events . A virus in an infected host therefore never changed antigenically or changed in mutation load during the course of infection . At a transmission event , the number of non-antigenic mutations was drawn from a Poisson distribution with mean λ . Each of the non-antigenic mutations was characterized as beneficial ( with probability εc ) or deleterious ( with probability 1 − εc ) , where εc << 1 . The net change in the number of deleterious mutations was then calculated and the virus's mutation load was appropriately increased or decreased . Our phylodynamic simulations required that a proportion of non-antigenic mutations were beneficial rather than deleterious because of the finite number of hosts we were computationally able to simulate . Specifically , in finite populations it is well known that the lowest-load mutation classes will at some point be stochastically lost , initiating Muller's ratchet ( Muller , 1964 ) . Recent work has indicated , however , that if a fraction of mutations are beneficial rather than deleterious , the average mutation load in a population can be maintained indefinitely over time ( Goyal et al . , 2012 ) . In our simulations , we set εc to a value of 0 . 16 , which resulted in a stable deleterious mutation load over time . The number of antigenic mutations occurring at a transmission event was drawn from a Poisson distribution with mean λantigenic , independently of the number of non-antigenic mutations . We used a λantigenic of 0 . 00075 . The size of each of these antigenic mutations was drawn from a gamma distribution with a mean of 0 . 012 and a shape parameter of 2 . The antigenic difference σij between the virus i in the infecting host and the virus j in the host becoming infected was calculated as the sum of the sizes of the antigenic mutations occurring at transmission . The simulations were run using an individual-based stochastic simulation algorithm based on Gillespie's tau-leap algorithm , using a time step τ of 1 day . Experimentation with smaller time steps yielded similar results , such that we used a τ of 1 day for computational efficiency . Over each time interval τ , the number of events occurring during that interval was drawn from a Poisson distribution . The modeled events were births , deaths , recoveries , and infectious contacts . Individuals born into the population were considered naive to infection ( such that their strain history lists were empty ) . Deaths occurred from both uninfected and currently infected hosts , indiscriminately . For each infected host death , an infected individual was randomly chosen to be discarded from the population , independent of the virus currently infecting the individual . Similarly , for each recovery , an infected individual was chosen at random to recover . For each infectious contact from individuals infected with a class-k virus , an infected individual was chosen at random from the current class-k infected population and a currently uninfected individual was also chosen at random . The uninfected host was then infected with probability pij , where , as described above , pij is given by min ( 1 . 0 , σij ) , where σij is the antigenic distance between infecting strain i and the strain j in the host's repertoire that is antigenically closest to strain i . We did not allow for coinfection , so our phylodynamic simulations did not include the possibility of viral reassortment leading to antigenic cluster transitions . For these individual-based simulations we used a population size of N = 40 million individuals , 1/100th of the human population size in 1980 . Our choice of this limited population size was due to the large amount of memory required to keep track of the immune histories of each individual host . The relationships of who-infected-whom and at what time were tracked such that the true phylogenetic tree could be reconstructed and compared against influenza A/H3N2's HA phylogeny reconstructed from empirical sequences . Preliminary simulations revealed highly volatile population dynamics , with large peaks in prevalence followed by extinction or long periods of low prevalence . To stabilize the dynamics we allowed a small number of infectious contacts to occur from outside of the focal population by having uninfected individuals experience an additional force of infection from an external pool of infected individuals ( M = 200 for all simulations ) . So as not to alter the evolutionary dynamics , this external pool of infected individuals was assumed to have the same mutational load distribution and antigenic type frequencies as the focal population . All remaining evolutionary and epidemiological parameters used in these individual-based simulations were the same as the ones listed in the model parameterization section above . The simulations were implemented in Java using a modified version of the program Antigen ( http://bedford . io/projects/antigen/ ) . The original code was modified to track the mutation load and the antigenic phenotypes of the viral population , which in our case requires us to track the pairwise distances between all antigenic types instead of viral locations on a two-dimensional antigenic surface . Source code for our full phylodynamic model is available on the GitHub repository at https://github . com/davidrasm/MutAntiGen . git . In our simulations , we tracked the fitness of individual viruses in terms of their net reproductive rate R . For a particular virus i carrying k deleterious mutations , R=βk ( μ+ ν ) SeffN , where Seff is the effective number of susceptible hosts available to a particular virus . Because the susceptibility of a host to a particular virus depends on the host's detailed immune history , to compute Seff we first compute the probability ρij of strain i infecting each host in the population upon contact , and then sum ρij over all uninfected hosts in the population to arrive at Seff . In practice , for computationally tractability , we only sum over a large number ( n = 10 , 000 ) of randomly sampled hosts to approximate Seff . Given the distribution of R in the viral population , we can then compute the fraction of viral fitness variation attributable to deleterious mutations ( i . e . , variation in βk ) vs antigenic differences among strains ( i . e . , variation in Seff ) . To decompose the total variance in fitness into these components , we first log-transform R , so thatln ( R ) =ln ( βk ) +ln ( SeffN ) +ln ( 1 ( μ+ν ) ) . Dropping the constant 1 ( μ+ν ) term , we can then decompose the total variance in log R into its component parts using the well-known relation that the variance of the sum of two random variables is equal to the sum of their variances plus their covariance , var ( R ) =var ( βk ) +var ( SeffN ) +2cov ( βk , SeffN ) . Note that the covariance term cannot be ignored because βk and SeffN are not independent random variables and can be correlated due to the shared common ancestry of viral strains ( i . e . , phylogenetic correlations ) . To compute the fraction of total viral fitness variation attributable to βk and thus deleterious mutations , we subtract the variance attributable to the covariance:var explained by βk= var ( βk ) var ( R ) −2cov ( βk , SeffN ) . We also conducted additional simulations that did not include any antigenic evolution but did include deleterious mutations to determine if the spindly phylogeny shown in Figure 6 arose simply due to the presence of purifying selection rather than being driven by antigenic change . Simulations without antigenic evolution were conducted with the same evolutionary and epidemiological parameters; average viral R0 was as in the simulations with antigenic evolution except λantigenic was set to zero . However , because there was no antigenic evolution , the average number of infected humans over time and the annual attack rates were significantly lower than in the simulations with antigenic evolution . To compensate for this , in Figure 9 , we allowed immunity to wane over time , as in standard SIRS epidemiological models . The rate of immune waning was set such that immunity from a previous infection lasted 12 years on average before the host became completely susceptible again . This rate of immune waning was chosen to produce an average annual attack rate of ∼5–10% , consistent with what was observed in the simulations with antigenic and non-antigenic mutations .	Each year , up to 15% of the world's population experience symptoms of an influenza infection , also commonly known as flu . The most common culprit is a strain of the virus called influenza type A subtype H3N2 . One reason that so many people become infected each year is that this virus evolves rapidly . Within a few years , proteins on the surface of the virus known as antigens become less recognizable to the immune system of a person who has been previously infected . This means that the person can become ill with the virus again because their immune system cannot mount an effective response to the evolved virus strain . Influenza virus strains evolve rapidly because their genetic material accumulates mutations quickly . Although some of these mutations are beneficial to the virus , other mutations are harmful and reduce the ability of the virus to spread . Sometimes beneficial mutations may occur alongside harmful ones , but it is not known how the harmful mutations affect the evolution of the virus . Here , Koelle and Rasmussen used computer models of H3N2 influenza to examine the effect of harmful mutations on the evolution of this virus population . The models show that harmful mutations limit how quickly the antigens can evolve . Also , the presence of these harmful mutations effectively acts as a sieve: they allow only large changes in the antigens to establish in the virus population . The models suggest that there are three routes by which large changes in the antigens on H3N2 viruses may occur . The first is by a single mutation that has a big effect on the antigens in viruses that only carry a few harmful mutations , but these large mutations would not happen very often . Another route may be through more common mutations that have only a small or moderate benefit , which would allow the virus to become more common in the population before it acquires a beneficial mutation with a much greater effect . The third possibility is that a large beneficial mutation may arise in viruses that have many harmful mutations . These harmful mutations may initially limit the ability of the virus to spread , but over time , some of these harmful mutations may then be lost . Koelle and Rasmussen found that the computer models could recreate the patterns of virus evolution that have been observed in real strains of H3N2 . Researchers use predictions of influenza evolution to help them decide which virus strains should be included in flu vaccines each year . Koelle and Rasmussen findings indicate that harmful mutations should be considered when making these predictions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "epidemiology", "and", "global", "health" ]	2015	The effects of a deleterious mutation load on patterns of influenza A/H3N2's antigenic evolution in humans
The number of children born since the origin of Assisted Reproductive Technologies ( ART ) exceeds 5 million . The majority seem healthy , but a higher frequency of defects has been reported among ART-conceived infants , suggesting an epigenetic cost . We report the first whole-genome DNA methylation datasets from single pig blastocysts showing differences between in vivo and in vitro produced embryos . Blastocysts were produced in vitro either without ( C-IVF ) or in the presence of natural reproductive fluids ( Natur-IVF ) . Natur-IVF embryos were of higher quality than C-IVF in terms of cell number and hatching ability . RNA-Seq and DNA methylation analyses showed that Natur-IVF embryos have expression and methylation patterns closer to in vivo blastocysts . Genes involved in reprogramming , imprinting and development were affected by culture , with fewer aberrations in Natur-IVF embryos . Methylation analysis detected methylated changes in C-IVF , but not in Natur-IVF , at genes whose methylation could be critical , such as IGF2R and NNAT . “Most fertility researchers are trying to improve Assisted Reproductive Technologies ( ARTs ) success as measured by a single , clear standard: the birth of an apparently healthy baby . Only a few are trying to discern whether in vitro fertilization ( IVF ) leaves a subtle legacy in children . What will happen to these kids when they are middle-aged ? ” ( Servick , 2014 ) . In humans , according to a study by the World Health Organisation ( WHO ) in 190 countries , infertility affects 20% of couples and it was estimated that at least 40 . 5 million women were seeking infertility medical care in 2007 ( Mascarenhas et al . , 2012 ) . ARTs provide a helpful alternative for a high proportion of infertility cases and the number of children born to date using these methods exceeds 5 million ( International Committee for Monitoring Assisted Reproductive Technology I , 2012 ) . Although the majority of them seem healthy , studies have reported higher rates of preterm births ( Rubens et al . , 2014 ) , non-chromosomal birth defects and adverse perinatal effects in ART pregnancies ( El Hajj and Haaf , 2013 ) , with long-term effects being under study in humans ( Kissin et al . , 2014 ) . Epidemiological data suggest that perturbed epigenetic gene regulation by the application of ART could be a contributory factor in these adverse outcomes ( El Hajj and Haaf , 2013; Whitelaw et al . , 2014 ) , although such alterations could also be considered as consequences of parental characteristics , gamete quality or other non-epigenetic technique-derived effects ( Simpson , 2014 ) . To clarify the impact of each of these factors , the use of an animal model that avoids , as much as possible , the effect of parental circumstances and the use of protocols minimizing the technique-derived effects would help to attain the goal of offering safer ART for patients . For modeling ART-related disorders in human , swine could be a good candidate for several reasons: their genetic , anatomical and physiological similarities with human ( Swindle et al . , 2012 ) ; their size and length of gestation; and the availability of individuals genetically selected by their excellent reproductive performance in artificial insemination centres . Importantly , this last trait could be useful to remove the paternal factor ( low-quality male gametes ) from studies as a possible reason for any epigenetic alterations found . However , most protocols for processing boar spermatozoa for IVF include their selection by density gradient centrifugations and just a few used the swim-up procedure to isolate highly motile spermatozoa which is the routine selection in human infertility clinics . Since it was observed that spermatozoa selected by swim-up show higher rates of normal morphology and motility , and decreased DNA fragmentation and methylation levels ( Kim et al . , 2015 ) , it would be necessary to adapt the sperm selection protocols in pig before using them to model ART-derived epigenetic alterations . In both mouse and human , accumulating evidence indicates that the embryo is sensitive to its very early environment and that culture media used in ART ( as factors involved in technique-derived effects ) may have long-lasting consequences ( Kleijkers et al . , 2014; Fernandez-Gonzalez et al . , 2004 ) . Several imprinting disorders and abnormal phenotypes have been linked to ART , but of special significance is the relationship between the presence of serum in culture media and the incidence of Large Offspring syndrome ( LOS ) in ruminants ( Young et al . , 1998 ) , which includes diverse pathologic alterations and shows phenotypic and epigenetic similarities with the imprinted disorder Beckwith-Wiedemann syndrome ( BWS ) in humans ( Chen et al . , 2013 ) . Since it was proposed that serum in the culture medium could be a crucial factor in LOS incidence , the tendency in the procedures for both human and livestock was to move toward the use of chemically defined media , limiting the presence of proteins in the culture medium to serum albumin . Although practical , this approach may have unpredictable consequences , because it ignores the fact that the reproductive fluids have a different composition to serum and are extremely rich in proteins other than serum albumin ( more than 150 have been described in the oviductal fluid [Avilés et al . , 2010] ) . If these proteins are physiologically present , they must play a variety of roles supporting the normal development of the embryo , roles that serum albumin alone cannot properly provide and serum cannot fully mimic . In addition , although ART in species such as cattle and sheep usually results in foetal overgrowth ( Young et al . , 2001; Chen et al . , 2015 ) , opposing phenotypes such as low birth weights ( excluding BWS ) are often seen in humans ( Schieve et al . , 2002 ) and pigs ( García-Vázquez et al . , 2010 ) . A study showing the relationship between child birth weight and the protein source in embryo culture media ( Zhu et al . , 2014 ) reinforces the hypothesis that the protein composition of the culture media plays a role in the correct regulation of epigenetic marks in the growing embryo . A similar conclusion can be reached from a clinical trial showing that protein enrichment of media compared with addition of serum albumin alone improved the blastocyst implantation rate and may increase human births by more than 8% ( Meintjes et al . , 2009 ) . Therefore , as with breast milk , which is so complex and so rich in bioactive factors that cannot be easily replaced with any artificial composition ( Hennet and Borsig , 2016 ) , the idea that reproductive secretions could be necessary in the culture media should not be underrated . At least , it should be explored under experimental conditions to unveil the relevance of these secretions . DNA and RNA sequencing have become affordable cutting-edge technologies that could help to understand the mechanisms underlying abnormalities observed in ART-derived offspring . However , so far , single blastocyst whole-genome DNA methylation profiles comparing in vivo and in vitro produced embryos have not been published for any mammalian species and we therefore aim to produce these in this study . We report here that modified swim-up protocols for the selection of spermatozoa in pigs and the use of reproductive secretions as additives in the culture media significantly increase the yield and quality of the blastocysts produced from a morphological , epigenetic and gene expression point of view . Using genome-wide analyses of gene expression by RNA-Seq and DNA methylation by Bisulfite-Seq in single blastocysts , we provide datasets of pig blastocysts produced in vitro with and without reproductive secretions as additives in the culture medium and show that the former are more similar to the in vivo specimens than the later . This suggests an alternative approach for conceiving healthier ART-derived children . In order to select spermatozoa before IVF , a swim-up protocol was set up and compared with a conventional selection system by density gradient centrifugations . To do this , it was necessary to design a suitable washing and sperm selection medium imitating , as far as possible , in vivo conditions ( NaturARTs PIG sperm washing medium and NaturARTs PIG sperm swim-up medium , EmbryoCloud , Murcia , Spain ) . The swim-up medium was supplemented either with bovine serum albumin ( BSA ) ( Swim-up BSA group ) or porcine oviductal fluid ( POF , Swim-up fluid group ) collected at the late follicular ( LF ) phase of the estrous cycle ( NaturARTs POF-LF , EmbryoCloud , Murcia , Spain ) ( Figure 1 ) . All the fluids used in this study were directly aspirated from the lumen of ovarian follicles , oviducts or uterus and processed according to the information described in the Materials and methods section , at http://embryocloud . com , and in previous references ( Coy et al . , 2008 ) . 10 . 7554/eLife . 23670 . 003Figure 1 . Schematic representation of three different sperm processing protocols used for in vitro fertilization . Swim-up-BSA: NaturARTs PIG medium + BSA; Swim-up-Fluid: NaturARTs PIG medium + POF-LF* . Density gradient centrifugation: centrifugation through a discontinuous Percoll: gradient ( 45% and 90% v/v ) . POF-LF: porcine oviductal fluid collected at the late follicular phase of the estrous cycle . Red box represents the portion of the reproductive tract whose conditions we tried to resemble in vitro . IVF results after using these three different sperm processing protocols are included in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 23670 . 003 Polyspermy after IVF is a major issue in the pig ( Coy and Avilés , 2010 ) . With these new protocols , we obtained significantly higher rates of monospermy than with conventional ones ( 49 . 6 ± 4 . 5 vs 17 . 4 ± 4 . 1 , Table 1 ) and the final percentage of putative zygotes ( evaluated at 24 hours post insemination , hpi ) was significantly higher ( 35 . 2 ± 0 . 2 vs 14 . 6 ± 0 . 1 , Table 1 ) . Moreover , the addition of POF-LF to the Swim-up media instead of BSA increased the final yield of the system ( 35 . 2 ± 0 . 2 vs 29 . 7 ± 0 . 2 , Table 1 ) . 10 . 7554/eLife . 23670 . 004Table 1 . IVF results after using three different sperm processing protocols ( Density gradient , Swim-up-BSA and Swim-up-Fluid ) as represented in Figure 1 . a , b: Different letters in the same column indicate values statistically different ( p<0 . 05 ) . Penetration: proportion of oocytes penetrated by one or more spermatozoa . Monospermy: Monospermy percentage , calculated from penetrated oocytes , represents the proportion of penetrated oocytes with only one spermatozoon inside the ooplasm . Spermatozoa/Oocyte: Mean number of sperm per penetrated oocyte . Spermatozoa/ZP: Mean number of spermatozoa attached to ZP per oocyte . Yield: Percentage of putative zygotes per oocyte . DOI: http://dx . doi . org/10 . 7554/eLife . 23670 . 004Sperm processing methodNPenetration ( % ) Monospermy ( % ) Spermatozoa/OocyteSpermatozoa/ZPZygote yield ( % ) Density gradient centrifugation10584 . 3 ± 3 . 6a17 . 4 ± 4 . 1a8 . 4 ± 0 . 7a17 . 3 ± 2 . 3a14 . 6 ± 0 . 1aSwim-up-BSA18069 . 6 ± 3 . 5b42 . 7 ± 4 . 6ab2 . 1 ± 0 . 1b7 . 2 ± 0 . 5b29 . 7 ± 0 . 2bSwim-up-Fluid18371 . 1 ± 3 . 4b49 . 6 ± 4 . 5b2 . 7 ± 0 . 1b8 . 6 ± 0 . 5b35 . 2 ± 0 . 2c In a second experiment , and using the Swim-up protocol for sperm selection , a new IVF/Embryo culture ( EC ) system ( Natur-IVF ) was developed , which included preincubation of oocytes in oviductal fluid ( NaturARTs PIG OF-LF ) and the presence of reproductive fluids as additives in the IVF and EC media ( 0–8 hr: NaturARTs POF-LF; 8–48 hr: oviductal fluid from the early luteal–EL- phase of the estrous cycle , NaturARTs POF-EL; 48–180 hr: uterine fluid -UF-from this same phase , NaturARTs PUF-EL ) ( Figure 2 ) . Corresponding controls with BSA instead of OF/UF for each step ( referred as C-IVF group ) were analyzed ( Figure 2 ) . Evaluation at 24 hpi revealed higher penetration rate ( 66 . 6 ± 0 . 1 vs 43 . 7 ± 0 . 1 , p<0 . 05 ) and similar monospermy rate ( 78 . 6 ± 0 . 1 vs 72 . 7 ± 0 . 1 , p<0 . 05 ) for the Natur-IVF and C-IVF groups , respectively . Regarding embryo development , more than 40% of cleaved embryos reached the blastocyst stage in both groups ( Table 2A ) . However , the Natur-IVF blastocysts showed a significant increase in their mean number of cells ( 81 . 8 ± 7 . 2 , Table 2A ) compared to the C-IVF ones ( 49 . 9 ± 3 . 7 ) , and this number was similar to that observed in the in vivo samples ( In-vivo group , 87 . 0 ± 7 . 2 ) . Moreover , at day 7 . 5 , embryos reaching the hatching or hatched stages were observed only in the Natur-IVF group ( Table 2B ) . Taken together , these data indicate a higher quality , in terms of cell number and ability to hatch , in the ART-derived blastocysts when reproductive fluids were added to the culture medium . 10 . 7554/eLife . 23670 . 005Figure 2 . Schematic representation of the different steps of the new IVF/EC system . Swim-up-BSA or Swim-up-Fluid protocols were used for IVF . Previously , oocytes were preincubated in OF-LF for 30 min . Then , each group of putative zygotes were incubated in different media ( 0–8 hr , 8–48 hr and 48 hr-7days ) as indicated in the diagram . O: ovary with hemorrhagic corpus luteum; O*: early corpus luteum; OF-LF: oviductal fluid-late follicular phase of the estrous cycle; OF-EL: oviductal fluid-early luteal phase of the estrous cycle; UF-EL: uterine fluid-early luteal phase of the estrous cycle . Swim-up-BSA: NaturARTs PIG medium + BSA; Swim-up-Fluid: NaturARTs PIG medium + POF-LF . TALP: culture medium used for IVF . NCSU23: culture medium used for embryo development in vitro supplemented with sodium lactate , pyruvate and non-essential amino acids ( NCSU23a ) or with glucose and essential and non-essential amino acids ( NCSU23b ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23670 . 00510 . 7554/eLife . 23670 . 006Table 2 . ( A ) Comparative results of IVF yield by using BSA ( C- IVF ) or reproductive fluids ( Natur-IVF ) as additives in the culture medium for 7 . 5 days . ( B ) Results of blastocyst development ( for each type ) using BSA ( C- IVF ) or reproductive fluids ( Natur-IVF ) as additives in the culture medium for 7 . 5 days . Columns from ‘Early blastocyst’ to ‘Hatched blastocyst’ indicate the percentage of each type of blastocyst from Total blastocyst ( Table 2A ) , classified according to Bo and Mapletoft25 . a , b: Different letters in the same column indicate values statistically different ( p<0 . 05 ) . Cleavage: Cleavage percentage from N . Total Blastocysts: Percentage of blastocysts calculated from cleaved embryos . Yield: Percentage of putative blastocysts from N . Cell/blastocyst: mean number of cells per blastocyst . DOI: http://dx . doi . org/10 . 7554/eLife . 23670 . 006A ) GroupNPenetration ( % ) Monospermy ( % ) Cleavage ( % ) Total blastocysts ( % ) Blastocyst Yield ( % ) Cell/ blastocystIn vivo41 87 . 0 ± 7 . 2bC- IVF903395 ( 43 . 7 ± 0 . 1a ) 656 ( 72 . 7 ± 0 . 1 ) 429 ( 47 . 5 ± 1 . 6a ) 178 ( 41 . 4 ± 2 . 4 ) 19 . 6 ± 1 . 349 . 9 ± 3 . 7aNatur-IVF961640 ( 66 . 6 ± 0 . 1b ) 755 ( 78 . 6 ± 0 . 1 ) 405 ( 42 . 1 ± 1 . 6b ) 180 ( 44 . 5 ± 2 . 5 ) 18 . 7 ± 1 . 281 . 8 ± 7 . 2b B ) GroupNEarly blastocyst ( % ) Blastocyts ( % ) Expanded blastocyst ( % ) Hatching blastocyst ( % ) Hatched blastocyst ( % ) C- IVF17857 ( 31 . 7 ± 6 . 1 ) a50 ( 28 . 3 ± 5 . 9 ) 71 ( 40 . 0 ± 6 . 4 ) 0 ( 0 ) a0 ( 0 ) aNatur -IVF18023 ( 12 . 8 ± 5 . 4 ) b55 ( 30 . 8 ± 7 . 5 ) 65 ( 35 . 9 ± 7 . 8 ) 28 ( 15 . 4 ± 5 . 9 ) b9 ( 5 . 1 ± 3 . 6 ) b In vitro culture systems significantly alter embryonic gene expression as previously observed in pooled pig blastocysts ( Bauer et al . , 2010 ) . Here , the transcriptomes from three individual day 7 . 5 blastocysts produced by C-IVF or Natur-IVF were compared with their in vivo counterparts ( Figure 3A–B ) . RNA libraries showed acceptable quality in all nine blastocysts . Mean number of raw reads was 14 . 24 ± 2 . 23 ( ±SD ) millions , and transcripts from 13 , 309 to 14 , 512 different genes ( from a total of 20 , 789 annotated pig mRNAs ) were detected in each individual . Principal Component Analysis ( PCA ) showed that , despite expected individual variability , the three embryos from each group clustered together ( Figure 3B ) , with the C-IVF embryos showing higher variability , which could represent high embryo plasticity in response to suboptimal culture conditions . Therefore , after combining the triplicates , data from both in vitro groups showed high correlation ( Pearson correlation coefficient [r] = 0 . 964 ) , but Natur-IVF was closer to the In vivo group ( [r] = 0 . 95 ) than C-IVF ( [r] = 0 . 938 ) . RNA-Seq data analysis ( DESeq2 p<0 . 05 after multiple testing correction ) identified 787 differentially expressed genes ( DEG ) between the C-IVF and In-vivo , and 621 DEGs between Natur-IVF and In vivo ( Figure 3—source data 1 , including also all the expression values for all the genes ) . Of the genes that were significantly different ( adjusted p-value < 0 . 05 , fold change > 1 . 5 ) in the pair-wise comparisons , there was a higher number of up-regulated ( 534/787–68%- in C-IVF embryos and 431/621–69%- in Natur-IVF ) than down-regulated ( 253 and 190 , respectively ) ( Figure 3C , Figure 3—source data 1 ) . 10 . 7554/eLife . 23670 . 007Figure 3 . Gene expressed analysis in blastocysts obtained in vivo , by the Natur-IVF system or by C-IVF system . ( A ) Heatmap of global gene expression ( with log2 fold change >1 . 5 and adjusted B-H p-value < 0 . 05 ) . Numbers denote ID of a specific embryo . ( B ) Principal Component Analysis ( PCA ) of the RNA-Seq samples: In vivo embryos ( IV , red ) , Natur-IVF ( N , green ) and C-IVF ( C , blue ) . Numbers denote ID of specific embryos . ( C ) Venn diagram with DEGs ( Figure 3—source data 1 ) . , # , § denotes DEGs exclusive for C-IVF , Natur-IVF and In vivo , respectively ( Figure 3—source data 2 ) . ( D ) Heat map of gene expression of key genes associated with embryo development/differentiation , epigenetic reprogramming , cell cycle/cell growth , gene expression and imprinting . DOI: http://dx . doi . org/10 . 7554/eLife . 23670 . 00710 . 7554/eLife . 23670 . 008Figure 3—source data 1 . Differentially expressed genes ( DEGs ) for pair-wise comparisons ( C-IVF vs Natur-IVF , In vivovs Natur-IVF , C-IVF vs In vivo ) and list of all gene expression values . This data relates to Figure 3C . DOI: http://dx . doi . org/10 . 7554/eLife . 23670 . 00810 . 7554/eLife . 23670 . 009Figure 3—source data 2 . DEGs exclusives for each group: 328 DEG exclusive In vivo , 7 DEGs exclusive Natur-IVF and 13 DEGs exclusive C-IVF . This data relates to Figure 3C . DOI: http://dx . doi . org/10 . 7554/eLife . 23670 . 009 Top Canonical Pathways , Physiological Systems and Molecular and Cellular Functions related to DEGs were identified ( summarized in Supplementary file 1 ) using the Ingenuity Pathway Analysis ( IPA ) software . Globally , down-regulated genes in C-IVF and in Natur-IVF were linked to similar Top-cellular functions ( Supplementary file 1 ) . Equally , top Canonical Pathways affected by up-regulated genes were similar for both groups . In contrast , two pathways were identified in down-regulated DEGs in C-IVF embryos , but not in Natur-IVF DEGs ( Supplementary file 1 ) . Increased pathways in Natur-IVF and C-IVF included cholesterol , mevalonate , serine and glycine biosynthesis and p53 signaling . Decreased pathways ( protein ubiquitination and 14-3-3 mediated signaling ) were detected only in C-IVF . Similarly , Physiological Systems and Functions over-represented by up-regulated or down-regulated DEGs were different between C-IVF or Natur-IVF . These results show that , in spite of similarity , there were differences that could influence specific pathways and affect key molecular and cellular functions in the embryos from each group . Natur-IVF and C-IVF blastocysts shared 334 genes that were aberrantly expressed in both groups vs In vivo ( Exclusive DEGs , Figure 3C- Figure 3—source data 2 ) . However , there were 440 genes ( from the 784 DEGs in C-IVF ) that showed aberrant expression only in C-IVF vs In vivo ( DEGs only in C-IVF , Figure 3C ) , while 40% fewer genes ( n = 281 from the 620 DEGs in Natur-IVF ) showed aberrant expression only in the Natur-IVF group vs In vivo ( DEGs only in Natur-IVF , Figure 3C ) . Importantly , several genes related to epigenetic reprogramming ( down: DNMT3B , DNMT1; up: HDAC5 , KDM5A ) , embryo development ( down: CTGF , ING2 , KIT , EZH2; up: BMP4 , TLN1 , ADAR ) , cell growth ( down: CDCA5 , SMC1A; up: RB1 , SMARCA2 ) or imprinting ( up: IGF2BP2 , GNAS; down: DIRAS3 ) were amongst the C-IVF-specific DEGs ( Figure 3D ) . Direct comparison between Natur-IVF vs In vivo and C-IVF vs In vivo DEGs revealed that only 29 genes reached significant expression differences between the two in vitro groups after DESeq2 analysis ( Figure 3—source data 1 ) . Interestingly , of these 29 DEGs , 13 were similarly expressed in Natur-IVF and In vivo , and only seven showed similar expression between C-IVF and In vivo groups ( Figure 3C , Figure 3—source data 2 ) . Although the number of these genes was low , they could be critical because among the 13 genes exclusively different in the C-IVF blastocysts ( Figure 3—source data 2 ) , those down-regulated ( n = 6 ) were KIT , MPPA6 , MTA3 , KIF4A , UBR2 and ISOC1 ( Log Fold Change from −5 . 9 to −54 . 18 ) . For all six genes data were available for the corresponding knock-out mice or knock-down studies , which showed phenotypes of altered/abnormal growth/size , reproduction/fertility , mortality/aging , hematopoietic system , homeostasis/metabolism and other abnormalities ( Supplementary file 2 ) . These data suggest that in vitro culture significantly alters embryonic gene expression to a lesser extent than previously proposed ( Bauer et al . , 2010 ) , and a better modulation of the blastocyst transcriptome was achieved by mimicking physiological conditions of fertilization and early embryo development by the addition of reproductive fluids ( Natur-IVF ) . In this study , for the first time , whole-genome DNA methylation profiles on individual porcine blastocysts were generated by a low-cell adaptation of the post-bisulphite adaptor-tagging ( PBAT ) method ( Miura et al . , 2012; Peat et al . , 2014 ) . Three blastocysts from each group were analyzed . The number of unique alignments in the samples ranged from 13 , 150 , 508 to 42 , 208 , 651 and the coverage of CpGs ( ≥1 read ) from 52% to 59 . 2% . The global methylation percentages of CpGs were 15 . 02 ± 3 . 3 , 11 . 09 ± 2 . 6 and 12 . 33 ± 3 . 6 for the C-IVF , Natur-IVF and In-vivo groups , respectively . The distribution of methylation levels in windows of 150 CpGs across the genome and a general view of the methylation profiles of the nine individual blastocysts are shown in Figure 4A–B . The generally low level of methylation suggests that the genome has experienced substantial loss of methylation from the gametes , analogous to that observed in other mammals ( Guo et al . , 2014; Kobayashi et al . , 2012 ) . The landscape of methylated cytosines suggests some structure across the genome , with regions with more methylation consistent between the individual blastocysts ( Figure 4B ) . What contributes to this structure , for example , the regions of relatively higher methylation , is not immediately obvious , as methylation was similar in different genomic contexts with no marked enrichment in repetitive elements , for example ( Table 3 ) . Regarding the different classes of blastocysts , methylation over specific genomic features followed the same tendency as the global differences , with higher values for C-IVF ( Table 3 ) . 10 . 7554/eLife . 23670 . 010Figure 4 . Distribution of methylation levels and general view of the methylation profiles of 9 individual pig blastocysts . ( A ) Distribution of methylation percentages across tiles of 150 CpGs on the pig genome for three groups of blastocysts ( In-vivo , C-IVF and Natur-IVF ) . ( B ) Random browser shot as example of methylation landscape of the nine individual blastocysts analysed ( Chr8:37027152–118458156 ) . The two first rows in the picture represent the genes and CpG islands annotated ( Ensembl , RRID:SCR_006773 Sus scrofa 10 . 2 ) in the pig genome , respectively . Color scale represents methylation levels from red ( highest methylation , up to 25% ) to blue ( lowest methylation-0% ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23670 . 01010 . 7554/eLife . 23670 . 011Table 3 . Percentages of methylation over genome features in porcine blastocysts produced in vitro ( C-IVF and Natur-IVF ) or collected in vivo ( In vivo ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23670 . 011% MethylationIn vivoC-IVFNatur-IVFCpG islands9 . 6911 . 8010 . 11Promoters9 . 2611 . 619 . 11TU12 . 8415 . 4712 . 36Intergenic11 . 7514 . 4811 . 37LINE112 . 6315 . 4312 . 02LTR12 . 7715 . 5312 . 06SINE12 . 4515 . 3011 . 94GLOBAL12 . 3315 . 0211 . 09 PCA revealed a good level of clustering for In-vivo and Natur-IVF embryos but not for C-IVF embryos ( Figure 5A ) . In particular , embryos C34 and C36 were far from the other seven embryos analyzed . 10 . 7554/eLife . 23670 . 012Figure 5 . DNA-methylation analysis in blastocysts obtained in vivo , by the Natur-IVF system or by C-IVF system . ( A ) Principal Component Analysis ( PCA ) of the DNA methylation samples: In vivo embryos ( red ) , Natur-IVF ( green ) and C-IVF ( blue ) . Numbers denote ID of specific embryo . ( B ) Venn diagram of DMRs by pair-wise comparison ( adjusted-p <0 . 05 ) . Number of DMRs with higher ( ↑ ) or lower ( ↓ ) methylation in each pair-wise comparison are indicated ( Figure 5—source data 1 ) . ( C ) Heatmap of the 417 DMRs between the C-IVF group and the other two groups ( In vivo and Natur-IVF ) . ( D ) Heatmap of the 324 DMRs between Natur-IVF group and the other two groups ( In vivo and C-IVF ) . ( E ) Heatmap of the 448 DMRs between the In vivo group and the other two groups ( Natur-IVF and C-IVF ) . For C , D and E ( Figure 5—source data 2 ) : Relative methylation measure as the difference in percent of methylation from the median methylation across all samples . DOI: http://dx . doi . org/10 . 7554/eLife . 23670 . 01210 . 7554/eLife . 23670 . 013Figure 5—source data 1 . All differentially methylated regions ( DMRs ) for each pair-wise comparison ( C-IVF vs Natur-IVF , In vivo vs Natur-IVF , C-IVF vs In vivo ) . This data relates to Figure 5B . DOI: http://dx . doi . org/10 . 7554/eLife . 23670 . 01310 . 7554/eLife . 23670 . 014Figure 5—source data 2 . Differentially methylated regions ( DMRs ) exclusive for each group ( C-IVF , Natur-IVF , In vivo ) . This data related to Figure 5B , C , D and E . DOI: http://dx . doi . org/10 . 7554/eLife . 23670 . 014 The low level of global methylation suggested that few differentially methylated regions ( DMRs ) could be found . For this reason , and to obtain an unbiased measure of differences in genome methylation , a fixed size of 150 CpGs was used for analysis , as this was found to give a modal tile size of around 3 kb with about 150 reads per tile for most individuals . To make the data comparable to enable the detection of DMRs , separately from the global changes , the tiles informative in all samples ( 258 , 885 ) were extracted and quantile normalized . To identify DMRs , the comparison was filtered to require a consistent ≥5% absolute methylation change between all replicates of the first and second condition , followed by a T-test ( B-H adjusted p<0 . 05 ) . Differences between the groups were observed with fewer than 4000 DMRs for each pair-wise comparison ( Figure 5—source data 1 ) . Globally , fewer DMRs showed higher methylation in In vivo vs Natur-IVF ( n = 1 , 660 ) than in In vivo vs C-IVF ( n = 2244 ) ( Figure 5B ) . To better characterize the changes in methylation exclusively affecting one of the groups ( p<0 . 05 for both comparisons ) , the corresponding subsets of DMRs ( ‘exclusive’ DMRs for each group ) were obtained by combining the previous lists ( Figure 5B , C , D and E; Figure 5—source data 2 ) , and the enrichment in specific features in those DMRs was evaluated ( Supplementary file 3 ) . For the three subsets of DMRs , there was a lower proportion of promoters compared to the global average ( p<0 . 001 ) . A lower proportion of LINE1s ( p<0 . 05 ) was also found for the C-IVF group , while the Natur-IVF blastocyst group showed a higher proportion of DMRs in transcription units ( defined over the annotated genes from 500 bp downstream of the annotated TSS , p<0 . 05 ) . Both C-IVF and Natur-IVF DMRs were less enriched in intergenic regions ( p<0 . 001 ) and at LTRs ( p<0 . 05 ) than In vivo blastocysts . These departures from the methylation state might reflect global differences in the DNA methylation and/or demethylation capacity of the different groups at a developmental time when DNA methylation is rather dynamic . Exclusive DMRs for each group were linked to Canonical Pathways ( p<0 . 01 ) and Diseases and Bio Functions ( adjusted p-value < 0 . 05; Figure 6 ) by IPA software . Representative genes for specific DMRs in each group are listed in Supplementary file 4 . A DMR overlapping IGF2R , a gene directly related with the LOS in ruminants and mouse , was found in the subset of exclusive C-IVF DMRs ( Figure 5—source data 2 ) . The methylation percentages for this region ( Chr1: 9 , 199 , 522–9 , 201 , 143 ) were 12 . 45% , 28 . 3% and 35 . 5% for C-IVF , Natur-IVF and In vivo , respectively ( Figure 7A ) . In addition , a CpG island ( oe = 0 . 89 , Chr1:9 , 200 , 658–9 , 202 , 276 ) that overlapped the DMR showed significant differences in methylation ( p<0 . 05 ) : 14 . 1% , 27 . 8% and 29 . 4% for C-IVF , Natur-IVF and In vivo groups , respectively ( Figure 7B ) , although we should be cautious about their significance since the CpG island distribution in the pig genome is very different to the human or mouse genome . 10 . 7554/eLife . 23670 . 015Figure 6 . Top Diseases and Bio Functions linked by Ingenuity Pathways Analysis to DMRs exclusive for each group with low or high methylation . DOI: http://dx . doi . org/10 . 7554/eLife . 23670 . 01510 . 7554/eLife . 23670 . 016Figure 7 . Methylation differences at IGF2R . ( A ) Methylation quantitation at IGF2R from the unbiased analysis of genome methylation in SeqMonk with a fixed size of 150 CpG windows . Mean percentages of methylation are shown by the bars for each group . Blue ( unmethylated ) and red ( methylated ) dots represent methylation reads . Asterisks indicate that methylation at the indicated region showed significantly different values ( p<0 . 05 ) in Natur-IVF ( * ) and In vivo ( ** ) vs C-IVF . TSS: transcription starting site . ( B ) Detailed view and methylation quantitation of the CpGi at the identified IGF2R DMR . Red rectangles represent , as indicated , CpG islands of the genes . Black boxes indicate the position of the targeted features , whose mean percentages of methylation are shown by the bars for each group . Blue ( unmethylated ) and red ( methylated ) dots represent methylation reads . DOI: http://dx . doi . org/10 . 7554/eLife . 23670 . 016 Top Diseases and Bio Functions linked by IPA to DMRs exclusive for each group with low or high methylation are represented in Figure 6 . Top Molecular and Cellular Functions and representative genes related to DMRs with higher or lower methylation in each group ( C-IVF , Natur-IVF and In vivo ) are listed in Supplementary file 4 . Following the finding of a DMR at IGF2R , targeted analysis of candidate imprinted genes was done , as the differentially methylated regions of imprinted gene ( igDMRs ) are expected to maintain constant methylation in preimplantation embryos to ensure faithful imprinted expression of the associated genes throughout development . Therefore , they represent sites of methylation in preimplantation of clear biological significance . To identify putative igDMRs in the pig genome , all mouse igDMRs were lifted-over onto the pig genome . Where this was not possible , a gene-by-gene approach was taken to find the best possible fit for a candidate igDMR based on the known organization of the corresponding mouse imprinted gene . All the genomic regions were then inspected manually to confirm that the correct regions had been found ( Table 4A ) . It is not possible to conclude that all regions were actually igDMRs ( as this would require methylation information from oocyte and sperm ) and , indeed , the methylation values indicated that for some of the genes there was no conserved DMR ( i . e . methylation in blastocysts was far below the theoretical 50% ) and the associated locus was unlikely to be imprinted . This would seem to be the case , for example , for the genes IMPACT , ZFP787 and ZFP777 . For some , there was difficulty in finding possible homologous igDMRs , probably because of gaps in the porcine genome assembly ( such as SNRPN , KCNQ1 and GRB10 ) , and there were a number of others that were excluded because the homologous pig region had no suggestion of a CpG island in the region equivalent to the igDMR in mouse ( e . g . U2AF1-RS1 , MCTS2/H13 ) . Comparison of methylation in the three groups of blastocysts for the resulting 14 candidate igDMRs ( with sufficient read coverage ) revealed differences for ZAC1 and PEG10 , which were more methylated ( p<0 . 05 ) in the C-IVF than in In vivo group , and PEG10 and NNAT , which were more methylated ( p<0 . 05 ) in the C-IVF than in Natur-IVF and In vivo groups ( Table 4B ) . No statistical differences were found between Natur-IVF and In vivo groups . Of these three igDMRs , the one at NNAT coincides with the promoter CpG island ( Kobayashi et al . , 2012 ) and , in addition , one 150 CpG tile overlapping NNAT had methylation higher than 50% in C-IVF in the unbiased analysis ( Figure 8 ) . 10 . 7554/eLife . 23670 . 017Figure 8 . Methylation quantitation at NNAT from the unbiased analysis of genome methylation in SeqMonk with a fixed size of 150 CpG windows . Black boxes indicate the position of the selected 150 CpG windows , whose mean percentages of methylation are shown by the bars for each group . Blue ( unmethylated ) and red ( methylated ) dots represent methylation reads . Asterisks indicate that methylation at the indicated region ( black box ) showed significantly different values ( p<0 . 05 ) in Natur-IVF ( * ) and In-vivo ( ** ) vs C-IVF . DOI: http://dx . doi . org/10 . 7554/eLife . 23670 . 01710 . 7554/eLife . 23670 . 018Table 4 . Targeted analysis of candidate imprinted genes . ( A ) Predicted imprinted regions in the pig genome by lifted-over mouse igDMRs the pig genome and manually inspected . ( B ) Pair-wise comparison of methylation by Analysis Chi-Square in the three groups of blastocysts for the resulting 14 candidate igDMRs . C-IVF vs In vivo: p<0 . 05 with 20 minimum observations and 10 minimum percentage of difference % methylation . * C-IVF vsNatur-IVF: Analysis Chi- Square p<0 . 05 with 20 minimum observations and 10 minimum percentage of difference % methylation . Natur-IVF vs In vivo: no statistical differences . DOI: http://dx . doi . org/10 . 7554/eLife . 23670 . 018A ) TileChromosomeStartEndIGF2R/AIR19 , 244 , 2399 , 248 , 054ZAC1123 , 638 , 88723 , 643 , 228SOCS5399 , 885 , 36099 , 887 , 132ZFP787655 , 574 , 08055 , 575 , 926ZIM2656 , 641 , 19056 , 644 , 823IMPACT6102 , 001 , 929102 , 002 , 533NAT1l58139 , 773 , 830139 , 775 , 461PEG10981 , 642 , 95781 , 644 , 146INPP5FV214141 , 186 , 219141 , 188 , 231NNAT1746 , 041 , 84346 , 045 , 629NESPAS1766 , 313 , 67366 , 320 , 932GNAS-exon1a1766 , 348 , 00966 , 352 , 062MEST1819 , 340 , 33519 , 345 , 549ZFP7771860 , 941 , 42160 , 943 , 096 B ) TileChromosomeC-IVFNatur-IVFIn-vivoZAC1142 . 4133 . 5523 . 87PEG10947 . 7536 . 9130 . 90NNAT1734 . 6319 . 22*23 . 28 The milieu in which fertilization and embryo development takes place is crucial for healthy fetal and offspring growth , as revealed by developmental and epigenetic alterations as a consequence of in vitro culture and ART ( Fernández-Gonzalez et al . , 2004; Kleijkers et al . , 2014; Lazaraviciute et al . , 2014; Song et al . , 2015 ) . However , the progress made by ART during the past two decades make a future without their use inconceivable , thus it is necessary i ) to characterize the real epigenetic cost of ART , separated from other factors and ii ) to develop new protocols to safeguard against possible negative impacts in offspring . Our study evaluated , by single blastocyst profiling , the genetic and epigenetic impacts of modified protocols to produce embryos in vitro that mimic , as far as possible , the physiological conditions of fertilization and early embryo development . This imitation of the natural environment was first approached in both gametes separately: in the male gamete , by using sperm selection procedures that avoided centrifugations , and sperm washing and processing media containing oviductal fluid from the pre-ovulatory phase of the cycle; and , in the female gamete , by preincubating the oocytes within the precise fluid they encounter when , after ovulation , they are transported through the ampulla of the oviduct to the fertilization site , at the ampullar-isthmic junction ( Halbert et al . , 1988 ) . Secondly , two experimental groups were established for a comparison with the in vivo specimens , where either BSA or reproductive fluids ( obtained sequentially at the corresponding phases of the cycle ) were added at every step of the IVF and EC procedures . The results showed that reproductive fluids improve the outcome of IVF and the quality of pig blastocysts produced in vitro . The approach used , with spermatozoa coming from boars selected by their excellent reproductive performance , avoids the possibility of aberrations due to a paternal factor , which cannot be avoided in the human model , and helps to elucidate the epigenetic cost of ART independently of any paternal pathology . The figure of >40% progression of the cleaved embryos to blastocysts in vitro means an improvement over the best previous results ( Redel et al . , 2016 ) . Nonetheless , the most remarkable findings were that Natur-IVF blastocysts attained a more advanced developmental stage and that the mean number of cells per blastocyst was the same as In-vivo embryos and 61% higher than C-IVF ones , which it is also above some of the best data previously reported in pigs ( Redel et al . , 2016 ) . These results indicate that the use of reproductive fluids as additives , even at the low dose used in this study ( 1% ) is beneficial for in vitro development of pig embryos so that it is now possible to obtain similar or even higher yields in the pig ( 45% ) than in the bovine species . Although the possibility of transferring these methods to the human clinic might seem far off , the fact that nowadays other natural fluids such as breast milk for baby feeding or blood serum for transfusions are collected and stored at biobanks , make it possible to predict the future availability of human reproductive fluids obtained from oocyte donors during interventions at human infertility clinics ( Coy and Yanagimachi , 2015 ) . In fact , the first samples of these fluids are already stored at Biobanc-Mur in Spain ( National Register of Biobanks N° B . 0000859 ) . Our study also showed that Natur-ART blastocysts are closer to the gene expression profile of the In vivo blastocysts than C-IVF blastocysts . Amongst the most striking differences found was the expression of genes related to epigenetic reprogramming . It has been shown in mice and human that during the transition from zygote to blastocyst there is a massive loss of DNA methylation , with the exception of imprinted genes and some repetitive elements ( Guo et al . , 2014; Reik and Kelsey , 2014 ) . In agreement with this observation , the global methylation level in the three groups of pig blastocysts analyzed was below 15% , suggesting that they had largely undergone a reprogramming event . This globally low methylation level compared to somatic cells or gametes , made it difficult to find high quantitative differences between embryos . Despite this , methylation percentage was higher in C-IVF embryos than in the other two groups , in agreement with previous studies indicating that ART-derived blastocysts displayed higher levels of methylation than in vivo derived ones ( Deshmukh et al . , 2011 ) . This difference appeared to be global , with all features affected and , no evidence of multiple sub-groups over different genomic regions; therefore , there was no indication of specific regions resisting reprogramming . At the same time , genes for DNMT1 and the binding protein of its crucial cofactor UHRF1 , which are considered responsible for maintenance of methylation patterns in replicating DNA and for maintaining imprints during preimplantation embryonic stages , were less expressed in C-IVF blastocysts , as was DNMT3B , required for de novo remethylation from this stage onwards . Differences in cell numbers , as a result of a probable additional round of cell division in In vivo and Natur-IVF embryos compared to C-IVF , is unlikely to explain a shift from ~11–12% to ~15% global methylation . All together , these data suggest an impaired demethylation in the C-IVF group . Analysis of hemimethylated CpG dyads by deep hairpin bisulfite sequencing , as recently reported in mouse ( Arand et al . , 2015 ) , could help to clarify this issue . A second key finding in this study was that the methylation levels in the samples analyzed showed much lower overall methylation levels ( mean across all samples was 13 . 1% ) than would be expected from somatic tissues . Furthermore , there were differences in the global mean methylation levels between different samples , ranging from 8 . 9% to 18 . 5% . Taken together , these observations suggest that the samples were collected during a time of global methylation reprogramming . The variability in global methylation levels would have confounded a direct comparison focussing on locus-specific methylation differences , so to account for this a quantile normalization was required to allow for a direct quantitative comparison of methylation levels . Given that these samples are undergoing active reprogramming , it is also not unreasonable to think that some previously reported DMRs may not be established yet , or that the strength of the DMRs would be reduced . Despite this , we were able to find candidate DMRs between the groups with a reasonable statistical significance , although the magnitude of the methylation differences was low . Considering that previous studies have shown extremely close correlations between qPCR and RNA-seq data ( Asmann et al . , 2009; Griffith et al . , 2010; Wu et al . , 2014 ) and that validation by qPCR has its own probe-bias based on what region of the cDNA is amplified , we deem , in contrast to microarrays data , that there is not solid evidence that validation of the RNA-Seq and DNA methylation results by qPCR will provide extra significance to our results . For this reason , we did not perform qPCR validation in this study . Another key observation in this study was that the in vitro culture affects imprinted gene expression and methylation . Plasticity of the preimplantation embryo could enable a recovery of alterations in methylation and further expression of non-imprinted genes during development , but any erosion of methylation marks at imprinted genes are unlikely to be corrected . In our data , from the 10 candidate imprinted regions retaining more than 30% of methylation in the pig blastocysts , we found three in C-IVF ( ZAC1 , PEG10 and NNAT ) with significantly different methylation compared to In vivo blastocysts , and two ( PEG10 and NNAT ) compared to Natur-IVF . Knock-out mice lacking PEG10 showed early embryonic lethality with placental defects , indicating the importance of this gene in embryonic development ( Ono et al . , 2006 ) . The protein encoded by NNAT , on the other hand , may be involved in the regulation of ion channels during brain development and may also play a role in forming and maintaining the structure of the nervous system . Defects in methylation at ZAC1 and IGF2R have been found in patients with the imprinted disorders transient neonatal diabetes mellitus ( TNDM ) or Silver-Russell syndrome ( SRS ) , respectively , including those born following the use of ART ( Le Bouc et al . , 2010 ) . In addition , genes related to the IGF axis , IGF2BP2 and IGF2BP2-IMP2 , were up-regulated in C-IVF , and IGF2R in both C-IVF and Natur-IVF embryos . Altered IGF2BP2 expression in C-IVF is of interest , since reduced abundance of IGF2 has been associated with lower fetal weight after in vitro culture ( El Hajj and Haaf , 2013 ) . The imprinting status of IGF2R in the pig is unclear ( Killian et al . , 2001; Braunschweig , 2012 ) but , independently of this uncertainty , our data indicated higher expression of this gene in the two in vitro groups of blastocysts , which would be in agreement with previous reports in other species and could indicate a possibility of LOS-related alterations observed in abnormal in vitro and cloned embryos ( Young et al . , 2001 ) . At the same time , the reduced methylation in IGF2R specifically in the C-IVF group could suggest that this group is more likely to be susceptible to sustained deregulation of IGF2R expression and a greater probability of LOS-like syndromes . Altered expression in both groups of blastocysts produced under in vitro conditions was observed in some genes related to embryonic development , but some aberrations were absent in Natur-IVF embryos . In human blastocysts , it has been observed that those with higher implantation rate and higher number of cells per embryo showed up-regulation of DNMT3A ( Kleijkers et al . , 2015 ) . In our data , the In vivo and Natur-IVF blastocysts showed a higher number of cells than those from the C-IVF group , in which expression of DNMT3A was decreased . We also observed higher expression of CDKN1A in the two in vitro groups , with an intermediate value in Natur-IVF . CDKN1A inhibits embryonic cell proliferation in response to DNA damage and it is considered one of the key genes responsible for the abnormalities in ART embryos since an aberrant increase of CDKN1A expression might be related to the growth-defect phenotype ( Ishimura et al . , 2016 ) . Methylation of the CDKN1A gene , however , was similar in all three groups , between 5 and 7% . Other genes involved in DNA repair and cell cycle regulation were found to be altered , such as MDM2 ( in C-IVF ) and TP53INP ( up-regulated in Natur-IVF and C-IVF ) and HSPA4L , HSP40B1 , HSPH1 , HSP90 ( down-regulated only in C-IVF ) . Altered expression of these genes may limit the ability of the embryo to respond to DNA damage , such that in vitro culture may lead to dysregulation of such genes , thus affecting long-term embryo viability ( Zheng et al . , 2005 ) . The same situation was found for SLC2A3 ( Glut-3 ) and SLC2A2 , which have been related to LOS ( Wrenzycki et al . , 2004 ) and were highly up-regulated in the two in vitro groups . Again , no differences at the methylation level were found for any of these genes . Although DNA methylation at the promoter/gene bodies is directly/indirectly correlated with gene expression , this is not strictly true during the periods of dramatic loss of DNA methylation , as occurs during early embryo development or primordial germ cells ( PGC ) formation . For example , Gkountela et al . ( 2015 ) showed a general uncoupling between DNA methylation and gene expression during demethylation of PGCs , commenting ‘Our data reveal a remarkable and pervasive loss of DNA methylation in human PGCs and AGCs during prenatal life that has almost no relationship to changes in gene expression’ . Comparative analyses between our methylation and gene expression data also showed this lack of correlation . In our opinion , at this stage of development and with this low level of methylation , this was an expected result . Finally , the exclusive alteration in C-IVF of genes such as KIT , whose knock-out in mouse results in multiple alterations including embryonic lethality ( Ro et al . , 2010 ) , UBR2 , whose deletion results in female embryonic lethality and growth arrest ( Kwon et al . , 2003 ) , or ISOC1 , whose mutation produces phenotypes with body weight loss ( Rainger et al . , 2013 ) , support the hypothesis that offspring produced with Natur-IVF conditions would be healthier than those produced with C-IVF , although additional studies are necessary to confirm this finding . In conclusion , we report here the first time genome-wide DNA methylation and transcription analysis in single blastocysts ( in vivo and in vitro ) of a mammalian species and propose a new strategy for prevention of aberrant epigenetic and gene expression profiles induced by ART . This strategy , based on the addition of reproductive fluids in the culture media used during the ART procedures , can be applied in other animals as well as in humans , after safety concerns of transmission of diseases have been properly addressed . The design of new culture media containing all the proteins that are naturally present in the original biological fluid , represents not only a technical challenge but a biomedical responsibility that must be addressed to prevent future pathologies both in animals and humans . In addition , we offer a new protocol for the in vitro production of pig embryos with a significant improvement over the previous data published . Our study represents a new form of thinking in the field , far from the chemically defined culture media , and could help to face one of the biggest milestones of the current reproductive medicine: safer ART . Unless otherwise indicated , all chemicals and reagents were purchased from Sigma-Aldrich Quimica S . A . ( Madrid , Spain ) . Fluids were obtained from animals raised at a commercial farm ( CEFU , S . A . , Murcia , Spain ) and slaughtered in an abattoir belonging to a food industry ( El Pozo , S . A ) near the University of Murcia . For the collection of follicular fluid , ovaries from 6-month-old Large White animals weighing 100–110 kg were transported to the laboratory in saline containing 100 μg/ml kanamycin sulfate , washed once in 0 . 04% cetrimide solution ( alkyltrimethylammoniumbromide ) and twice in saline within 30 min of slaughter . The content of follicles between 3 and 6 mm diameter , from at least 50 ovaries ( 25 females ) , was quickly aspirated , centrifuged at 1800 g for 30 min at 4°C and the supernatant filtered through 0 . 22 µm diameter filter ( Naito et al . , 1988 ) . One ml follicular fluid ( FF ) aliquots were stored at −80°C until their use as additives for the IVM medium . For the collection of oviductal ( OF ) and uterine ( UF ) fluids , genital tracts from cyclic Large White sows ( 2–4 years old ) were obtained at the abattoir and transported to the laboratory on ice within 30 min of slaughter . The cyclic stage of animals was assessed once in the laboratory , on the basis of ovarian morphology on both ovaries from the same female . Oviducts and uteri were classified as early follicular , late follicular , early luteal or late luteal phase ( Carrasco et al . , 2008 ) . Both oviducts and uteri coming from the same genital tract were classified as in the same stage of the cycle . Once classified , oviducts and uteri were separated and quickly washed once with 0 . 4% v/v cetrimide solution and twice in saline . Oviducts and uteri were dissected on Petri dishes or trays , respectively , sitting on ice . Once dissected , OF were collected by aspiration with an automatic pipette by introducing a 200 µl pipette tip into the ampulla and manually making an increasing pressure gradient from the isthmus to the ampulla . The UF was collected by making a manual increasing pressure gradient from the proximal end to the distal end ( utero-tubal junction ) of the uterine horn and letting the fluid drop into a sterile 50ml Falcon tube . Once recovered , samples ( OF and UF ) were centrifuged twice at 7000 g for 10 min at 4°C to remove cellular debris . Then the supernatant was immediately stored at −80°C until use . Oviducts from animals at the late follicular phase ( POF-LF ) and at the early luteal phase ( POF-EL ) gave a mean volume of around 50 µl and 40 µl , respectively per oviduct . At the early luteal phase , approximately 10 ml of UF per uterine horn were collected each time . Aliquots of 50 µl OF and 50 ml UF of pooled samples from at least 20 animals for OF and five animals for UF were used . Only samples that passed quality controls ( pH 7 . 0–7 . 6 , osmolality 280–320 mOsm/kg , endotoxin <0 . 10 EU/mL , a minimum 90% of Metaphase II oocytes after IVM with FF and ZP hardening for oocyte preincubation in POF-LF >1 hr ) were used for experiments . Ovaries from 6 months old animals weighing 100–110 kg were transported to the laboratory in saline containing 100 µg/ml kanamycin sulfate at 38°C , washed once in 0 . 04% cetrimide solution and twice in saline within 30 min of slaughter . Cumulus–oocyte complexes ( COCs ) were collected from antral follicles ( 3–6 mm diameter ) , washed twice with Dulbecco’s PBS ( DPBS ) supplemented with 1 mg/ml polyvinyl alcohol ( PVA ) and 0 . 005 mg/ml red phenol , and twice more in maturation medium previously equilibrated for a minimum of 3 hr at 38 . 5°C under 5% CO2 in air . Maturation medium was NCSU37 supplemented with 0 . 57 mM cysteine , 1 mM dibutyryl cAMP , 5 mg/ml insulin , 50 µM β-mercaptoethanol , 10 IU/ml equine chorionic gonadotropin ( eCG; Foligon; Intervet International BV , Boxmeer , Holland ) , 10 IU/ml human chorionic gonadotropin ( hCG; Veterin Corion; Divasa Farmavic , Barcelona , Spain ) , and 10% porcine follicular fluid ( v/v ) . Only COCs with complete and dense cumulus oophorus were used for the experiments . Groups of 50 COCs were cultured in 500 µl maturation medium for 22 hr at 38 . 5°C under 5% CO2 in air . After culture , oocytes were washed twice in fresh maturation medium without dibutyryl cAMP , eCG and hCG and cultured for an additional period of 20–22 hr . Before IVF , mature oocytes were preincubated in 100% porcine oviductal fluid ( POF ) from the late follicular ( LF ) phase ( NaturARTs POF-LF ) for 30 min ( Coy et al . , 2008 ) and then washed three times in TALP medium . TALP medium consisted of 114 . 06 mM NaCl , 3 . 2 mM KCl , 8 mM Ca-lactate . 5H2O , 0 . 5 mM MgCl2 . 6H2O , 0 . 35 mM NaH2PO4 , 25 . 07 mM NaHCO3 , 1 . 85 mM Na-lactate , 0 . 11 mM Na-pyruvate , 5 mM glucose , 2 mM caffeine , 1 mg/ml PVA and 0 . 17 mM kanamycin sulfate . Either 3 mg/ml BSA-FAF ( A-6003 ) or 1% of NaturARTs POF-LF was included as additives in the IVF medium for the first 8 hr of coincubation ( C-IVF and Natur-IVF groups , respectively ) . Ejaculated spermatozoa from boars of proven fertility ( 1–2 years old ) were transported to the laboratory and 1 ml of semen was lay below 1 ml of NaturARTs PIG sperm swim up medium ( http://embryocloud . com ) at the bottom of a conical tube . After 20 min of incubation at 37°C ( with the tube at a 45° angle ) , 0 . 75 ml from the top of the tube were aspirated and used for insemination of the IVF dishes ( 105 cells/ml ) with the oocytes . For the density gradient group , aliquots of the semen samples ( 0 . 5 ml ) were centrifuged ( 700 g , 30 min ) through a discontinuous Percoll ( Pharmacia , Uppsala , Sweden ) gradient ( 45% and 90% v/v ) and the resultant sperm pellets were diluted in TALP medium and centrifuged again for 10 min at 100 g . Finally , the pellet was diluted in TALP and 250 μl of this suspension were added to the wells containing the oocytes , giving a final concentration of 105 cells/ml . Spermatozoa and oocytes were incubated at 38 . 5°C under 5% CO2 for 8 hours . Later on , the putative zygotes were transferred to embryo culture medium . At this point , a sample of the putative zygotes from each group was collected , fixed and stained as previously described ( Coy et al . , 2008 ) to assess the fertilization rates ( percentage of penetrated oocytes , percentage of monospermy , mean number of spermatozoa penetrating each oocyte and mean number of spermatozoa attached to the zona pellucida ) . Penetration rate was defined as the proportion of oocytes penetrated by one or more spermatozoa . Media for embryo culture were NCSU23 supplemented with sodium lactate ( 5 mM ) , pyruvate ( 0 . 5 mM ) and non-essential amino acids ( NCSU23a , for the first 48 hr ) or NCSU23 supplemented with glucose ( 5 . 5 mM ) and essential and non-essential amino acids ( NCSU23b , 48–180 hr ) . At 8 hr post insemination ( hpi ) , putative zygotes were transferred to culture dishes containing NCSU23a medium and 0 . 4% BSA in the C-IVF group or 1% POF from the early luteal ( EL ) phase of the estrous cycle ( NaturARTs POF-EL ) in the Natur-IVF group . At 48 hpi , the cleavage was assessed under the stereomicroscope and the 2–4 cell stage embryos were transferred to NCSU23b with 0 . 4% BSA ( C-IVF group ) or 1% of porcine uterine fluid ( PUF ) from early luteal phase ( NaturARTs PUF-EL , Natur-IVF group ) . On day 7 . 5 ( 180 hpi ) , blastocyst stage morphology was assessed under the stereomicroscope and later on a sample was fixed and stained ( Coy et al . , 2008 ) and the remaining blastocyst were washed in PBS and frozen in PCR tubes in the minimum volume of medium . The parameters assessed in the stained blastocysts were development stage ( 2–4 cells , 8–16 cells , morula or blastocyst ) , mean number of cells per blastocyst , and ability for hatching ( rhythmic movements of expansion and contraction before going out of the zona pellucida ) . The blastocysts frozen for genetic and epigenetic study were passed through liquid nitrogen vapours for 5 s and immediately introduced in the freezer at -80°C until the day of use for RNA extraction or bisulphite treatment . Data are presented as mean ± SEM , and all percentages were modeled according to the binomial model of variables and arcsin transformation to achieve normal distribution . The variables in all the experiments were analyzed by one-way or two-way ANOVA . When ANOVAs revealed a significant effect , values were compared by the Tukey test . A pvalue < 0 . 05 was taken to denote statistical significance . Ten sows 18-month old were weaned 21 days after second parturition and five days later showed signs of standing estrous . Animals were inseminated in the collaborative farm and slaughterhoused 7 . 5 days after . Genital tracts were collected and transported to the laboratory where uterine horns were briefly dissected and washed with PBS within 2 hr from slaughtering . Blastocysts were identified under the stereomicroscope , collected and immediately frozen as described for the in vitro produced embryos . A portion of these blastocysts was fixed in glutaraldehyde and stained with Hoechst for cell counting . C-IVF group ( C-IVF ) : six blastocysts classified as 7A according to Bo and Mapletoft ( Bo and Mapletoft , 2013 ) ( #34 , 35 , 36 , 93 , 94 and 96 ) were produced in vitro with BSA as the only protein source . Sperm were processed by swim up in NaturARTs sperm medium with BSA ( Swim-up-BSA ) . IVF medium consisted of TALP ( 0–8 hr ) and embryo culture medium consisted of NCSU23a ( 8–48 hr ) and NCSU23b ( 48–180 hr ) . Natur-IVF group: six blastocysts classified as 7A ( #55 , 85 , 86 , 27 , 54 and 60 ) were produced in vitro with NaturARTs POF and PUF as the protein source . Sperm were processed by swim up in NaturARTs sperm medium with NaturARTs POF-LF ( Swim-up-Fluid ) . IVF medium consisted of TALP +1% NaturARTs POF-LF ( 0–8 hr ) and embryo culture medium consisted of NCSU23a + 1% NaturARTs POF-EL ( 8–48 hr ) and NCSU23b + 1% NaturARTs PUF-LL ( 48–180 hr ) . For both groups , before IVF oocytes were pre-incubated for 30 min in preovulatory oviductal fluid ( NaturARTs POF-LF ) . In vivo group: six blastocysts classified as 7A ( #186 , 193 , 197 , 189 , 190 and 191 ) were collected by flushing the uteri of animals within 2 hr of slaughtering . The animals were under natural heat after weaning and insemination was performed 7 days before slaughtering . ARCTURUS PicoPure RNA Isolation Kit ( KIT0204 , Life Technologies ) was used to extract the RNA from individual blastocysts . RNA-Seq libraries were generated using Ovation RNA-Seq System V2 ( NuGEN , Cat . 7102–08 ) for low amount of starting material and further amplified with NEB Next DNA Library Prep Master Mix for Illumina ( NEB , Cat . E6040S ) . All steps were performed according to manufacture guidelines . iPCRTag reverse primer with individual index was used to generate three independent biological replicates from each condition . 100 bp single end reads were sequenced on Illumina HiSeq 1000 . Sequencing data were processed . For RNA-Seq libraries , raw sequence reads were trimmed using Trim Galore to remove adapter contamination and reads with poor quality defined by low PHRED score . Mapping was performed using Tophat software ( http://tophat . cbcb . umd . edu/ ) and data were visualized with Seqmonk ( RRID:SCR_001913 , http://www . bioinformatics . babraham . ac . uk/projects/seqmonk/ ) . RNA quality was assayed by Bioanalyzer and even though each sample came from a single blastocyst , RIN score was between 6 . 1–8 . 2 . Annotated pig mRNA features were quantitated with raw read counts in SeqMonk and these were fed into DESeq2 for differential expression analysis using a p-value cutoff of 0 . 05 and not applying independent filtering . Reads were subsequently re-quantitated as log2RPM ( reads per million reads of library ) and globally normalized to the 75th percentile of the data . Significant effect sizes were selected using the Seqmonk intensity difference filter where the difference in expression in each gene was compared to the set of differences in the 1% of the data with the most similar average expression level as the gene being tested . Only genes with significantly higher changes ( p<0 . 05 after Benjamini and Hochberg correction ) were kept . An adaptation of whole genome bisulfite sequencing that involves post-bisulfite adapter tagging ( PBAT ) ( Miura et al . , 2012 ) was used to analyze the methylome of individual pig blastocysts at single-base resolution on a genome-wide scale . Further modification of the method described in Smallwood et al . ( Smallwood et al . , 2014 ) was used to generate BS-seq libraries . Briefly , an individual blastocyst was lysed for 1 hr in 1% SDS with proteinase K and treated with bisulfite reagent using Imprint DNA modification kit ( Sigma , MOD50 ) . DNA was eluted in EB buffer and one round of first strand synthesis was performed using a biotinylated oligo 1 ( 5-[Btn]CTACACGACGCTCTTCCGATCTNNNNNNNNN-3 ) . Samples were further treated with Exonuclease I , washed and eluted in 10 mM Tris-Cl and incubated with washed M-280 Streptavidin Dynabeads ( Life Technologies ) to pull down the biotinilated fraction of DNA . Second strand synthesis was performed using oligo 2 ( 5’-TGCTGAACCGCTCTTCCGATCTNNNNNNNNN −3’ ) and samples were amplified for 15 PCR cycles using indexed iPCRTag reverse primer ( Smallwood et al . , 2014 ) with KAPA HiFi HotStart DNA Polymerase ( KAPA Biosystems ) and purified using 0 . 8× Agencourt Ampure XP beads ( Beckman Coulter ) . Libraries were assessed for quality and quantity using High-Sensitivity DNA chips on the Agilent Bioanalyser , and the KAPA Library Quantification Kit for Illumina ( KAPA Biosystems ) . Three libraries generated from individual blastocysts for each experimental condition were prepared for 100 bp single-end sequencing on Illumina HiSeq 1000 and sequenced at three samples per lane . For the unbiased analysis , tiles were defined in SeqMonk using the read position tile generator tool and selecting one read count per position and 150 valid positions per window , in all the nine individual data sets ( 286 , 136 tiles ) . Then , the bisulphite quantitation pipeline was run over existing tiles , one minimum count to include position and 20 minimum observations to include feature . To remove the tiles without data , the filter on values for individual tiles was applied , where values had to be between 0 and 100 for exactly 9 of the nine selected data stores . Then , tiles with data for all the samples were obtained ( N = 258 , 885 tiles ) . Bisulphite quantitation pipeline was run again over the new tiles and data were normalized by the match distribution quantile normalization tool . Finally , every pair-wise comparison was filtered to require a consistent 5% change between all replicates of the first and second condition , and then replicate sets stats was applied where every comparison had a significance below 0 . 05 after Benjamini and Hochberg correction . For the targeted analysis of the candidate imprinted regions a Chi-Square test ( p<0 . 05 ) was applied for every comparison .	Infertility has become more common in many countries , particularly those where many people delay having children until later in life . To help individuals experiencing infertility conceive a child , scientists have developed treatments called assisted reproductive technologies ( or ARTs for short ) . So far , more than 5 million children have been born with the help of these treatments . Most of the children seem healthy; however , birth defects are more common in ART-conceived babies than those conceived without treatment . The cause of these birth defects is not known , though scientists suspect it may have something to do with techniques used in ART . One possible culprit is the liquid that is used in the laboratory to help the parents’ sperm and egg come together for fertilization . This same liquid is also used to bathe the developing embryo for the first few days after fertilization before it is implanted into its mother’s womb . Some scientists wonder whether adding the fluids normally found in the reproductive tract of their mother to this liquid could reduce defects in children conceived via ART . Now , Canovas et al . have shown that fertilizing and growing pig embryos in liquids supplemented with fluid from the wombs of female pigs results in embryos that are closer to naturally conceived pig embryos than in non-supplemented liquids . In the experiments , naturally conceived embryos were compared to ART embryos exposed to the usual liquids and with ART embryos grown in liquids with fluid collected from the pig’s reproductive tract added . Cutting edge technologies were used to sequence the entire genomes of all of the embryos and compare which genes were active in each case . Canovas et al . also looked at chemical markers on the DNA – called epigenetic changes – that turn on or off the expression of genes without changing the DNA code itself . The analysis showed that ART-conceived embryos grown in the usual liquid had different patterns of gene expression and epigenetic changes compared to naturally conceived embryos . Gene expression and epigenetic changes in the ART embryos grown with the pig reproductive fluid was more similar to the naturally conceived embryos . These findings suggest that abnormal gene expression in the ART-liquid exposed embryos may lead to birth defects , and that using natural reproductive fluids may be safer . To confirm this , scientists will have to implant embryos conceived in these three different conditions into mother pigs and assess the health and gene expression patterns of the resulting piglets . If successful , these new insights might one day lead to improvements in ART techniques used to treat infertility in people .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2017	DNA methylation and gene expression changes derived from assisted reproductive technologies can be decreased by reproductive fluids
Protein-protein interactions involving intrinsically disordered proteins are important for cellular function and common in all organisms . However , it is not clear how such interactions emerge and evolve on a molecular level . We performed phylogenetic reconstruction , resurrection and biophysical characterization of two interacting disordered protein domains , CID and NCBD . CID appeared after the divergence of protostomes and deuterostomes 450–600 million years ago , while NCBD was present in the protostome/deuterostome ancestor . The most ancient CID/NCBD formed a relatively weak complex ( Kd∼5 µM ) . At the time of the first vertebrate-specific whole genome duplication , the affinity had increased ( Kd∼200 nM ) and was maintained in further speciation . Experiments together with molecular modeling using NMR chemical shifts suggest that new interactions involving intrinsically disordered proteins may evolve via a low-affinity complex which is optimized by modulating direct interactions as well as dynamics , while tolerating several potentially disruptive mutations . While the majority of proteins fold into well-defined structures to function , a substantial fraction of the proteome is made up by intrinsically disordered proteins ( IDPs ) ( Uversky and Dunker , 2012 ) . These IDPs , which can be fully disordered or contain disordered regions of variable size , play pivotal roles in biology , usually by participating in protein-protein interactions that govern key functions such as transcription and cell-cycle regulation ( Uversky et al . , 2008; Wright and Dyson , 2015 ) . IDPs are present in all domains of life , but they are more common and have a unique profile in eukaryotes as compared to archea and bacteria ( Peng et al . , 2015 ) . One particular problem with analyzing structure-function relationships in IDPs with regard to evolution is that IDPs appear to evolve faster than structured proteins and are more permissive to substitutions that do not apparently modulate function in either a positive or negative way ( Brown et al . , 2011 ) . In addition , insertion and deletion of amino acids are more common in IDPs ( Brown et al . , 2010; Light et al . , 2013 ) , further complicating structure-function analysis . Thus , while analyses of protein sequences have shed light on the evolution of IDPs , few biophysical studies have directly addressed the evolution of IDPs on a molecular level . Here , we use an ‘evolutionary biochemistry’ approach ( Bridgham et al . , 2006; Harms and Thornton , 2013; McKeown et al . , 2014 ) , in which phylogenetic reconstruction of ancestral sequences is combined with biophysical experiments ( Figure 1 ) , to reconstruct the evolution of a particular protein-protein interaction . This interaction involves disordered protein domains from two different transcriptional coactivators: ( i ) the molten-globule-like ( Kjaergaard et al . , 2010a ) nuclear co-activator binding domain ( NCBD ) present in the two paralogs CREB-binding protein ( CREBBP , also known as CBP ) and p300 , and ( ii ) the highly disordered CREBBP-interacting domain ( CID ) , found in the three mammalian paralogs NCOA1 , 2 and 3 ( also called SRC1 , TIF2 and ACTR , respectively ) . The interaction between CID from NCOA3 ( ACTR ) and NCBD from CREBBP represents a classical example of coupled folding and binding of disordered protein domains ( Demarest et al . , 2002 ) . 10 . 7554/eLife . 16059 . 003Figure 1 . General approach to investigate the evolution of a protein-protein interaction involving intrinsically disordered domains . Multiple sequence alignment forms the basis for the phylogeny , which is used to predict ancient variants of two interacting protein domains , CID and NCBD , respectively . The ancient variants are then resurrected by expression in Escherichia coli and purified to homogeneity . Finally , the resurrected as well as present-day variants of CID and NCBD are subjected to biophysical and computational characterization to assess the evolution of structure-function relationships . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 003 The two paralogs CREBBP and p300 , as well as the three paralogs NCOA1 , 2 and 3 , likely result from whole genome duplications . In general , a gene family is created by speciation events and genomic duplications . The general conclusion is that the ancestral vertebrate ( s ) went through two rounds of whole genome duplications ( denoted 1R and 2R , respectively ) prior to the origin of jawed vertebrates , and that teleost ( bony ) fish experienced a third whole genome duplication ( 3R ) ( Hughes and Liberles , 2008; Van de Peer et al . , 2009 ) . Thus , if no gene losses had occurred , vertebrates would have four copies and teleost fish eight copies of every gene . Genomic duplications are beneficial for inventing new biochemical functions ( Ohno , 1970; Näsvall et al . , 2012 ) . Therefore , the abundance of new genes following whole genome duplications creates multiple possibilities to evolve new functions on a large scale through point mutations and natural selection , and the 1R and 2R genome duplications have likely contributed to the broad repertoire of IDPs in vertebrates . In the present paper , we track the evolutionary history during 600 million years ( Myr ) of two interacting IDPs , the older NCBD domain and the younger CID domain , with regard to sequence , affinity and structural properties . The two domains established a molecular interaction involved in regulation of transcriptional activation in the lineage leading to present day deuterostome animals , and may serve as a paradigmatic example of evolution of protein-protein interactions involving IDPs . As a first step in our study , ancestral versions of the amino acid sequences of the NCBD domain of CREBBP/p300 ( NCBD ) and the CID domain of NCOA ( CID ) were reconstructed by phylogenetic analyses ( Figure 2 , Figure 2—figure supplements 1–4 and Figure 2—source data 1 and 2 ) using a maximum likelihood method ( Materials and methods ) . This analysis shows that NCBD is older than CID , as NCBD can be traced in deuterostomes , protostomes ( including extant arthropods , nematodes and molluscs ) and cnidarians , but not in more distantly related eukaryotes such as the choanoflagellate Monosiga brevicollis . Hence , NCBD arose in the animal lineage as a domain within the ancestral CREBBP/p300 protein . CID , on the other hand , appeared as an IDP domain within NCOA after the split of deuterostomes and protostomes , since it could not be identified in protostome NCOA . Furthermore , CID was present in an early deuterostome , before the first whole-genome duplication ( 1R ) , since it could be traced in extant sea urchins and acorn worms , in addition to vertebrates . Thus , we conclude that the CID/NCBD interaction emerged at the beginning of the deuterostome lineage around the Cambrian period , an era with a very dramatic evolutionary history . 10 . 7554/eLife . 16059 . 004Figure 2 . Reconstruction of the evolution of the interacting NCBD and CID domains . ( A ) Sequence alignments of extant and reconstructed ancient NCBD ( top ) and CID domains ( bottom ) . The positions of helices are according to the NMR structure of the complex between extant CREBBP NCBD ( blue ) and NCOA3 CID ( yellow ) . Free NCBD ( protein data base code 2KKJ ) and the CID/NCBD complex ( 1KBH ) are NMR structures , whereas the picture of free CID is a hypothetical modified structure made from the NCOA1 CID/NCBD complex ( 2C52 ) . The first residue in the NCBD alignment is referred to as position 2062 in the text and the first residue in the CID alignment as 1040 . The color coding of the sequences reflects similarities in chemical properties of the amino acid side chains and is a guide for the eye to see patterns of conservation . ( B ) Schematic tree of life with selected animal groups depicting the evolution of the NCBD domain ( blue ) in both protostomes and deuterostomes and the CID domain ( yellow ) in the deuterostome lineage only . See Figure 2—figure supplements 1–4 for detailed alignments and trees . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 00410 . 7554/eLife . 16059 . 005Figure 2—source data 1 . Probabilities of resurrected amino acid residues at the respective position ( 2062–2109 ) in the NCBD domain . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 00510 . 7554/eLife . 16059 . 006Figure 2—source data 2 . Probabilities of resurrected amino acid residues at the respective position ( 1040–1081 ) in the CID domain . The gap in Figure 2—figure supplement 4 created by one of the Takifugu rubripes sequences was removed in this table to make it easier to understand the numbering . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 00610 . 7554/eLife . 16059 . 007Figure 2—figure supplement 1 . Sequence alignment of NCBD domains of CREBBP/p300 used in the phylogenetic reconstruction . The whole CREBBP/p300 gene alignment was used to create the phylogenetic tree and then with the use of that tree the NCBD domain could be resurrected . The categories to the right should only be seen as guidelines to which group the individual species sequences belongs . Amino acids are colored based on chemical properties of the side chain according to eBioX standard . The tree in Figure 2—figure supplement 3 has the correct grouping information . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 00710 . 7554/eLife . 16059 . 008Figure 2—figure supplement 2 . Sequence alignment of the CID domains of NCOA1-3 used in the phylogenetic reconstruction . The whole NCOA1-3 gene alignment was used to create the phylogenetic tree and then with the use of that tree the CID domain could be resurrected . The categories to the right should only be seen as guidelines to which group the individual species sequences belongs . Amino acids are colored based on chemical properties of the side chain according to eBioX standard . The tree in Figure 2—figure supplement 4 has the correct grouping information . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 00810 . 7554/eLife . 16059 . 009Figure 2—figure supplement 3 . Phylogenetic tree of CREBBP/p300 proteins that contain the NCBD domain . The tree follows what is generally accepted regarding species evolution and whole genome duplications . Cnidarian CREBBP/p300 contains the NCBD domain . Since these species are distantly related to the other animals in the tree Cnidarian CREBBP/p300 proteins are used as outgroup in the analysis . After the divergence of Cnidaria from other metazoans , deuterostomes and protostomes diverged from each other . The node between the deuterostomes and protostomes ( marked with a dark blue circle ) is the oldest time point that we resurrect . Protostomes contain groups such as insects and molluscs , while deuterostomes contain all vertebrates . At the beginning of vertebrate evolution two whole genome duplications occurred . Thus , at this point all vertebrate genes were duplicated twice resulting in four copies , but many were rapidly lost . For CREBBP/p300 , two copies remained , namely CREBBP and p300 . The node corresponding to the ancestral CREBBP/p300 protein is marked with a green circle and referred to as 1R/2R in the text , since we cannot distinguish these events for CREBBP/p300 . The most recent node that we resurrect is the separation of Teleost fish CREBBP from Tetrapod CREBBP ( marked with a light blue circle ) . A third round of whole genome duplications later occurred in the fish lineage ( not indicated ) resulting in the two variants CREBBP1 and CREBBP2 . The indicated resurrected nodes correspond to the same nodes as in the simplified Figure 2 in the main text . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 00910 . 7554/eLife . 16059 . 010Figure 2—figure supplement 4 . Phylogenetic tree of NCOA1-3 proteins that contain the CID domain . The tree follows what is generally accepted regarding species evolution and whole genome duplications . The CID domain could be identified in Hemicordata and Echinodermata NCOA proteins , and since these species are more distantly related to the other animals in the tree they were used as outgroups in the analysis . After the divergence of these groups from the rest of the deuterostomes , the two vertebrate-specific whole genome duplications occurred . Our analysis suggests that the ancestral NCOA gene was first split into NCOA-1 and the ancestor of NCOA2 and 3 , which were subsequently split into NCOA2 and NCOA3 in the second whole genome duplication . ( The second copy of NCOA1 was lost from the genomes ) The nodes are marked with a dark ( 1R ) and light green circle ( 2R ) , respectively . The most recent node that we resurrected is the separation of Teleost fish NCOA3 from Tetrapod NCOA3 ( marked with a light blue circle ) . A third round of whole genome duplications occurred later in the fish lineage ( not marked ) but the resulting NCOA proteins were not retained in the genome . The indicated resurrected nodes correspond to the same nodes as in the simplified Figure 2 in the main text . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 010 For the oldest reconstructed version of NCBD , deuterostome/protostome ( D/P ) NCBD , there were several relatively uncertain positions with regard to amino acid identity ( Figure 2—source data 1 ) . This was experimentally resolved by testing the effect of alternative residues at these positions in the context of one of the most likely D/P variants . The first position included in the NCBD domain ( 2062 ) was particularly problematic in this respect , with several amino acid identities with similar and low probability: Ile ( 0 . 19 ) , Val ( 0 . 16 ) , Thr ( 0 . 15 ) , Met ( 0 . 13 ) , etc . In addition , the probability numbers at this position were very sensitive to which extant sequences were included in the phylogenetic analysis . Because of this and to avoid a hydrophobic side-chain at the N-terminus , we chose Thr at position 2062 as the ‘wild type’ ancestral NCBD ( referred to as D/P NCBD in the paper ) . Previous studies have suggested that the amino acid sequence of IDPs display a higher substitution rate ( following selection ) than those of ordered proteins ( Brown et al . , 2011 ) . Although limited to two proteins , NCOA and CREBBP/p300 , our data set allowed comparison between ordered and disordered regions within the same protein and during time . We therefore assessed the number of substitutions in selected ordered and disordered domains , respectively , in both NCOA and CREBBP/p300 . Clearly , amino acid sequences within linker regions between interaction domains have evolved such that no sequence similarity can be detected in most cases . For example , outside of the interacting regions of CID and NCBD , as defined by the NMR structures , it is impossible to align the sequences , even for closely related species . Thus , the domain boundaries were defined based on sequence conservation and available crystal or NMR structures . Based on how well we could define these boundaries and the confidence in the resurrection , which in some cases was low due to poor alignment quality , we selected four folded CREBBP domains ( HAT , KIX , RING/PHD , TAZ1 ) and one folded NCOA domain ( Pas-A ) for the analysis . To make the comparison between folded and disordered protein interaction domains simple , we used the same alignments and phylogenetic trees used to reconstruct ancient CID and NCBD domains and reconstructed other domains from NCOA and CREBBP/p300 , respectively . We then counted the number of amino acid substitutions and insertions/deletions in a particular sequence as compared to its predecessor . Thus , for domains from the CREBBP lineage , we compared the 1R/2R sequence with the D/P sequence , the ancestral fish/tetrapod ( F/T ) CREBBP with the 1R/2R , and present day human and zebrafish CREBBP with F/T CREBBP ( Figure 3 ) . In general , the amino acid substitution rates in NCOA and CREBBP/p300 are higher for all domains during the early times of animal evolution and up to the common ancestor of fish and tetrapods 390 Myr ago . Within CREBBP , the substitution rate of NCBD is similar to those of the ordered histone acetyltransferase ( HAT ) and RING/PHD domains , whereas KIX and TAZ1 have remained more conserved . For NCOA3 it was more difficult to define domain boundaries and we only compared CID to one folded domain , Pas-A . The overall substitution rate is only slightly higher for the CID domain , and the two domains have similar profiles . Thus , for CID and NCBD , functional constraints of the domains rather than disorder per se is probably determining amino acid substitution rate , similarly to a subset of disordered regions highlighted in earlier studies ( Brown et al . , 2011 , 2010 , 2002 ) . 10 . 7554/eLife . 16059 . 011Figure 3 . Amino acid substitutions in different domains in CREBBP/p300 and NCOA as a function of time . The predicted ancient sequences for distinct domains in CREBBP/p300 ( A and B ) and NCOA ( C ) were used to calculate the number of substitutions and indels between each evolutionary node ( Deuterostome/protostome , D/P; 1R; 2R; Fish/tetrapod , F/T; and present day ) in a particular lineage ( human and zebrafish CREBBP and human and zebrafish NCOA3 , respectively ) . The alignment and trees used to resurrect HAT , KIX , RING/PHD and TAZ1 were the ones optimized for NCBD . Similarly , the alignment and trees used to resurrect Pas-A were the ones optimized for the CID domain . The number of substitutions plus indels were normalized against the number of amino acid residues in each domain and the accumulated fraction of sequence changes plotted versus historical time . Both 1R and 2R occurred around 450 million Myr ago and the distance between them in panel C ( 10 Myr ) is arbitrary . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 011 To bring the analysis from the sequence level to the structural level , we considered the NMR structures of extant mouse CREBBP NCBD in complex with human NCOA1 CID ( Waters et al . , 2006 ) and NCOA3 CID ( Demarest et al . , 2002 ) , respectively . A structural alignment shows that the NCBD domain aligns well in the two complexes but that only the first α-helix ( Cα1 ) of NCOA1 CID and NCOA3 CID occupy similar positions ( Figure 4 ) . In fact , the third α-helix ( Cα3 ) of NCOA1 CID aligns with the second α-helix ( Cα2 ) of NCOA3 CID , while Cα2 of NCOA1 CID is non-binding and split into two smaller α-helices . The first α-helix of NCBD ( Nα1 ) interacts with Cα1 in both complexes , but the conformation of the third α-helix ( Nα3 ) is slightly different in the two complexes . Residues forming the Nα1/Cα1 interface in the CID/NCBD complex are conserved in the deuterostome lineage , while Nα3 has accumulated five substitutions ( Figure 2A ) . 10 . 7554/eLife . 16059 . 012Figure 4 . Structural alignment of two CID/NCBD complexes . ( A ) Superimposition of the structures of two complexes solved by NMR: CREBBP NCBD ( Light blue ) -NCOA1 CID ( Yellow ) ( 2C52 ) and CREBBP NCBD ( Dark blue ) -NCOA3 CID ( Red ) ( 1KBH ) . The complexes contain a hydrophobic core formed by residues from the respective protein domain . ( B ) Superimposition of NCBD from the complexes shows that in particular Nα1 and Nα2 align very well . ( C ) Superimposition of the NCBD-bound conformations of NCOA1 CID and NCOA3 CID . Whereas Cα1 from both complexes align well , the C-terminal regions of the CID domains occupy different positions . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 012 The next step in our approach was to ‘resurrect’ the sequences identified through our evolutionary analysis to analyze their biophysical properties . In these experimental studies , ancient and selected extant variants of CID and NCBD were expressed in E . coli , purified to homogeneity ( see Materials and methods ) and subjected to binding studies using isothermal titration calorimetry ( ITC ) to monitor changes in affinity over historical time . Strikingly , the affinity of deuterostome/protostome ( D/P ) NCBD for 1R CID , which is the oldest CID domain that we were able to resurrect with good confidence , was relatively low ( 5 µM ) as compared to younger NCBD variants ( Figure 5A and B , Table 1 ) . Importantly , all tested ancestral D/P NCBD ( Thr2062 ) variants measured with 1R CID yielded a relatively low affinity ( Kd values = 1 . 5 to 18 µM ) . The Kd values for D/P NCBD with Ile and Val at position 2062 were 2 . 0 and 2 . 2 µM , respectively , which is very close to that of 'wild-type' D/P NCBD with Thr2062 ( 3 . 0 µM ) ( Figure 5A and Figure 6 ) . Thus , our conclusions in the paper hold irrespective of the nature of the amino acid residues at the uncertain positions in D/P NCBD . 10 . 7554/eLife . 16059 . 013Figure 5 . Biophysical characterization of ancient and extant CID and NCBD domains . ( A ) Affinity of CID/NCBD complexes was measured by isothermal titration calorimetry ( three examples are shown including the low-affinity D/P NCBD , 1R CID interaction ) . ( B ) The affinities ( Kd values ) were normalized against the interaction between extant human NCOA2 CID and p300 NCBD . The relative affinity for D/P NCBD , 1R CID was calculated using the average Kd values of all D/P NCBD variants ( 5 ± 2 µM ) . ( C ) Propensity for helix formation for ancient and extant CID domains as measured by circular dichroism at 222 nm upon addition of the helix stabilizer 1 , 1 , 1-trifluoroethanol . ( D ) Global stability of NCBD domains as measured by circular dichroism at 222 nm ( reflecting the fraction folded NCBD ) upon addition of the denaturant urea . Hsa , Homo sapiens; Dre , Danio rerio ( zebrafish ) ; Pme , Petromyzon marinus , ( sea lamprey ) ; Dmel , Drosophila melanogaster ( fruit fly ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 01310 . 7554/eLife . 16059 . 014Figure 6 . Characterization of alternative variants at position 2062 in D/P NCBD . Isothermal titration calorimeter and circular dichroism experiments of D/P NCBD with ( A ) Ile and ( B ) Val at position 2062 . See Figure 5A for Thr2062 and Table 1 for Kd values . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 01410 . 7554/eLife . 16059 . 015Table 1 . Equilibrium dissociation ( Kd±standard error ) values for the interaction between NCBD and CID variants as determined by ITC . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 015Hsa NCOA1 CID ( SRC1 ) Hsa NCOA2 CID ( TIF2 ) Hsa NCOA3 CID ( ACTR ) Fish/Tetrapod NCOA3 CID2R CID2R CID N1043S2R CID G1080S1R CID1R CID S1058N1R CID G1080S1R CID S1078QHsa p53TADHsa ETS-2 PNTKd ( µM ) Hsa CREBBP NCBD0 . 33 ± 0 . 0390 . 13 ± 0 . 0110 . 35 ± 0 . 0310 . 65 ± 0 . 0240 . 38 ± 0 . 02084 ± 2 . 30 . 76 ± 0 . 071Hsa p300 NCBD0 . 18 ± 0 . 0150 . 071 ± 0 . 0100 . 11 ± 0 . 0100 . 28 ± 0 . 0120 . 22 ± 0 . 0249 . 2 ± 2 . 21 . 5 ± 0 . 077Dre CREBBP NCBD0 . 29 ± 0 . 0320 . 23 ± 0 . 0130 . 63 ± 0 . 0570 . 57 ± 0 . 025Pma NCBD0 . 19 ± 0 . 0230 . 044 ± 0 . 0170 . 23 ± 0 . 0301 . 0 ± 0 . 10Dmel NCBD5 . 2 ± 0 . 2022 ± 1 . 637 ± 2 . 84 . 1 ± 0 . 939 . 7 ± 1 . 6No detectable bindingFish/Tetrapod CREBBP NCBD0 . 41 ± 0 . 04052 ± 5 . 21 . 3 ± 0 . 0831R/2R NCBD0 . 11 ± 0 . 0420 . 045 ± 0 . 0180 . 23 ± 0 . 0400 . 28 ± 0 . 0210 . 290 ± 0 . 0350 . 33 ± 0 . 0230 . 20 ± 0 . 0160 . 22 ± 0 . 0270 . 24 ± 0 . 0240 . 25 ± 0 . 02134 ± 4 . 0 nM0 . 85 ± 0 . 0461R/2R NCBD N2065S0 . 11 ± 0 . 0200 . 15 ± 0 . 0130 . 13 ± 0 . 0121R/2R NCBD N2065S K2107R0 . 18 ± 0 . 0210 . 160 ± 0 . 0110 . 17 ± 0 . 0230 . 13 ± 0 . 018D/P NCBD1 . 5 ± 0 . 0880 . 52 ± 0 . 0325 . 0 ± 0 . 223 . 0 ± 0 . 133 . 9 ± 0 . 164 . 8 ± 0 . 205 . 5 ± 0 . 2143 ± 3 . 91 . 4 ± 0 . 051D/P NCBD T2062I2 . 0 ± 0 . 2D/P NCBD T2062V2 . 2 ± 0 . 6D/P NCBD P2063L7 . 7 ± 0 . 53D/P NCBD Q2088H1 . 5 ± 0 . 080D/P NCBD Q2088N2 . 2 ± 0 . 070D/P NCBD H2107Q18 ± 1 . 2Hsa CREBBP NCBD A2106Q0 . 10 ± 0 . 02Hsa CREBBP NCBD Y2108Q0 . 21 ± 0 . 06Hsa CREBBP NCBD A2106Q/Y2108Q0 . 22 ± 0 . 06 Already around the time of the two vertebrate-specific whole genome duplications ( 1R and 2R , respectively ) , NCBD had evolved an affinity toward the CID domain similar to that found today ( ∼200 nM ) . All tested alternative variants of 1R/2R NCBD , 1R CID and 2R CID yielded Kd values in the same range ( 100–300 nM ) ( Table 1 ) . Moreover , the affinities between extant NCBD domains and human CID domains were found to be surprisingly similar across several species . For example , CREBBP NCBD from sea lamprey and human , respectively , has similar affinity toward human NCOA3 CID . We also determined the affinity between ancient and extant NCBD domains and extant versions of other protein ligands for NCBD , namely the disordered transactivation domain ( TAD ) of human p53 ( p53TAD ) and the globular pointed ( PNT ) domain from human ETS-2 . In contrast to the CID domain , the affinities of p53TAD and PNT , respectively , were similar for ancient and extant NCBD ( Table 1 ) . The NCOA3 CID domain from human displays a high degree of disorder ( Ebert et al . , 2008; Kjaergaard et al . , 2010b ) and all tested CID variants in the present study displayed a strikingly similar far-UV CD spectrum , typical of IDPs ( Figure 7A–C ) . However , the propensity to form α-helices may differ between variants , which could result in higher affinity ( Iešmantavičius et al . , 2014 ) . Addition of 1 , 1 , 1-trifluoroethanol ( TFE ) induces α-helix formation and can be used to experimentally assess the α-helical propensity of any given polypeptide when there is a helix-coil equilibrium ( Jasanoff and Fersht , 1994 ) . TFE titrations ( 0–50% ) were performed for ancient and extant CID domains ( Figure 5C and Table 2 ) . The similar shape and midpoints of the resulting sigmoidal curves ( reflecting the helix-coil transition ) demonstrates that the overall α-helical propensity for ancient and modern CID variants is virtually identical . Increased α-helical propensity would likely have increased the affinity for NCBD ( Iešmantavičius et al . , 2014 ) , but at the cost of lower plasticity and flexibility . We note , however , that the stability for individual α-helices as predicted by the software AGADIR ( Muñoz and Serrano , 1994 ) has changed during evolution . For example , α-helix 1 of CID from human NCOA1 and 2 displays a higher α-helix propensity than α-helix 1 from human NCOA3 and the ancestral versions ( Figure 8 ) . Such local changes in α-helical propensity is a likely route to evolve an optimal affinity for short recognition motifs , which are common in intrinsically disordered regions ( Fuxreiter et al . , 2004 ) . 10 . 7554/eLife . 16059 . 016Figure 7 . Far-UV Circular dichroism experiments . ( A–C ) CD spectra of CID variants display a profile typical for disordered proteins . ( D–E ) CD spectra of NCBD variants show a qualitatively similar shape for all variants . ( G–I ) Thermal denaturations of NCBD variants show a similar apparent non-cooperative transition . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 01610 . 7554/eLife . 16059 . 017Figure 8 . The helical propensity of CID variants as predicted by AGADIR . ( A ) CID domains from extant human NCOA1 , 2 and 3 . ( B ) Ancestral CID domains: 1R , 2R and the fish/tetrapod ancestor . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 01710 . 7554/eLife . 16059 . 018Table 2 . Equilibrium parameters for CD-monitored trifluoroethanol ( TFE ) induced helix formation of CID variants determined in 20 mM sodium phosphate , pH 7 . 4 , 150 mM NaCl , at 25°C . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 018CID variant[TFE]50%* ( % ) [TFE]50%† ( % ) mD-N† ( %−1 ) 1R‡8 . 5 ± 1 . 37 . 6 ± 2 . 30 . 15 ± 0 . 022R10 . 7 ± 0 . 912 . 0 ± 0 . 20 . 22 ± 0 . 01Fish/tetrapod NCOA39 . 9 ± 0 . 710 . 0 ± 0 . 60 . 17 ± 0 . 01Hsa NCOA1-§-§0 . 15 ± 0 . 03Hsa NCOA29 . 5 ± 1 . 7-§0 . 11 ± 0 . 02Hsa NCOA35 . 6 ± 1 . 56 . 5 ± 0 . 90 . 18 ± 0 . 01The mD-N value was shared among the datasets in the curve fitting; mD-N = 0 . 17 ± 0 . 01 %−1 . †Free fitting of both [TFE]50% and mD-N . ‡1R , the node around the time of the first whole genome duplication in the vertebrate lineage; 2R , the node around the time of the second whole genome duplication in the vertebrate lineage; Fish/tetrapod , the node where fish diverged from tetrapods; Hsa , Homo sapiens; Dre , Danio rerio ( zebrafish ) ; Pme , Petromyzon marinus , ( sea lamprey ) ; Dmel , Drosophila melanogaster ( fruit fly ) . §Not well determined in the curve fitting . While extant mammalian CREBBP NCBD has a small but well-defined hydrophobic core , it is a very dynamic protein domain in the absence of a bound CID domain as previously shown by NMR measurements ( Ebert et al . , 2008 ) . We were therefore particularly interested in whether the global stability of NCBD changed when CID was recruited as binding partner . CD spectra ( Figure 7D–F ) and thermal denaturations ( Figure 7G–I ) were similar for all NCBD variants . The most ancient D/P NCBD variant was slightly less stable than historically subsequent variants in the deuterostome lineage , as determined by the midpoint from far UV CD-monitored urea denaturation experiments ( Figure 5D and Table 3 ) . However , despite high precision in experimental data , the broad unfolding transition of small protein domains such as NCBD leads to low accuracy in the estimates of the free energy of denaturation . Therefore , we refrain from making strong conclusions regarding the stability despite a clear shift in the urea midpoint for unfolding . 10 . 7554/eLife . 16059 . 019Table 3 . Equilibrium parameters for CD-monitored urea denaturation of NCBD variants determined in 20 mM sodium phosphate , pH 7 . 4 , 150 mM NaCl , 1 M TMAO at 10°C . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 019NCBD variant[Urea]50% ( M ) △GD-N* ( kcal mol−1 ) [Urea]50%† ( M ) mD-N† ( kcal mol−1 ) △GD-N† ( kcal mol−1 ) D/P‡2 . 4 ± 0 . 41 . 5 ± 0 . 32 . 2 ± 0 . 20 . 56 ± 0 . 041 . 2 ± 0 . 2D/P T2062I3 . 3 ± 0 . 32 . 0 ± 0 . 33 . 4 ± 0 . 10 . 70 ± 0 . 082 . 4 ± 0 . 31R/2R4 . 4 ± 0 . 32 . 7 ± 0 . 34 . 4 ± 0 . 10 . 67 ± 0 . 053 . 0 ± 0 . 3Fish/tetrapod CREBBP4 . 0 ± 0 . 32 . 5 ± 0 . 34 . 0 ± 0 . 10 . 62 ± 0 . 052 . 5 ± 0 . 2Hsa CREBBP3 . 8 ± 0 . 32 . 3 ± 0 . 33 . 7 ± 0 . 20 . 46 ± 0 . 091 . 7 ± 0 . 4Hsa p3004 . 4 ± 0 . 32 . 7 ± 0 . 34 . 4 ± 0 . 30 . 66 ± 0 . 172 . 9 ± 0 . 8Dre CREBBP1§3 . 4 ± 0 . 32 . 1 ± 0 . 32 . 2 ± 1 . 60 . 33 ± 0 . 160 . 7 ± 0 . 6Pma4 . 1 ± 0 . 22 . 5 ± 0 . 34 . 2 ± 0 . 60 . 50 ± 0 . 222 . 1 ± 1 . 0Dmel1 . 6 ± 0 . 51 . 0 ± 0 . 32 . 6 ± 0 . 41 . 2 ± 0 . 73 . 3 ± 1 . 9The mD-N value was shared among the datasets in the curve fitting; mD-N = 0 . 61 ± 0 . 05 kcal mol−1M−1 . †Free fitting of both [Urea]50% and mD-N‡D/P , Deuterostome/protostome node; 1R/2R , the node ( s ) around the time of the two whole genome duplications in the vertebrate lineage; Fish/tetrapod , the node where fish diverged from tetrapods; Hsa , Homo sapiens; Dre , Danio rerio ( zebrafish ) ; Pme , Petromyzon marinus , ( sea lamprey ) ; Dmel , Drosophila melanogaster ( fruit fly ) . §The bony fish lineage experienced a third whole-genome duplication and has two variants of CREBBP NCBD . We next asked what happens on a molecular level when a new binding partner is recruited . To shed light on this question , we first subjected two ancient complexes ( 1R CID with D/P NCBD and 1R CID with 1R/2R NCBD , respectively ) and one extant complex ( human NCOA3 CID with human CREBBP NCBD ) to NMR experiments and then used the chemical shifts assigned for Cα , Cβ , H and N as restraints in molecular dynamics simulations using Metadynamic Metainference ( Bonomi et al . , 2016a , 2016b ) , a recently developed scheme that can optimally balance the information contents of experimental data with that of a priori physico-chemical information . Structural ensembles based on molecular dynamics simulations and NMR chemical shifts were obtained starting from the two available NMR structures , mouse CREBBP NCBD in complex with human NCOA1 CID ( 2C52 ) ( Waters et al . , 2006 ) and NCOA3 CID ( 1KBH ) ( Demarest et al . , 2002 ) , respectively . By using Metadynamic Metainference simulations ( see Materials and methods ) , three complexes were analyzed: ( i ) 1R CID with , D/P NCBD , ( ii ) 1R CID with 1R/2R NCBD and ( iii ) extant human NCOA3 CID with CREBBP NCBD ( Figure 9 ) . The NMR data used as restraints in the simulations were of high quality and the peak assignment was close to 100% for all the three complexes ( Figure 9—figure supplement 1 and Figure 9—source data 1 ) . Comparison of the structural ensembles for the three complexes and their relative free-energy projections indicates that evolution of higher affinity correlates with a somewhat reduced conformational heterogeneity ( Figure 9A ) . Indeed , the free-energy surfaces as a function of the fraction of the overall helix-content and the radius of gyration ( Rg ) show that the ancient complex is slightly less structured and more compact , with an average helical fraction of 0 . 41 . This is reflected in the ensemble with more disordered N- and C-terminal helices for CID and a more disordered C-terminal helix Nα3 for NCBD . The younger 1R/2R complex is a little more structured in particular with respect to the C-terminal helix of NCBD , with an average helical fraction of 0 . 44 . The extant human complex , finally , is again slightly more structured at the N- and C- terminal helices of CID , with an average helical fraction of 0 . 47 . These results were confirmed by an independent analysis of the chemical shifts by δ2D ( Camilloni et al . , 2012a ) ( Figure 9B ) that shows how the terminal helices obtain more structure in going from the ancient ( blue ) , via the 1R/2R complex ( green ) and to the extant human complex ( red ) ; yet , the helix content of the second helix of CID decreases . The increase in helical structure at the terminal helices corresponds to a decrease in the average fluctuations as was confirmed by the analysis of the root mean square fluctuations ( RMSF ) ( Figure 9C ) . The changes in the helical structure have an effect in terms of inter-domain contacts . Indeed , the most ancient complex appears to have more but less populated contacts than younger complexes , in line with a marginally more disordered interface ( Figure 10 ) . During evolution , the total number of possible contacts between the domains decreases while the average population of the formed contacts increases . To visualize the extent to which different residues in NCBD interact in the respective CID complex ( most ancient , 1R/2R , and extant CREBBP NCBD/human NCOA3 CID ) , the normalized number of interface contacts per residue was analyzed ( Figure 11 ) . In this analysis , we also included two other complexes involving extant CREBBP NCBD , those with p53TAD and the ‘interferon regulatory factor ( IRF ) interaction domain from IRF-3’ ( denoted as IRF-3 in the paper ) , respectively . Overall , the main effect observed along the evolution of the CID/NCBD complex is a decreased fraction of contacts formed by N-terminal residues correlating with increased helical structure of the N-terminal helix of CID ( Figure 9 ) and an increased fraction of contacts formed by C-terminal residues of NCBD , correlating with increased helical structure of this region ( Figure 9 ) . In the C-terminal , there is an increase in fraction of contacts in particular at position 2108 that joins Arg2104 in binding Asp1068 from CID , and at position 2105 whose side chain forms multiple hydrophobic contacts with CID . However , there is a decrease in fraction of contacts at position Gln2103 while other positions show a less clear trend reflecting the complexity of the changes in the interaction surface . Of note is the similarity of the interface used by NCBD to bind CID with that of the NCBD/p53TAD complex , where for example Arg2104 makes a salt-bridge with Asp49 while the Tyr2108 makes a hydrogen bond with the backbone of Met44 , and its difference with the structurally distinct NCBD/IRF-3 complex , where the same two residues are exposed to the solvent . 10 . 7554/eLife . 16059 . 020Figure 9 . The CID/NCBD complex displays minor structural changes upon evolution . ( A ) Free-energy surfaces ( in kJ/mol ) as a function of the fraction of helix content and the Rg , for the most ancient complex ( D/P NCBD and 1R CID ) , the 1R/2R complex and one extant complex ( human NCOA3 CID/CREBBP NCBD ) . For each free-energy surface , the position of the minimum and a set of representative structures are shown: CID in yellow and NCBD in blue . N- and C- termini ( NT and CT , respectively ) are labeled for the central ensemble . ( B ) Per residue helix population of the protein ensembles of the most ancient ( blue circles ) , 1R/2R ( green squares ) and extant ( red bars ) variants as predicted by δ2D from the chemical shifts . ( C ) Average root-mean-square fluctuation for the three variants showing a weak correlation between historical age and conformational heterogeneity of the complex . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 02010 . 7554/eLife . 16059 . 021Figure 9—source data 1 . Chemical shift data of CID/NCBD complexes used in the molecular dynamics simulationsDOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 02110 . 7554/eLife . 16059 . 022Figure 9—figure supplement 1 . Heteronuclear single quantum correlation ( 1H/15N ) spectra for the ancient complex between 1R CID and D/P NCBD ( red peaks ) and the extant complex between human NCOA3 CID and CREBBP NCBD ( blue peaks ) . The spectra were recorded such that either the CID domain was NMR active ( panel A ) or the NCBD was NMR active ( panel B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 02210 . 7554/eLife . 16059 . 023Figure 10 . Contact analysis for ancient and extant CID/NCBD complexes . The probability contact maps are shown for each pair of residues for ( upper left ) the most ancient complex ( 1R CID and D/P NCBD ) , ( upper right ) the 1R/2R complex and ( lower right ) the extant NCOA3 CID/CREBBP NCBD complex . Inter-domain contacts are framed by gray rectangles . Given two residues in a certain conformation , a contact is defined as a distance within 0 . 5 nm ( excluding hydrogen atoms ) . Lower left panels: The total number of inter-domain contacts ( left ) and the inter-domain average contact formation ( right ) are reported as the number of residues with a contact populated more than 5% and the average over population for the same contacts , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 02310 . 7554/eLife . 16059 . 024Figure 11 . NCBD Interface contact analysis . The normalized number of interface contacts per residue is calculated from the simulations of the three historical CID/NCBD complexes ( upper three panels ) and compared with two extant complexes formed by CREBBP NCBD and alternative protein ligands , p53TAD ( pdb code 2L14 ) ( Lee et al . , 2010 ) and a binding domain from IRF-3 ( pdb code 1ZOQ ) ( Qin et al . , 2005 ) , respectively . In the IRF-3 complex ( bottom panel ) , NCBD adopts a distinct tertiary structure as compared to complexes with CID and p53 . The Gly-Ser residues at the N-terminus of the NCBD sequences result from the expression construct used in the study . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 024 The Metadynamic Metainference analysis highlighted a few positions that appeared prominent in the evolution of higher affinity in the CID/NCBD interaction and in particular position 2108 , where the Q2108Y mutation makes the interaction stronger . To test this result and further probe the less conserved region at the end of Nα3 in NCBD , we made the two reverse mutations in human CREBBP NCBD ( i . e . A2106Q , Y2108Q and the double mutant A2106Q/Y2108Q ) and measured the affinity to human NCOA2 CID using ITC ( Figure 12 ) . The A2106Q mutation did not change the affinity for NCOA2 CID . The Y2108Q and the double mutation A2106Q/Y2108Q resulted in a slightly lower affinity ( twofold ) toward NCOA2 CID ( Table 1 ) , in support of the simulation . However , the mutations made the NCBD protein less soluble , which led to precipitation and less reliable data due to ill-defined native baselines in the titration experiments . Nevertheless , the limited effect of these reverse mutations in one extant complex underscores the high degree of plasticity in this region of the CID/NCBD complex and illustrates the permissive nature of IDPs toward mutations . 10 . 7554/eLife . 16059 . 025Figure 12 . Isothermal titration calorimeter experiments between human NCOA2 CID and 'reverse mutants' in human CREBBP NCBD . ( A ) A2106Q , ( B ) Y2108Q and ( C ) A2106Q/Y2108Q . Below are CD spectra of the respective NCBD variant . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 02510 . 7554/eLife . 16059 . 026Figure 13 . Analysis of the convergence of the simulations . Free-energy plots along the dRMSD collective variable for the second half of the simulations are shown for ( A ) the most ancient complex ( 1R CID and D/P NCBD ) , ( B ) the 1R/2R complex and ( C ) the extant NCOA3 CID/CREBBP NCBD complex . The solid black curve is the final free energy obtained by averaging over the second half of the simulations . Overall , all the simulations are converged within 3 kJ/mol . DOI: http://dx . doi . org/10 . 7554/eLife . 16059 . 026 We have used evolutionary biochemistry to reconstruct the evolutionary process by which the interaction between two disordered proteins has emerged from a low-affinity complex . By identifying and resurrecting the early ancestor partners , we show how a combination of direct interactions and structural heterogeneity contributed to optimizing the affinity of the CID/NCBD complex . In accordance , we observe a remarkably high tolerance to mutation in the interface of this particular protein-protein interaction in extant complexes . For example , in NCBD there are a number of substituted residues at the end of Nα3 forming the interface to CID , ( Figure 2A ) . For the CID domain , there are also non-conserved residues in the CID/NCBD interface , for example Phe1055 in NCOA1 CID , which is Ala in NCOA2 CID , Leu in NCOA3 CID and Leu in the ancestor . Moreover , Lys1059 in NCOA1 CID forms part of the interface in the complex ( PDB code 2C52 ) , whereas the corresponding position in NCOA3 CID is a solvent exposed Thr ( the ancestral residue ) ( PDB code 1KBH ) . In NCOA2 CID this residue is a Phe , which is likely part of the hydrophobic core of the NCBD/NCOA2 CID complex . A third non-conserved interacting residue is position 1053 in Cα1 . Here , the ancestral residue is Asp , while in the human domains it is Val in NCOA1 CID , Tyr in NCOA2 CID and His in NCOA3 CID . These are all examples of substitutions , which could be expected to have a large impact on the protein-protein interaction and also on the structure of the complex . Nevertheless , these mutations in the CID domain have minor impact on the affinity for NCBD , illustrating the high tolerance to mutations in IDPs such as the CID domain , with regard to affinity . The tolerance to mutation is also reflected in the surprisingly similar affinities of human NCBD for CID domains from sea lamprey , zebra fish and human , respectively ( Table 1 ) . Based on the two available NMR structures of CID/NCBD complexes ( Figure 4 ) , one explanation is that the CID domain may adopt distinct conformations in the C-terminal region ( corresponding to Cα2 and Cα3 in 1KBH ) depending on amino acid sequence; however , the most populated simulated structures ( Figure 9 ) are all relatively similar to the 1KBH structure . Some structural features appear important for maintaining the CID/NCBD interaction . We have previously shown that the initial native contacts formed during the binding reaction between human NCOA3 CID ( ACTR ) and CREBBP NCBD are made between residues in the respective α1 of the two domains ( Iešmantavičius et al . , 2014; Dogan et al . , 2013 ) . Both Cα1 and Nα1 contain leucine-rich motifs , which are very common in protein-protein interactions ( Kobe and Deisenhofer , 1994 ) and which are conserved in the CID/NCBD interaction ( Figure 2 ) . In contrast , in extant fruit fly ( Drosophila melanogaster ) , a proteostome , the NCBD domain is preserved but , similarly to D/P NCBD it has a low affinity for the human CID domains ( Table 1 ) . The leucine-rich motif in Cα1 is mutated in fruit fly NCBD ( Figure 2 ) and the C-terminal region is more similar to D/P NCBD . Thus , it is likely the combination of several amino acid substitutions , affecting both direct interactions and structural heterogeneity , that made the 1R/2R NCBD a high-affinity CID binder . On a functional level there is much more to the interaction than affinity , for example cellular concentrations , solubility , spatial organisation of the proteins and competing ligands . NCBD is particularly interesting in this respect because it has several physiological ligands including the PNT domain from the transcription factor ETS-2 ( Jayaraman et al . , 1999 ) , IRF-3 ( Lin et al . , 2001 ) , and p53TAD . Any mutation that increases the affinity toward the CID domain must maintain the affinity for the other ligands . Intriguingly , in the complex between CREBBP NCBD and IRF-3 ( Qin et al . , 2005 ) , NCBD adopts a completely different conformation as compared to the conformation with CID domains and p53TAD ( Lee et al . , 2010 ) . Furthermore , NMR and folding experiments suggest that NCBD can adopt two different conformations in absence of ligand , one of which is the CID-bound conformation and the other possibly the IRF-3 bound conformation , although this has not been experimentally confirmed ( Kjaergaard et al . , 2010a , 2013; Dogan et al . , 2016 ) . Such functional conformational heterogeneity in NCBD would obviously put extra constraints on the evolutionary process . With these observations in mind , we note that the affinity of the extant human PNT domain of ETS-2 is similar for ancient and extant NCBD domains ( ∼1 µM , Table 1 ) . Although we have not resurrected the ancient versions of the PNT domain , these results suggest that the PNT/NCBD interaction was present before , and preserved during the evolution of the CID/NCBD interaction . Consistently , the PNT domain is present in protostomes ( e . g . D . melanogaster ) , while the CID domain is not . We observe a similar pattern for the affinity between extant human p53TAD and ancient and extant human NCBD variants . The limit to a further increase in affinity between CID and NCBD in the animal lineage leading to the species that experienced the 1R whole genome duplication could well be due to constraints imposed by the ancient versions of PNT , IRF-3 and p53TAD . On a residue level , it is clear that NCBD employs distinct residues for the interaction with IRF-3 , which allows this interaction ( and the alternative conformation of NCBD ) to coexist with those of CID and p53TAD ( Figure 11 ) . For CID and p53TAD , similar residues have been used throughout evolution for interactions , albeit with different relative influences . CREBBP and p300 nucleotide sequences were downloaded from the Gene tree ENSGT00730000110623 , automatically generated by Ensembl ( Flicek et al . , 2013 ) ( www . ensembl . org ) . This collection was complemented with hits from tblastn ( RRID:SCR:011822 ) searches from preEnsembl ( RRID:SCR_006766 ) ( pre . ensembl . org ) , EnsemblMetazoa ( RRID:SCR_000800 ) ( metazoa . ensembl . org ) , NCBI ( www . ncbi . nlm . nih . gov ) ( Metazoa specific searches ) , Skatebase ( RRID:SCR_005302 ) ( Wang et al . , 2012 ) ( skatebase . org ) , JGI Genome Portal ( RRID:SCR_002383 ) ( Grigoriev et al . , 2012 ) ( genome . jgi . doe . gov ) , MOSAS amphixus ( http://mosas . sysu . edu . cn/genome/ ) , EchinoBase ( RRID:SCR_007441 ) ( http://spbase . org ) ( Cameron et al . , 2009 ) , Japanese lamprey genome project ( http://jlampreygenome . imcb . a-star . edu . sg/ ) ( Mehta et al . , 2013 ) and OrcAE ( http://bioinformatics . psb . ugent . be/orcae/ ) ( Sterck et al . , 2012 ) . The same was done for NCOA1 , 2 and 3 ( Gene tree ENSGT00530000063109 ) . A number of cycles with alignments and maximum likelihood tree analysis were performed to detect erroneous sequences or alignments . The erroneous sequences were manually edited where possible , by for instance removing or adding exons . If manual editing still did not result in a good alignment for that individual gene , the sequence was removed from the analysis . Sequences shorter than 500 amino acids were excluded from the CREBBP analysis ( the whole length of the protein is roughly 2400 residues ) , while sequences shorter than 300 amino acids were excluded from the analysis for the roughly 1400 residues long NCOA proteins . Sequences lacking a complete NCBD or CID domain were also removed . The final sequences were aligned at Guidance ( Penn et al . , 2010 ) ( guidance . tau . ac . il ) using MAFFT ( RRID:SCR_011811 ) and ClustalW ( RRID:SCR_002909 ) with the codon model , and Muscle with amino acid model . The best alignment according to guidance alignment score was MAFFT and this alignment was therefore used for the continued analysis . Less reliable residues according to guidance ( confidence score below 0 . 5 ) were masked as X . Masking residues is preferable to removing whole columns since more good data will be retained in the analysis ( Privman et al . , 2012 ) . All columns without a guidance confidence score , as well as regions with few aligned sequences were also removed with Gap Strip/Squeeze v 2 . 1 . 0 , which kept columns with less than 95% gap . The amino acid and nucleotide models that best describe the resulting alignment according to the Bayesian information criterion ( BIC ) ( Schwarz , 1978 ) , calculated with Mega 5 . 2 . 2 ( Tamura et al . , 2011 ) , were JTT+ G+ I and GTR+ G+ I , respectively for CREBBP . For NCOA , the best amino acid model was JTT+G . Using the alignment , maximum likelihood ( ML ) trees were calculated both for the best nucleotide and amino acid model with PhyML3 . 0 ( Guindon et al . , 2010 ) with SPR and NNI tree improvement and with SH-aLRT branch support , which has been shown to perform better than traditional bootstrap . The cnidarian species were chosen as outgroups for the CREBBP tree and non-chordate deuterostomes as outgroups for the NCOA tree . The branches in the tree generated by amino acid alignment follow what has previously been suggested for the evolution of species ( Letunic and Bork , 2011 ) and whole genome duplications , while a few branches in the nucleotide tree diverge from the species tree . Our conclusion is therefore that the amino acid model and tree best describe the actual evolution and further analyses were performed using this tree . For ancestral sequence reconstruction , it is very important to have the correct alignment and therefore we cut out the NCBD and CID domains , which should be resurrected , as well as 30 adjacent amino acid residues and realigned the respective set of sequences with MAFFT , Muscle , ClustalW and PRANK with the codon model at Guidance . Sequences that did not have a confidence score above 0 . 5 , for example the urochordata ( tunicates ) were removed from the CREBBP tree . The final alignments contained 181 CREBBP protein sequences and 184 NCOA protein sequences . The highest guidance alignment score for both the NCBD and CID domains , respectively , were obtained with Muscle . However , for NCBD we obtained an even better alignment score by manually editing the alignment ( alignment score of 0 . 97 versus 0 . 96 and lowest column score 0 . 85 versus 0 . 76 ) . The manually edited NCBD alignment and the CID alignment ( Figure 2—figure supplements 1–2 ) were inserted into the respective original full length protein alignments and together with the previously obtained ML trees were used to resurrect the ancestral sequences at Mega 5 . 2 . 2 , which uses an ML method that correctly deals with indels ( Hall , 2011 ) . The cDNA corresponding to the most likely sequences at each evolutionary node ( for both NCBD and CID ) were purchased and subcloned into the pRSET vector used previously for expression of both NCBD and CID variants ( Dogan et al . , 2012 ) . All NCBD and CID variants were expressed and purified as previously described using nickel affinity and reversed phase chromatography ( Dogan et al . , 2012 ) . The selected nodes were ( i ) the fish/tetrapod ( CREBBP NCBD and NCOA3 CID ) , ( ii ) 2R , ( iii ) 1R and ( iv ) deuterostome/proteostome ( D/P ) , respectively . When the probability of the resurrected amino acid was lower than 0 . 7 ( Figure 2—source data 1 and 2 ) a construct with the second , third etc . most likely amino acid was produced to investigate possible effects of a different amino acid . In all cases in the present work , there was no significant effect of alternative residues ( Table 1 , Figure 6 ) . The numbering of residues for CID domains is according to the splice variant NCOA3-002 ( ENST00000371997 ) and the numbering of residues for NCBD domains according to human CREBBP . The PNT domain of ETS-2 and p53TAD were expressed and purified as previously described ( Dogan et al . , 2015 ) . All ITC experiments were performed in 20 mM sodium phosphate , pH 7 . 4 , 150 mM NaCl . Protein concentrations were measured either using the absorbance at 280 nm and calculated extinction coefficients or ( if the variant lacked Trp and Tyr residues ) with a direct detect IR spectrometer . The ITC measurements were performed at 25°C on an iTC200 ( Malvern instruments ) according to the instructions of the manufacturer . For each experiment , the respective NCBD and CID variant was dialysed against the same buffer to minimize artifacts due to buffer mismatch . The baseline of the experimental data was adjusted to get the lowest possible chi value in the curve fitting . For all high-affinity interactions ( Kd<1 µM ) , good baselines were obtained and the binding stoichiometry was generally around 1 . For the low-affinity interactions ( Kd = 1–10 µM ) , only one baseline was obtained and the stoichiometry thus not well determined . For certain variants , the ITC measurements were repeated twice with the same sample ( i . e . a technical replication ) and in two cases the experiments were repeated with new samples ( biological repetition ) . Both the biological and technical repetitions yielded highly similar results . ITC experiments are very sensitive to factors such as buffer mismatch and experiments with poor baselines jeopardizing curve fitting were excluded from the study . In the cases where a certain amino acid position was reconstructed with low probability ( <0 . 7 ) , the most likely variants were subjected to ITC experiments to rule out that the uncertainty affected the conclusions . For example , seven different variants of D/P NCBD were made and their affinities reported in Table 1 . Three additional variants of 1R CID were also tested with D/P NCBD with virtually identical Kd values . In the case of D/P NCBD/1R CID , the average value of all 10 reported Kd values in Table 1 ( 5 . 1 ± 1 . 6 µM ) were used to calculate the relative affinity shown in Figure 5B . Importantly , all variants tell the same story , namely that the affinity between 1R/2R NCBD and 1R CID ( and 2R CID and later variants ) is significantly higher than that between the ancestral D/P NCBD variants and 1R CID . All experiments were performed in 20 mM sodium phosphate , pH 7 . 4 , 150 mM NaCl . Circular dichroism ( CD ) experiments were performed on a JASCO-810 spectropolarimeter with a Peltier temperature control system . Far-UV spectra of NCBD and CID variants were recorded from 260 nm to 200 nm at 4°C using 30 µM protein . To assess the global stability of NCBD , urea ( 0–8 M ) was added to protein ( 30 µM ) -buffer solutions containing 1 M trimethylamine N-oxide ( TMAO ) at 10°C and the CD signal at 222 nm was measured by taking the average of 61 individual recordings at each urea concentration . To assess the helix propensity of CID variants 1 , 1 , 1-trifluoroethanol ( TFE ) was added to protein ( 30 µM ) -buffer solutions at 25°C and the CD signal at 222 nm was measured by taking the average of 61 individual recordings at each TFE concentration . Data from the urea and TFE experiments were fitted to the equation for a two-state process ( folded state in equilibrium with the denatured state ) to obtain the midpoint of the transition and the associated cooperativity factor ( m value ) ( Fersht , 1999 ) . Regarding the TFE titrations ( little difference between variants ) and urea titrations ( perhaps a small difference ) , selected experiments were repeated ( for example the D/P NCBD variant ) . The reported standard errors in Tables 2 and 3 come from the curve fitting but in this type of biophysical experiments , which we perform routinely in the lab , such errors are generally very similar to the error from three independent experiments . Figure 5D contains data from two independent experiments for urea-induced D/P NCBD denaturation , showing the high precision in these types of experiment . The largest source of error for the transition midpoint in this type of experiment is the concentration of TFE or urea , which are both determined with high accuracy . The urea concentration is double checked by measuring the refractive index . The largest source of error in the free energy of unfolding △GD-N derives from the error in the mD-N value ( describing the 'cooperativity' of the transition ) . The error in the mD-N value is large because of the small size of the proteins in the present study , resulting in short baselines and low accuracy in the parameters derived from the curve fitting . However , similar to the ITC experiments , the results are clear insofar that there are only small insignificant differences between most variants . The D/P NCBD ( and D . melanogaster NCBD ) have a slightly lower midpoint for urea denaturation , and lamprey NCBD a slightly higher one , but given the error associated with mD-N , and the lack of well-defined baselines ( due to the broad transition ) , we do not stress this apparent difference in △GD-N as a major finding . The respective CID and NCBD domains were expressed as unlabeled ( in rich medium ) and 15N-13C doubly labeled ( in minimal medium containing 15NH4Cl and 13C-D-glucose ) . Purification was as previously described for unlabeled protein ( Dogan et al . , 2012 ) . After purification , the samples were lyophilized and stored at −20°C . Prior to use the lyophilized samples , they were dissolved in a buffer containing 10 mM sodium phosphate pH 6 . 8 , 150 mM NaCl and dialyzed against the same buffer overnight at 25°C . Protein complexes were formed by titrating saturating concentrations of the unlabeled ( CID or NCBD ) to labeled fractions ( NCBD or CID ) . The final NMR samples contained 300–500 µM of the labeled protein ( bound ) to which 0 . 01% NaN3 and 5% D2O was added . NMR experiments were acquired on Bruker 600 , 700 and 900 MHz spectrometers equipped with triple resonance cryogenic probes at 25°C . For assignment purposes , standard 3D HNCACB and 15N-resolved NOESY-HSQC ( Cavanagh et al . , 2007 ) were recorded . All experiments were processed with NMRPipe ( Delaglio et al . , 1995 ) and analyzed with CcpNmr ( Vranken et al . , 2005 ) . The backbone chemical shifts were used in the simulations as described in the next section . Structural ensembles of the CID/NCBD complexes were obtained using Metadynamic Metainference ( Bonomi et al . , 2016a , 2016b ) . Prior information contained in the Charmm22 force field ( Piana et al . , 2011 ) with explicit solvent . Chemical shifts ( Cα , Cβ , N and H ) were included using Metainference in its Gaussian form ( Bonomi et al . , 2016a ) using CamShift ( Kohlhoff et al . , 2009; Camilloni et al . , 2012b ) over N = 10 replicas , r . The Metainference energy was calculated as:E=∑r=1N{Eff+kBT∑i=1Nd[ ( fi ( X ) −di ) 22σr , i2+0 . 5log2πσr , i+0 . 5logσr , i]} where the first term is the sum the over the replicas and the second term is the sum over the experimental data . Eff is the energy of the force-field , kBT is the Boltzmann constant times the temperature , f ( X ) is the calculated chemical shift averaged over the replicas , d is the reference experimental value , σ is the error estimated on-the-fly that includes the standard error of the mean resulting from the averaging over a finite number of replicas as well as the error estimate for random , systematic and the intrinsic error of the forward model ( i . e . CamShift ) . All simulations were run in GROMACS ( Pronk et al . , 2013 ) using PLUMED 2 ( Tribello et al . , 2014 ) . Van der Waals and Coulomb interactions were implemented with a cutoff at 0 . 9 nm , and long-range electrostatic effects were treated with the particle mesh Ewald method on a grid with a mesh of 0 . 1 nm . All simulations were carried out in the canonical ensemble at constant volume and by thermosetting the system using a stochastic velocity rescaling ( Bussi et al . , 2007 ) . The starting conformations were taken from the available NMR structures ( PDB code 1KBH and 2C52 ) and mutated accordingly using Scwrl4 ( Krivov et al . , 2009 ) . The structures were solvated with 5000 water molecules and ions were added to neutralize the total charge . Two preliminary 100 ns long simulations were run for each structure to equilibrate the system . Metadynamic Metainference simulations were performed using 10 replicas . The sampling of each replica was enhanced by Parallel Bias Metadynamics ( Pfaendtner and Bonomi , 2015 ) along five collective variables ( CVs ) namely , the helix content of CID , the helix content of NCBD , the radius of gyration of the complex , the dRMSD from 1KBH calculated using only the Cα carbons and the AlphaBeta collective variable defined as one half of the sum over all residues of one plus the cosine of χ1 angles for all hydrophobic residues , except alanine . Gaussians deposition was performed with an initial rate of 0 . 2 kJ/mol/ps , where the σ values were set to 0 . 2 , 0 . 1 , 0 . 01 , 0 . 05 , and 0 . 5 for the five CVs , respectively . In order to keep under control the convergence of the simulations , we rescaled the height of the Gaussians using the well-tempered scheme with a bias-factor of 16 ( Barducci et al . , 2008 ) . Furthermore , in order to limit the extent of accessible space along each collective variable and correctly treat the problem of the borders , intervals were set to 8–28 , 18–38 , 1 . 2–1 . 6 , 0–0 . 6 and 0–24 for the five CVs , respectively ( Baftizadeh et al . , 2012 ) . We set the bias as constant outside a defined interval for each CV . Each replica has been evolved for 150 ns . The sampling of the 10 replicas was combined using a simple reweighting scheme based on the final bias B where the weight of a conformation is given by w=exp ( +B ( X ) /kBT ) , consistently with the quasi static behavior at convergence of well-tempered metadynamics . An analysis of the convergence of the simulations is shown in Figure 13 .	Proteins are an important building block of life and are vital for almost every process that keeps cells alive . These molecules are made from chains of smaller molecules called amino acids linked together . The specific order of amino acids in a protein determines its shape and structure , which in turn controls what the protein can do . However , a group of proteins called 'intrinsically disordered proteins' are flexible in their shape and lack a stable three-dimensional structure . Yet , these proteins play important roles in many processes that require the protein to interact with a number of other proteins . At multiple time points during evolution , new or modified proteins – and consequently new potential interactions between proteins – have emerged . Often , an interaction that is specific for a group of organisms has evolved a long time ago and not changed since . As intrinsically disordered proteins lack a specific shape , it is harder to study how their structure ( or lack of it ) influences their purpose; until now , it was not known how their interactions emerge and evolve . Hultqvist et al . analyzed the amino acid sequences of two specific intrinsically disordered proteins from different organisms to reconstruct the versions of the proteins that were likely found in their common ancestors 450-600 million years ago . The ancestral proteins were then ‘resurrected’ by recreating them in test tubes and their characteristics and properties analyzed with experimental and computational biophysical methods . The results showed that the ancestral proteins created weaker bonds between them compared to more ‘modern’ ones , and were more flexible even when bound together . However , once their connection had evolved , the bonds became stronger and were maintained even when the organism diversified into new species . The findings shed light on fundamental principles of how new protein-protein interactions emerge and evolve on a molecular level . This suggests that an originally weak and dynamic interaction is relatively quickly turned into a tighter one by random mutations and natural selection . A next step for the future will be to investigate how other protein-protein interactions have evolved and to identify general underlying patterns . A deeper knowledge of how this molecular evolution happened will broaden our understanding of present day protein-protein interactions and might aid the design of drugs that can mimick proteins .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "computational", "and", "systems", "biology" ]	2017	Emergence and evolution of an interaction between intrinsically disordered proteins
Acquisition of distinct neuronal identities during development is critical for the assembly of diverse functional neural circuits in the brain . In both vertebrates and invertebrates , intrinsic determinants are thought to act in neural progenitors to specify their identity and the identity of their neuronal progeny . However , the extent to which individual factors can contribute to this is poorly understood . We investigate the role of orthodenticle in the specification of an identified neuroblast ( neuronal progenitor ) lineage in the Drosophila brain . Loss of orthodenticle from this neuroblast affects molecular properties , neuroanatomical features , and functional inputs of progeny neurons , such that an entire central complex lineage transforms into a functional olfactory projection neuron lineage . This ability to change functional macrocircuitry of the brain through changes in gene expression in a single neuroblast reveals a surprising capacity for novel circuit formation in the brain and provides a paradigm for large-scale evolutionary modification of circuitry . Animals display a wide repertoire of complex and adaptable behaviours executed by equally complex nervous systems . Understanding how the vast number of diverse cell types is assembled into functional neural circuits in complex brains during development is a major challenge . Studies of lineage tracing and circuit mapping reveal that heterogeneous pools of neural progenitors sequentially generate series of neuronal progeny , and that such lineally related neurons with shared developmental histories often share functional connectivity in the brain . In consequence , neural lineages can be considered to form neuroanatomical units of projection that represent the developmental basis of the functional circuitry of the brain ( Pearson and Doe , 2004; Cardona et al . , 2010; Pereanu et al . , 2010; Custo Greig et al . , 2013; Franco and Muller , 2013; Gao et al . , 2013; Kohwi and Doe , 2013 ) . This is exemplified in Drosophila where the tens of thousands of neurons that comprise the adult brain are generated during development by a set of approximately 100 pairs of individually identifiable neural stem cells called neuroblasts ( Truman and Bate , 1988; Urbach et al . , 2003; Urbach and Technau , 2003a , 2003b; Technau et al . , 2006 ) . Each neuroblast gives rise to a specific , invariant lineage of post-mitotic neural cells in a highly stereotyped manner and in many cases , lineally related neurons share functional connectivity and many neuroanatomical features such as innervation of common neuropiles in the brain and common axon tract projection patterns ( Pereanu and Hartenstein , 2006; Ito et al . , 2013; Lovick et al . , 2013; Wong et al . , 2013; Yu et al . , 2013 ) . Examples of this are the four neuroblast lineages that give rise to the intrinsic cells of the mushroom body , a neuropile compartment involved in learning and memory , or the five neuroblast lineages that innervate the antennal lobe , the primary olfactory processing centre in the fly brain ( Ito and Hotta , 1992; Ito et al . , 1997; Stocker et al . , 1997; Jefferis et al . , 2001; Das et al . , 2008 , 2011 , 2013; Lai et al . , 2008; Chou et al . , 2010 ) . Thus , neuroblast lineages are considered to form neuroanatomical units of projection that represent the developmental basis of the functional macrocircuitry of the fly brain ( Pereanu and Hartenstein , 2006; Cardona et al . , 2010; Lovick et al . , 2013; Wong et al . , 2013 ) . Comparable principles are manifested in the developing cerebral cortex of vertebrates , which consists of diverse neurons organized into six distinct layers , each of which is laid in place sequentially during development . Neural progenitors in the cortex are known to be multipotent , capable of generating neurons that populate each of the layers . Lineage tracing experiments in mice suggest that lineally related neurons occupy columns spanning across the layers of the cerebral cortex , as proposed in the ‘radial unit hypothesis’ ( Rakic , 1988; Yu et al . , 2009b; Li et al . , 2012; Ohtsuki et al . , 2012 ) . Furthermore , lineally related neurons also show a propensity to interconnect and have functional similarity , for example , similar orientation preferences in the visual cortex ( Li et al . , 2012; Ohtsuki et al . , 2012 ) . Thus in vertebrates as in invertebrates , developmental history and lineage relationships govern the assembly of functional circuits . In order to understand how lineally specified circuits develop in the brain , it is critical to understand the molecular mechanisms that confer unique identities to neural progenitors and their lineages . Studies on the molecular genetics of brain development indicate that neural progenitors emerge from the embryonic neuroectoderm where unique spatial information represented by unique combinations of gene expression specifies unique identities to the progenitors . For example , in Drosophila , the embryonic neuroectoderm becomes spatially regionalized due to the action of embryonic patterning genes , which define the anterior-posterior and dorsal-ventral body axes . Their combined expression creates a Cartesian coordinate-like gene expression system in the neuroectoderm , resulting in unique domains of expression of developmental control genes along the neuroectoderm ( Skeath and Thor , 2003; Urbach and Technau , 2004; Technau et al . , 2006 ) . Changes in the combinatorial expression pattern of these genes in specific domains can lead to changes in the identities of the neuroblasts that delaminate from the neuroectoderm during embryogenesis ( e . g . , Deshpande et al . , 2001 ) . The process of spatial regionalization of the embryonic neuroectoderm is very similar in vertebrates . Homologous embryonic patterning genes result in unique domains of combinatorial gene expression along the vertebrate neuroectoderm ( Reichert and Simeone , 2001; Lichtneckert and Reichert , 2005 , 2008; Reichert , 2009 ) . Thus , in both vertebrates and invertebrates , the cells of the neuroectoderm acquire unique spatial information in the form of a combinatorial code of gene expression , which is conferred by embryonic patterning genes . It is noteworthy that neural progenitors also use temporal information ( typically a series of sequentially expressed transcription factors ) to generate neuronal diversity within lineages ( Pearson and Doe , 2004; Lin and Lee , 2012; Kohwi and Doe , 2013 ) . While spatial cues convert a homogenous pool of progenitors into heterogeneous populations , temporal cues result in the ordered production of different neural subtypes from each progenitor . Given that spatial information in the neuroectoderm , in the form of embryonic patterning genes , imparts heterogeneity to neural progenitor populations , it is likely that these genes might also act as intrinsic determinants in the progenitor to give lineages their unique identities and hence determine their place in neural circuitry . This implies that spatially encoded intrinsic factors determine the identity of the progenitor and , as a consequence , the unique circuit features of its neural lineage . According to this , removal or addition of one or more of these genes could lead to a change in neuroblast identity resulting in transformation of the neuronal lineage and the lineal circuitry that derives from it . However , the extent to which individual transcription factors can contribute to this specification of neuroblast identity is not well understood . In order to test this , it is important to be able to uniquely identify individual neural progenitors and their lineages in the brain . The complexity of the vertebrate brain makes it difficult to conduct such an analysis at the resolution of single progenitors and single lineages . However , each of the neuroblasts in the Drosophila brain has been identified and their lineages characterized in the larval and adult brains ( Ito et al . , 2013; Lovick et al . , 2013; Wong et al . , 2013; Yu et al . , 2013 ) . Furthermore , each of these neuroblasts has also been characterized by the expression of a specific combination of spatial genes , which could act as cell intrinsic determinants in the specification of unique neuroblast identity and hence control lineage-specific neuronal cell fate ( Skeath and Thor , 2003; Urbach and Technau , 2004; Technau et al . , 2006 ) . This allows an investigation of the role of putative intrinsic determinants by changing their expression in identified stem cells and assessing its effect at the lineage level in an otherwise normal brain . Here , we focus on two identified neuroblast lineages in the Drosophila brain , LALv1 and ALad1 , which develop in close spatial proximity to each other in the larval brain but become spatially segregated in the adult brain . While the ALad1 neuroblast generates olfactory projection interneurons that innervate the antennal lobe , the LALv1 neuroblast generates wide-field interneurons that innervate the central complex . We show that orthodenticle ( otd ) , an embryonic patterning gene involved in specifying the anterior-most regions of the neuroectoderm and embryonic brain ( Lichtneckert and Reichert , 2008; Reichert and Bello , 2010 ) , is expressed during development in the LALv1 neuroblast lineage but not in the ALad1 neuroblast lineage . Remarkably , loss of otd from the LALv1 neuroblast results in a complete transformation in the identity of the neurons that derive from this lineage . The otd null LALv1 neurons transform into antennal lobe projection interneurons similar to the ALad1 lineage , and this transformation includes a complete change in the neuroanatomy of the neurons , a change in their molecular properties as well as in their functional connectivity . This remarkably complete respecification of a neuroblast lineage upon the mutation of a single gene in the brain demonstrates that intrinsic determinants acting in the neuroblast during development specify the identity of its neural progeny and the macrocircuitry that these progeny establish . This large-scale modification of functional circuits in the brain by a single transcription factor in a single stem cell is unprecedented and reveals a surprising capacity for novel neural circuit formation in the developing brain , which may provide a paradigm for large-scale evolutionary modification of brain connectivity . We focused our analysis on two identified neuroblast lineages referred to as LALv1 and ALad1 ( Pereanu and Hartenstein , 2006; Lovick et al . , 2013 ) ( see ‘Materials and methods’ for lineage nomenclature ) . During postembryonic development in the larval brain , the adult-specific ( postembryonically generated ) neural progeny of these lineages have their cell bodies clustered close to each other , dorsal to the larval antennal lobe ( Figure 1A , B ) . Although their cell body clusters are closely apposed , the two lineages can be easily identified based on their distinct and unique axon tracts that project to different brain regions ( Pereanu et al . , 2010; Das et al . , 2013; Lovick et al . , 2013 ) . 10 . 7554/eLife . 04407 . 003Figure 1 . Development , morphogenesis , and differential Otd expression in two identified central brain neuroblast lineages , LALv1 and ALad1 . ( A and B ) show anterior and lateral views of 3D reconstructions of the LALv1 ( green ) and the ALad1 ( magenta ) lineages in the larval brain . ( A ) shows that the cell bodies are closely apposed to each other and lie above the larval antennal lobe ( AL , yellow ) , ( B ) shows their tracts diverge—the ALad1 tract ( magenta ) projects dorsally and the LALv1 tract ( green ) projects posteriorly behind the AL and splits . ( C and D ) show WT MARCM clones of the larval ALad1 and LALv1 lineages , respectively . Their cell bodies are outlined by white dotted lines and the white arrows trace their tracts . Insets in C and D show that while LALv1 cells ( cyan arrow in D ) express Otd , ALad1 cells ( cyan arrow in C ) do not . ( E–L ) is a third larval instar brain ( CS ) immunolabelled with neurotactin ( green , to identify lineages ) , Otd ( red ) and TOPRO-3 ( to label nuclei ) . The LALv1 lineage is documented in E–H , and the ALad1 lineage is documented in I–L . The neuroblasts are marked with yellow dotted lines and the lineages are marked with white dotted lines . The LALv1 neuroblast expresses Otd ( yellow arrow in F ) and the ALad1 neuroblast does not ( yellow arrow in J ) . ( M and N ) show anterior and lateral views of 3D reconstructions of the LALv1 ( green ) and the ALad1 ( magenta ) lineages in the adult brain . Note that the adult antennal lobe ( AL , yellow in M and N ) is situated between the ALad1 lineage ( antero-dorsal to AL ) and the LALv1 lineage ( ventral to AL ) and the cell bodies of these lineages are not closely apposed anymore . The arrows in M and N indicate the ALad1 tract ( magenta ) , which projects dorsally towards the protocerebrum and the LALv1 tract ( green ) , which projects posterior to the AL . ( O and P ) show WT clones of the adult LALv1 and ALad1 lineages , respectively . Their cell bodies are outlined by white dotted lines and the AL is outlined by yellow dotted lines . White arrows trace the tracts of these lineages . The LALv1 lineage innervates the lateral accessory lobe ( LAL ) and the central complex ( CC ) . The ALad1 lineage innervates the calyx of the mushroom body ( MB ) and lateral horn ( LH ) . The midline is represented by a yellow line in all images . Scale bars in C ( applicable to D ) and in L ( applicable to E–L ) are 20 µm . Scale bar in P ( applicable to O ) is 50 µm . Genotypes in C and D: FRT19A/FRT19A , Tub-Gal80 , hsFLP; Tub-Gal4 , UAS-mCD8::GFP/+ . Genotype in O: FRT19A/FRT19A , Tub-Gal80 , hsFLP; Per-Gal4 , UAS-mCD8::GFP/+ . Genotype in P: FRT19A/FRT19A , Tub-Gal80 , hsFLP; GH146-Gal4 , UAS-mCD8::GFP/+ . DOI: http://dx . doi . org/10 . 7554/eLife . 04407 . 003 The anatomical features of these two wild-type lineages can be visualized by MARCM clonal labelling ( randomly induced neuroblast clones; ubiquitous Tub-Gal4 driver ) . The ALad1 lineage initially projects its axon tract medially , dorsal to the larval antennal lobe , then turns posteriorly and projects towards the protocerebrum via the medial antennal lobe tract ( Figure 1B , C ) ( Das et al . , 2013; Lovick et al . , 2013 ) . The LALv1 lineage initially projects its axon tract ventro-medially , posterior to the larval antennal lobe , then loops dorsally and splits into two secondary axon tracts ( Figure 1B , D ) ( Spindler and Hartenstein , 2011; Lovick et al . , 2013 ) . In addition to the differences in axonal trajectories , we found that these two lineages also differed in their expression of the transcription factor Otd . Co-immunolabelling for the homeodomain transcription factor Otd and for Neurotactin ( to identify lineage-specific axon tracts ) shows that the LALv1 neuroblast ( Figure 1F ) and all of its lineal progeny ( white dotted lines in Figure 1F and inset in Figure 1D ) express Otd . In contrast , neither the ALad1 neuroblast nor its lineal progeny are found to express Otd ( Figure 1I–L and inset in Figure 1C ) . In the adult brain , the neural progeny of the ALad1 lineage are olfactory projection neurons , which innervate the glomeruli of the antennal lobe and the neural progeny of the LALv1 lineage are wide-field interneurons that innervate the central complex , a sensorimotor integration centre in the fly brain ( Ito et al . , 2013; Wong et al . , 2013; Yu et al . , 2013 ) . To study the neuroanatomical features of the two lineages in the mature brain , we took advantage of the fact that they are differentially labelled by four enhancer-Gal4 driver lines in the adult brain . Thus , the adult LALv1 lineage is labelled by the OK371-Gal4 ( a glutamatergic neuron label ) and Per-Gal4 driver lines ( Figure 1O , Table 1 ) , while the ALad1 lineage is not . Conversely , the adult ALad1 lineage is labelled by the Cha7 . 4-Gal4 ( a cholinergic neuron label ) and GH146-Gal4 lines , while the adult LALv1 lineage is not ( Figure 1P , Table 1 ) . 10 . 7554/eLife . 04407 . 004Table 1 . Summary of the specific molecular changes in the LALv1 and Alad1 lineagesDOI: http://dx . doi . org/10 . 7554/eLife . 04407 . 004WT LALv1otd−/− LALv1WT ALad1OK371-Gal4+−−Per-Gal4+−−Lim+−−Cha7 . 4-Gal4−++GH146-Gal4−++LN1-Gal4−−−Acj6+++Note that the molecular signature of the otd−/− LALv1 lineage is similar to that of the wild-type ALad1 lineage . MARCM clonal labelling of the neurons in the adult ALad1 lineage using GH146-Gal4 or Cha7 . 4-Gal4 drivers shows that their cell bodies are positioned dorsal to the adult antennal lobe , their dendrites innervate the antennal lobe glomeruli and their axons exit the lobe via the medial antennal lobe tract ( Figure 1M , N , P ) . The axons then project dorso-posteriorly to innervate the calyx of the mushroom body and the lateral horn ( Video 1 ) ( Ito et al . , 2013; Wong et al . , 2013; Yu et al . , 2013 ) . 10 . 7554/eLife . 04407 . 014Video 1 . Projection pattern of the wild-type ALad1 lineage . DOI: http://dx . doi . org/10 . 7554/eLife . 04407 . 014 MARCM clonal labelling of the neurons in the adult LALv1 lineage using Per-Gal4 or OK371-Gal4 drivers shows that their cell bodies are positioned ventral to the adult antennal lobe and their axons project into the loVM tract , which courses posteriorly behind the adult antennal lobe , then loops dorsally creating the prominent LEp fascicle , which innervates the central complex neuropiles and the lateral accessory lobe ( Video 2 ) ( Figure 1O ) ( Spindler and Hartenstein , 2011; Ito et al . , 2013; Lovick et al . , 2013; Wong et al . , 2013; Yu et al . , 2013 ) . As in the corresponding larval lineages , the adult LALv1 neuroblast lineage expresses Otd while the adult ALad1 neuroblast lineage does not ( insets in Figure 1O , P ) . It is noteworthy that in the adult brain the cell bodies of the LALv1 neurons are located ventral to the adult antennal lobe , whereas in the larval brain the position of the cell bodies is dorsal to the larval antennal lobe ( compare Figure 1A , M ) . This change in cell body position of the LALv1 lineage occurs as a consequence of the morphogenetic changes associated with the de novo development of the adult antennal lobe ( Jefferis et al . , 2004; Spindler and Hartenstein , 2011; Lovick et al . , 2013; Wong et al . , 2013 ) . 10 . 7554/eLife . 04407 . 013Video 2 . Projection pattern of the wild-type LAlv1 lineage . DOI: http://dx . doi . org/10 . 7554/eLife . 04407 . 013 In summary , the LALv1 neuroblast and its progeny , which innervate the central complex , express otd throughout brain development as well as in the adult brain , while the ALad1 neuroblast and its progeny neurons , which innervate the antennal lobe , do not . As otd is expressed in all of the cells of the central complex lineage—the neuroblast and the postmitotic neurons—we tested its possible function in both these cell types . In order to do this , we used MARCM-based clonal mutational methods to generate GFP-labelled otd−/− clones in the LALv1 lineage in an otherwise heterozygous background ( Lee and Luo , 2001 ) . Using this technique , it is possible to genetically inactivate otd in the postmitotic neurons , the GMC , or the neuroblast , thus allowing us to assess its role in each of these cell types ( see schematic in Figure 2A ) . We generated such otd−/− clones early in larval development and analyzed them in the adult brain . 10 . 7554/eLife . 04407 . 005Figure 2 . otd−/− postmitotic clones of the LALv1 lineage . ( A ) shows a schematic of clone generation by the MARCM method . ( B and C ) document two wild-type single cell MARCM clones of the LALv1 lineage . These neurons skirt around the antennal lobe ( white dotted lines ) and innervate the LAL . ( D–F ) show that these wild-type single cell clones express Otd . ( G and H ) document two single cell otd−/− MARCM clones of the LALv1 lineage . ( I–K ) shows one such cell ( I ) , which does not express Otd ( J ) . Note that the otd−/− single cell clones of the LALv1 lineage shown in G and H skirt around the antennal lobe and innervate the LAL and one of the central complex nodule ( yellow arrowhead in G and H ) , like the wild-type single cell clones shown in B and C . Genotypes in B–F: FRT19A/FRT19A , Tub-Gal80 , hsFLP; Per-Gal4 or OK371-Gal4 , UAS-mCD8::GFP/+ . Genotypes in G–K: FRT19A , oc2/FRT19A , Tub-Gal80 , hsFLP; Per-Gal4 or OK371-Gal4 , UAS-mCD8::GFP/+ . Scale bar is 50 µm . Yellow line represents the midline . DOI: http://dx . doi . org/10 . 7554/eLife . 04407 . 005 Using this technique , we first investigated a possible requirement of otd in the postmitotic neurons of the central complex lineage . In these experiments , in which we used the OK371-Gal4 and Per-Gal4 driver lines to label the MARCM clones , we obtained a total of seven wild-type single cell clones and 11 otd−/− single cell clones . Although we have not dated the birth of these clones precisely ( matched the time of clone generation ) , the wild-type single cell clones that we obtained in our experiments were very similar to those described previously ( Yu et al . , 2009a ) . Six of these single cell wild-type clones consisted of neurons that innervated both the lateral accessory lobe as well as one of the noduli of the central complex ( Figure 2B ) and one clone only innervated the lateral accessory lobe ( Figure 2C ) . All of the 11 otd−/− single cell clones we recovered also displayed a similar neuroanatomy . Their cell bodies were located ventral to the antennal lobe , their axons coursed through the loVM and LEp tracts and they all innervated the lateral accessory lobe and one of the noduli of the central complex ( Figure 2G , H ) . Thus , loss of otd function from the postmitotic neurons did not result in any gross defects in the neuroanatomy of these neurons . It is however possible , that there are fine-scale changes in the arborisation of these neurons within the lateral accessory lobe and the central complex that we were unable to identify . It is also possible that otd function is required in the GMC for the targeting of the postmitotic neurons ( see schematic in Figure 2A ) . However , in our experiments , we never obtained two cell GMC clones in order to be able to address this possibility . We then asked if otd might be required in the neuroblast of the LALv1 lineage for its proper development . In order to test this , we inactivated otd in the neuroblast during early larval development and analyzed the neuroanatomy of the resultant labelled wild-type and otd−/− mutant neurons in the adult brain . In these experiments , in which we used the OK371-Gal4 and Per-Gal4 driver lines to label the MARCM clones , we recovered 6 and 14 wild-type clones , respectively . As expected , all the wild-type neuroblast clones displayed the neuroanatomy of the central complex lineage as described above ( ventrally position cell bodies , axon projection via the loVM and LEp tracts and innervations in the lateral accessory lobe and central complex . Figure 3B–E , J–M ) . However , when we generated otd−/− neuroblast clones in the LALv1 lineage ( identifiable by the loss of Otd staining in the corresponding cell cluster ventral to the antennal lobe; white arrowhead in Figure 3G , O ) , neither of these drivers labelled the mutant LALv1 lineage ( Figure 3F–I , N–Q ) . 10 . 7554/eLife . 04407 . 006Figure 3 . Loss of otd from the LALv1 lineage results in the suppression of the OK371 and Per enhancers . ( A ) schematises the experimental logic . In neuroblast clones of the LALv1 lineage-specific enhancer Gal4s label the wild-type LALv1 neurons because they are active in the lineage ( for example , OK371 and Per ) . Inability to label the LALv1 neurons in otd−/− neuroblast clones of the LALv1 lineage ( detectable by absence of Otd immunolabelling ) will suggest that the enhancers become suppressed in the mutant neurons . The brains in B–E and J–M show wild-type MARCM clones of the LALv1 lineage labelled by OK371-Gal4 and Per-Gal4 , respectively . The insets in B–E and J–M show that these cells express Otd . The brains in F–I and N–Q show otd−/− clones of the LALv1 lineage . Otd expression is lost in one hemisphere in these brains ( white arrowhead in G and O; compare with the Otd expression within white dotted lines in the other brain hemisphere ) . Neither the OK371-Gal4 ( F ) nor the Per-Gal4 ( N ) enhancers drive the expression of UAS-mCD8::GFP in these cells and the transformed lineage is not labelled in these experiments . Genotype in B–E: FRT19A/FRT19A , Tub-Gal80 , hsFLP; OK371-Gal4 , UAS-mCD8::GFP/+ . Genotype in F–I: FRT19A , otdYH13/FRT19A , Tub-Gal80 , hsFLP; OK371-Gal4 , UAS-mCD8::GFP/+ . Genotype in J–M: FRT19A/FRT19A , Tub-Gal80 , hsFLP; Per-Gal4 , UAS-mCD8::GFP/+ . Genotype in N–Q: FRT19A , otdYH13/FRT19A , Tub-Gal80 , hsFLP; Per-Gal4 , UAS-mCD8::GFP/+ . Midline is represented by a yellow line . Scale bar is 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04407 . 006 In order to investigate the neuroanatomy of the otd−/− LALv1 lineage further , we utilized the ubiquitously expressed Tub-Gal4 driver to label neuroblast clones and recovered 19 WT and 37 otd−/− neuroblast MARCM clones in the LALv1 lineage . While the wild-type neurons displayed all the features of the LALv1 lineage described above ( Figure 4—figure supplement 1A ) , the otd−/− LALv1 neuroblast clones had dramatic changes in its neuroanatomy ( Figure 4—figure supplement 1B–D ) . Mutant neurons no longer innervated the central complex or lateral accessory lobe neuropiles; instead they innervated the antennal lobe neuropile ( asterisk in Figure 4—figure supplement 1B–D ) and sent projections via the medial antennal lobe tract towards the protocerebrum ( arrowhead in Figure 4—figure supplement 1B–D ) . These changes in dendritic and axonal innervation patterns were reversed by targeted expression of the full-length otd coding sequence in mutant neuroblast clones using the Tub-Gal4 ( Figure 4—figure supplement 2 ) . Taken together , these findings indicate that the mutant LALv1 neurons have acquired a transformed identity . Moreover these data suggest that this transformed identity has features characteristic of antennal lobe projection neurons . As the neuroanatomy of the otd−/− LALv1 lineage shows such a dramatic transformation , we wanted to confirm that the transformed neurons did indeed belong to the LALv1 lineage . We used three different approaches to determine that it was indeed the LALv1 lineage that transformed into an antennal lobe-like lineage upon the loss of otd from its neuroblast . First , we showed that the appearance of the transformed otd−/− LALv1 lineage corresponds to the appearance of an extra antennal lobe lineage . Second , we showed that the transformed otd−/− LALv1 lineage results in the corresponding loss of the LEp tract specific to the wild-type LALv1 lineage . Third , we used an independent molecular marker to unambiguously identify the wild-type and otd−/− LALv1 lineage . We next asked whether otd gain-of-function in the antennal lobe lineage , ALad1 , might result in a reciprocal anatomical transformation of this lineage into one resembling the wild-type central complex lineage , LALv1 . We used the MARCM system to misexpress the full-length otd coding sequence in the antennal lobe neuroblast clones using a Tub-Gal4 driver . In all Otd misexpression clones of the antennal lobe lineage ( 15/15 ) , we found a partial transformation of this lineage towards the central complex identity ( Figure 8 ) . All 15 clones comprised a few cells that retained neuroanatomical features of the wild-type antennal lobe lineage such as antero-dorsal cell body position , innervation of the antennal lobe , and axonal projections via the medial antennal lobe tract ( yellow asterisk and arrowhead in Figure 8A–C ) . However , most of the cells in the clones displayed neuroanatomical features of the central complex lineage . These cell bodies were positioned ventral to the adult antennal lobe , they projected their axons via the loVM and LEp tracts and they innervated the lateral accessory lobe ( magenta asterisk and white arrowheads in Figure 8A–C ) . Thus , otd gain-of-function was able to cause a partial , albeit incomplete transformation , of the antennal lobe lineage into a central complex-like lineage . 10 . 7554/eLife . 04407 . 016Figure 8 . Overexpression of Otd in the ALad1 lineage results in a partial reciprocal transformation . ( A–C ) ALad1 neuroblast clone misexpressing Otd . ( B and C ) are the anterior and lateral views of the 3D reconstructions of the clone in A . Few of the Otd misexpressing cells ( yellow asterisk in A ) retain the wild type neuroanatomy of the ALad1 lineage ( magenta cells in the reconstructions ) —they have innervations in the AL glomeruli ( yellow arrowheads in A–C ) and project via the antennal-cerebral tract ( yellow arrowhead in C ) . Most of the Otd misexpressing ALad1 neurons are seen ventral to the AL ( magenta asterisk in A; green cells in the reconstructions in B and C ) . They do not innervate the AL and instead project towards the LAL ( white arrowheads in A–C ) . Genotype in A–C: FRT19A/FRT19A , Tub-Gal80 , hsFLP; Tub-Gal4 , UAS-mCD8::GFP/UAS-otd . Midline is represented by a yellow line . Scale bar is 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04407 . 016 The innervation pattern of the otd−/− LALv1 neuroblast lineage strikingly resembled that of the antennal lobe lineage , ALad1 . Furthermore , the projection neuron-specific GH146 driver , which does not label the wild-type central complex lineage , labelled the transformed otd−/− LALv1 neurons . This ectopic activation of the GH146 driver in the otd−/− neurons indicates that the anatomical transformation of the mutant neurons might be accompanied by corresponding molecular transformations that reflect a transformation to the antennal lobe lineage identity . To test if this is indeed the case , we analyzed the activity of select molecular markers ( the Gal4 driver lines described above ) in the central complex lineage ( summarized in Table 1 ) . In order to test the activity of these enhancers in the otd−/− LALv1 lineage , we generated otd−/− MARCM clones and used these selected Gal4 lines to label the mutant lineage ( see schematic in Figure 3A ) . Finally , we also tested the expression of the homeodomain transcription factor LIM1 , which is expressed in the wild-type LALv1 lineage and is not expressed in the wild-type ALad1 lineage . As described above , when we generated wild-type MARCM clones and used either OK371-Gal4 or Per-Gal4 driver lines to label the clones , we found that both lines were able to drive reporter expression in the wild-type LALv1 lineage ( Figure 3B–E , J–M ) . In contrast , neither of these driver lines was able to drive reporter expression in the otd−/− LALv1 lineage ( Figure 3F–I , N–Q ) . This suggests that the OK371-Gal4 and Per-Gal4 driver lines are suppressed in the transformed otd−/− LALv1 neurons . However , in these experiments , the transformed neurons were not labelled at all because the activity of the drivers was suppressed in the otd−/− LALv1 lineage . In order to confirm this finding , we decided to positively label the otd−/− LALv1 neurons and in this background assay the activity of the Gal4 drivers . In order to do this , we utilized the dual MARCM method ( Lai and Lee , 2006 ) , which uses two independent binary expression systems ( Gal4-UAS and LexA–LexA operator ) to label the MARCM clones . In these experiments , we used the GH146-LexA driver to label the otd−/− LALv1 neurons positively and combined it with OK371-Gal4 and Per-Gal4 driver lines to assay their activities . We first tested if the GH146-LexA driver , like the GH146-Gal4 and the GH146-QF driver , was active in the otd−/− LALv1 neurons . Thus , in the first set of dual MARCM experiments , we combined the GH146-LexA with Tubulin-Gal4 ( Figure 9A–D ) . Under these conditions , when we generated otd−/− neuroblast clones in the LALv1 lineage , we found that the transformed neurons were labelled by both the Tubulin-Gal4 ( magenta dotted lines and inset in Figure 9A ) and GH146-LexA ( magenta dotted lines and inset in Figure 9B ) drivers , confirming that GH146-LexA , like GH146-Gal4 and the GH146-QF , is active in the otd−/− LALv1 neurons and thus able to label it . In the following dual MARCM experiments , we used the GH146-LexA to positively label the transformed otd−/− LALv1 neurons and combined it with either OK371-Gal4 or Per-Gal4 driver lines to assay for their activity in the transformed neurons . In both cases , while the GH146-LexA positively labelled the transformed otd−/− LALv1 neurons ( magenta dotted lines and insets in Figure 9E , I ) , neither of the Gal4 driver lines was able to drive reporter expression in these mutant neurons ( magenta dotted lines and insets in Figure 9F , J ) . This indicates that enhancers that are normally active in the wild-type central complex lineage and inactive in the antennal lobe lineage become suppressed in the transformed otd−/− LALv1 lineage . 10 . 7554/eLife . 04407 . 017Figure 9 . Specific molecular changes occur in the transformed otd−/− LALv1 lineage . ( A–D ) documents an otd−/− LALv1 neuroblast clone that has been dual labelled with Tubulin-Gal4 ( A ) and GH146-LexA ( B ) using the dual MARCM technique . ( E–H and I–L ) document otd−/− LALv1 neuroblast clones that have been dual labelled with GH146-LexA ( E and I ) to positively label the mutant neurons and either OK371-Gal4 ( E–H ) or Per-Gal4 ( I–L ) using the dual MARCM technique . In these clones , while the GH146-LexA driver labels the transformed otd−/− LALv1 neurons ( E and I ) , neither the OK371-Gal4 ( F ) nor the Per-Gal4 ( J ) do . ( M–P ) documents a transformed otd−/− LALv1 lineage ( inset and magenta dotted lines in N ) labelled with Cha7 . 4-Gal4 . While Cha7 . 4-Gal4 is not active in the wild-type LALv1 lineage ( not shown ) , it becomes activated in the otd−/− LALv1 lineage ( inset and magenta dotted lines in M–P ) . Genotypes in A–D: FRT19A , otdYH13/FRT19A , Tub-Gal80 , hsFLP; Tubulin-Gal4 , UAS-mCD8::GFP/FRTG13 , GH146-LexA::GAD , LexAop::rCD2::GFP . Genotype in E–H: FRT19A , otdYH13/FRT19A , Tub-Gal80 , hsFLP; OK371-Gal4 , UAS-mCD8::GFP/FRTG13 , GH146-LexA::GAD , LexAop::rCD2::GFP . Genotype in I–L: FRT19A , otdYH13/FRT19A , Tub-Gal80 , hsFLP;Per-Gal4 , UAS-mCD8::GFP/FRTG13 , GH146-LexA::GAD , LexAop::rCD2::GFP . Genotype in H: FRT19A , otdYH13/FRT19A , Tub-Gal80 , hsFLP; Cha7 . 4-Gal4 , UAS-mCD8::GFP/+ . Midline is represented by a yellow line . DOI: http://dx . doi . org/10 . 7554/eLife . 04407 . 017 Might the converse be true; do enhancers that are normally inactive in the wild-type central complex lineage and active in the antennal lobe lineage become activated in the transformed otd−/− LALv1 lineage ? To test this , we used the Cha7 . 4-Gal4 in MARCM clonal experiments . As expected , we never recovered Cha7 . 4-Gal4 labelled LALv1 neuroblast clones in wild-type MARCM experiments ( data not shown ) . In contrast , when we generated otd−/− neuroblast clones in the LALv1 lineage ( identifiable by the loss of Otd staining ventral to the antennal lobe; magenta dotted lines in Figure 9N ) the Cha7 . 4-Gal4 driver robustly drove reporter expression in the transformed otd−/− LALv1 lineage ( magenta dotted lines and insets in Figure 9M–P ) . This suggests that the Cha7 . 4-Gal4 becomes activated in the transformed otd−/− LALv1 lineage . Furthermore , the concomitant loss of the OK371-Gal4 driver ( a putative glutamatergic label ) and ectopic activation of the Cha7 . 4-Gal4 driver ( a putative cholinergic label ) suggest that there might be a change in neurotransmitter identity of the LALv1 lineage from wild-type glutamatergic to transformed cholinergic identities , similar to the cholinergic identity of wild-type antennal lobe neurons . Finally , immunolabelling with an anti-LIM antibody indicates that the wild-type central complex lineage expresses the homeodomain transcription factor LIM ( Figure 10G–I ) while the transformed otd−/− LALv1 lineage , like the wild-type antennal lobe lineage , does not ( Figure 10J–L ) . Interestingly , the LN1-Gal4 driver , which is inactive in both the wild-type LALv1 and ALad1 lineages , remains inactive in the otd−/− LALv1 lineage ( data not shown ) . 10 . 7554/eLife . 04407 . 018Figure 10 . Specific molecular changes occur in the transformed otd−/− LALv1 lineage . ( A–C and G–I ) document neuronal cell bodies of the wild-type LALv1 lineage immunolabelled with Acj6 ( A–C ) and Lim1 ( G–I ) . The WT LALv1 neurons express both these molecular markers . ( D–F and J–L ) document neuronal cell bodies of the otd−/− LALv1 lineage immunolabelled with Acj6 ( D–F ) and Lim1 ( J–L ) . While otd−/− LALv1 neurons continue to express Acj6 ( E ) , they downregulate Lim1 expression ( K ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04407 . 018 Taken together , these findings indicate that the otd−/− LALv1 lineage acquires the molecular signature of a wild-type antennal lobe lineage ( see Table 1 ) implying that otd loss-of-function in the LALv1 neuroblast lineage results in a molecular as well as an anatomical transformation of this lineage into one resembling the ALad1 lineage . Given the extent of the anatomical and molecular transformations seen in the otd−/− LALv1 neuroblast lineage , might the neurons in the transformed lineage receive functional input from olfactory sensory neurons ? To address this question , we specifically expressed the calcium sensor G-CaMP3 in the transformed otd−/− LALv1 lineage by MARCM clonal labeling ( Wang et al . , 2003 ) and used two-photon microscopy to monitor calcium activity in the dendrites of these transformed neurons in the antennal lobe . We first tested if the transformed otd−/− LALv1 neurons established functional connectivity with other neurons in the antennal lobe . Typically , olfactory sensory neurons bring odour information to the antennal lobe via the antennal nerve . Here , they make synaptic connections with projection neurons and local interneurons; projection neurons , take the odour information to higher brain centres ( mushroom body and lateral horn ) and local interneurons process the odour information locally in the antennal lobe . We reasoned that if the transformed neurons did made functional connections within the antennal lobe they would be postsynaptic to the olfactory sensory neurons and would be activated upon the activation of the antennal nerve . We therefore electrically stimulated the antennal nerve while simultaneously monitoring calcium activity from the transformed neurons . We found that electrical stimulation of the antennal nerve , which contains the axons of the olfactory sensory neurons , resulted in an increase in calcium activity in the dendrites of the otd−/− LALv1 lineage . Moreover , a greater number of electrical stimulus pulses applied to the antennal nerve resulted in a corresponding increase in the amplitude of the calcium signal recorded in the mutant transformed neurons ( Figure 11A–C ) . These results demonstrate that the transformed otd−/− LALv1 neurons were able to make functional connections in the antennal lobe and were able to receive functional input from sensory afferents . 10 . 7554/eLife . 04407 . 019Figure 11 . The otd−/− transformed LALv1 lineage has functional synapses in the AL and can respond to odour stimulation . ( A ) Two-photon calcium imaging from the otd−/− LALv1 neurons in response to electrical stimulation of the antennal nerve . Gray-scale image shows average pre-stimulation fluorescence of one transformed lineage . Colour images show fluorescence changes in response to different numbers of electrical stimulation ( duration , 1 ms; amplitude , 10 V; frequency , 100 Hz ) . Graph in B shows peak amplitude of ΔF/F against the number of stimulations . Error bars represent S . E . M . n = 5 . ( C ) Average traces that plot ΔF/F over time at different stimulus intensities . Shaded region in each trace represents S . E . M . ( D ) Two-photon calcium imaging from the otd−/− LALv1 neurons in response to odour stimulation of the antenna . Gray-scale image shows the structure of the transformed lineage and three identifiable glomeruli . The colour images show glomerular activation patterns evoked by isoamyl acetate ( IAA ) , ethyl butyrate ( EB ) , 3-octanol ( 3-O ) and 3-heptanol ( 3-H ) . IAA activated VM2 , DM2 and DM3 . EB activated both VM2 and DM2 . 3-O and 3-H activated DM2 and VM2 , respectively . ( E ) Average traces that plot ΔF/F over time in VM2 , DM2 and DM3 . The shaded region in each trace represents S . E . M . n = 3–5 . The false colour scales ( ΔF ) and scale bars are shown at the right of each panel . Genotype: FRT19A , otdYH13/FRT19A , Tub-Gal80 , hsFLP;GH146-Gal4 , UAS-GCaMP3/GH146-Gal4 , UAS-GCaMP3 . DOI: http://dx . doi . org/10 . 7554/eLife . 04407 . 019 We further investigated if the transformed otd−/− LALv1 neurons could respond to specific odour stimuli . To do this , we performed calcium imaging experiments similar to those described above but replaced the electrical stimulation of the antennal nerve with odour stimulation of the intact antenna ( olfactory sensory neurons ) . Four different odorants ( isoamyl acetate , ethyl butyrate , 3-octonal , 3-heptanol ) were selected based on their ability to excite all or some of the VM2 , DM2 and DM3 glomeruli ( Dacks et al . , 2009; Semmelhack and Wang , 2009 ) in the antennal lobe , which are innervated by the otd−/− LALv1 neurons . Imaging calcium activity from the dendrites of the otd−/− LALv1 neurons in these glomeruli in response to the selected odours show that each of the four odorants evoked a unique pattern of glomerular activity . Isoamyl acetate excited all three glomeruli , whereas ethyl butyrate excited only the VM2 and DM2 glomeruli . 3-octanol and 3-heptanol , however , excited just the DM2 and VM2 glomeruli , respectively ( Figure 11D , E ) . These patterns of glomerular activation in the otd−/− LALv1 neurons are strikingly similar to that of wild-type antennal lobe olfactory projection neurons ( Wang et al . , 2003; Dacks et al . , 2009; Semmelhack and Wang , 2009 ) . Taken together , these functional studies indicate that the otd−/− LALv1 neurons receive specific input from olfactory sensory neurons that results in glomerulus-specific activation patterns to different odorants . This in turn implies that otd loss-of-function in a single neuroblast leads to a remarkably extensive reconfiguration of the macrocircuitry in the brain , which includes anatomical , molecular as well as functional transformation of neurons in the central complex lineage into neurons with properties of olfactory projection neurons . During neuronal development in both vertebrates and invertebrates neural progenitors use spatial and temporal information to generate diverse neuronal subtypes . For example , in Drosophila , unique spatial information imparts heterogeneity to the neuroblast pool and then temporal cues acting in the neuroblasts generate further diversity . In this way diverse neuronal subtypes are produced by the neuroblast lineages , which consequently create the diverse functional macrocircuitry of the brain . In addition , the neuroblasts in the central brain of Drosophila are characterized by the expression of a specific combination of cell intrinsic determinants ( Urbach and Technau , 2003a ) that are thought to act in the specification of unique neuroblast identity and hence control lineage-specific neuronal cell fate . In this study , we show that such intrinsic determinants present in the neuroblast are essential for the proper specification of the entire lineage that derives from the neuroblast . Our data demonstrate that a remarkable rewiring of the functional macrocircuitry of the brain occurs due to the manipulation of one intrinsic factor , otd , acting in an identified neuroblast during development . This transformation affects molecular properties , anatomical projection patterns ( dendritic and axonal ) , and functional inputs in all of the neurons in the lineage ( summarized in Figure 12 ) such that a central complex lineage is transformed into a functional olfactory projection neuron lineage . This otd-dependent , lineage-specific respecification of interneurons has implications for our understanding of the development and evolution of the circuitry in the brain . 10 . 7554/eLife . 04407 . 020Figure 12 . Transformation of the LALv1 lineage . ( A ) The WT LALv1 is a wide-field central complex lineage that expresses Otd ( orange ) . ( B ) The loss of otd function from the LALv1 neuroblast transforms these neurons into antennal lobe projection neurons . This neuroanatomical transformation is accompanied by molecular changes—markers that are active in the WT LALv1 ( green box in A ) are suppressed in the mutant ( B; with the exception of Acj6 ) ; those that are suppressed in the WT ( red box in A ) , are activated in the mutant ( B; with the exception of LN1-Gal4 ) . The exceptions are in agreement with a transformation to antennal lobe projection neuron fate ( see Table 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04407 . 020 The ability of a neuroblast lineage to transform completely into another upon the loss of a single intrinsic determinant suggests that many of the other putative members of a potential neural identity code might be shared between these lineages . The observation that the neuroblasts of the central complex and antennal lobe lineages develop in such close spatial proximity to each other during early development suggests that these two neuroblasts may experience similar spatial cues as they develop on the neuroectoderm . If this is the case , then by manipulating a single differentially expressed factor , otd , we might have been able to uncover the underlying similarity in the intrinsic spatial code between the two neuroblast lineages . Importantly , this neuroblast-specific transformation of lineage identity resulted in an alteration of the brain's circuitry such that an entire neuroanatomical unit of projection to the central complex was lacking while a novel and functional ectopic unit of projection was added to the antennal lobe . Implicit in these findings are the notions of ‘coded’ and ‘soft’ properties of circuit assembly . On the one hand , the neuroanatomical and molecular transformation described above demonstrates that circuitry in the brain is ‘hard-wired’ or ‘coded’ by the spatially encoded intrinsic factors—the presence or absence of otd from the central complex neuroblast determines the identity of the resultant neurons . On the other hand , the resulting functional transformation suggests that circuit assembly involves substantial ‘soft-wiring’—the olfactory sensory neurons and interneurons indigenous to the antennal lobe are able to make functional connections with the extraneous transformed otd−/− LALv1 neurons , which they are normally not ‘hard-wired’ to connect with . Thus , while genetically encoded properties might ‘lock’ lineages into particular circuit states ( central complex or antennal lobe ) it is their ‘soft’ properties ( developmental plasticity ) that allow circuits to functionally incorporate changes as dramatic as extraneous neurons . As both these wiring strategies operate simultaneously in the brain , it bestows upon the brain the huge potential of evolvability of functional circuits . Many interesting questions emerge as a result of our findings . How does the developmental plasticity of a functional circuit support these large-scale rearrangements ? Do developing circuits acquire a propensity for exuberant connectivities , or do they try and maintain a homeostasis in their connections and therefore make compensatory changes in the number of synapses with their normal partners ? It has been shown in some cases that neuronal activity can mediate such ‘soft’ properties of synaptic connections ( Tripodi et al . , 2008; Singh et al . , 2010 ) . It will be interesting to test if this is also the case for the transformed neurons and the olfactory circuit . Finally , do all parts of the brain display such striking developmental plasticity such that they can be remodelled to this extent and incorporate extraneous neurons into existing circuitry ? The ability to change the functional macrocircuitry of the brain through changes in the expression of a single transcription factor in a single neuroblast lineage may provide a simple paradigm for large-scale modification of brain connectivity during evolution . The otd−/−transformed LALv1 lineage functionally integrates into the antennal lobe circuitry and participates in olfactory information processing . This suggests that a functional rewiring of the olfactory circuitry can occur due to the addition of an entire neuroblast lineage to the normal olfactory circuit . In more general terms , this type of lineage-specific rewiring might fuel the evolutionary modification of neural circuitry in the brain . It provides an elegant and simple solution to the evolution of complex circuitry in that a ‘microevolutionary’ molecular change ( changing the expression of one gene in one cell ) can have ‘macroevolutionary’ consequences on the brain's circuitry ( changing an entire macrocircuit or an information processing module ) . This simple strategy suggests that large-scale changes in the brain's wiring do not need to come about through many minor , sequentially accumulating changes at the cellular level . Instead , large-scale wiring changes can occur in response to remarkably simple changes in gene expression in single cells . In this context , our data may have provided evidence for the mechanistic ease with which circuitry in the brain can evolve . The brain lineages have been named by various groups in the past . Here , we list the various names by which the two lineages , which are the focus of our study , have been called . ALad1 ( Ito et al . , 2013; Yu et al . , 2013 ) /BAmv3 ( Pereanu and Hartenstein , 2006; Lovick et al . , 2013; Wong et al . , 2013 ) /adNB ( Jefferis et al . , 2001 ) . LALv1 ( Ito et al . , 2013; Yu et al . , 2013 ) /BAmv1 ( Pereanu and Hartenstein , 2006; Lovick et al . , 2013; Wong et al . , 2013 ) . Fly stocks were obtained from the Bloomington Stock Centre ( IN , USA ) and , unless otherwise stated , were grown on cornmeal media , at 25°C . UAS-miRNA otd-1 was kindly provided by Henry Sun , Taiwan . To generate WT MARCM clones females of the genotype FRT19A or FRT19A;UAS-mCD8::GFP were crossed to males of the following genotypes: FR19A , Tub-Gal80 , hsFLP; Tub-Gal4 , UAS-mCD8::GFP/CyO or FR19A , Tub-Gal80 , hsFLP; OK371-Gal4 , UAS-mCD8::GFP/CyO or FR19A , Tub-Gal80 , hsFLP; Per-Gal4 , UAS-mCD8::GFP/CyO or FR19A , Tub-Gal80 , hsFLP; GH146-Gal4 , UAS-mCD8::GFP/CyO or FR19A , Tub-Gal80 , hsFLP; LN1-Gal4 , UAS-mCD8::GFP/CyO or FR19A , Tub-Gal80 , hsFLP; Cha7 . 4-Gal4 , UAS-GFP/CyO . To generate otd−/− MARCM clones , females of the genotype FRT19A , otdYH13/FM7c or FRT19A , oc2/FM7c or FRT19A , otdYH13;UAS-mCD8::GFP or FRT19A , oc2;UAS-mCD8::GFP were crossed to males of the above-mentioned genotypes . To knockdown otd in neuroblasts using RNAi and simultaneously visualise GH146-labelled cells , females of the genotype UAS-dicer; Insc-Gal4/CyO were crossed to males of the genotype UAS-miRNA otd-1/CyO; GH146-QF , QUAS-mtdTomato-HA/TM6B and the crosses were grown at 25°C . To generate Dual MARCM clones , females of the genotype FRT19A , otdYH13/FM7a; FRTG13 , GH146-LexA::GAD , LexAop-rCD2::GFP/CyO were crossed to males of the following genotypes: FRT19A , Tub-Gal80 , hsFLP; Tub-Gal4 , UAS-mCD8::GFP/CyO or FRT19A , Tub-Gal80 , hsFLP; OK371-Gal4 , UAS-mCD8::GFP/CyO or FRT19A , Tub-Gal80 , hsFLP; Per-Gal4 , UAS-mCD8::GFP/CyO . To generate Otd overexpression clones , females of the genotype FRT19A;UAS-otd were crossed to males of the genotype FR19A , Tub-Gal80 , hsFLP; Tub-Gal4 , UAS-mCD8::GFP/CyO . To generate rescue clones , females of the genotype FRT19A otdYH13/FM7a;UAS-otd were crossed to males of the genotype FR19A , Tub-Gal80 , hsFLP; Tub-Gal4 , UAS-mCD8::GFP/CyO . To generate clones for functional imaging , females of the genotype FRT19A , otdYH13/FM7a;GH146-Gal4 , UAS-GCaMP3/CyO were crossed to males of the genotype FR19A , Tub-Gal80 , hsFLP; GH146-Gal4 , UAS-GCaMP3/CyO . Embryos were collected from the progeny of all the clonal crosses and these were given a heat shock for 1 hr at 37°C at either the embryonic stage , 0–4 hr after larval hatching or 24 hr after larval hatching . To knock down Otd in neuroblasts using RNA interference , females of the genotype UASdicer; Insc-Gal4 were crossed to males of the genotype UAS-miRNA-otd-1;GH146-QF , QUAS-mtdTomato-HA/TM6B and grown at 25°C . Brains were dissected in 1× PBS and fixed in freshly prepared 4% PFA for 30 min at room temperature . The fixative was removed , and the brains were washed with blocking solution ( 1× PBS with 0 . 3% TritonX and 0 . 1% BSA ) . Primary antibodies were diluted in blocking solution . The samples were incubated in a moist chamber on horizontal shaker at 4°C for 24 hr . Samples were then washed with 0 . 3% PTX ( 1× PBS with 0 . 3% TritonX ) and secondary antibody diluted in 0 . 3% PTX was added . The samples were incubated in this at 4°C in a moist chamber on horizontal shaker overnight , after which they were washed and mounted in vectashield on glass slides . Primary antibodies used were: rabbit anti-GFP ( 1:10 , 000; Molecular Probes , Invitrogen , Delhi , India ) ; chick anti-GFP ( 1:10 , 000 , AbCam , Cambridge , UK ) ; mouse anti-Bruchpilot ( mAbnc82 , 1:20; DSHB , Iowa , USA ) ; Rb anti-Otd ( 1:1 , 500 , gift from Henry Sun , IMB , Sinica , Taiwan ) ; guinea pig anti-Otd ( 1:7 , 500 , gift from Tiffany Cook , Cincinnati , OH ) ; mouse anti-Acj6 ( 1:25 , DSHB , Iowa , USA ) ; rabbit anti-dLim1 ( 1:1000; a gift from J Botas ) ; rat anti-mCD8 ( 1:100 , Caltag , Burlingname , Californina ) ; mouse anti-rCD2 ( AbD Serotec , Raleigh , NC , USA ) ; mouse anti-Neurotactin ( BP106 , 1:10 , DSHB , Iowa , USA ) ; mouse anti-Neuroglian ( BP102 , 1:10 , DSHB , Iowa , USA ) ; rabbit anti-HA ( 1:100 , AbCam ) . Secondary antibodies used were Alexa Fluor-488- , Alexa Fluor-568- and Alexa Fluor-647-coupled antibodies generated in goat ( 1:400; Molecular Probes , Invitrogen , Delhi , India ) and TO-PRO -3 iodide-661 ( Thermo Fischer Scientific , MA , USA ) . All samples were imaged on Olympus Fluoview ( FV1000 ) laser scanning confocal microscope . Optical sections were acquired at 1-µm intervals with a picture size of 512 × 512 pixels . Images were digitally processed using Adobe Photoshop CS3 . 3-D reconstructions were made using Amira . Brains of clonal animals were dissected in Ca2+-free AHL saline , which contains 108 mM NaCl , 5 mM KCl , 4 mM NaHCO3 , 1 mM NaH2PO4 , 8 . 2 mM MgCl2 , 5 mM HEPES , 5 mM trehalose , and 10 mM sucrose , with pH adjusted to 7 . 4 . Live brains that contain only the otd null transformed neurons were selected for imaging experiments .	The cells in the brain—including the neurons that transmit information—work together in groups called neural circuits . These cells develop from precursor cells called neuroblasts . Each neuroblast can produce many cells , and it is likely that cells that develop from the same neuroblast work together in the adult brain in the same neural circuit . How the adult cells develop into their final form plays an important role in creating a neural circuit , but this process is not fully understood . In many animals , the complexity of their brain makes it difficult to follow how each individual neuroblast develops . However , all of the neuroblasts in the relatively simple brain of the fruit fly Drosophila have been identified . Furthermore , the genes responsible for establishing the initial identity of each neuroblast in the Drosophila brain are known . These genes may also determine which adult neurons develop from the neuroblast , and when each type of neuron is produced . However , the extent to which a single gene can influence the identity of neurons is unclear . Sen et al . focused on two types of neuroblasts , each of which , although found next to each other in the developing Drosophila brain , produces neurons for different neural circuits . One of the neuroblasts generates the olfactory neurons responsible for detecting smells; the other innervates the ‘central complex’ that has a number of roles , including controlling the fly's movements . A gene called orthodenticle is expressed by the central complex neuroblast , but not by the olfactory neuroblast , and helps to separate the two neural circuits into different regions of the fly brain . Sen et al . found that deleting the orthodenticle gene from the central complex neuroblast causes it to develop into olfactory neurons instead of central complex neurons . Tests showed that the modified neurons are completely transformed; they not only work like olfactory neurons , but they also have the same structure and molecular properties . Sen et al . have therefore demonstrated that it is possible to drastically alter the circuitry of the fruit fly brain by changing how one gene is expressed in one neuroblast . This suggests that new neural circuits can form relatively easily , and so could help us to understand how different brain structures and neural circuits evolved .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2014	Genetic transformation of structural and functional circuitry rewires the Drosophila brain
Sterol homeostasis is essential for the function of cellular membranes and requires feedback inhibition of HMGR , a rate-limiting enzyme of the mevalonate pathway . As HMGR acts at the beginning of the pathway , its regulation affects the synthesis of sterols and of other essential mevalonate-derived metabolites , such as ubiquinone or dolichol . Here , we describe a novel , evolutionarily conserved feedback system operating at a sterol-specific step of the mevalonate pathway . This involves the sterol-dependent degradation of squalene monooxygenase mediated by the yeast Doa10 or mammalian Teb4 , a ubiquitin ligase implicated in a branch of the endoplasmic reticulum ( ER ) -associated protein degradation ( ERAD ) pathway . Since the other branch of ERAD is required for HMGR regulation , our results reveal a fundamental role for ERAD in sterol homeostasis , with the two branches of this pathway acting together to control sterol biosynthesis at different levels and thereby allowing independent regulation of multiple products of the mevalonate pathway . Sterols , such as cholesterol in animals or ergosterol in yeast , are essential components of cellular membranes and their concentration to a large extent determines many of the membrane properties , such as fluidity and rigidity . Therefore , cells evolved sophisticated mechanisms to precisely regulate their sterol levels ( Goldstein et al . , 2006; Brown and Goldstein , 2009 ) . These are critical not only for adjusting membrane properties to diverse cellular environments but also for preventing the accumulation of free sterols , which is toxic both to individual cells and to whole organisms ( Goldstein et al . , 2006; Espenshade and Hughes , 2007; Maxfield and van Meer , 2010 ) . The regulation of cellular sterol levels occurs primarily during their biosynthesis in the endoplasmic reticulum ( ER ) by the mevalonate pathway ( Espenshade and Hughes , 2007; Brown and Goldstein , 2009 ) . This highly conserved pathway produces isoprenoids , precursors not only for sterols but also for other essential molecules such as dolichol or ubiquinone ( Figure 1A; Goldstein and Brown , 1990 ) . While a constant supply of these molecules is required , cells must avoid overaccumulation of sterols , a balance that is achieved by a number of feedback systems operating at the transcription , translation , and post-translational levels . Remarkably , decades of work demonstrated that many of these homeostatic control systems converge on the regulation of 3-hydroxy-3-methylglutaryl-coenzyme A reductase ( HMGR ) , an enzyme involved in an early and rate-limiting step of the mevalonate pathway ( Brown and Goldstein , 2009; Burg and Espenshade , 2011 ) . 10 . 7554/eLife . 00953 . 003Figure 1 . Erg1 is a substrate of the Doa10 complex . ( A ) Schematic representation of the mevalonate pathway and its different end products . The steps catalyzed by HMG-CoA reductase and the squalene monooxygenase Erg1 are indicated . Adapted from Goldstein and Brown ( 1990 ) . ( B ) Erg1 abundance in the indicated mutants relative to wt cells , as detected by mass spectrometry upon SILAC labeling . All strains used are lysine auxotrophs and were grown in the presence of either heavy L-lysine ( wt cells ) or light L-lysine ( deletion mutants ) . Note that high steady state levels of Erg1 result in a low heavy/light ratio . ( C ) The degradation of endogenous Erg1 was followed after inhibition of protein synthesis by cycloheximide in wt cells or in cells with the indicated deletions . Whole-cell extracts were analyzed by SDS–PAGE and western blotting . Erg1 was detected with α-Erg1 antibody . Phosphoglycerate kinase ( Pgk1 ) was used as loading control and detected with α-Pgk1 antibodies . A representative gel of three independent experiments is shown . ( D ) The degradation of endogenous Erg1 was analyzed as in ( C ) in wt cells or the temperature sensitive cdc48-3 and npl4-1 cells either at the permissive temperature of 25°C or after a 2 hr shift to 37°C , the restrictive temperature . DOI: http://dx . doi . org/10 . 7554/eLife . 00953 . 00310 . 7554/eLife . 00953 . 004Figure 1—figure supplement 1 . Abundance of Erg1 but not of other components of the Erg pathway is altered in Doa10 complex mutants . ( A ) Abundance of the indicated Erg pathway components in endoplasmic reticulum-associated protein degradation ( ERAD ) mutants relative to wt cells , as detected by mass spectrometry upon SILAC labeling . All strains used are lysine auxotrophs and were grown in the presence of either heavy L-lysine ( wt cells ) or light L-lysine ( deletion mutants ) . Only proteins in which more than one peptide was detected by mass spectrometry were included . Note that high steady state levels of a protein result in a low heavy/light ratio . ( B ) The degradation of Erg1 was followed after inhibition of protein synthesis by cycloheximide in wt cells or in cells with the indicated deletions . Expression of Erg1 was driven from the strong constitutive promoter glyceraldehyde-3-phosphate dehydrogenase ( GAPDH ) replacing the endogenous ERG1 promoter . Whole-cell extracts were analyzed by SDS–PAGE and western blotting . Erg1 was detected with α-Erg1 antibody . Phosphoglycerate kinase ( Pgk1 ) was used as loading control and detected with α-Pgk1 antibodies . A representative gel of three independent experiments is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00953 . 004 A key mechanism in sterol homeostasis involves the proteasomal degradation of HMGR in a sterol-dependent manner ( Burg and Espenshade , 2011 ) . In fact , an increase in sterol biosynthetic intermediates , such as geranylgeranyl pyrophosphate in yeast and lanosterol or its immediate product 24 , 25-dihydrolanosterol in mammals , strongly accelerates the degradation of HMGR , thereby lowering the flux through the mevalonate pathway . Both in yeast and in mammals , the targeting of HMGR to the proteasome for degradation is mediated by a branch of the ER-associated protein degradation ( or ERAD ) pathway , which is primarily studied for its role in the elimination of misfolded ER proteins ( Smith et al . , 2011; Brodsky , 2012 ) . Importantly , ERAD factors involved in the recognition of misfolded ER proteins are not required for HMGR degradation . Instead specific chaperones called Insigs , in the presence of the right sterol signal , regulate the interaction of HMGR with the central ERAD components , a ubiquitin ligase complex in the ER membrane , called Hrd1 in yeast and Gp78 in mammals ( Hampton et al . , 1996; Bays et al . , 2001; Sever et al . , 2003; Flury et al . , 2005; Song et al . , 2005b ) . The Hrd1/Gp78-dependent ubiquitination of HMGR leads to its membrane extraction , facilitated by the Cdc48/p97 ATPase , and release in the cytoplasm for degradation by the proteasome . In mammalian cells , a second ER-bound ubiquitin ligase , Trc8 , can also promote the sterol-dependent degradation of HMGR ( Jo et al . , 2011a ) . Other branches of ERAD have never been implicated in sterol homeostasis . Here , we report a novel role of ERAD in the regulation of sterol biosynthesis . A screen for substrates of the ERAD ubiquitin ligase Doa10 identified the squalene monooxygenase Erg1 , an enzyme required for a sterol-specific step of the mevalonate pathway in Saccharomyces cerevisiae . We show that Doa10-dependent degradation of Erg1 is regulated by the levels of lanosterol and that , together with sterol esterification , it is essential for preventing the accumulation of toxic sterol intermediates . Moreover , in mammalian cells , sterol-dependent degradation of squalene monooxygenase requires the Doa10 homologue Teb4 . Altogether , our findings reveal an evolutionarily conserved , central role of ERAD in sterol homeostasis . To identify novel substrates of the ERAD ubiquitin ligase Doa10 ( Swanson et al . , 2001 ) , we used SILAC ( Stable Isotope Labeling by Amino acids in Culture ) labeling followed by quantitative proteomics ( de Godoy et al . , 2008 ) . We found that , at steady state , several proteins were overrepresented in doa10Δ mutant when compared to wild type ( wt ) cells ( data not shown ) . Among these potential Doa10 substrates was the yeast squalene monooxygenase Erg1 ( Figure 1A , B ) . Other components of the yeast sterol biosynthetic pathway ( ergosterol or Erg pathway ) were present at similar levels in doa10Δ and wt cells ( Figure 1—figure supplement 1A ) . The high levels of Erg1 in doa10Δ mutants are not due to increased ERG1 transcription as the abundance of Erg1 mRNA was indistinguishable between doa10Δ and wt cells ( data not shown ) . Interestingly , cells lacking Ubc6 or Ubc7 , the ubiquitin-conjugating enzymes required for Doa10-dependent ubiquitination and members of the Doa10 complex ( Carvalho et al . , 2006 ) , also showed increased steady state levels of Erg1 when compared to wt cells ( Figure 1B ) . In contrast , deletion of Hrd1 , Der1 , or Usa1 , involved in a different branch of ERAD as part of the Hrd1 complex ( Carvalho et al . , 2006 ) , had no effect on the steady state levels of Erg1 ( Figure 1B ) . To directly test for a role of Doa10 in the turnover of Erg1 , we performed cycloheximide shut-off experiments . We found that in wt cells Erg1 is an unstable protein with a half-life of <120 min ( Figure 1C ) . Deletion of the ubiquitin ligase Doa10 or any of the components of the Doa10 complex strongly impaired the degradation of Erg1 expressed from its own promoter ( Figure 1C ) or from the heterologous glyceraldehyde-3-phosphate dehydrogenase ( GAPDH ) promoter ( Figure 1—figure supplement 1B ) . Similarly , Erg1 degradation was virtually blocked in cdc48-3 and npl4-1 cells , expressing temperature sensitive alleles in essential subunits of the Cdc48 ATPase complex that pulls substrates out of the ER membrane after Doa10-dependent ubiquitination ( Figure 1D ) . On the other hand , in hrd1Δ cells the kinetics of Erg1 degradation was similar to that in wt cells ( Figure 1C and Figure 1—figure supplement 1B ) . Together , these data show that Erg1 is an ERAD substrate of the Doa10 complex . In the vast majority of cases , protein ubiquitination occurs on the ε-amino group of lysine residues . We therefore searched for Erg1 lysine residues required for its Doa10-dependent degradation . We focused our attention on a cluster of residues proximal to the C-terminal region ( K278 , K284 , K311 , and K360 ) that were reported to be ubiquitinated in several large scale studies ( Hitchcock et al . , 2003; Peng et al . , 2003; Beltrao et al . , 2012 ) . We generated Erg1-derivatives in which individual or pairs of the lysine residues were mutated to arginine . These lysine mutant alleles supported the growth of yeast cells , indicating that the substitutions did not significantly affect the essential enzymatic function of Erg1 ( Figure 2A ) . To test the effect of the lysine mutations on Erg1 stability , we performed cycloheximide shut-off assays in otherwise wt cells . Degradation of Erg1 ( K278 , 284R ) and Erg1 ( K360R ) was indistinguishable from degradation of wt Erg1 ( Figure 2B , C ) . In contrast , Erg1 ( K311R ) was strongly stabilized either when expressed from the endogenous ERG1 promoter ( Figure 2B , C ) or from the strong constitutive GAPDH promoter ( Figure 2—figure supplement 1A ) . Importantly , Erg1 ( K311R ) and Erg1 ( K278 , 284 , 311 , 360R ) , with simultaneous mutations on four lysine residues , were only slightly stabilized by additional deletion of DOA10 ( Figure 2B , C ) . Thus , the lysine residue at position 311 is essential for Doa10-dependent degradation of Erg1 . 10 . 7554/eLife . 00953 . 005Figure 2 . Doa10-dependent degradation of Erg1 depends on a single lysine residue . ( A ) Expression of ERG1 or ERG1-derivatives with the indicated lysine mutations from a plasmid rescues the growth of yeast cells upon repression of the endogenous ERG1 . A yeast strain expressing endogenous ERG1 from the regulated GAL1 promoter was transformed with plasmids encoding different lysine mutants or a control empty vector . The growth of serial dilutions of cells was tested under conditions of induced ( galactose-containing media ) or repressed ( glucose-containing media ) endogenous ERG1 . ( B ) The degradation of Erg1 or the indicated Erg1 lysine mutant expressed from the endogenous ERG1 promoter was followed after inhibition of protein synthesis by cycloheximide in wt or doa10Δ cells . Samples were analyzed as in Figure 1C . ( C ) Quantitation of two independent experiments performed as described in ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00953 . 00510 . 7554/eLife . 00953 . 006Figure 2—figure supplement 1 . Doa10-dependent degradation of Erg1 depends on the lysine residue at position 311 . ( A ) The degradation of Erg1 or the indicated Erg1 lysine mutants expressed from the constitutive glyceraldehyde-3-phosphate dehydrogenase ( GAPDH ) promoter was followed after inhibition of protein synthesis by cycloheximide . Samples were analyzed as in Figure 1C . DOI: http://dx . doi . org/10 . 7554/eLife . 00953 . 006 We then tested whether Doa10-dependent degradation of Erg1 was affected by the levels of cellular sterols . To manipulate the levels of sterols and sterol biosynthetic intermediates , we took advantage of several well characterized small molecule inhibitors of the Erg pathway enzymes ( Figure 3A ) . Reduction of ergosterol synthesis in wt cells by a brief treatment with zaragozic acid , an inhibitor of the squalene synthetase Erg9 , led to a strong stabilization of Erg1 when compared to controls ( Figure 3B ) . Similar treatments did not affect the degradation kinetics of the Doa10 substrates Vma12-Ndc10902–956-HA and Pca11–392-DHFR-HA ( Figure 3C and Figure 3—figure supplement 1A–C ) ( Adle et al . , 2009; Furth et al . , 2011 ) . These experiments indicate that sterol depletion specifically affects the Doa10-dependent degradation of Erg1 but not other Doa10 substrates , such as misfolded proteins . Similar results were obtained upon treatment of wt cells with Ro48-807 , an inhibitor of the squalene cyclase Erg7 and whose product ( lanosterol ) is the first sterol of the pathway ( Figure 3A , D , E and Figure 3—figure supplement 1D–F ) . The next step in ergosterol biosynthesis is catalyzed by Erg11 , the lanosterol demethylase , which is inhibited by fluconazole ( Figure 3A ) . In contrast to zaragozic acid and Ro48-807 , a short treatment of wt cells with fluconazole induced a marked acceleration of Erg1 degradation ( Figure 3F , G ) . Importantly , the acceleration of Erg1 degradation upon fluconazole treatment was completely dependent on Doa10 , as it was blocked in doa10Δ cells ( Figure 3F ) . Moreover , fluconazole treatment did not have any effect on the kinetics of degradation of the Doa10 substrates Pca11–392-DHFR-HA ( Figure 3G and Figure 3—figure supplement 1G ) and Vma12-Ndc10902–956-HA ( Figure 3—figure supplement 1H ) . Therefore , lanosterol accumulation induced by fluconazole treatment stimulates Doa10-dependent degradation of Erg1 . In agreement with these results , we found that Erg1 degradation was also accelerated in a erg11DAmP mutant bearing a hypomorphic allele of ERG11 ( Figure 3H ) . Mutation of genes required for the very last steps of ergosterol biosynthesis ( ERG2 to ERG6 ) did not have a major impact on the kinetics of Erg1 degradation ( Figure 3—figure supplement 1I ) . Taken together , these data demonstrate that Doa10-dependent degradation of Erg1 is regulated by the levels of sterols , most likely lanosterol . 10 . 7554/eLife . 00953 . 007Figure 3 . Flux through the sterol pathway regulates Erg1 degradation . ( A ) Schematic representation of the ergosterol biosynthetic pathway highlighting the enzymatic steps affected by the small inhibitor zaragozic acid , Ro48-807 , and fluconazole . The step catalyzed by the squalene monooxygenase Erg1 is also indicated . ( B ) The degradation of endogenous Erg1 was followed after inhibition of protein synthesis by cycloheximide in wt control cells ( DMSO ) or wt cells treated for 2 hr with 10 µg/ml zaragozic acid ( ZA ) . Samples were analyzed as described in Figure 1C . ( C ) Quantitation of three independent experiments as described in ( B ) in cells expressing the Doa10 substrate Vma12-Ndc10902–956-HA , a model misfolded protein . Vma12-Ndc10902–956-HA was detected with anti-HA antibodies . ( D ) The degradation of endogenous Erg1 was followed after inhibition of protein synthesis by cycloheximide in wt control cells ( DMSO ) or wt cells treated for 2 hr with 40 µg/ml Ro48-807 ( Ro48 ) . Samples were analyzed as described in Figure 1C . ( E ) Quantitation of three independent experiments as described in ( D ) in cells expressing the Doa10 substrate Vma12-Ndc10902–956-HA , a model misfolded protein . Vma12-Ndc10902–956-HA was detected with anti-HA antibodies . ( F ) The degradation of endogenous Erg1 was followed after inhibition of protein synthesis by cycloheximide in wt and doa10Δ cells . Cells were incubated for 1 hr with DMSO or with 10 µg/ml fluconazole ( Fluco ) . Samples were analyzed as described in Figure 1C . ( G ) Quantitation of three independent experiments as described in ( F ) in cells expressing the Doa10 substrate Pca11–392-DHFR-HA , a model misfolded protein . Pca11–392-DHFR-HA was detected with anti-HA antibodies . ( H ) The degradation of endogenous Erg1 was followed after inhibition of protein synthesis by cycloheximide in wt cells or in cells with the indicated mutations . Samples were analyzed as described in Figure 1C . DOI: http://dx . doi . org/10 . 7554/eLife . 00953 . 00710 . 7554/eLife . 00953 . 008Figure 3—figure supplement 1 . Sterol depletion affects Doa10-dependent degradation of Erg1 but not of other Doa10 substrates . ( A ) The degradation of the Doa10 substrates Vma12-Ndc10902–956-HA and Erg1 was followed after inhibition of protein synthesis by cycloheximide in wt control cells ( DMSO ) or in wt cells treated for 2 hr with 10 µg/ml zaragozic acid ( ZA ) . Samples were analyzed as described in Figure 1C . Vma12-Ndc10902–956-HA was detected with anti-HA antibodies . ( B ) The degradation of the Doa10 substrates Pca11–392-DHFR-HA and Erg1 was followed as in ( A ) . ( C ) Quantitation of three independent experiments as described in ( A ) in cells expressing the Doa10 substrate Pca11–392-DHFR-HA . ( D ) The degradation of the Doa10 substrates Vma12-Ndc10902–956-HA and Erg1 was followed after inhibition of protein synthesis by cycloheximide in wt control cells ( DMSO ) or in wt cells treated for 2 hr with 40 µg/ml Ro48-807 ( Ro48 ) . Samples were analyzed as described in Figure 1C . Vma12-Ndc10902–956-HA was detected with anti-HA antibodies . ( E ) The degradation of the Doa10 substrates Pca11–392-DHFR-HA and Erg1 was followed as in ( D ) . ( F ) Quantitation of three independent experiments as described in ( D ) in cells expressing the Doa10 substrate Pca11–392-DHFR-HA . ( G ) The degradation of the Doa10 substrates Pca11–392-DHFR-HA and Erg1 was followed after inhibition of protein synthesis by cycloheximide in wt control cells ( DMSO ) or in cells treated for 1 hr with 10 µg/ml fluconazole ( Fluco ) . Samples were analyzed as described in Figure 1C . ( H ) The degradation of the Doa10 substrates Vma12-Ndc10902–956-HA and Erg1 was followed after inhibition of protein synthesis by cycloheximide in wt control cells ( DMSO ) or in cells treated for 1 hr with 10 µg/ml fluconazole ( Fluco ) . Samples were analyzed as described in Figure 1C . ( I ) The degradation of endogenous Erg1 was followed after inhibition of protein synthesis by cycloheximide in wt cells or in cells with the indicated mutations . Samples were analyzed as described in Figure 1C . DOI: http://dx . doi . org/10 . 7554/eLife . 00953 . 008 We then asked whether the Doa10-dependent degradation of Erg1 had an effect on sterol homeostasis . To address this issue , we analyzed the lipid composition of doa10Δ cells by shotgun lipidomics ( Ejsing et al . , 2009 ) . While the overall sterol levels in doa10Δ and wt cells were comparable , the relative abundance of individual sterol species was significantly different ( Figure 4A ) . When compared to wt cells , the doa10Δ mutant showed a reduction in the levels of ergosterol ( by 13% ) with a concomitant fivefold increase in the ergosterol precursor lanosterol . Moreover , doa10Δ cells had small amounts of ergostadienol , a sterol intermediate virtually undetectable in wt cells ( Figure 4A ) . Ergosterol concentration is a major determinant of membrane fluidity , which is tightly regulated according to the cellular environment ( Brown and Goldstein , 2009; Burg and Espenshade , 2011 ) . The changes in sterol composition observed in doa10Δ mutants are expected to affect the membrane properties in these cells . Accordingly , we noticed that mutations in components of the Doa10 complex have a mild growth phenotype when grown at low temperatures ( 10°C ) , as previously reported ( Loertscher et al . , 2006 and data not shown ) . Altogether , the data presented so far indicate that sterol-dependent degradation of Erg1 by Doa10 is part of a feedback system essential for sterol homeostasis . 10 . 7554/eLife . 00953 . 009Figure 4 . Doa10-dependent degradation of Erg1 affects sterol homeostasis and , together with sterol esterification , is essential to prevent buildup of sterol intermediates . ( A ) Relative amounts of ergosterol , lanosterol , and ergostadienol in cells with the indicated genotype . Cells were grown in synthetic complete ( SC ) media until early stationary phase , and lipids were extracted and analyzed by shotgun lipidomics . ( B ) Analysis of sterol esters in wt and doa10Δ cells by mass spectrometry . Cells were grown in SC media until early stationary phase , and lipids were extracted and analyzed by shotgun lipidomics . ( C ) Serial dilutions of cells with the indicated genotype were spotted on YPD or YPD + 0 . 3% of benzyl alcohol ( BA ) and incubated for 2 ( 25°C ) , 4 ( YPD + 0 . 3% BA ) or 5 days ( 14°C ) . ( D ) Serial dilutions of cells with the indicated genotype were spotted on YPD and incubated for 2 ( 25°C ) or 5 days ( 14°C ) . ( E ) The degradation of endogenous Erg1 was followed after inhibition of protein synthesis by cycloheximide in cells with the indicated genotype . Samples were analyzed as described in Figure 1C . ( F ) The degradation of Erg1 or the indicated Erg1 lysine mutants expressed from the constitutive glyceraldehyde-3-phosphate dehydrogenase ( GAPDH ) promoter was followed after inhibition of protein synthesis by cycloheximide in wt or are1Δ are2Δ cells . Samples were analyzed as described in Figure 1C . DOI: http://dx . doi . org/10 . 7554/eLife . 00953 . 009 A mechanism preventing the accumulation of sterol intermediate metabolites , which are toxic , is their esterification into sterol esters ( Chang et al . , 2006; Jacquier and Schneiter , 2012 ) . Interestingly , doa10Δ cells also accumulated significantly higher amounts of sterol esters ( approximately a 40% increase in relation to wt cells; Figure 4B ) . This increase prompted us to investigate a potential connection between the Doa10-dependent feedback regulation of Erg1 and sterol esterification . In yeast , esterified sterols are produced by two partially redundant acyl-CoA sterol acyl transferases ( ASATs ) , Are1 and Are2 , that are not essential for growth under laboratory conditions ( Yang et al . , 1996; Zweytick et al . , 2000 ) . We then again used shotgun lipidomics to evaluate the effect of the DOA10 mutation on sterol composition in esterification-deficient are1Δ are2Δ cells . Compared to wt cells , are1Δ are2Δ mutant has ∼20% less ergosterol ( Figure 4A ) , as previously reported ( Yang et al . , 1996; Zweytick et al . , 2000 ) . Moreover , are1Δ are2Δ cells show significantly higher amounts of lanosterol and ergostadienol ( Figure 4A ) , indicating that loss of esterification also leads to the accumulation of sterol intermediates . Strikingly , in are1Δ are2Δ doa10Δ mutant the levels of ergosterol dropped by more than 50% , while the levels of the intermediates lanosterol and ergostadienol increased dramatically ( Figure 4A ) . The changes observed in the are1Δ are2Δ doa10Δ mutant , although of a much larger magnitude , perfectly mirror those observed in doa10Δ cells . Importantly , the effects appear specific to doa10Δ , as deletion of HRD1 did not significantly change the sterol profile of esterification-deficient cells ( data not shown ) . These data suggest that sterol esterification and the ubiquitin ligase Doa10 play parallel , redundant functions in preventing accumulation of sterol intermediate metabolites . Does the massive increase in sterol intermediates at the expense of the final product , ergosterol , have any impact in the cellular physiology of are1Δ are2Δ doa10Δ cells ? To address this issue , we have performed growth assays under conditions of membrane stress such as low temperature or the presence of benzyl alcohol , a membrane fluidizing agent . Although are1Δ are2Δ doa10Δ mutant grows normally at 25°C , these cells barely grew when incubated at 14°C or in the presence of benzyl alcohol ( Figure 4C ) . The same phenotype was observed when are1Δ are2Δ mutation was combined with deletion of UBC7 , encoding for the ubiquitin-conjugating enzyme in complex with Doa10 ( data not shown ) . In contrast , the are1Δ are2Δ hrd1Δ mutant cells , lacking the ubiquitin ligase of the other ERAD branch , did not display any growth defect at 14°C or in the presence of benzyl alcohol ( Figure 4C ) . These results indicate that accumulation of sterol intermediates with a concomitant decrease in ergosterol levels renders are1Δ are2Δ doa10Δ cells vulnerable to membrane stress , perhaps due to the inability to adjust membrane fluidity . To rule out the possibility that the growth phenotypes of are1Δ are2Δ doa10Δ are due to pleiotropic effects of doa10Δ , for example due to its role in the clearance of misfolded proteins , we took advantage of the degradation resistant ERG1 allele , ERG1 ( K311R ) . Remarkably , constitutive expression from the GAPDH promoter of ERG1 ( K311R ) in are1Δ are2Δ cells led to a cold phenotype comparable to that observed in are1Δ are2Δ doa10Δ mutant ( Figure 4D ) . In contrast , expression of ERG1 or ERG1 ( K360R ) under the same conditions did not cause any apparent defect ( Figure 4D ) , suggesting that Erg1 is a key Doa10 substrate for sterol homeostasis . In sum , these data show that Are1 , Are2-dependent sterol esterification and Doa10-dependent degradation of Erg1 are parallel mechanisms preventing abnormal accumulation of potentially toxic sterol intermediates . Interestingly , these two mechanisms appear to be tightly integrated in cells . As suggested by previous studies ( Sorger et al . , 2004 ) , we found that degradation of Erg1 is accelerated in are1Δ are2Δ cells ( Figure 4E , F ) , likely due to the higher levels of lanosterol in this mutant ( Figure 4A ) . Importantly , this accelerated degradation still requires the lysine at position 311 ( Figure 4F ) and is completely dependent on Doa10 as it is blocked in doa10Δ cells ( Figure 4E ) . The homologue of Erg1 in mammals , SM , is ubiquitinated and degraded by the proteasome ( Gill et al . , 2011 ) . Interestingly , SM proteasomal-dependent degradation is stimulated by sterols , in this case by the final product of the pathway , cholesterol ( Gill et al . , 2011 ) . These similarities prompted us to test whether the mammalian homologue of Doa10 , the ubiquitin ligase Teb4 ( Swanson et al . , 2001; Hassink et al . , 2005 ) , was involved in the turnover of SM . We transfected human embryonic kidney ( HEK ) 293 cells with a pool of siRNA against Teb4 , which led to a 57% ( ±0 . 044% ) reduction in TEB4 mRNA levels compared to control treated cells , as detected by qPCR . We noticed that although transcription of SM in Teb4-depleted and control cells was indistinguishable , the steady state levels of SM were 1 . 8-fold ( ±0 . 232 ) higher in cells treated with Teb4 siRNA ( Figure 5A , ‘untreated’ lanes ) . This stabilization is consistent with a role of Teb4 in the degradation of SM . To directly examine SM half-life upon Teb4 depletion we performed cycloheximide shut-off experiments with or without the addition of cholesterol to stimulate SM degradation . These experiments were performed in sterol-deprived cells , as previously described ( Gill et al . , 2011; Jo et al . , 2011b ) . In these conditions , SM appears to have a relatively long half-life that is even longer in Teb4 siRNA-treated cells ( Figure 5A , B ) . Cholesterol treatment of control cells induces very rapid degradation of SM ( half-life <4 hr ) , as previously shown ( Gill et al . , 2011 ) . In contrast , cholesterol treatment in Teb4-depleted cells has a much milder effect on the degradation of SM ( Figure 5A , B ) and its half-life remains longer than 4 hr ( Figure 5B ) . Both in control and in Teb4-depleted cells , SM sterol-dependent degradation is significantly attenuated by the proteasome inhibitor MG132 ( Figure 5A ) . To independently assess the role of Teb4 in the regulated degradation of SM , HEK293 cells were transfected with constructs overexpressing Teb4-myc or Teb4 ( C9A ) -myc , a dominant negative mutant that lacks ubiquitin ligase activity ( Hassink et al . , 2005 ) . While SM degradation occurred with normal kinetics in Teb4-overexpressing cells , expression of Teb4 ( C9A ) strongly inhibited the cholesterol-dependent acceleration of SM degradation , as assayed by cycloheximide shut-off experiments ( Figure 5—figure supplement 1 ) . Altogether , these data indicate that the Teb4 ubiquitin ligase promotes the degradation of SM in a cholesterol-dependent manner . Moreover , it emphasizes the remarkable conservation of the role of ERAD in regulating sterol homeostasis in yeast and mammals . 10 . 7554/eLife . 00953 . 010Figure 5 . Doa10 homologue Teb4 promotes degradation of human squalene monooxygenase . ( A ) The degradation of endogenous squalene monooxygenase ( SM ) was followed after inhibition of protein synthesis by cycloheximide in sterol-deprived HEK293 cells treated with control siRNA or siRNA targeting the ubiquitin ligase Teb4 . The degradation of SM was monitored under basal conditions ( methyl-β-cyclodextrin vehicle , +CD ) and upon addition of cholesterol ( +CD/Ch ) . Where indicated DMSO or the proteasome inhibitor MG132 ( 10 µM ) was included . Samples were analyzed by SDS–PAGE and immunoblotting . SM was detected with rabbit polyclonal anti-SM antibodies and α-tubulin with mouse monoclonal anti-α-tubulin antibodies . The arrow indicates the band corresponding to SM . ( B ) Quantitation of three independent experiments performed as described in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00953 . 01010 . 7554/eLife . 00953 . 011Figure 5—figure supplement 1 . Overexpression of dominant negative Teb4 ( C9A ) , but not of wt Teb4 , strongly delays degradation of human squalene monooxygenase . The degradation of endogenous squalene monooxygenase ( SM ) was followed after inhibition of protein synthesis by cycloheximide in sterol-deprived HEK293 cells transfected with Teb4-myc or Teb4 ( C9A ) -myc . The degradation of SM was monitored under basal conditions ( methyl-β-cyclodextrin vehicle , +CD ) and upon addition of cholesterol ( +CD/Ch ) . Where indicated , DMSO or the proteasome inhibitor MG132 ( 10 µM ) was included . Samples were analyzed by SDS–PAGE and immunoblotting . SM was detected with rabbit polyclonal anti-SM antibodies and α-tubulin with mouse monoclonal anti-α-tubulin antibodies . The arrow indicates the band corresponding to SM . DOI: http://dx . doi . org/10 . 7554/eLife . 00953 . 011 Here we characterize a novel role of the ERAD pathway in the homeostatic control of sterol levels and demonstrate that loss of this feedback system leads to sterol deregulation . We show that the yeast ubiquitin ligase Doa10 , along with the other components of the Doa10 complex , promotes the regulated degradation of the squalene monooxygenase Erg1 . The Doa10-mediated degradation of Erg1 is stimulated by lanosterol and loss of Doa10 leads to accumulation of lanosterol and sterol esters . Moreover , we found that the requirement for Doa10 function increases in sterol esterification-deficient cells , indicating that feedback regulation of Erg1 and sterol esterification are redundant pathways preventing accumulation of sterol intermediates . Lanosterol is the first sterol produced by the pathway and , in mammalian cells , its levels or the levels of an immediate product ( 24 , 25-dihydrolanosterol ) are key in the regulation of sterol biosynthesis . Specifically , an increase in lanosterol/24 , 25-dihydrolanosterol levels stimulates the Gp78-dependent ubiquitination of HMGR , its proteasomal degradation , and consequently a decrease in the rate of sterol biosynthesis ( Song et al . , 2005a; Lange et al . , 2008; Nguyen et al . , 2009 ) . We have now found that in yeast , high levels of lanosterol also lead to a decrease in flux through the sterol pathway , in this case by the Doa10-dependent degradation of Erg1 . Together these observations suggest that the regulation of the rate of sterol biosynthesis according to the lanosterol levels might be a unifying principle of sterol homeostasis . Besides being the first sterol of the pathway , lanosterol is also a poor substrate for the sterol esterification enzyme Are2 , the most active of the yeast ASATs ( Zweytick et al . , 2000 ) . Therefore , lanosterol levels might provide an accurate measurement of the flux into the sterol branch of the mevalonate pathway ( Zweytick et al . , 2000; Espenshade and Hughes , 2007 ) . High Erg1 levels in doa10Δ cells lead to accumulation of lanosterol , that is particularly prominent in cells also lacking the ASATs Are1 and Are2 . This huge increase in lanosterol levels is somewhat surprising because these cells have an intact Erg11 , the lanosterol 14-α-demethylase that consumes lanosterol . This indicates that Erg11 activity is either rate limiting or regulated . Erg11 is an enzyme of the cytochrome P450 family that uses heme as a co-factor . Therefore , it is possible that lanosterol accumulation in these mutants is a consequence of changes in the availability of heme , also a product of the mevalonate pathway . Alternatively , the lanosterol accumulation might result from a still unknown regulatory system acting on Erg11 itself or another downstream enzyme . This possibility is supported by the observation that simultaneous overexpression of Erg1 and Erg11 leads to a significant increase in the production of ergosterol in yeast ( Veen et al . , 2003 ) . Besides the Doa10-dependent degradation of Erg1 in yeast , we show that the Doa10 homologue Teb4 also promotes the degradation of SM in mammalian cells . While the degradation of yeast Erg1 is accelerated by the sterol intermediate lanosterol , SM degradation is stimulated by the end product of the sterol pathway , cholesterol ( Gill et al . , 2011 ) . Moreover , Doa10-dependent degradation of Erg1 requires a single lysine residue that is poorly conserved outside fungi , while proteasomal-dependent degradation of SM depends on an N-terminal fragment conserved only in higher animals ( Gill et al . , 2011 ) . Despite these differences , it appears that the post-translation feedback regulation of squalene monooxygenase by the Doa10/Teb4 ubiquitin ligase is a conserved mechanism of sterol homeostasis . In fact , this regulatory system might also operate in plants , as suggested by the observation that mutations of Arabidopsis thaliana Doa10 homologue suppress a hypomorphic allele of squalene monooxygenase ( Doblas et al . , 2013 ) . HMGR , the major target for Hrd1/Gp78 regulation , acts at a very early step of the mevalonate pathway and is required for the synthesis of all its products ( Figure 1A ) . On the other hand , squalene monooxygenase is required for the synthesis of sterols but not of other key metabolites of the mevalonate pathway , such as dolichol or ubiquinone ( Figure 1A ) . Therefore , the novel feedback system described here allows cells to independently regulate the biosynthesis of the different metabolites derived from mevalonate . Remarkably , regulation of HMGR and squalene monooxygenase is mediated by two branches of the ERAD pathway , both in yeast and in mammals . As previously shown , the ubiquitin ligase Hrd1/Gp78 promotes the degradation of HMGR ( Bays et al . , 2001; Song et al . , 2005b ) . Our results now show that the ubiquitin ligase Doa10/Teb4 promotes the degradation of squalene monooxygenase , which place ERAD at the center of cellular sterol homeostasis , with multiple branches of ERAD acting together to regulate sterol biosynthesis at different levels ( Figure 6 ) . More broadly , our findings raise the possibility that regulated degradation of folded , active proteins might be a more prominent feature of ERAD , a pathway primarily studied for its role in the elimination of misfolded proteins . Whether regulated ERAD substrates are all related to sterol homeostasis or ERAD plays a more general role in regulating the ER proteome remains to be determined . 10 . 7554/eLife . 00953 . 012Figure 6 . A central role of endoplasmic reticulum-associated protein degradation in sterol homeostasis . ( A ) Schematic representation of the feedback inhibition systems required for sterol homeostasis in yeast ( left ) and mammals ( right ) previously characterized ( dotted lines ) and described here ( solid lines ) . Endoplasmic reticulum-associated protein degradation ( ERAD ) ubiquitin ligases are in bold and the enzymes targeted by ERAD-regulated degradation are enclosed in gray boxes . DOI: http://dx . doi . org/10 . 7554/eLife . 00953 . 012 Gp78-mediated degradation of HMGR involves proteins called Insigs ( Song et al . , 2005b; Espenshade and Hughes , 2007 ) . These proteins function as adaptors: in a sterol-dependent manner they bind to the sterol-sensing domain of HMGR and recruit Gp78 ligase complex ( Song et al . , 2005b ) . However Insigs are not required for the degradation of squalene monooxygenase either in yeast ( AR and PC unpublished results ) or in mammals ( Gill et al . , 2011 ) . Therefore , future studies should elucidate the mechanisms by which squalene monooxygenase is recognized by the ubiquitin ligase Doa10/Teb4 in a sterol-dependent manner . Heavy lysine [13C615N2] was purchased from Sigma-Aldrich ( St Louis , MO ) . Cycloheximide ( CHX; Sigma-Aldrich ) stock was stored at 12 . 5 mg/ml in water at −20°C and was used at 250 µg/ml . Zaragozic acid ( Sigma-Aldrich ) stock was stored at 10 mg/ml in DMSO at 20°C and used at 10 µg/ml . Ro48-8071 ( Sigma-Aldrich ) stock was stored at 40 mg/ml in DMSO and used at 40 µg/ml . Fluconzole ( Sigma-Aldrich ) stock was stored at 5 mg/ml in DMSO and used at 10 µg/ml . Rat monoclonal anti-HA high-affinity antibody ( 3F10 ) was purchased from Roche . Mouse monoclonal anti-phosphoglycerate kinase ( Pgk1 ) was purchased from Invitrogen . The polyclonal rabbit anti-Erg1 antibody was a generous gift from Dr Chao-Wen Wang , Academia Sinica , Taipei , Taiwan . Mouse monoclonal antibodies against human α-tubulin were purchased from Sigma-Aldrich . Rabbit polyclonal antibodies against human SM were purchased from Proteintech ( Chicago , IL ) . All other reagents and chemicals were purchased from Sigma-Aldrich unless otherwise stated . Tagging of proteins and individual gene deletions were performed by standard PCR-based homologous recombination ( Longtine et al . , 1998 ) . Strains with multiple gene deletions and/or genomically encoded fusion proteins were made by PCR-based homologous recombination ( Longtine et al . , 1998 ) or by crossing haploid cells of opposite mating types , followed by sporulation and tetrad dissection using standard protocols ( Guthrie and Fink , 1991 ) . The strains used are isogenic either to BY4741 ( Mata ura3Δ0 his3Δ1 leu2Δ0 met15Δ0 ) , BY4742 ( Matα his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0 ) , or FY251 ( Mata ura3-52 his3Δ200 leu2Δ1 trp1Δ63 ) and are listed in Supplementary file 1A . Plasmids and primers used in this study are listed in Supplementary files 1B and C , respectively . ERG1 gene ( including endogenous promoter and terminator regions ) was cloned by PCR amplification from genomic DNA isolated from the strain BY4741 . Primers were designed to anneal approximately 500 bp upstream and downstream of ERG1 ORF and to introduce NotI and XhoI restriction sites for cloning in the polylinker of pRS316 plasmid originating the plasmid pPC943 . Erg1-derivatives with specific lysine to arginine mutations were made by site-directed mutagenesis using pPC943 as template ( Supplementary file 1B ) . To replace the endogenous copy of Erg1 with the lysine mutant versions , the C-terminal portion of Erg1 was amplified by PCR from pPC943 including the Erg1 3´UTR region . This fragment was digested with XbaI and NotI restriction sites ( introduced by the primers ) and cloned in a pRS303 plasmid originating pPC995 . The resulting plasmid was linearized with EcoRI and integrated into the ERG1 locus of the recipient strain . Transformants were selected based on HIS prototrophy and confirmed by DNA sequencing . In all cases , the recombinant plasmids were able to restore sterol synthesis , with the engineered Erg1 protein as the sole source of Erg1 activity . Wild type strain BY4741 was transformed by PCR-based homologous recombination to introduce the GAL promoter and HA-tag upstream of ERG1 ORF generating strain yPC6475 . PCR product was obtained from amplification of plasmid pYM-N24 ( Janke et al . , 2004 ) using primers 1242 and 1243 . Strain yPC6475 was transformed with plasmids encoding Erg1-derivatives with specific lysine to arginine mutations as specified in Supplementary files 1A , B . The plasmid encoding for the misfolded protein Pca11–392-DHFR-HA ( pPC860 ) was generated by cloning the DNA sequence coding for FLAG-Pca11–392 under the constitutive PRC1 promoter into pRS316 and a sequence coding Escherichia coli DHFR was fused in frame , followed by a 3× HA tag-coding sequence and the PRC1 terminator . The plasmid encoding for the modular misfolded protein Vma12-Ndc10902–956-HA ( pPC926 ) was generated by cloning the sequence coding full-length Vma12 fused to a fragment of Ndc10 that is sufficient to promote its Doa10-dependent degradation , Ndc10902–956 ( Furth et al . , 2011 ) , and to a GGS–3× HA tag-coding sequence . The resulting fusion protein was expressed under the constitutive PRC1 promoter from the plasmid pRS316 . For SILAC experiments , the yeast strain BY4742 , which is an auxotroph for lysine , was used to generate the strains doa10Δ , ubc6Δ , ubc7Δ , hrd1Δ , usa1Δ , and der1Δ by PCR-based homologous recombination . Cells were grown in 5 ml of synthetic complete media containing either 1 mM L-lysine ( light ) or 1 mM L-lysine [13C615N2] ( heavy ) until stationary phase . Cultures were then diluted to OD600 ∼0 . 005 in 50 ml fresh medium of the same composition . Cells were harvested at OD600 of 1 . 0–1 . 2 . Then 20 OD of cells grown in light lysine media were mixed with 20 OD of cells grown in heavy lysine , harvested by centrifugation , and washed twice with cold water . Total cell extracts were prepared by glass-bead disruption in lysis buffer ( 6 M urea , 50 mM Tris–Cl pH 8 . 0 , 0 . 5% SDS , 0 . 5% NP-40 , 10 mM DDT ) . After removal of cell debris by centrifugation at 2000×g , proteins were precipitated by the addition of TCA to a final concentration of 20% . The protein pellet was washed with ice-cold acetone which was removed by drying the pellet at 50°C . Proteins were resuspended in a small amount of buffer ( 6 M urea , 50 mM Tris–Cl pH 7 . 5 ) and quantified by the BCA quantification method ( Bio-Rad , Hercules , CA ) . Urea concentration was reduced to 4 M and the pH was adjusted to 8 . 8 with 25 mM Tris–HCl . Samples were digested with Lys-C in a 1:10 enzyme–protein ratio and desalted using an Oasis Plus HLB cartridge ( Waters , Milford , MA ) . Digestions were fractionated using electrostatic repulsion-hydrophilic interaction chromatography ( ERLIC ) . Each peptide fraction was separated by nanoLC in an EasyLC system ( Proxeon ) prior to mass spectrometric analysis on an LTQ-Orbitrap Velos Pro ( Thermo Fisher Scientific , Waltham , MA ) fitted with a nanospray source ( Thermo Fisher Scientific ) . Data analysis was performed using the Proteome Discoverer software suite ( v1 . 3 . 0 . 339; Thermo Fisher Scientific ) and the Mascot search engine ( v2 . 3; Matrix Science ) was used for peptide identification . Data were searched against an in-house generated database containing all proteins in the Saccharomyces Genome Database plus the most common contaminants . The identified peptides were filtered using an FDR lower than 1% . Peptide areas were used to calculate the heavy–light ratios . Cycloheximide shut-off experiments were performed in exponentially growing cells , as described ( Carvalho et al . , 2010 ) . cdc48-3 and npl4-1 temperature sensitive strains were grown in synthetic complete medium at permissive temperature ( 25°C ) to exponential phase and then shifted to 37°C for 2 hr before addition of cycloheximide . Sterol synthesis inhibitors were added to culture medium 2 hr ( ZA and Ro-48 ) or 1 hr ( Fluco ) before addition of cycloheximide . DMSO at the same final concentration was added to control strains under the same experimental conditions . For each time point , samples corresponding to 1 OD of yeast cells were collected and protein extracts were prepared and analyzed as previously described ( Carvalho et al . , 2010 ) . Quantification of lipid species was performed essentially as previously described ( Ejsing et al . , 2009; Klose et al . , 2012 ) . In short , yeast cell pellets were resuspended in 1 ml 155 mM ammonium acetate and disrupted by vigorous shaking in the presence of 300 µl glass beads in 1 . 5 ml Eppendorf tubes . Cell lysates equivalent to 0 . 4 OD600 in 200 µl 155 mM ammonium acetate were spiked with a cocktail of internal lipid standards followed by two-step lipid extraction . Lipid extracts were analyzed using a LTQ Orbitrap XL mass spectrometer ( Thermo Fisher Scientific ) equipped with a robotic TriVersa NanoMate ion source ( Advion Biosciences , Ithaca , NY ) . Lipid species were identified and quantified using MSFileReader ( Thermo Fisher Scientific ) , ALEX software , and Orange software 2 . 6 ( Curk et al . , 2005 ) . Lipid species were annotated using sum composition nomenclature as previously described ( Klose et al . , 2012 ) . Tissue culture cells were grown in monolayer at 37°C in an atmosphere of 8–9% CO2 . HEK293 cells were cultured in DMEM high-glucose medium ( Dulbecco’s modified Eagle’s medium containing 100 U/ml penicillin and 100 mg/ml streptomycin sulfate ) supplemented with 10% ( vol/vol ) fetal calf serum ( FCS ) . Cells were seeded in 6-well plates at 2 × 105 cells/well . The next day , they were transfected using Fugene HD reagent ( Promega , Madison , WI ) with a DNA–transfection reagent ratio of 1:3 , according to the manufacturer’s instructions . Myc-tagged TEB4- or TEB4 ( C9A ) -encoding plasmids were previously described in Hassink et al . ( 2005 ) . On the second day after transfection , cells were sterol deprived by an overnight statin treatment as described below . RNAi was carried out as described previously with minor modifications ( Sever et al . , 2003 ) . HEK293 cells were set up on day 1 at 2 × 105 cells/well in DMEM high-glucose medium containing 10% FCS . Cells were incubated with 20 µM of SMARTpool ON-TARGETplus MARCH6 siRNA ( L-006925-00-0005 ) or 20 µM ON-TARGETplus Non-targeting Pool siRNA ( D-001810-10-05 ) mixed with HiPerFect Transfection Reagent ( Qiagen , Hilden , Germany ) that was diluted in Opti-MEM I-reduced serum medium ( Invitrogen , Carlsbad , CA ) according to the manufacturer’s instructions . After 24 hr incubation at 37°C , fresh DMEM high-glucose medium containing 10% FCS was added . On day 3 , cells were incubated for 16 hr at 37°C in DMEM high-glucose medium containing 5% fetal bovine lipoprotein-deficient serum ( LPDS ) , the HMGR inhibitor compactin ( 5 µM ) , and a low level of mevalonate ( 50 µM ) that allows synthesis of essential non-sterol isoprenoids but not of cholesterol ( Hartman et al . , 2010 ) . After 16 hr incubation , residual compactin was removed by washing the cells with PBS . Cells were treated as indicated in the figure legends . Test agents comprised 10 µg/ml cycloheximide , 20 µg/ml methyl-β-cyclodextrin ( CD ) , or cholesterol/methyl-β-cyclodextrin complex ( Chol/CD ) ( prepared as described previously in Brown et al . [2002] ) , and 10 µM MG-132 or an equivalent amount of DMSO . Cell lysates were prepared in 100 µl sample buffer ( 100 mM Tris–Cl pH 6 . 8 , 3% SDS , 15% glycerol ) after a wash step with PBS . Samples ( usually 40 µg of proteins ) were analyzed by 4–15% gradient SDS–PAGE and immunoblotted with the following antibodies: rabbit polyclonal anti-SM ( 1:5000 ) and mouse monoclonal anti-α-tubulin ( 1:1000 ) . The relative intensities of bands were quantified using Quantity One software ( Bio-Rad ) . RNA was harvested using the RNeasy RNA extraction kit ( Qiagen ) and reverse transcribed to yield complementary DNA ( cDNA ) with the SuperScript III First Strand cDNA Synthesis kit ( Invitrogen ) . Levels of SM and MARCH6/Teb4 mRNA were determined relative to the housekeeping gene hypoxanthine phosphoribosyltransferase 1 ( HRPT1 ) by quantitative real-time PCR using LightCycler 480 SYBR Green mix ( Roche ) .	All cells are enclosed by a membrane that is made up of fatty molecules called lipids and is studded with proteins . This membrane allows cells to detect and react to outside events . Since external conditions , such as temperature , can vary dramatically , membranes need to be able to adjust their properties . For example , lipids become more fluid as the temperature rises , so membranes respond to heat stress by incorporating molecules called sterols to increase their rigidity . In fact , sterols have profound effects on membrane properties and are essential to regulate a number of cellular processes . But high levels of sterols can become toxic , so it is essential that they are carefully controlled . Sterols , such as ergosterol in yeast or cholesterol in mammals , are synthesized in a tightly regulated multi-step process; some of the early steps in sterol production also make common building blocks for other key molecules in the cell . A mechanism to control sterol levels is the regulated destruction of an enzyme that carries out an early step of their synthesis . This occurs via one branch of the ER-associated degradation ( ERAD ) pathway , which also destroys non-functional proteins . Now , Foresti et al . have found that sterol synthesis is also regulated by another branch of the ERAD pathway . This second control point , which occurs later in the biosynthetic process , allows cells to regulate sterol levels independent of the other products of the pathway that are derived from the same preliminary compounds . In yeast , the two ERAD branches are directed by Hrd1 and Doa10 . These are both ubiquitin ligases—proteins that attach a tag called ubiquitin to other proteins , thus labeling them for recycling by the proteasome ( essentially a waste-disposal complex in the cell ) . To identify the proteins that are tagged by Doa10 , Foresti et al . compared protein levels in strains lacking Doa10 with those in wild type yeast . Unexpectedly , the enzyme Erg1 , which helps to synthesize ergosterol , was more abundant in cells lacking Doa10 . Foresti et al . found that Doa10 tagged Erg1 for destruction when levels of the building blocks of ergosterol rose inside the cell . These ergosterol intermediates are toxic to yeast , which converts them into less harmful molecules known as sterol esters using the proteins Are1/2 . When the DOA10 or ARE1/2 genes were deleted , these intermediates were more abundant; strikingly , they became even more prevalent when all three genes were knocked out in the same strain . In contrast , blocking the other ERAD branch by deleting HRD1 did not cause ergosterol intermediates to accumulate , nor did it exacerbate the effects of ARE1/2 knockout . When combined with previous findings , these results provide evidence that the different branches of the ERAD pathway regulate ergosterol synthesis at distinct steps . The same mechanism is observed in human cells when high levels of cholesterol are detected . By identifying parallel routes to control sterol levels , this work reinforces the importance of membrane integrity to life .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2013	Sterol homeostasis requires regulated degradation of squalene monooxygenase by the ubiquitin ligase Doa10/Teb4
Planarian flatworms regenerate every organ after amputation . Adult pluripotent stem cells drive this ability , but how injury activates and directs stem cells into the appropriate lineages is unclear . Here we describe a single-organ regeneration assay in which ejection of the planarian pharynx is selectively induced by brief exposure of animals to sodium azide . To identify genes required for pharynx regeneration , we performed an RNAi screen of 356 genes upregulated after amputation , using successful feeding as a proxy for regeneration . We found that knockdown of 20 genes caused a wide range of regeneration phenotypes and that RNAi of the forkhead transcription factor FoxA , which is expressed in a subpopulation of stem cells , specifically inhibited regrowth of the pharynx . Selective amputation of the pharynx therefore permits the identification of genes required for organ-specific regeneration and suggests an ancient function for FoxA-dependent transcriptional programs in driving regeneration . Many organs in the human body have the potential to repair themselves after injury . For example , the hematopoietic system can replenish the entire blood in an animal from a single stem cell after bone marrow transplants ( Lagasse et al . , 2001; Shizuru et al . , 2005 ) , and hair follicle and epidermis can regenerate in mammals following injury ( Fuchs and Segre , 2000; Seifert et al . , 2012 ) . To initiate regeneration , stem cells must sense an injury , proliferate and differentiate appropriately , and replace the missing organs ( Poss , 2010 ) . Stem cells ( and in particular iPS cells ) represent enormous potential for developing therapeutic treatments for disease . However , effective implementation of these technologies will require an improved understanding of stem cell activation and regulation in vivo . Planarians are a classical system for studying regeneration . After amputation , even small fragments can support regrowth of entire animals ( Morgan , 1898; Reddien and Sánchez Alvarado , 2004 ) , indicating that the resident cells have the capacity to self-renew , and can replace all of the missing tissues comprising the animal ( including muscle , nervous system , digestive system , excretory system and epithelial cells ) . This regenerative capacity depends on a population of stem cells termed neoblasts ( Reddien and Sánchez Alvarado , 2004 ) . These pluripotent cells are constantly dividing , driving replenishment of all cell types during homeostasis ( Newmark and Sánchez Alvarado , 2000; Pellettieri and Sánchez Alvarado , 2007 ) . Upon amputation , neoblasts are stimulated to divide rapidly ( Baguñà , 1976; Reddien et al . , 2005a; Wenemoser and Reddien , 2010 ) and begin to differentiate , but how these stem cells are regulated to produce only the tissues that need to be replaced is unclear . One hypothesis for how stem cells can produce any tissue in the planarian body on demand is that these cells exhibit heterogeneity across the population , in terms of both gene expression and cell cycle status ( Rink , 2012; Reddien , 2013 ) . Heterogeneity has been observed molecularly by gene expression-profiling of neoblasts purified via fluorescence-activated cell sorting ( FACS ) ( Hayashi et al . , 2010; Shibata et al . , 2012 ) . Functional evidence of such heterogeneity is supported by single-cell transplantation experiments in which some , but not all , stem cells can repopulate and rescue animals lacking stem cells ( Wagner et al . , 2011 ) . However , it is unclear what percentage of the stem cell population is in fact pluripotent , or if these cells produce lineage-restricted stem cells . Recent studies have also demonstrated that discrete subpopulations of neoblasts express markers of differentiated tissues ( Scimone et al . , 2011; Lapan and Reddien , 2012; Cowles et al . , 2013 ) . Therefore , cell fate decisions can be established within neoblasts , but how this happens is unknown . Normally contained within an internal cavity referred to as the pharyngeal pouch , the pharynx protrudes through a ventral opening upon sensing food or prey and ingests food by contractile peristalsis ( Wulzen , 1917 ) . The planarian pharynx serves as both the entrance and exit to the digestive system and is a complex organ consisting of multiple tissues including neurons , muscle , epithelial cells and secretory cells ( Hyman , 1951; Ishii , 1962; MacRae , 1963; Kido , 1964 ) . The pharynx is a large cylindrical structure that clearly lacks dividing stem cells ( Hay and Coward , 1975; Newmark and Sánchez Alvarado , 2000; Orii et al . , 2005 ) . Previous experiments describing de novo pharynx regeneration in head and tail pieces have shown that mesenchymal cells adopt pharyngeal fate prior to accumulation in the nascent pharynx ( Asai , 1991; Bueno et al . , 1997; Cebrià et al . , 1999; Kobayashi et al . , 1999 ) . These observations suggest that neoblasts respond to the absence of the pharynx and produce cells of the pharyngeal lineage shortly after amputation . Pharynx regeneration is therefore an excellent model for understanding organ regeneration in general , beginning with pluripotent stem cells that differentiate into distinct cell types , which then integrate with pre-existing tissues to form a functional organ . Here we describe a novel strategy for amputation of a single organ , the pharynx , and for studying its regeneration . Using feeding behavior as a quantitative assay for regeneration , we screened a library of transcripts upregulated during pharynx regeneration . We show that RNAi of these genes causes a wide range of regeneration phenotypes , and that the pioneer transcription factor FoxA , which functions in many organisms to specify endodermal organogenesis , is required for regeneration of the planarian pharynx . Taken together , this novel amputation strategy offers a defined context in which to measure and understand discrete changes in a stem cell population during regeneration . In order to dissect the specific response to organ amputation , we sought to develop a method to selectively remove a single organ . We found that soaking animals in sodium azide for a brief period of time caused the pharynx to be extruded , and then dislodged completely from the rest of the animal following gentle agitation ( Figure 1A , B ) . Because this amputation does not require surgery , the wound produced after chemical treatment is indistinguishable between animals . Within several days the pharynx regenerated , as visualized either in live animals or by hybridization with a pharynx-specific riboprobe ( Figure 1C; Cebrià et al . , 2007 ) . Other organs , including the gastrointestinal system , were unaffected at a gross morphological level ( Figure 1D ) , indicating that sodium azide treatment causes selective amputation of the pharynx without noticeably perturbing other organs . We term this treatment ‘chemical amputation’ . 10 . 7554/eLife . 02238 . 003Figure 1 . Sodium azide selectively removes the pharynx . ( A ) Live animals before and after sodium azide treatment , showing pharynges ( arrows ) . ( B ) Schematic of chemical amputation . ( C ) Pharynx ( labeled with Smed-laminin ) reappears 2–3 days after pharynx removal . ( D ) Intestine ( labeled with Smed-porcupine ) before and immediately after chemical amputation . ( E ) Representative hematoxylin/eosin sections of the regenerating pharynx ( white arrowheads ) . Yellow lines outline mesenchyme and white arrows highlight intestine . Scale bars , A–D: 500 μm , E: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02238 . 00310 . 7554/eLife . 02238 . 004Figure 1—figure supplement 1 . Histological analysis of regenerating pharynx . ( A ) Neurons and epithelial cells appear in the blastema 24 hr after amputation but are disorganized . 2 days after amputation , epithelial cells are arranged at the periphery and neurons and muscle coalesce interiorly . Stains: Neurons ( Smed-PC2 , pink ) ; muscle ( Tmus , blue ) , and epithelia ( acetylated tubulin , green ) . Dashed white line highlights regeneration pharynx . Scale bars , 50 μm . ( B ) Pharynx markers are expressed in the regenerating pharynx 3 days after amputation . Animals shown 1 day and 3 days after amputation . DOI: http://dx . doi . org/10 . 7554/eLife . 02238 . 00410 . 7554/eLife . 02238 . 005Figure 1—figure supplement 2 . Effects of sodium azide exposure . ( A ) Schematic of experiment . Amputations were performed in the presence of sodium azide , and fragments were maintained in sodium azide until the pharynx was ejected from the trunk pieces . ( B and C ) Quantification of mitotic activity in head ( B ) and tail ( C ) pieces at specified times . Error bars = standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 02238 . 005 The pharynx is ensheathed by a ciliated epithelium that covers layers of muscle , an extensive neural network , and secretory gland cells ( Cebrià et al . , 1997 , 1999; Okamoto et al . , 2005 ) . Histological analysis of the anatomical changes that occur during pharynx regeneration confirmed that chemical amputation causes detachment of the pharynx without affecting the intestine ( Figure 1E , white arrows ) . Within 24 hr of amputation , small , undifferentiated cells accumulate at the entry to the intestine , within the mesenchyme ( Figure 1E , white arrowheads ) . Two days after amputation , these cells begin to organize , characterized by a layer of ciliated epithelial cells , a concentration of neurons , and a subepidermal muscle layer ( Figure 1—figure supplement 1A ) . Three days after amputation the lumenal connection to the intestine has been restored , radial symmetry has been re-established , and most genes expressed in the mature pharynx are present , including the secreted Frizzled-related protein sFRP-1 , ciliary dynein heavy chain DNAH-β3 , and two members of the Nou Darake family of FGF-receptor-like proteins ndk and ndl-3 ( Rink et al . , 2011; 2009; Figure 1—figure supplement 1B ) . Therefore , following selective removal of the pharynx , animals regenerate all of the component tissues , in a similar sequence to what has been observed previously for de novo pharynx regeneration ( Bueno et al . , 1997; Cebrià et al . , 1999; Kobayashi et al . , 1999 ) . Even though the regeneration of chemically amputated pharynges appears to proceed normally by all histological and molecular measures , and 100% of animals regenerated pharynges after amputation ( n > 1000 ) , we wished to further test whether the brief exposure to sodium azide during chemical amputation might cause secondary effects in regeneration , particularly soon after the treatment . To test the possibility that sodium azide broadly compromised regenerative potential , we performed transverse amputations in sodium azide . After washout , wound healing occurred normally , indicating that animals recovered rapidly from sodium azide treatment . Furthermore , in these regenerating fragments , the mitotic profile triggered by amputation during the early phase of regeneration was indistinguishable from controls ( Figure 1—figure supplement 2; Wenemoser and Reddien , 2010 ) . Altogether , these data demonstrate that sodium azide exposure does not significantly perturb the kinetics of regeneration in general and likely has minimal effects on pharynx regeneration in particular . Exposure of animals to lethal doses of gamma-irradiation completely prevents stem cell division and regeneration ( Bardeen and Baetjer , 1904 ) . To confirm that pharynx regeneration also requires neoblasts , animals were lethally irradiated ( 10 , 000 rads γ-irradiation ) prior to pharynx amputation . Radiation completely prevented pharynx regeneration ( Figure 2A ) in 100% of animals ( n = 100 animals ) indicating that , as expected , stem cells are required for regeneration . Furthermore , lethal irradiation inhibited the accumulation of cells at the wound site 24 hr after amputation ( Figure 2—figure supplement 1 ) , indicating that the first cells to arrive at the wound site are either neoblasts or their descendants . Similarly , RNAi knockdown of the planarian piwi/Argonaute protein Smedwi-2 phenocopies radiation by inhibiting stem cell function ( Reddien et al . , 2005b ) . Indeed , Smedwi-2 ( RNAi ) animals failed to regenerate the pharynx ( 0/33 animals , compared to 24/24 control animals ) ( Figure 2B ) . These results indicate that pharynx regeneration , like all other regeneration in planaria , depends on functional neoblasts and that large reserves of post-mitotic cells competent to become pharyngeal tissues are unlikely to exist . 10 . 7554/eLife . 02238 . 006Figure 2 . Local proliferation of stem cells drives regeneration . ( A ) Irradiated animals fail to regenerate the pharynx ( 100%; n >50 ) , as indicated by Smed-laminin ISH . ( B ) Smedwi-2 ( RNAi ) inhibits pharynx regeneration ( 100% , n >30 ) . ( C ) Representative confocal images of animals during pharynx regeneration , stained with anti-phosphoH3-Ser10 . Circles are representative of those used for quantification in ( E ) . ( D ) Quantification of phosphoH3-Ser10 staining in whole animals . Error bars = SD . ( E ) Local proliferation measured in two equal-sized circles , ( 1 ) centered over the pharynx and ( 2 ) centered in the tail as marked in ( C ) . Error bars = SD; *** equals p< . 0001; significance determined with Student’s t test . ( F ) Schematic of strategy for expression profiling . Scale bars , A and B: 200 μm , C: 500 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02238 . 00610 . 7554/eLife . 02238 . 007Figure 2—figure supplement 1 . Irradiation prevents accumulation of cells at the blastema . 48 hr prior to chemical amputation , animals were irradiated with 6 , 000 rads and fixed 24 hr or 3 days afterward . Paraffin sections were stained with hematoxylin and eosin . Scale bars , 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02238 . 00710 . 7554/eLife . 02238 . 008Figure 2—figure supplement 2 . Body-wide mitotic activity after chemical amputation . Body-wide mitotic activity at specified times after pharynx removal measured by anti-H3-Ser10phos staining . In the first 24 hr after sodium azide treatment , there is a general , transient suppression of mitotic activity . Error bars = Standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 02238 . 00810 . 7554/eLife . 02238 . 009Figure 2—figure supplement 3 . Validation of microarray by in situ timecourses . ( A ) Summary of 21/42 genes tested that showed regional expression increases after chemical amputation . Each timepoint on the microarray is represented by a unique color , to indicate significant upregulation in the microarrays at the indicated timepoints ( adjusted P<0 . 05 and log2 fold change>0 . 4 ) . Right panel summarizes results from whole-mount in situ hybridizations . 6 hr and 18 hr timepoints were not examined . ( B ) Selected in situ timecourses at specified times after amputation with the accompanying graph of microarray results . Asterisks highlight significance ( adjusted P<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02238 . 009 In planaria , amputation stimulates two characteristic waves of proliferation: within hours of any wound , mitotic events increase throughout the body , and 2 days later proliferation is localized to the wound ( Baguñà , 1976; Wenemoser and Reddien , 2010 ) . Because chemical amputation produces an internal wound but leaves the epithelium intact , we wondered whether it would elicit similar proliferation kinetics to other types of surgically-induced wounds . We quantified the number of mitoses in the animal during pharynx regeneration by staining planarians with an antibody recognizing phosphorylated histone H3 at serine 10 ( Hendzel et al . , 1997; Newmark and Sánchez Alvarado , 2000 ) . In the first 24 hr after amputation , we observed a sharp but transient decrease in overall mitotic activity ( Figure 2—figure supplement 2 ) , presumably due to the metabolic suppression effects of sodium azide . Overall , body-wide mitotic activity did not significantly change as regeneration progressed ( Figure 2C , D ) . However , mitotic activity in the vicinity of the wound site appeared to increase 24 hr after amputation , although this effect diminished as regeneration proceeded . To confirm this observation , we quantified mitoses in each of two defined regions: one around the wound site and one posterior to the wound site ( Figure 2E , example in Figure 2C ) . Indeed , we found a significant enrichment of mitotic nuclei around the pharynx wound site , indicating that the proliferative response induced by pharynx removal generates a sufficiently powerful signal to induce and maintain proliferation where regeneration is necessary . We sought to define the molecular mechanisms driving pharynx regeneration by expression-profiling experiments . We designed custom oligonucleotide microarrays representing 43 , 806 predicted S . mediterranea transcripts and isoforms from various sources ( Robb et al . , 2008; Blythe et al . , 2010; Adamidi et al . , 2011 ) . Based on our observations that pharynx regeneration triggered a localized stem cell proliferative response , we isolated a plug of tissue surrounding the pharynx wound site in order to enrich for those transcripts most directly relevant to this process ( Figure 2F ) . We extracted RNA at specific times during a window of time up to 72 hr post-amputation , and compared these samples to plugs isolated immediately after amputation ( time 0 ) . Because we aimed to identify transcripts that were essential for the initiation of pharynx regeneration , we first focused on genes upregulated during the first 24 hr after amputation . During this window , 718 genes were enriched upon pharynx removal ( log2 fold change>0 . 4 , adjusted p<0 . 05 ) and we cloned 274 of these genes . We also cloned a second group of genes that were significantly upregulated at 48 and 72 hr after amputation , but not prior . An additional set of 82 genes were included based on consistent upregulation at both 48 and 72 hr after amputation , for a total of 356 genes ( Supplementary file 1A ) . To validate our microarray data , we examined the expression patterns of several upregulated genes during pharynx regeneration with whole-mount in situ hybridization . We found that 21/42 of these genes showed distinct upregulation in the area around the regenerating pharynx , as expected ( Supplementary file 1B; Figure 2—figure supplement 3 ) . Interestingly , expression of several of these genes was undetectable prior to amputation , but increased significantly afterward , mimicking the marked upregulation in transcription observed for wound-response genes ( Wenemoser et al . , 2012 ) and suggesting that chemical amputation does in fact stimulate a wound response . Two genes in this category ( PDZ ring finger protein 4 and rhomboid ) showed dramatic increases in the region surrounding the pharynx after amputation ( Figure 2—figure supplement 3 ) , suggesting that these cells may broadly respond to pharynx removal . Because the pharynx is required for ingestion of food , we developed an assay that measures the recovery of feeding behavior after selective pharynx removal as a rapid and quantitative method to gauge the extent of pharyngeal regeneration ( Ito et al . , 2001 ) . When presented with food , planarians normally chemotax toward it and ingest it ( Figure 3A ) . If the pharynx is missing or is incompletely regenerated , animals are unable to eat and will not attempt to swim toward the food , implying that the pharynx has a sensory role in stimulating movement towards food . Animals regained the ability to eat 7 days after amputation ( Figure 3B ) , indicating that all of the tissues comprising the pharynx were present , functional , and integrated with the rest of the animal by this point in regeneration . 10 . 7554/eLife . 02238 . 010Figure 3 . RNAi screen for genes affecting pharynx regeneration . ( A ) Schematic of feeding assay . ( B ) Animals recover ability to ingest food 7 days after chemical amputation . For each timepoint , n = 10 animals , repeated in triplicate . Error bars = SD . ( C ) Quantification of feeding behavior of RNAi-treated animals 10 days after amputation . Shown are averages of three independent experiments; error bars = SEM , n ≥30 animals . Smed-laminin in situ hybridization shows extent of pharynx regeneration defects in RNAi-treated animals . Scale bars = 250 μm . ( E ) Quantification of pharynx length in RNAi animals 11 days after amputation . For each bar , n = 6–10 animals; error bars = SD . ( F ) Mitotic activity of whole animals 3 days after pharynx amputation measured by phosphoH3-Ser10 staining . Error bars represent SD , and n = 8 animals for each condition . DOI: http://dx . doi . org/10 . 7554/eLife . 02238 . 01010 . 7554/eLife . 02238 . 011Figure 3—figure supplement 1 . Candidate gene summary . ( A ) Whole-mount in situ hybridizations showing distributions of 20 candidate genes during pharynx regeneration . Red arrows highlight enrichment in mesenchyme surrounding the pharynx . Scale bars = 500 μm . ( B ) Summary of expression changes of candidate genes on microarray . DOI: http://dx . doi . org/10 . 7554/eLife . 02238 . 011 Using feeding behavior as a proxy for pharynx regeneration , we screened the 356 cloned genes by RNA-interference . We considered a 50% defect in food uptake as our initial threshold and identified 20 genes ( 5 . 6% of total ) that caused reproducible defects in this assay upon knockdown ( Figure 3C ) . Analysis of pharynx length by in situ hybridization with the pharynx marker Smed-laminin ( Figure 3D , E ) allowed classification of the RNAi phenotypes into groups ( see below ) with predicted functions during pharynx regeneration . In addition , we measured both the extent of regeneration after head and tail amputation ( Table 1 ) and mitotic activity ( Figure 3F ) . 10 . 7554/eLife . 02238 . 012Table 1 . Summary of RNAi phenotypesDOI: http://dx . doi . org/10 . 7554/eLife . 02238 . 012RNAi IDFull nameAbbr . Putative functionHead/Tail RegPhx LengthLysis ? Other phenotypesFoxA expressionReference7D04FoxA1FoxA1transcription factorphx defective+–dorsal spikeN/A5B08Mothers against decapentaplegic homolog 4Madd4Bmp signalingnormal+++––wt6F08fos/BZIP transcription factorFos-1immediate early genenormal+++–bump over pharynxwtWenemoser et al . , Genes and Development5B06novelnovelnormal+++––wt1B10Heterogeneous nuclear ribonucleoprotein KhnRNP KmRNA binding/p53 signalingsmall blst++–HRFoxA OK; small phx5C01jagunaljagunalER organization , membrane traffickingvariable+++/−partial3C09G1/S-specific cyclin-D1cyclin D1phosphorylates and inhibits Rbblst ( − ) ++–no FoxAZhu and Pearson , Development2E03WD repeat-containing protein 36WDR36glaucoma disease genesmall blst++–aggregates5C12Squamous cell carcinoma antigen recognized by T-cells 3SART-3snRNP complex/interacts with Ago proteinssmall blst++–aggregates2D06WD repeat-containing protein 3WDR3unclearsmall blst++–aggregates5G05ribonucleoside-diphosphate reductase M2RRdeoxyribonucleotide biosynthesisblst ( − ) +–no FoxAEisenhoffer et al . , Cell Stem Cell2A11DNA replication licensing factor MCM7MCM7MCM complexblst ( − ) +–no FoxA1D01Polyadenylate-binding protein 2PABP-23′ end processing of mRNAblst ( − ) ++curlingreduced FoxA; aggregates1C06Decaprenyl-diphosphate synthase subunit 2PDSS-2required for biosynthesis of Coenzyme Q10blst ( − ) +–curlingreduced FoxA1D07COP9 signalosome complex subunit 5COP9protease subunit of CSN complexblst ( − ) ++FoxA OK; small phx1B06Cleavage and polyadenylation specificity factor 3CPSF-33′ end processing of mRNAblst ( − ) ––curlingno FoxA2G03Histone-binding protein RBBP4RBBP4associates with chromatin-regulatory complexesblst ( − ) –+ndBonuccelli et al . , J . Cell Sci . ; Wagner et al . , Cell Stem Cell; Zeng et al . , JCB1C07RuvB-like 2RuvBDNA helicase involved in Holliday junction formationblst ( − ) –++no FoxALabbe et al . , Stem Cells1E02Zinc finger MYM-type protein1ZMYM1cell morphology regulationblst ( − ) –+no FoxAWagner et al . , Cell Stem Cell1E04SUMO-activating enzyme subunit 2SAE-2SUMO ligaseblst ( − ) –+no FoxAPhx Length:normalized to controlAbbreviations-0–25%nd = not determined+25–50%blst = blastema++50–75%HR = head regression+++75–100%phx = pharynxReg = regeneration The first category of molecules we expected to uncover was general regulators of stem cell function ( Figure 3C , Stem Cell Effectors ) , based on the requirement for neoblasts in regeneration . The RNAi screen identified eight genes that exhibited phenotypes consistent with a general function in stem cells . In addition to causing a strong inhibition of feeding and pharynx regeneration ( Figure 3C , D ) , knockdown of these eight genes severely compromised regeneration ( Table 1 ) and decreased mitotic activity ( Figure 3F ) . Moreover , in situ hybridization for these transcripts demonstrated that they were expressed primarily in stem cells ( Figure 3—figure supplement 1 ) . These genes include ribonucleotide reductase ( Eisenhoffer et al . , 2008; Böser et al . , 2013 ) , the chromatin assembly factor Rbbp4 ( Bonuccelli et al . , 2010; Wagner et al . , 2012; Zeng et al . , 2013 ) , the G1/S-specific cyclin D1 ( Zhu and Pearson , 2013 ) , the zinc finger-containing protein zmym-1 ( Wagner et al . , 2012 ) and the RuvB DNA helicase ( Labbé et al . , 2012 ) . Three other genes in this group have not been previously implicated in stem cell function in planarians . These genes include the DNA licensing factor MCM7 , cleavage and polyadenylation specificity factor 3 ( CPSF-3 ) , and the Sumo-activating enzyme subunit 2 ( SAE-2 ) . Interestingly , SAE-2 was recently identified as a component of the stem cell proteome ( Böser et al . , 2013 ) . Therefore , our screen captured phenotypes for novel stem cell genes , indicating that this strategy can successfully identify genes acting at discrete and early steps in the process of regeneration . The next group of genes ( Figure 3C , Specific Effectors ) caused profound defects in feeding , but unlike the previous category , most of these animals produced some pharyngeal tissue ( Figure 3C–E ) , indicating that pharynx regeneration was either delayed or stalled . The inability to feed was most pronounced following knockdown of the Forkhead transcription factor FoxA and the heterogeneous nuclear protein hnRNPK . Other genes that caused strong pharynx phenotypes included the poly-A binding protein PABP-2 , a component of the ubiquitin proteasome COP9 signalosome complex ( COP9 ) , two WD-repeat containing proteins ( WDR3 and WDR36 ) , the ER membrane protein jagunal , and decaprenyl diphosphate synthase subunit 2 ( DDSS-2 ) . Overall , mitotic activity of the RNAi knockdown animals in this group was comparable to controls ( Figure 3F ) , indicating that these genes were unlikely to affect general stem cell function . However , PABP-2 ( RNAi ) animals exhibited elevated mitotic activity ( Figure 3F ) , a phenotype observed in EGFR1 ( RNAi ) and p53 ( RNAi ) animals ( Fraguas et al . , 2011; Pearson and Sánchez Alvarado , 2010 ) and reflecting a failure of proliferative control of the stem cell population . The final category of phenotypes ( Figure 3C , Other ) contains animals that fail to feed but regenerate full-length pharynges ( Figure 3D , E ) . To accomplish successful feeding after pharynx amputation , animals need to retain the ability to properly sense food and move towards it . Because our screening strategy relied on feeding behavior , it allowed for the discovery of genes required for motility , chemosensation , or other defects in organ function that do not accompany obvious morphological defects . Indeed , we uncovered several genes in this category , including the immediate early gene fos-1 , the SMAD protein Madd4 , the RNA-binding protein SART-3 , and a novel protein . Following knockdown , these animals regenerated pharynges indistinguishable from controls at the morphological level , and had normal levels of mitotic activity ( Figure 3D–F ) . None of these knockdown animals exhibited motility problems , raising the possibility that these genes may be required for sensory function or for proper integration of the newly regenerated pharynx with the rest of the animal . We then performed in situ hybridizations to determine the distribution of each of these transcripts during pharynx regeneration . Most of these genes demonstrated a striking upregulation in the vicinity of the pharynx as soon as 24 hr after amputation ( Figure 3—figure supplement 1 ) . Together , these results show that our combined approach of expression profiling followed by RNAi screening successfully identified genes that are functioning at different steps in the regeneration process . Forkhead transcription factors are critical determinants of foregut development throughout evolution , in both protostomes ( Mango , 2009 ) and deuterostomes ( Fritzenwanker et al . , 2004; Lee et al . , 2005 ) . Expression studies have demonstrated that the planarian homolog of FoxA1 localizes to the nascent pharynx during embryogenesis and regeneration ( Koinuma et al . , 2000; Martín-Durán and Romero , 2011 ) , demonstrating that the transcript localizes to developing pharyngeal tissue . In our screen , after three doses of FoxA ( RNAi ) , animals developed a lesion on their dorsal side through which they subsequently ejected their pharynx ( Figure 4A , 35/50 animals ) . 10 . 7554/eLife . 02238 . 013Figure 4 . FoxA is required for pharynx regeneration . ( A ) FoxA ( RNAi ) animals develop dorsal lesions ( arrow ) . ( B and C ) Confocal images of cryosections stained with antibodies recognizing muscle ( α-Tmus ) , epithelial cells and protonephridia ( α-acetylated tubulin ) , and nuclei ( DAPI ) . Control ( B ) and FoxA ( RNAi ) animals ( C ) are shown 3 days after pharynx removal . Dashed green lines highlight the regenerating pharynx . ( D ) Tail fragments amputated at dashed red line regenerate brain tissue ( Smed-PC2 , red arrowheads ) but not a pharynx ( Smed-PKD2 , green arrowheads ) . ( E ) Head fragments regenerate posterior intestinal branches ( Smed-porcupine , red arrowheads ) despite the absence of a pharynx ( Smed-PKD2 , green arrowhead ) . ( F ) Whole-mount ISH for Wnt11-5 in control and FoxA ( RNAi ) tail fragments 7 days after amputation . Green boxes highlight insets shown below . ( G ) Ratio of Wnt11-5 expression to total length of tail fragment . Significance determined by Student’s t test . Error bars = SEM . N = 14 fragments . Scale bars , ( A ) , 500 μm , ( B and C ) , 50 μm , ( D–F ) , 200 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02238 . 01310 . 7554/eLife . 02238 . 014Figure 4—figure supplement 1 . FoxA is not required for anterior/posterior patterning during regeneration . ( A ) Schematic of amputations . ( B and C ) Regenerating pieces shown 7 days after amputation . ( B ) Tail fragments showing distribution of the anterior marker Sfrp-1 . ( C ) Head and tail fragments showing distribution of ndl-3 . Scale bars = 200 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02238 . 014 To examine in greater detail the anatomical defects caused by FoxA knockdown , we analyzed the cellular architecture of the regenerated pharynx after chemical amputation . Three days after amputation , control animals had regenerated the arrayed muscle fibers covered by epithelial cells , along with the ciliated cells that extend to the distal tip of the pharynx ( Figure 4B ) . By contrast , FoxA ( RNAi ) pharynges appeared as a disorganized mass of cells , lacking obvious muscle fibers and any epithelial covering ( Figure 4C ) , indicating both a failure to produce the cells comprising the pharynx and to pattern them properly . This defect in specification suggests that FoxA is required for coordinating the differentiation of cell types comprising the regenerating pharynx . Because chemical amputation selectively removes the pharynx without noticeably affecting other organs , we asked whether regeneration of other tissues required FoxA . Tail fragments normally regenerate heads containing a central nervous system , which can be stained with the neuronal markers Smed-ChAT and Smed-PC2 , as well as a new pharynx . FoxA ( RNAi ) tail pieces successfully regenerated nervous system tissue that was indistinguishable from controls , but failed to form a new pharynx ( Figure 4D ) . Conversely , head pieces normally regenerate posterior tissue containing two intestinal branches that extend posteriorly around the new pharynx ( Figure 4E ) . FoxA ( RNAi ) head fragments had no defect in intestinal growth or patterning ( Figure 4E ) , demonstrating that FoxA controls regeneration of the pharynx but not other organs . This is consistent with its role in other organisms as an organ-specific transcription factor . Transverse amputation of planarians initiates re-patterning of tissues that re-establishes anterior/posterior polarity . Successful regeneration depends on the Wnt pathway and is accompanied by dynamic changes in expression domains of various transcripts ( Gurley et al . , 2008; 2010; Petersen and Reddien , 2008; 2009 ) . To determine if FoxA is required for establishment of anterior/posterior polarity during regeneration , we examined the patterning of these markers in regenerating fragments . The anterior marker Sfrp1 and the posterior marker nou darake-like-3 ( ndl-3 ) were re-established properly 7 days after amputation in FoxA knockdown animals ( Figure 4—figure supplement 1 ) , indicating that FoxA is not required for specifying expression of these markers . Wnt11-5 ( also known as WntP2 ) is expressed in a posterior-to-anterior gradient in planarians with its anterior boundary at the pharynx . Initially after removing large regions of the body during amputation , Wnt11-5 expression is more uniform , but as regeneration progresses the gradient resets to its strong posterior-to-anterior bias ( Petersen and Reddien , 2009; Gurley et al . , 2010 ) . This dynamic re-establishment does not depend initially on stem cells , but requires them indirectly for rescaling ( Gurley et al . , 2010 ) . Interestingly , the expression of Wnt11-5 in FoxA ( RNAi ) tail fragments does not reset the anterior boundary at the pharynx as sharply and reproducibly as controls ( Figure 4F , G ) . However , TOR ( RNAi ) animals successfully reset the anterior Wnt11-5 boundary despite their lack of a pharynx ( Tu et al . , 2012 ) . Therefore , FoxA may participate in specifying the central body region or in regulating stem cells in the vicinity of the pharynx . In adult planarians , FoxA is expressed strongly in the mature pharynx ( Koinuma et al . , 2000; Umesono et al . , 2013 ) , and in scattered cells in the mesenchyme surrounding the pharynx ( Figure 5A ) . These cells were arrayed in a branched pattern reminiscent of neoblasts ( Orii et al . , 2005; Reddien et al . , 2005b ) . We observed that after pharynx amputation , FoxA+ cells accumulated both in the nascent pharynx and tissue surrounding the pharyngeal pouch , adjacent to the wound site ( Figure 5B , top row ) . This region contained the densest concentration of FoxA+ cells 3 days after amputation . Interestingly , this pattern was similar to the expression of 10 other genes from the screen ( Figure 3—figure supplement 1 ) . These results suggest the possibility that FoxA+ cells in the mesenchyme may represent pharyngeal progenitors and that our screen uncovered markers for this cell population . 10 . 7554/eLife . 02238 . 015Figure 5 . FoxA expression in neoblasts increases after amputation . ( A ) Whole-mount ISH for Smed-FoxA in intact animals . Boxed region highlights areas shown in ( B ) . ( B ) Smed-FoxA expression in pharyngeal region during regeneration , in unirradiated animals ( top ) and lethally irradiated animals ( bottom ) . Yellow arrowheads point to accumulation of FoxA+ cells in mesenchyme surrounding pharynx and red arrows highlight pharyngeal pouch . ( C ) Schematic of mesenchymal pouch surrounding the pharynx , where FoxA+ cells concentrate during regeneration . ( D ) Double-FISH with smedwi-1 and Smed-FoxA at different times after pharynx removal . Arrowheads highlight positive cells . Scale bars = 10 μm . ( E ) Quantification of percentage of smedwi-1+ cells that co-express FoxA during regeneration . For each timepoint , n = 100–150 smedwi-1+ cells . DOI: http://dx . doi . org/10 . 7554/eLife . 02238 . 015 Previous experiments combining partial irradiation of planarians with partial transection of the pharynx have localized pharynx progenitors to the mesenchyme anterior to the pharynx ( Ito et al . , 2001 ) . To determine whether these cells were neoblasts or their descendants that might be incorporated into the regenerating pharynx , we lethally irradiated animals 2 days prior to pharynx amputation . Radiation exposure causes progressive loss of stem cells and their progeny by preventing the production of these cells , making it an effective strategy for identifying lineage relationships in planaria ( Eisenhoffer et al . , 2008 ) . Radiation completely inhibited accumulation of FoxA+ cells and caused a depletion of the mesenchymal cells surrounding the pharynx ( Figure 5B , bottom row ) . Expression of FoxA persisted for several days in the epithelial cells lining the pouch and in the central body region . This suggests that pharynx progenitors are neoblast-derived , FoxA+ cells; it also demonstrates that these progenitors localize to the regenerating pharynx rapidly after amputation . Several transcription factors marking differentiated cell types including photoreceptors , protonephridia and neurons have been shown to initiate their expression in smedwi-1+ stem cells ( Lapan and Reddien , 2011; Scimone et al . , 2011; Cowles et al . , 2013; Currie and Pearson , 2013 ) , suggesting that in planaria , lineage decisions can be made within the dividing stem cell population . The expression of progenitor markers within the pluripotent stem cell population is indicative of heterogeneity of the neoblasts and is thought to contribute to the ability of planarians to regenerate all of their organs equally well ( Reddien , 2013 ) . Based on our observation that FoxA marks a population of irradiation-sensitive pharynx progenitors , we wondered if any or all of them were smedwi-1+ stem cells . We performed double fluorescent in situ hybridization with FoxA and smedwi-1 . Although FoxA is expressed relatively weakly compared to smedwi-1 , in intact animals we found a subset of smedwi-1 cells that also expressed FoxA in the vicinity of the pharynx ( Figure 5D ) . This result is consistent with FoxA being present in irradiation-sensitive stem cells . One of the unanswered questions in planarian biology is how pluripotent stem cells sense the absence of particular organs and how they mount the appropriate regenerative responses to replace those organs . Pairing selective amputation of the pharynx with expression of progenitor markers in the stem cell population allows us to begin addressing this question . We quantified the percentage of FoxA+smedwi-1+ cells following pharynx amputation . Indeed , following pharynx removal , the percentage of FoxA+smedwi-1+ cells increased significantly , with the peak occurring 3 days after pharynx removal ( Figure 5D , E ) . This result demonstrates that a specific , FoxA+ portion of the stem cell population responded to pharynx removal . Interestingly , when small fragments lacking a pharynx initiate regeneration ( i . e . , from head or tail amputation ) , they rapidly increase FoxA expression in the pharynx rudiment ( Koinuma et al . , 2000 ) . Therefore , mesenchymal cells , even those residing in the anterior or posterior extremes of the animal , can be stimulated to express FoxA when regeneration of a pharynx is required . In Caenorhabditis elegans , FoxA is expressed in all pharyngeal precursors during embryogenesis ( Mango et al . , 1994 ) . By contrast , in adult animals the expression fades , and FoxA is preferentially expressed in the intestine ( Panowski et al . , 2007 ) . Embryonic expression studies in flatworms suggest that FoxA may not be expressed throughout the definitive organ once development is complete ( Martín-Durán et al . , 2010 ) . To explore whether planarian FoxA is expressed throughout the regenerating pharynx , we performed double fluorescent in situ hybridization with markers for known subsets of pharyngeal cell types ( muscle , neurons and epithelial cells ) . 3 days after amputation , when the pharynx has regenerated its cylindrical structure with regularly arrayed muscle fibers and concentrated neurons at its distal tip ( Figure 4B ) , FoxA was strongly expressed in the epithelial cells and only weakly in the muscle cells and neurons ( Figure 6A ) . Interestingly , as regeneration proceeded , FoxA expression in the epithelial cells diminished and became more pronounced throughout the interior part of the pharynx ( Figure 6B ) , which is enriched for muscle and neurons . This expression data suggests that FoxA transcripts are present in several subtypes of pharyngeal tissue during regeneration and that expression may diminish once the organ is restored . 10 . 7554/eLife . 02238 . 016Figure 6 . FoxA functions as a master regulator of the pharyngeal lineage . ( A and B ) Confocal images of animals stained for FoxA , myosin ( muscle ) , PC2 ( neurons ) , α-acetylated tubulin ( epithelial cells ) and nuclei showing FoxA enrichment in epithelial cells 3 days after pharynx amputation , and shifting to mesenchyme 7 days after amputation . ( C ) Confocal images of Smedwi-1 FISH ( 7 days after pharynx amputation ) showing distribution of stem cells . ( D ) Body-wide phosphoH3-Ser10 staining in FoxA ( RNAi ) animals during pharynx regeneration . Error bars = SD . ( E ) Local phosphoH3-Ser10 staining during pharynx regeneration in FoxA ( RNAi ) animals . Error bars = SD . ( F ) Dorsal outgrowths in FoxA ( RNAi ) animals ( day 20 ) lack pharyngeal tissue . Tissue-specific markers include: laminin , npp-1 ( pharynx ) , porcupine ( intestine ) , ndk , PC2 ( neurons ) , collagen ( muscle ) . In all images , anterior is up; in side views , dorsal is to the right . Scale bars ( A and B ) 50 μm , ( C ) 500 μm , ( F ) 250 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02238 . 01610 . 7554/eLife . 02238 . 017Figure 6—figure supplement 1 . Dorsal outgrowths in FoxA ( RNAi ) are disorganized . Transverse sections of paraffin-embedded animals fixed 20 days after RNAi administration and stained with hematoxylin and eosin . Dashed yellow line highlights dorsal outgrowth . Dorsal is up . Scale bars = 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02238 . 017 Neoblasts produce all cell types required to replace missing or dying tissues during normal tissue homeostasis and in response to amputation ( Pellettieri and Sánchez Alvarado , 2007; Reddien , 2011; Rink , 2012 ) . One model for how this is accomplished is that upon amputation , cells direct their output into the particular lineages that need to be regenerated . Based on the expression of FoxA in the stem cells , we asked whether it was required for the early steps in regeneration , including maintenance of the neoblast population and stimulation of local proliferation ( Figure 2 ) . Examination of the pattern of Smedwi-1+ neoblasts showed that the overall distribution of stem cells was comparable in FoxA ( RNAi ) and control animals ( Figure 6C ) , demonstrating that the neoblasts were not detectably affected by knockdown of FoxA . Consistent with this observation , body-wide and local proliferation during regeneration was indistinguishable from controls ( Figure 6D , E ) indicating that FoxA is not required for stimulation of proliferation . The finding that FoxA is dispensable for proliferation highlights a key question about the phenotype . If neoblasts divide normally but fail to produce a pharynx , what is the ultimate fate of the division progeny produced by the stem cells ? One possibility is that FoxA expression in the stem cells is required to direct the division progeny into pharyngeal fates . Interestingly , approximately 3 weeks after initiating RNAi , a pronounced dorsal outgrowth formed over the pharynx in 70% of animals ( n > 100 ) . Histological analysis of FoxA ( RNAi ) animals ( Figure 6—figure supplement 1 ) demonstrated that this dorsal outgrowth lacked any recognizable anatomical features of control pharynges , such as radial symmetry or clear laminar structure . Consistent with this observation , these dorsal outgrowths also failed to express either of two pharynx-specific markers ( Smed-laminin and Smed-npp-1 ) ( Collins et al . , 2010 ) or the intestine-specific marker Smed-porcupine ( Figure 6F ) , indicating that this aberrantly produced tissue does not acquire pharyngeal or endodermal fate . However , we did identify enrichment of neuronal markers ( Smed-PC2 and Smed-ndk ) and the muscle marker Smed-collagen , which suggests that in the absence of FoxA , differentiation into neurons and muscle remains intact . To eliminate the possibility that this dorsal outgrowth is an incomplete pharynx formed by restoration of FoxA levels after the dsRNA inhibition has weakened , we examined FoxA expression 20 days after the final administration of RNAi , and found that it remained strongly suppressed ( Figure 6F ) . Because these types of abnormal outgrowths were never observed in chemically amputated animals or in any other RNAi contexts , we conclude that FoxA likely acts to restrict differentiation of new tissues into the pharynx lineage . FoxA appears to function specifically in a subset of stem cells to drive pharynx regeneration in adult animals . Based on its conserved functions controlling endoderm development in other organisms , we decided to use it as a landmark to dissect the function of novel genes uncovered in our screen and implicated in the pharynx regeneration pathway . Following amputation of the pharynx , FoxA is upregulated in the nascent pharynx and in the mesenchymal stem cells surrounding it ( Figure 5B ) . We took advantage of this feature of FoxA to establish a molecular pathway for genes identified in our screen , reasoning that knockdown of genes functioning upstream of FoxA would inhibit accumulation of FoxA+ progenitors , while genes functioning downstream of FoxA would not affect FoxA distribution . Indeed , failure to accumulate FoxA+ cells was observed in irradiated animals ( Figure 5B ) and in Smedwi-2 ( RNAi ) animals ( Figure 7A ) , confirming that inhibiting stem cell function alters accumulation of FoxA+ progenitors . 10 . 7554/eLife . 02238 . 018Figure 7 . FoxA expression resolves a molecular pathway for pharynx regeneration . ( A ) Smed-FoxA expression 7 days after amputation in Smedwi-2 ( RNAi ) animals . ( B ) Smed-FoxA expression 7 days after amputation following knockdown of the indicated genes . ( C ) Gene-specific in situ hybridization in FoxA ( RNAi ) . ( D ) Model for molecular control of pharynx regeneration . Scale bars = 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02238 . 018 In situ hybridization of FoxA following RNAi knockdown of the remaining 19 candidate genes from our screen allowed clear categorization into distinct phases of pharynx regeneration . Based on the severe phenotypes of MCM7 , ribonucleotide reductase , RuvB , cyclin D1 , SAE-2 , CPSF-3 , and zmym-1 in regeneration of head and tail structures after transverse amputation ( Table 1 ) , along with their enriched expression in stem cells ( Figure 3—figure supplement 1 ) , we had predicted that these genes were likely to act in stem cells . As expected , knockdown of these genes caused complete inhibition of FoxA progenitor accumulation , as well as a failure to maintain the mesenchymal population of FoxA+ cells ( Figure 7B ) . Therefore , we conclude that this category of genes is required for the maintenance of pharyngeal progenitors , and confirmed that they function upstream of FoxA . Knockdown of other genes caused significant , but less profound , defects in FoxA accumulation during pharynx regeneration . For example , PABP-2 ( RNAi ) caused a severe decrease in the number of FoxA+ precursors accumulating in or around the pharynx ( Figure 7B ) . Given the important role for translational regulation of planarian stem cell function ( Solana et al . , 2013 ) , it is possible that PABP-2 ( as well as CPSF-3 ) may act in stem cells or in their descendants . Knockdown of WDR3 and WDR36 caused an unusual distribution of FoxA+ precursors ( Figure 7B , white arrows ) , with streams and clumps of FoxA+ cells present in the mesenchyme but directed toward the nascent pharynx . These phenotypes suggest the possibility that FoxA+ cells may be specified at a distance from the wound site , and then migrate towards it , and that WDR3 or WDR36 may function in either cell migration or maintenance of the pharynx regeneration program . Given the slight regeneration defects in head and tail regeneration caused by knockdown of WDR3 and WDR36 ( Table 1 ) , these genes may function broadly during regeneration of other organs . Knockdown of other genes ( e . g . , Fos-1 , Madd4 , and SART-3 ) caused subtle or undetectable defects in FoxA expression dynamics , suggesting that they function downstream of FoxA . Our in situ hybridization timecourses demonstrated that expression of many candidate genes increased around the pharynx during regeneration ( Figure 3—figure supplement 1 ) . Because FoxA functions as a regulatory node in specifying pharynx regeneration , we analyzed the expression of candidate genes in FoxA ( RNAi ) animals to confirm their function either upstream or downstream of FoxA . This analysis was restricted to genes that showed clear expression changes in the vicinity of the pharynx during regeneration . Interestingly , ribonucleotide reductase , cyclin D1 , and RuvB accumulated strongly around the nascent pharynx in FoxA ( RNAi ) animals ( Figure 7C , white arrows ) , providing further evidence that in the absence of FoxA , stem cells are prevented from differentiating into the pharyngeal lineage . Chemical amputation has several advantages over current surgical methods . First , the amount and types of tissues amputated is consistent among animals , stimulating the same regeneration program in each worm . Because planaria lack clear anatomical landmarks and vary in body proportion , surgical amputations are inherently variable among animals , removing different amounts of each tissue with each amputation . Second , chemical amputation produces wounds of exactly the same size , which normalizes both the extent of mitotic activity and the degree of apoptosis , both known to correspond directly with wound size ( Pellettieri et al . , 2010; Wenemoser and Reddien , 2010 ) . Third , regeneration of the pharynx after chemical amputation can be quantified by measuring feeding behavior , facilitating rapid screens . By reducing the complexity of amputation , we have simplified the challenge faced by neoblasts , requiring them to sense the absence of only one organ , and to channel their output into the pharyngeal lineage . This is in contrast to essentially all other types of surgical amputation performed in planaria , which introduce epithelial wounds and damage multiple underlying tissues that are broadly distributed throughout the body . Therefore , chemical amputation allows us to isolate the response of stem cells to the loss of a single organ in a potentially high-throughput , quantifiable manner . Utilizing chemical amputation as a foundation for an RNAi screen , we were able to identify 20 genes that are required at multiple stages for pharynx regeneration , as measured by feeding behavior . However , given the small size of this screen , we have probably uncovered only a portion of the genes acting at each stage of pharynx regeneration . For example , two of these genes ( Rhomboid and PDZ ring finger 4 ) were undetectable immediately after amputation , but strongly upregulated 12–24 hr later ( Figure 2—figure supplement 3B ) . Even though these genes failed to produce an RNAi phenotype , their specific transcriptional activation after wounding indicates that mechanisms exist that allow animals to distinguish between homeostatic versus regenerative events . Nonetheless , we uncovered a pharyngeal regeneration molecular pathway for genes identified in this screen ( Figure 7D ) . Pharynx regeneration begins with activation of stem cells , an increase in expression of a pharynx-specific progenitor marker , and migration of these progenitors to the blastema . Some of the basic mechanisms driving regeneration ( e . g . , stem cell proliferation ) are likely shared between the pharynx and other organs , and we identified several genes in this category , including ribonucleotide reductase , MCM7 , and CPSF-3 . In addition , we expected to identify genes specific to pharynx regeneration , and we identified at least one gene ( FoxA ) that appears to be highly specific for this organ . Interestingly , in C . elegans , the DNA helicase RuvB functions in a genetic pathway with FoxA to regulate pharynx organogenesis ( Updike and Mango , 2007 ) . Our results demonstrate that in planaria , RuvB is required for stem cell function , but it may also play a role in properly specifying progenitors during regeneration . A further category of genes identified were those that potentially affect the migration of progenitors into the nascent pharynx ( WDR3 and WDR36 ) , highlighting the fact that migration of stem cells and/or their progeny is an essential step in pharynx regeneration . FoxA functions as a pioneer transcription factor , opening chromatin due to its structural similarity to linker histones and activating transcription of endoderm-specific genes ( Cirillo et al . , 2002; Gaudet and Mango , 2002; Eeckhoute et al . , 2009 ) . Although it has been previously characterized in development and cancer ( Lupien et al . , 2008 ) , our study represents the first indication that FoxA functions in regeneration . FoxA is known to define organ identity during development ( Gaudet and Mango , 2002 ) and to directly modify chromatin structure ( Cirillo et al . , 2002 ) , leading to a hypothesis that during planarian regeneration , FoxA plays an analogous role . Because FoxA ( RNAi ) animals still established expression of anterior/posterior patterning molecules and regenerated head and tail structures normally , we conclude that FoxA is not required for regenerating or patterning organs besides the pharynx . We note , however , that some expression of pan-pharyngeal markers is maintained in FoxA ( RNAi ) animals , raising the possibility that organ specification is only partially compromised . Although this may be due to incomplete knockdown of FoxA , alternatively , it may reflect a block in differentiation of pharyngeal tissue , which requires FoxA activity . The pharynx consists of neurons , muscle , mesenchyme , and epithelial cells , but lacks neoblasts . Each of these differentiated tissues has a distinct morphology and identity from the rest of the animal , suggesting that these different cell types may share a common pharyngeal identity . Based on our results , it is possible that FoxA may act to initiate the pharynx differentiation hierarchy to establish organ identity during regeneration , with additional layers of cell-specific differentiation occurring later . FoxA mRNA was expressed in multiple tissue types as regeneration progressed , suggesting that FoxA activation in stem cells is the first step toward differentiation of several pharynx-specific cell types . However , the exclusion of stem cells from the pharynx indicates that a boundary within the mesenchyme prevents pluripotent stem cells from invading this organ . FoxA expression bridges this spatial boundary . Our data demonstrate that FoxA transcript is present in the stem cell population , like other transcription factors that are critical for brain , photoreceptor , and excretory system development in planaria ( Lapan and Reddien , 2011; Scimone et al . , 2011; Cowles et al . , 2013 ) . However , in the case of pharynx regeneration , we can monitor the percentage of FoxA+ stem cells in response to complete organ amputation , demonstrating that the stem cell population alters its output in response to organ amputation . The patterning and regeneration that occurs after amputation implies that signaling within the animal provides instructive cues guiding neoblast differentiation into particular fates . An interesting question raised by this work is how FoxA expression is triggered in stem cells . In other animals , FoxA recruitment to chromatin is controlled by trans-acting factors including T-Box , GATA , and lef transcription factors ( Mango , 2009 ) , and these types of proteins may function cell-autonomously in neoblasts . Alternatively , signals acting distantly from the neoblasts are likely to stimulate FoxA expression in neoblasts . One possibility is that these signals originate in the pharynx , and normally limit the production of FoxA+ stem cells . These kinds of molecules , known as chalones and best typified by myostatin/GDF11 ( Bullough , 1965; McPherron et al . , 1997 ) , have been characterized in mammalian muscle and are thought to limit organ size in adult animals . In planarians , induction of supernumerary pharynges or engraftment of transplanted pharynges only occurs at a distance from the resident pharynx , suggesting that an inhibitory activity is present in the peripharyngeal region ( Ziller-Sengel , 1967a , 1967b; Schilt , 1972 ) . Gaining a molecular handle on these aspects of whole-body regeneration will enhance our understanding of organismal homeostasis in animals . In sum , our findings uncovered a new role for FoxA in adult animal regeneration and demonstrates that our organ-selective screening strategy can identify genes with distinct and specific functions during regeneration . Pharynges were removed from animals 4–5 mm in size , and starved for 7 days . Planarian water was replaced with 100 mM sodium azide ( diluted in Montjuïc water ) . After 5–7 min , the pharynx was visibly extended out of the body . Vigorous pipetting often dislodged the pharynx from the body; if necessary , fine serrated forceps ( #5441; Ted Pella , Reading , CA ) were used to remove the pharynx , followed by several washes in Montjuïc . For tricaine treatment , 2 g/l tricaine was diluted in 10 mM Tris pH 7 . 5 . Animals soaked in tricaine display their pharynx but never eject it . Schmidtea mediterranea asexual clonal line CIW4 was maintained and used as previously described ( Newmark and Sánchez Alvarado , 2000 ) . Animals were exposed to 6000 or 10 , 000 rads on a GammaCell 40 Exactor irradiator . At specified times after amputation , plugs were extracted using 1 mm microcapillary pipets ( FHC , catalog # 30-30-0 , Bowdoin , ME ) , and transferred directly into Trizol ( Life Technologies , Grand Island , NY ) using a mouth pipet . For each replicate , 25 plugs were homogenized together , and then chloroform-extracted . The pellet was then precipitated with isopropanol , washed , and resuspended in water . RNA was then purified on an RNEasy column with DNase-treatment ( Qiagen , Germany ) . The experiment was performed in triplicate . RNA quality was assessed on a Bioanalyzer 2100 machine ( Agilent , Santa Clara , CA ) . Starting with 100 ng total RNA , amplification and labeling with Cy3 or Cy5 was performed using the Low Input Quick Amp Labeling Kit Two-Color from Agilent Technologies ( #5190-2306 ) . Custom Agilent 4x44k arrays with design id: 033226 were hybridized according to the manufacturer protocols , and scanned on an Agilent G2505C scanner . Data was analyzed in the R environment using the Limma library ( Smyth , 2004 ) for loess normalization and calculation of p-values between treatments . p-values were adjusted for multiple hypothesis testing by the method of ( Benjamini and Hochberg , 1995 ) . The data have been deposited in GEO with accession number: GSE56181 . Primers with overhangs homologous to pPR-T4P vector ( J Rink ) were used for PCR amplification from a cDNA library generated with SuperScriptIII ( Life Technologies ) . PCR products were treated with T4 polymerase , mixed with linearized vector ( digested with SmaI and treated with T4 polymerase ) and incubated for 15 min at room temperature . Ligations were transformed directly into Escherichia coli strain HT115 , then verified by PCR and sequencing . For screening , overnight cultures of individual cDNAs were grown in 2XYT , and 2X RNAi food was prepared ( 30 ml bacterial culture was pelleted and resuspended in 150 µl 3:1 liver:water paste ) . 15 animals were fed three times , 3 days apart , and amputations were done the following day . Feeding assays were performed 9 days after amputation . All RNAi experiments used this same timing strategy , with day 0 representing the time of amputation . Sequences of genes used in this study are deposited in GenBank with accession numbers KJ573350-KJ573369 . For the feeding assay , animals were transferred into a new petri dish and kept in the dark for at least an hour . Diluted liver paste consisting of approximately 4:1 liver:planarian water and 20 µl red food coloring was mixed and 25 µl was pipetted into the petri dish . Percentage of animals with red intestines were scored after approximately 30 min food exposure . For FoxA ( RNAi ) , we used 4X RNAi food ( 100 ml overnight culture resuspended in 250 µl liver paste ) . In situ hybridizations used the protocol in Pearson et al . ( 2009 ) for colorimetric development and the protocol in King and Newmark ( 2013 ) for fluorescent development except that animals were fixed for 45 min in a solution containing 4% PFA , 0 . 5% Triton X-100 , and 1X PBS . For mounting , we soaked fluorescently stained animals overnight in modified ScaleA2 solution for improved optical clarity ( Hama et al . , 2011 ) containing 40% glycerol , 2 . 5% DABCO ( Sigma–Aldrich , St . Louis , MO ) and 4M urea . For cryosectioning and immunohistochemistry , after completion of WISH animals were fixed overnight at 4°C in 4% paraformaldehyde ( in PBS ) , washed three times in PBS , equilibrated in 30% sucrose , frozen in OCT and cryosectioned ( 10 µm thick ) . To stain sections , slides were incubated in 1X Powerblock ( Biogenex , Fremont , CA ) for 30 min , then incubated with rabbit monoclonal acetylated tubulin at 1:1000 ( #5335; Cell Signaling Technology , Danvers , MA ) and Tmus ( kind gift of Rafael Romero , used 1:1000 ) for at least 1 hr , and developed with Alexa-conjugated secondary antibodies diluted 1:1000 ( Abcam , Cambridge , MA ) . All antibodies were diluted in Antibody Diluent Solution ( Life Technologies ) . Anti-H3Ser10Phos ( Millipore , Billerica , MA ) was used at 1:1000 and developed with Alexa-conjugated goat anti-rabbit secondary antibodies ( 150086; Abcam ) . For hematoxylin and eosin staining , animals were fixed overnight at 4°C in 4% paraformaldehyde ( in PBS ) , then washed 3X in PBS and dehydrated by an ethanol series through washing in 30% , 50% , 70% , 80% , 95% , and 100% ethanol ( 7 min each ) . To embed in paraffin , animals were soaked in ethanol with 5% glycerol , washed in xylene ( 7 min ) and clear-rite ( 2 × 7 min ) , and soaked in paraffin ( 2 × 14 min , then 2 × 30 min ) . After serial sectioning ( 10 µm thickness ) , slides were heated to 60°C for 20 min , deparaffinized with three 2-min washes in xylene , washed 3 × 1 minute in 100% ethanol , then 80% ethanol , rinsed in tap water , and then incubated 30 s in hemalast , 2 min in hematoxylin , rinsed 2 min in tap water . Staining utilized the Leica Infinity system and was performed in a Leica Autostainer . Colorimetric WISH images were captured on either Zeiss Lumar or Leica M205 stereoscopes . Confocal images were captured on a Zeiss LSM510-VIS inverted microscope with a 20X or 40X objective . For quantification of phosphohistone H3 , full slide or individual worm tiled image sets were acquired on a Perkin Elmer Ultraview spinning disk microscope . Stitching was performed using stitching plugins in FiJi with customized batch processing macros or wrapper plugins where necessary . Custom plugins were used to segment the DAPI labeled worms and the ‘Find Maxima’ function was used to count spots , both wrapped in batch processing macros . All macros and plugins are available at https://github . com/jouyun .	Some animals can regrow whole limbs or organs after amputation . Flatworms called planaria , for example , can regenerate their whole body from small pieces . This remarkable ability depends on neoblasts—a type of stem cell found in planaria that can detect damaged or lost organs , migrate to the site of damage , produce the required cells , and integrate into the remaining tissues . Researchers hope that studying these animals will reveal ways to use stem cells to regenerate injured limbs or organs in humans . Planaria have been used in many studies of regeneration . However , manually amputating organs from the flatworms is time-consuming and the resulting wounds vary , which makes it hard to compare regeneration between animals treated in different ways . Now , Adler et al . have developed a new technique for studying regeneration in planaria . Placing the flatworms briefly into a solution of sodium azide causes the pharynx—an organ that is used for both eating and excretion–to drop off . Using chemicals in this way means the loss of the pharynx leaves a uniform wound , with no damage to the adjacent digestive system , and that large numbers of planaria with identical wounds can be produced rapidly . To ensure that treatment with sodium azide did not alter normal regeneration processes in planaria , Adler et al . carried out manual amputation of tissue in sodium azide-treated flatworms; regeneration in these flatworms was identical to that seen in untreated planaria . Planaria in which the pharynx had been removed by sodium azide exposure showed rapid recruitment of neoblasts to the wound site , where they formed epidermal , muscle and nerve cells , and organized into a functioning pharynx within a few days . Adler et al . then identified 20 genes that were required for various stages of regeneration . These experiments revealed that a transcription factor ( a protein that controls gene expression ) called FoxA was specifically required for the regrowth of the pharynx . This is a previously unknown function for FoxA .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine" ]	2014	Selective amputation of the pharynx identifies a FoxA-dependent regeneration program in planaria
The human visual system is tasked with recovering the different physical sources of optical structure that generate our retinal images . Separate research has focused on understanding how the visual system estimates ( a ) environmental sources of image structure and ( b ) blur induced by the eye’s limited focal range , but little is known about how the visual system distinguishes environmental sources from optical defocus . Here , we present evidence that this is a fundamental perceptual problem and provide insights into how and when the visual system succeeds and fails in solving it . We show that fully focused surface shading can be misperceived as defocused and that optical blur can be misattributed to the material properties and shape of surfaces . We further reveal how these misperceptions depend on the relationship between shading gradients and sharp contours , and conclude that computations of blur are inherently linked to computations of surface shape , material , and illumination . An image of a surface is a product of its three-dimensional ( 3D ) shape , the reflectance and transmittance behavior of its material , the surrounding light sources ( the ‘light field’ ) , and the focal parameters of the imaging lens . All of these sources are conflated in the light that reaches the eyes , yet we nevertheless perceive distinct impressions of shape , material , illumination , and focus . One of the fundamental goals of mid-level vision research is to understand how the visual system extracts these different sources of structure . Most research into this problem has focused on how the visual system extracts environmental sources of structure – 3D shape , reflectance ( color , lightness , gloss ) , and surface opacity ( transparency , translucency , and subsurface scattering ) . But the focal properties of single-chambered eyes also contribute to the optical structure projected to the retinae . Eyes that utilize refraction to generate a focused image are subject to depth of field defocus , which causes some image regions to be blurred . Optical defects can also cause images to be globally blurred for the increasing number of people that require optical correction . Although there has been a significant body of research into cues that affect the severity of perceived blur ( Ciuffreda et al . , 2006; Crete et al . , 2007; Ferzli and Karam , 2006; Pentland , 1987; Tadmor and Tolhurst , 1994; Webster et al . , 2002 ) and the role of depth of field defocus as a cue to depth ( Held et al . , 2010; Marshall et al . , 1996; Mather , 1996; Mather , 1997; Mather and Smith , 2000; Mather and Smith , 2002; O'Shea et al . , 1997; Watt et al . , 2005 ) , it is still unknown how the visual system distinguishes blur from environmental sources of low spatial frequency image structure . The computational problem of discriminating optical defocus from environmental sources of low frequency structure does not appear to have been explicitly addressed previously . This may be due to the absence of empirical evidence that the visual system can misattribute image gradients produced by environmental sources to defocus or misattribute gradients produced by defocus to environmental sources . Here , we provide evidence of both . We show that defocus can be experienced in fully focused images and that optical defocus can be misperceived as distortions in the perceived 3D shape of smoothly shaded surfaces . Consider the surface depicted in Figure 1 , which was created by illuminating a smooth ( i . e . , differentiable ) Lambertian ( ‘matte’ ) surface with a collimated light source . The surface has shallow surface relief to avoid the formation of sharp attached shadows and is viewed along the axis of relief to avoid the formation of self-occluding contours . Although rendered as a fully focused surface , this image elicits a strong perception of blur; while some 3D shape from shading may be perceived , the surface appears ‘contaminated’ by optical defocus . The perceptual conflation of low frequency shading gradients and optical focus does not appear to have been previously reported . Most research into the perception of shading has attempted to understand how shading provides information about 3D shape using images where it was assumed or somehow ‘known’ that the intensity gradients were caused by focused patterns of shading . The potential conflation of shading and blur may have been overlooked because of the particular surface geometries and viewing conditions that were used in these studies . Most previous work on shape from shading has studied images that contained sharp contours generated by either smooth self-occlusions or abrupt bounding contours ( e . g . Horn and Brooks , 1989; Koenderink et al . , 2001; Mingolla and Todd , 1986; Pentland , 1984; Ramachandran , 1988; Todd et al . , 1996 ) , while experiments that have used ‘terrain’ surfaces similar to Figure 1 have predominantly been investigations of illumination perception ( Koenderink et al . , 2004; Koenderink et al . , 2007 ) . This suggests that the visual system may exploit sharp contours in generating percepts of focus when viewing low frequency shading gradients . If so , it should be possible to eliminate the perception of illusory blur in Figure 1 by introducing sharp bounding contours . We informally tested this hypothesis by constructing the images depicted in Figure 2 . The image on the left ( 2A ) was generated by intersecting the surface in Figure 1 with a gray plane parallel to the direction of its relief and occluding all surface regions beyond the depth of that plane ( a particular form of ‘planar cut’ dubbed a ‘level cut’ ) . Informal observation suggests that this manipulation enhances the perceived 3D shape of the surface and completely eliminates the perception of blur . Note also that the contour appears to be unambiguously ‘owned’ by ( attached to ) the shading gradients rather than the homogeneous gray regions . Now consider the image in Figure 2B , which was created by treating the gray regions in Figure 2A as parts of a single gray ‘mask’ and rotating it 180 degrees over the shaded surface . This image contains the exact same contours as Figure 2A , but the gray regions occlude different portions of the shaded surface . There are a number of striking perceptual differences evoked by comparison of Figure 2B and Figure 2A . First , whereas the perceived shape from shading is enhanced in Figure 2A , the perception of shape from shading is impaired by the contours in Figure 2B . A second difference is that the border ownership of the contour in Figure 2B is ambiguous; whereas the contours in Figure 2A appear unambiguously attached to the shading gradients , the contours in Figure 2B can appear attached to either side . The way the border ownership is perceived can have a dramatic impact on how the gradients are perceived . When the contours appear attached to the shading gradients , the gradients are perceived as unstructured 2D ‘noise’ ( such as variations in pigment ) without any clear sense of 3D shape or optical focus . But when the contours appear attached to the gray mask , the shaded surface appears as a partially occluded shaded surface that is just as blurred as the surface in Figure 1 . What is responsible for the striking perceptual differences observed in Figure 2A and Figure 2B ? Why does one set of contours eliminate the perceived blur , enhance perceived 3D shape , and unambiguously determine the side of the contour that is figure , while the other does not ? We suggest here that the perceived focus and enhanced perception of 3D shape arises because the level cut images approximate the geometric and photometric image properties generated by smooth self-occluding contours . More specifically , we will argue that there are two constraints that play a causal role in eliciting the perception of focus in images of shaded surfaces: photogeometric constraints that arise generically along smooth self-occlusions; and the attachment of shading gradients to convex surfaces . We consider each in turn . The first constraint arises from the physics of surface reflectance and the projective geometry of smooth self-occluding rims . Both shading intensity and bounding contour shape depend on the same environmental property – the local 3D shape of the surface – and are therefore inherently linked . This link makes it possible to combine two well-known constraints about the shape of the contours generated by smooth self-occlusions and the shading intensity into a novel constraint ( Marlow et al . , 2019 ) . First , it is known that the 3D shape along a smooth self-occluding rim can be derived from its 2D image contour: the local slant of the contour relative to the observer ( i . e . , how much it deviates from fronto-parallel ) is constant ( approximately 90 degrees ) , and the tilt of the surface ( the direction in which it slants away from the observer ) is specified by the orientation of the rim’s image contour ( Barrow and Tenenbaum , 1978 ) . Second , it is known that local 3D surface orientation is primarily responsible for shading intensity ( Horn and Brooks , 1989 ) , subject to some modification by vignetting and/or interreflections ( Langer and Zucker , 1994 ) . The mutual dependence of contour orientation and shading intensity on the same local 3D surface orientation causes shading to covary with the orientation of the rim’s projected image contour: for a Lambertian surface illuminated by a collimated light source , intensity will decline as a cosine function of contour orientation relative to the brightest point along the contour ( i . e . , the orientation most closely aligned with the illumination direction ) . This relationship holds exactly for surfaces illuminated by collimated light sources ( see Materials and methods ) , and we have previously shown that it is statistically robust in natural light fields that contain multiple sources of illumination ( Marlow et al . , 2019 ) . We refer to this relationship as the orientation-intensity covariation of the contour and its adjacent shading . The covariation of contour orientation and intensity along smooth self-occluding rims generalizes to other types of bounding contours that have been shown to affect perceived shape from shading , such as ‘planar cuts’ ( i . e . , contours formed by ‘slicing’ a shaded surface with a plane; Marlow et al . , 2019 ) . The level cut image in Figure 2A is one example of a planar cut . The orientation-intensity covariation that arises along planar cuts of shaded surfaces is similar to that generated by smooth self-occluding rims ( see Materials and methods ) . The relationship between contour orientation and shading intensity along the contours in Figure 2 is depicted in the plot below each image . Note that Figure 2A exhibits a clear cosine-like relationship , whereas Figure 2B exhibits no covariation at all . However , this covariation is generally weaker for planar cuts than for self-occluding rims: more than one 3D surface orientation can generate the same 2D contour orientation , which means that identically oriented planar cut segments can project different shading intensities in the image ( as can be seen upon close examination of Figure 2A ) . Nonetheless , we previously showed that planar cuts exhibit a robust orientation-intensity covariation ( apart from a few degenerate cases; Marlow et al . , 2019 ) , which suggests that this covariation could provide a reliable cue that the visual system uses to identify the bounding contours of shaded surfaces . The second property that links the level cut image in Figure 2A to images containing smooth self-occlusions is that they are both globally convex ( Koenderink , 1984 ) . If the covariation of intensity and contour orientation along sharp bounding contours is sufficient to explain the perception of focus in shaded surfaces , then it should not matter whether these contours bound a convex or concave surface . The importance of convexity can be assessed by constructing a level cut surface that removes all of the convex surface regions that appear in front of the cut and displaying only the ‘valleys’ ( Figure 2C ) . This stimulus exhibits the same orientation-intensity covariation , but is inherently ambiguous: it can be perceived as a convex surface illuminated from below or as a concave surface illuminated from above ( e . g . Ramachandran , 1988; see Liu and Todd , 2004 ) . There are two well-established biases that determine how such ambiguities are resolved: a bias to perceive the illumination as coming from above ( Belhumeur et al . , 1999; Koenderink et al . , 2001 ) and a bias to perceive surfaces as convex ( Hill and Bruce , 1994; Langer and Bülthoff , 2001 ) . These two biases are aligned in Figure 2a , which is presumably why this surface is perceived as a stable convex surface illuminated from above . However , these biases are in conflict in Figure 2C , causing some perceptual bistability . Informal observations suggest that perceived focus depends on the perceived convexity or concavity of the surface . When the surface appears convex , no clear percept of blur is experienced; but when it appears concave , the surface appears blurred . This implies that the strength of the orientation-intensity covariation along the contours is not the sole determinant of the perception of shading , focus , and contour attachment; the convexity of the surface is also critical . Our previous work showed that photogeometric cues along smooth self-occlusions and planar cuts of 1D luminance profiles provide information that predicts when identical luminance gradients are perceived as surface shading of 3D surfaces . The experiments described below were designed to psychophysically assess the relationship between bounding contour orientation , shading intensity , and contour sharpness on the perception of surface shading , optical defocus , and border ownership . The goal of Experiment 1 was to test if sharp image contours cause bounded shading gradients to appear more focused when contour orientation covaries with the intensity of the bounded gradients . This covariation is typically exhibited along contours generated by self-occlusions . Self-occlusions feature prominently in prior work on shape from shading , but their contours are difficult to manipulate without altering the geometry of the entire shaded surface ( and hence the shading gradients ) . However , it is possible to approximate the photogeometric behavior along smooth self-occlusions with planar cuts . This was accomplished by slicing the bumpy plane depicted in Figure 1 with a plane oriented perpendicular to the axis of surface relief ( a ‘level cut’; see Materials and methods for details ) . The resulting level cut contours exhibit an orientation-intensity covariation similar to the covariation along self-occluding contours , but unlike self-occlusions , level cuts can be generated anywhere on the surface and place no constraints on local surface curvature . The photogeometric constraints of self-occlusions and level cuts are described in further detail in the Materials and methods . To assess the role of the photogeometric covariation along bounding contours in the perception of 3D shape , border ownership , and optical focus , three sets of level cut contours were created by intersecting the surface in Figure 1 with a fronto-parallel plane at different depths along the axis of surface relief . Each set of contours was used to generate four homogenous gray masks . In the first condition , each mask occluded all surface regions that lay at a greater depth than the level cut , leaving only the shaded peaks visible ( e . g . Figure 2A ) . The visual system’s biases toward interpreting shaded surfaces as convex and top-lit are aligned in this image , and the visible surface regions consequently appear unambiguously convex; we therefore refer to this condition as the ‘convex’ level cut condition . In a second condition , the same contours were used to generate a complementary mask that removed all regions in front of the level cut , leaving only the shaded valleys visible ( e . g . Figure 2C ) . The visual system’s convexity and illumination biases conflict in this image; the shaded regions may appear as concave dents illuminated from above or convex bumps illuminated from below . This conflict can result in some bistability in the perceived illumination direction and convexity/concavity , so we refer to this as the ‘bistable’ level cut condition . The masks in the two ‘rotated’ conditions were created by rotating the level cut masks in the convex and bistable conditions ( respectively ) by 180° over the underlying gradients , which breaks the orientation-intensity covariation that occurs along the level cuts ( e . g . Figure 2B ) . The depth of the intersecting plane used to define the contours determined the relative proportion of visible gradients in each masked image; the three depth values were chosen to create masks that preserved 25% , 50% , and 75% of the gradients . Note that the masks in the ‘convex’ conditions with 25% gradients visible are the complements of the masks in the ‘bistable’ conditions with 75% gradients visible , and vice versa . The luminance of each gray mask was set to the average of the original gradient image . The full set of twelve stimuli can be seen in Figure 2—figure supplement 1 and Figure 2—figure supplement 2 . If the visual system uses the orientation-intensity covariation between contours and shading to distinguish attached bounding contours from arbitrary edges , then the level cut contour conditions should induce stronger percepts of focus than the rotated contours , which have no relationship with the shading gradients . However , if the visual system is better at inferring contour attachment when the adjacent surface appears convex , then the covariation alone ( which provides no curvature cues ) may not fully eliminate the perception of blur . If this is true , then the contours in the ‘bistable’ level cut condition should produce weaker percepts of focus than the ‘convex’ level cut condition , as the bistable stimuli may sometimes appear concave . Observers judged perceived focus in a paired comparison task and judged perceived surface curvature ( convex , concave , or neither ) in a separate three-alternative classification task ( N = 20 ) . The results confirm our hypotheses and informal observations of the stimuli ( Figure 3 ) and were analyzed with an ANOVA and appropriate two-sided contrasts . Observers perceived the level cut stimuli ( solid lines in Figure 3A ) as appearing more focused than the rotated stimuli ( dotted lines in Figure 3A ) , F ( 1 , 19 ) =64 . 68 , p<0 . 001 , 95% CI [25 . 33 , 43 . 15] , Cohen’s d=1 . 85 , and further perceived the ‘convex’ level cut stimuli as more focused than the ‘bistable’ level cut stimuli , t ( 19 ) =7 . 05 , p<0 . 001 , 95% CI [20 . 45 , 37 . 73] , Cohen’s d=1 . 62 . All observers perceived the ‘convex’ level cut surfaces as convex bumps on all trials ( first plot in Figure 3B ) , whereas the gradients occluded by rotated masks were most likely to be perceived as neither bumps nor dents ( third and fourth plots in Figure 3B ) . Perceived focus decreased as a function of increasing gradient visibility in the ‘bistable’ level cut condition , t ( 19 ) =7 . 07 , p<0 . 001 , and the rotated conditions , t ( 19 ) =5 . 05 , p<0 . 001 , but not the convex level cut condition , t19=1 . 73 , p=0 . 100 , which suggests that the convex level cut masks were the only stimuli that were perceived as ( equally ) fully focused . In the other three mask conditions , it is likely that observers simply preferred to select images in which less of the perceptually blurred gradients were visible . The data suggest that the effects of the different contours on perceived focus are mediated by the perception of convexity , which is strongest when the contours exhibit an orientation-intensity covariation with the shading and the convex interpretation is consistent with percepts of top-down illumination . Notably , perceived focus and perceived convexity simultaneously decreased in the ‘bistable’ conditions – but not the ‘convex’ conditions – as the ratio of visible gradients increased . There was a significant correlation between perceived focus and the proportion of observers who rated each stimulus as appearing convex in shape , ρ=0 . 904 , p< . 001 . This correlation remained significant even when the unambiguous ‘convex’ mask conditions were excluded from the analysis , ρ=0 . 777 , p=0 . 014 . This finding supports our informal observation from Figure 2: contours generate stronger cues to contour attachment when they bound surfaces regions that appear convex , and these contours are therefore more likely to propagate focus cues ( produced by their sharpness ) to the shaded surface . This may also explain why prior studies using globally convex shaded 3D shapes have not observed any illusory gradient blur: the self-occluding contours of these stimuli not only exhibit a strong orientation-intensity covariation ( Marlow et al . , 2019 ) but also necessarily bound surface regions that have convex curvature in at least one direction . The bounding contours of surface regions that appear concave , however , can appear as occluding edges of the gray mask , similar to the percept that arises from the rotated contours ( Figure 2B ) when the gray regions appear to ‘own’ the contours and the surface appears occluded . The results of Experiment 1 suggest that sharp contours that exhibit an orientation-intensity covariation with nearby shading ( such as level cuts ) provide information about image focus and enhances percepts of 3D shape within the shaded surface . In Experiment 2 , we tested the importance of this covariation directly by parametrically varying the strength of the orientation-intensity covariation along contours . This was accomplished with images of smoothly shaded ‘ribbons’ on a gray background . These images were created by displaying only the shading gradients immediately adjacent to the mask contours used in Experiment 1 . Thus , the only source of information about 3D shape is the relationship between the intensity and orientation along the ribbon . Sixteen unique ribbon paths ( two pixels wide ) were created by generating more smoothly deformed 3D surfaces and intersecting them with planes to produce level cut contours , but the ribbon was manually shaded in MATLAB according to the orientation of its path . This allowed us to create ribbons that exhibited perfect orientation-intensity covariation that could be parametrically decreased to any arbitrary value . Ribbons were constructed such that intensity decreased in relative intensity from 0 . 8 to 0 . 2 as a linear function of its orientation relative to 90° ( the brightest point ) . The covariation was then progressively weakened by gradually adding increasing amounts of random low-frequency noise to the ribbon gradients . The gray background had a relative intensity of exactly 0 . 5 . Figure 4 depicts three example stimuli . The ribbons in the left panel exhibit a perfect linear orientation-intensity covariation , which induces a vivid percept of surface relief: some of the gray regions are perceived as ‘plateaus’ raised above the adjacent recessed regions . The Pearson correlation coefficient between shading intensity and contour orientation ( relative to 90° ) measured directly from each image is shown in the lower-right corner . In the center panel , the intensity of a different perfectly covarying ribbon has been mixed in equal proportion with low-frequency noise . A moderate degree of covariation remains , but it is not consistent across the image , and the overall impression of stepped relief is substantially weaker . In the right panel , ribbon intensity has been generated entirely by noise; no covariation is present , and no impression of 3D relief is apparent . Observers ( N = 15 ) were shown randomly-generated ribbon images with thirteen values of noise proportion ranging from 0% ( as in the left panel of Figure 4 ) to 100% ( as in the right panel of Figure 4 ) . These values were not evenly spaced but were instead selected to produce an approximately uniform distribution of correlation coefficients between 1 and 0 when the covariation was measured in each image . Each observer viewed all possible pairs of these thirteen noise values and were instructed to select the image that appeared more three-dimensional in each pair . The results confirm our informal experience of Figure 4: 3D shape percepts monotonically decreased in strength as a function of increasing ribbon noise ( left plot in Figure 5 ) . This relationship was verified with a linear regression , with the likelihood of being selected as appearing more three-dimensional decreasing by approximately one percentile for every percentile increase in ribbon noise , b=−1 . 025 , R2=0 . 923 , p<0 . 001 . Further analysis revealed that this effect was mediated by the amount of orientation-intensity covariation present in the images: mean perceived 3D shape strength also decreased monotonically with the value of the Pearson correlation coefficient when the correlations measured from all presented stimuli were sorted into bins of width 0 . 1 ( right plot in Figure 5 ) . These findings do not directly address the issue of perceived defocus , but do support our hypothesis that the photo-geometric behavior occurring at the very edge of covarying contours ( such as the level cut contours in Figure 2A ) is sufficient to generate the vivid impressions of contour attachment observed in the ‘convex’ conditions of Experiment 1 . The data also reinforce our previous findings on the importance of this covariation in generating percepts of 3D shaded shape ( Marlow et al . , 2019 ) . The results of Experiments 1 and 2 suggest that the absence of an orientation-intensity covariation along contours can lead to the misperception of optical defocus and impair the perception of 3D shape . In Experiment 3 , we investigated whether image structure that has been optically blurred by defocus can be misperceived as focused if there are sharp contours present nearby that exhibit an orientation-intensity covariation; that is we tested whether the presence of such contours can effectively mask the visibility of optically induced blur . We tested this by constructing two variants of the shaded terrain in Figure 1 . One variant was similar to the level cut image , which contained sharp bounding contours a given relief height ( middle of the bottom row of Figure 6 ) , and therefore referred to as the level cut image . The other image was constructed by geometrically smoothing the sharp level cut contours of this image , resulting in the image depicted in the middle of the top row of Figure 6 . As in Figure 1 , the absence of sharp , intensity-correlated contours causes this surface to appear blurred even though it was rendered in full focus . We refer to this surface as the ‘smoothed’ condition . The difference in perceived blur experienced with these two images reinforce the importance of contour sharpness for inducing percepts of focused gradients: the absence of sharp edges in the smoothed condition greatly reduces perceived focus , even though the changes in surface curvature at the edges of the bumps are relatively abrupt . To assess sensitivity to optical focus , both shaded surfaces were subject to different degrees of optical defocus , with the focal length set to either the background ( so the contours surrounding the bumps in the level cut condition remained sharp ) or the tips of the bumps ( so the contours in the level cut condition were blurred by defocus ) . The effects of increasing background blur and increasing foreground blur can be seen to the left and right of the center column in Figure 6 , respectively . The effects of foreground blur on the level cut surface ( bottom-right panels ) are particularly striking: when the peaks of the surface are affected by defocus blur , the resulting changes in the image gradients do not appear to significantly change the perceived focus of the surface . In the absence of the level cut contours , however , the same defocus manipulation appears to increase apparent blur ( top-right panels ) . These informal observations were bolstered by psychophysical experiments that measured perceived focus for all ten stimuli using a paired comparison task ( N = 10 ) . The results ( depicted in Figure 7 ) align with our informal observations of the stimuli in Figure 6 and were analyzed with appropriate contrasts within an ANOVA ( the main effects of which are not relevant here ) . The vertical axis in Figure 7 represents the percentage of trials in which each stimulus was selected as appearing more focused , and the horizontal axis represents the different blur conditions . The blue and red lines depict perceived focus for the smoothed and level cut conditions , respectively . The data reveal that defocus blur can be misperceived as shading: the three stimuli with sharp level cut contours ( i . e . the level cut condition with no blur or foreground blur ) were perceived as the most focused , even when the peaks of the surface were actually blurred by moderate or severe foreground defocus . Statistical analysis revealed that foreground blur had a significantly larger ( more negative ) effect on perceived focus than background blur for the level cut surface , t ( 9 ) =20 . 74 , p<0 . 001 , 95% CI [45 . 29 , 56 . 38] , Cohen’s d=6 . 91 , but not the smoothed surface , t9=0 . 31 , p=0 . 763 , 95% CI [-11 . 49 , 8 . 71] . Furthermore , the fully-focused smoothed surface with no contours did not significantly differ in perceived focus to the level cut stimulus with weak background blur , t9=0 . 71 , p=0 . 494 , 95% CI [-9 . 28 , 4 . 84] . These results demonstrate that the presence or absence of sharp contours attached to shading gradients can induce both misperceptions of optical focus or defocus ( respectively ) when they exhibit a systematic intensity-orientation covariation . The results of the preceding experiments suggest that the most important mediating factor in the perception of focused shading gradients is the presence of sharp bounding contours that exhibit covariation between their orientation and adjacent shading intensity . In Experiment 3 , we found that optical blur was overestimated when these sharp , covarying contours were absent from the image . Actual changes in image focus , however , were effectively undetected by observers when the sharpness of nearby covarying contours was preserved . In Experiment 4 , we tested whether this underestimation of optical blur occurs because the smooth gradient structure is misattributed to environmental sources: that is whether the optical effects of defocus are attributed to the 3D shape and reflectance properties of the defocused surface . Surfaces with higher microscopic roughness ( e . g . glossy plastic or matte materials ) will scatter incoming light in more directions and produce smoother image gradients instead of the sharp , detailed specular reflections produced by low-roughness surfaces ( e . g . mirrors ) . This effect of surface roughness on image structure is further mediated by curvature: high-curvature surfaces vary in surface orientation more rapidly than low-curvature surfaces , and the shading or reflections they produce are therefore more compressed ( i . e . , have higher spatial frequency ) in the image . We have previously demonstrated that the visual system can misperceive low-curvature , low-roughness surfaces as high-curvature , high-roughness surfaces ( Mooney and Anderson , 2014 ) , as both combinations of surface properties generate similar gradient structure . The local effects of optical blur on image gradients are similar to the effects of increasing surface roughness or decreasing surface curvature: all three of these physical transformations typically reduce the sharpness of image gradients exhibited by the surface . It is therefore likely that the visual system will have difficulty distinguishing these optical and environmental influences on gradient appearance when they do not produce simultaneous changes in contour sharpness or covariation strength . We tested this hypothesis by creating three identically shaped surfaces with different reflectance properties: ‘matte’ , ‘rough gloss’ , and ‘smooth gloss’ . All three materials have an identical diffuse shading component , but the two gloss conditions also contain a specular reflectance component . The amount of scattering in this specular component is greater in the ‘rough gloss’ condition , which consequently produces reflections with less detail than the ‘smooth gloss’ condition . The surface’s 3D shape had higher curvature than the shape used in Experiment 3 to increase the likelihood of generating measurable misperceptions of 3D shape . Level cut contours were created by intersecting the surface with a plane , as in previous experiments , but the matte gray planar surface itself was here included as part of the rendered scene rather than a mask added to the image afterward . The combined surface was rendered in a natural light field with cast shadows , inter-reflections , and chromatic information . This was done to test whether misperceptions of optical blur occur in more realistic viewing conditions , which are particularly important for the appearance of glossy materials ( Fleming et al . , 2003; Pellacini et al . , 2000 ) . Each material condition was rendered with five optical defocus conditions , identical to the conditions in Experiment 3: the ‘foreground blur’ conditions blurred the peaks of the surface but preserved the sharp contours , and the ‘background blur’ conditions blurred the sharp contours , leaving the peaks unaffected . The fifteen stimuli are depicted in Figure 8 , which has a similar layout to Figure 6 . Each row depicts a different material and each column depicts a different focus condition . As background blur increases to the left of the center column , perceived focus appears to decrease for all three materials . As foreground blur increases to the right , the changes in gradient appearance are misattributed to transformations in material ( the surface appears more matte ) and 3D shape ( the surface appears less curved ) . We measured perceived image focus , perceived surface gloss , and perceived 3D shape in three distinct tasks . Perceived focus and perceived surface gloss were each measured from the same group of observers ( N = 10 ) in two separate paired comparison tasks . Perceived 3D shape was measured from expert observers ( N = 5 ) for six of the stimuli ( outlined in red in Figure 8 ) using a line of twenty ‘gauge figure’ probes across the prominent ridge on the left side of each image ( red dots in central stimulus of Figure 8 ) . These probe settings were integrated to form cross-sectional profiles of perceived relief ( see Koenderink et al . , 1992; Koenderink et al . , 2001 ) . The results accord with our findings in Experiment three and support our informal observations of the stimuli , which are described in separate sections below for each of the three measured properties . Observer’s reports of perceived focus are depicted in Figure 9 and were analyzed with appropriate ANOVA contrasts , as in Experiment 3 . The vertical axis is the percentage of trials in which each image was selected as appearing more focused . The horizontal axis plots the different blur conditions and each colored line represents a different surface reflectance type . The data indicate that background blur ( moving from the central ‘no blur’ condition to the left ) significantly reduced perceived focus for all three materials , t ( 9 ) =−39 . 62 , p<0 . 001 , 95% CI [-71 . 86 , -64 . 10] , Cohen’s d=13 . 21 . Foreground blur ( moving from the ‘no blur’ condition to the right ) also reduced perceived focus , t ( 9 ) =−9 . 75 , p<0 . 001 , 95% CI [-29 . 92 , -18 . 65] , Cohen’s d=3 . 25 , but to a significantly lesser extent than background blur , t ( 9 ) =−34 . 80 , p<0 . 001 , 95% CI [-46 . 53 , -40 . 85] , Cohen’s d=11 . 60 . The data also reveal interactions between blur type and material: in the background blur conditions , the surface with smooth gloss was perceived as significantly more focused than the surface with rough gloss , t ( 9 ) =3 . 58 , p<0 . 005 , 95% CI [3 . 16 , 13 . 98] , Cohen’s d=1 . 19 , which was in turn perceived as more focused than the matte surface , t9=4 . 27 , p=0 . 002 , 95% CI [4 . 03 , 13 . 11] , Cohen’s d=1 . 42 , but there were no significant differences in perceived focus between materials in the foreground blur conditions , t9=-0 . 13 , p=0 . 901 , 95% CI [-6 . 67 , 5 . 95] ( smooth vs . rough gloss ) and t9=-1 . 59 , p=0 . 146 , 95% CI [-19 . 89 , 3 . 47] ( rough gloss vs . matte ) . That is , perceived focus decreased as surfaces exhibited more scattering in their reflectance function , but only in the background blur conditions . The effect of material in the background blur conditions suggests that the sharp ridge gradients generated by the more specular materials may have provided useful cues to the focus of the surface peaks . Most of these cues would have been destroyed by defocus in the foreground blur conditions , which may explain why the effect of material disappeared . Perceived gloss is depicted in Figure 10 and was analyzed with analogous contrasts to perceived focus . The layout of the plot is identical to Figure 9 , but the vertical axis now represents the percentage of trials in which each image was selected as appearing glossier . The data indicate that foreground blur ( right side ) had a large negative effect on perceived gloss relative to the ‘no blur’ condition , t ( 9 ) =−21 . 57 , p<0 . 001 , 95% CI [-42 . 09 , -34 . 10] , Cohen’s d=7 . 19 . This effect was more severe for the materials with less scattering in their reflectance functions , which is likely because they were accurately perceived as being glossier in the no blur condition . The steep slopes for the smooth and rough gloss conditions indicate that optical defocus in the foreground rapidly destroyed the cues to gloss generated by the sharp surface ridge , but these large decrements in perceived gloss were not accompanied by large decrements in perceived focus . Together , these findings imply that observers partially misattributed the reduction in image focus in the foreground blur conditions to a change in surface material . Increasing background blur ( left side ) had a small significant negative effect on perceived gloss for the smooth gloss condition , t9=-4 . 43 , p=0 . 002 , 95% CI [-12 . 95 , -4 . 20] , Cohen’s d=1 . 48 , but no significant effect for the rough gloss ( t9=-0 . 098 , p=0 . 924 , 95% CI [-8 . 57 , 7 . 85]] ) or matte ( t9=2 . 52 , p=0 . 033 , 95% CI [0 . 917 , 16 . 940] ) conditions after the significance criterion was corrected with the Bonferroni method . This is likely due to the loss of gloss cues generated by the sharp specular reflections near the defocused base of the smooth glossy ridge . Cross-sectional depth profiles of perceived 3D shape constructed from the gauge figure settings are depicted in Figure 11 . The profiles represent the average across observers after normalizing the mean height of each observer’s reconstructed profiles to the overall mean height . Differences in shape between the mean profiles were analyzed with an ANOVA in which each of the twenty probe positions was considered an independent sample; significant main effects of this ANOVA represent systematic changes in shape between pairs of focus conditions . The profiles for both the matte ( top ) and smooth gloss ( bottom ) conditions reveal that foreground blur had a significant effect on the perceived 3D shape of the ridge relative to the fully focused conditions , F ( 19 , 38 ) =34 . 48 , p<0 . 001 ( matte ) and F ( 19 , 38 ) =38 . 84 , p<0 . 001 ( smooth gloss ) . Background blur had no significant impact on perceived shape for either the matte surface , F19 , 38=1 . 42 , p=1 . 78 , or the smooth gloss surface , F19 , 38=0 . 92 , p=0 . 569 . The transformations in perceived 3D shape in the foreground blur condition involve a systematic reduction in ridge curvature , height , and position , which suggests that these perceptual distortions were not simply caused by a loss of information . As with perceived gloss , these findings indicate that observers partially misattributed the change in gradient structure induced by foreground blur to a reduction in surface curvature . Taken together , the results of the three tasks in Experiment 4 reveal that misperceptions of optical defocus are closely coupled with misperceptions of surface properties that also contribute to the smoothness of image gradients . The image features that appear to modulate these misperceptions in our stimuli are the sharp level cut contours that bound the surface ridge . When these contours were defocused by background blur , observers accurately reported a decrease in image focus , but when the ridge was defocused by foreground blur , the largest perceptual changes were instead in material ( less gloss ) and shape ( less curvature ) . This suggests that the visual system was directly misattributing the smoothness of the image gradients in the foreground blur conditions to the wrong physical sources . The demonstrations and experiments presented herein were designed to assess how the visual system disentangles the blurred , low-frequency image gradients generated by optical defocus from low-frequency gradients generated by focused shaded surfaces . The experiments were designed to assess the importance of photometric and geometric constraints in resolving this ambiguity . Our results revealed the existence of two types of misperceptions: illusory percepts of optical defocus when none is present ( Experiments 1 and 3 ) , and misperceptions of 3D shape and material when defocus is present but not detected ( Experiment 4 ) . We found that the perception of optical defocus arose in all shaded stimuli that lacked sharp bounding contours consistent with geometrically-correlated contours such as self-occlusions or level ( planar ) cuts of the surface . Taken together , our results indicate that this class of bounding contours plays a critical role in the modulating our experience of optical defocus and 3D shape in otherwise ambiguous images of shaded surfaces . The main theoretical idea that shaped our experiments was that there are specific photogeometric constraints exhibited by the bounding contours of shaded surfaces that play a critical role in identifying low spatial frequency intensity gradients as surface shading , determining the contour’s border ownership , and establishing whether the surface is optically focused . We considered two primary constraints . The first was the covariation of contour orientation and shading intensity , which occurs generically for both smooth self-occluding contours and planar cuts ( Figure 2A ) . The results of Experiment 1 confirm our informal observations that the rotated level cut mask ( Figure 2B ) not only failed to generate a clear perception of gradient focus , they actually interfered with the perception of 3D shape . We attribute this difference to the fact that the contours of level cut masks exhibit a strong orientation-intensity covariation , whereas the contours of the rotated masks do not . We directly assessed the importance of this constraint in the perception of shading in Experiment 2 , where thin shaded ‘ribbons’ tracing the paths of level cut contours were presented and the intensity-orientation covariation was directly manipulated . Our findings showed that this covariation strongly predicts the perception of 3D shape . The second constraint we considered involved the direction of curvature of the shading adjacent to a contour ( i . e . , whether its perceived as convex or concave ) , and its role in the perceived attachment and focus of the shading gradients . The results of Experiment 1 support our informal observations that concavities appear less focused than convexities . This demonstrates that the intensity-orientation covariation cannot fully explain the perception of focus in images of shaded surfaces , as identical covariation can elicit percepts of both vivid focus and moderate defocus depending on whether the shaded surface is perceived as convex or concave ( respectively ) . The difference in perceived focus may be caused by differences in perceived contour attachment in these two configurations . The shading gradients of the convex surface appear clearly attached to the contours of the level cut , but the same gradients do not appear clearly attached to the contour when the surface appears concave; the level cut contour can appear as the edge of a ‘cliff’ , with the shading appearing at a more distant depth . This suggests that the shading gradients must appear clearly attached to the sharp contour to make full use of the information its sharpness provides about optical focus . This result is also consistent with arguments that the visual system has a bias to interpret abrupt discontinuities in luminance as self-occluding contours rather than sudden changes in surface orientation ( Howard , 1983; Liu and Todd , 2004 ) , which may explain why the contours appear more attached to the shading ( and the shading more focused ) when the surface appears convex . Our findings have implications for the existing literature on both the perception of 3D shape from shading and the perception of optical blur . Prior studies of perceived shape have predominantly used globally convex shaded objects as stimuli ( e . g . Fleming et al . , 2004; Mingolla and Todd , 1986; Nefs et al . , 2006 ) . The self-occluding contours that are invariably exhibited by these stimuli may explain why their shading gradients are always perceived as fully focused and vividly three-dimensional . Our data also imply that attempts to ‘eliminate’ these self-occlusions will only impair perceived 3D shape to the extent that the orientation-intensity covariation exhibited in the image is actually reduced . Artificially cropping a self-occluding contour out of the image , for example , may simply create a new bounding contour that still exhibits enough covariation to induce percepts of focused 3D shading . This may explain why manipulations that rotate shaded terrains ( which often exhibit no self-occlusions at all; Reichel and Todd , 1990; Todd and Reichel , 1989 ) have been found to negatively affect perceived shape more than cropping the self-occlusions of convex objects ( Fleming et al . , 2004; Egan and Todd , 2015 ) . Our results suggest that manipulations involving planar cuts may be a more effective method of investigating the role of contours in surface perception in future work . The literature on focus perception has predominantly investigated how the severity of perceived blur varies with the spatial frequency and contrast properties of images ( Mather , 1997; O'Shea et al . , 1997; Tadmor and Tolhurst , 1994 ) . It has been established that relative changes in apparent blur magnitude can provide information about scene depth ( Pentland , 1987 ) and that sharp bounding contours can resolve the depth ordering of ambiguous surfaces ( Marshall et al . , 1996 ) , but to our knowledge , no prior studies have examined how the visual system distinguishes gradients produced by optical blur from gradients produced by environmental sources such as shading . Cases of source misattribution , such as those reported here , are likely difficult to produce with the 2D textures and simple contours employed in past work on focus perception . Our data reveal that identical low-frequency shading gradients can be perceived as vividly focused in some contexts and highly blurred in others , which implies that models of focus perception that rely entirely on local gradient features ( or even the presence of sharp contours ) are not sufficient . Our findings instead suggest that optical focus may be better characterized as a mid-level perceptual category that interacts with the visual system’s estimation of other mid-level properties such as contour attachment , 3D surface orientation and curvature , surface reflectance , and scene illumination . The experiments and demonstrations reported herein have focused on the role of sharp contours that approximate smooth self-occluding rims in providing information about the 3D shape , depth of field , and optical focus of low frequency shading gradients . It seems unlikely , however , that sharp bounding contours are the sole means by which the visual system estimates the optical focus of shaded surfaces . Indeed , we carefully avoided other sources of image contours that could provide information about optical focus , such as the sharp contours generated by either shadows or specular reflections , and only evaluated low-curvature surfaces to avoid generated regions of high spatial frequency shading . Shaded surfaces with regions of high curvature could exhibit enough high spatial frequencies to eliminate percepts of blur , which could explain why some stimuli from prior studies of shading appear focused even in the absence of any contours ( e . g . the ‘crater’ in Figure 7 of Todd et al . , 2014 ) . Specular reflections generated by low-curvature glossy surfaces also have similar spatial frequency properties to high-curvature shading ( Mooney and Anderson , 2014 ) . The efficacy of specular reflections in eliminating perceived blur can be experienced directly in the left panel of Figure 12 . This surface is identical to that depicted in Figure 6 but was rendered in a natural light field with a specular reflectance component in addition to shading . This image appears as a fully focused , glossy , shaded surface . However , the perception of focus in this image requires that the specular reflections appear linked to the same surface geometry as the shading gradients; if the specular reflections are rotated to arbitrary positions , the shaded surface again appears blurred , and the specular reflections appear as overlaid pigment or a second , independent layer ( right panel ) . Thus , specular reflections must also respect photogeometric constraints that link the specular and diffuse components of reflectance in order to provide information about optical focus , as has been shown previously in the perception of gloss ( Anderson and Kim , 2009; Kim et al . , 2011; Marlow et al . , 2012 ) . The underlying issue in both instances is source attribution: when the relevant image features ( image highlights or sharp edges ) are attributed to environmental sources ( attached specular highlights or attached bounding contours ) , they simultaneously modulate the perception of environmental properties ( gloss or 3D shape ) and optical focus . Note that , in general , realistic specular reflections are not optically attached to the surface; when the depth of field is reduced , different regions of specular reflections may only remain sharp when viewed with different focal lengths , and may not be in focus at the same focal length as the attached surface shading . The interactions between specular reflectance , depth of field , and perceived focus are worth consideration in future work . The demonstrations and experiments presented here have focused on a previously unappreciated computational problem: distinguishing low-frequency structure caused by optical blur from the shading gradients of smooth surfaces . We showed that the visual system exploits specific forms of geometric and photometric covariation to generate percepts of optical focus and vivid surface shading , which arise generically along both smooth self-occlusions and planar cuts . Our results indicate that the presence or absence of intensity-orientation correlated contours is a powerful cue to focus: fully focused shaded surfaces can appear blurred in their absence , and actual defocus can be mistaken for transformations in material and shape when correlated contours are nearby . The findings reported herein provide evidence that our perceptions of material , 3D shape , and optical defocus are inherently coupled , which suggests that optical defocus perception does not occur in a completely independent visual pathway to the perception of surface and scene properties . Future work is required to understand the neural processes that exploit these sources of covariation , and the other sources of information that the visual system utilizes to distinguish environmental sources of image structure from optical artifacts induced by the imaging properties of single-chambered eyes . The equations for Lambertian shading reveal the correlations between contour orientation and shading intensity that are likely to arise along a level cut . Equation 1 shows the equation for Lambertian luminance in observer-centric spherical coordinates of surface orientation and illumination direction ( Mamassian , 1993 ) : ( 1 ) L=max⁡ 0 , ricos ( ϕscos⁡ϕi+sin⁡ϕssin⁡ϕicos⁡θs-θi }where L is observed luminance , r is Lambertian surface albedo , and i is the illumination intensity . The sum in square brackets expresses Lambert’s cosine law of shading in terms of surface tilt θs , surface slant ϕs , illumination azimuth θi , and illumination elevation ϕi , all relative to the observer . The subscripts distinguish which pair of spherical coordinates specifies surface orientation ( s ) and which pair specifies the illumination direction ( i ) . The maximum function maps negative values of L to zero , which represents surface regions that receive no illumination . If albedo r , illumination strength i , and illumination direction ( θi , ϕi ) are approximately constant across the image , Equation 1 can be simplified to a function of surface orientation ( θs , ϕs ) only: ( 2 ) L=max⁡{ 0 , Acos⁡ϕs+Bsin⁡ϕscos⁡θs-C }where A and B are constants determined by surface albedo , illumination strength , and illumination elevation , and C is a constant determined by illumination azimuth . Note that this function is a cosine of surface tilt with phase C whose amplitude and vertical offset may vary with surface slant across the image . At a self-occluding rim , surface slant approaches 90° ( which causes the tilt cosine’s vertical offset Acos⁡ϕs to approach zero and its amplitude B sin ( ϕs ) to approach B ) and surface tilt is equal to the rim contour’s orientation . This further constrains the equation for Lambertian luminance to Equation 3: ( 3 ) L=max⁡{ 0 , Bcos⁡θrim-C }where θrim is the orientation of the self-occluding rim and B and C are constants that determine the amplitude and phase of the cosine function . Note that the rim’s 2D orientation is specified in a full 360° range and not a double-angle 180° range: parallel rim segments bound the surface from opposite sides have opposite orientation ( i . e . ± 180° ) . This equation indicates that at a self-occluding rim , shading luminance decreases as a cosine function of the angular separation between contour orientation and the orientation corresponding to maximal luminance ( i . e . where θrim=C=θi , the illumination azimuth ) . The phase of the cosine is determined by the illumination azimuth and its amplitude is determined by surface slant , Lambertian albedo , illumination elevation , and illumination intensity . For non-Lambertian diffuse reflectance functions with roughness parameters ( e . g . Oren and Nayar , 1993 ) , this falloff function will not be an exact cosine , but will at least be monotonic and continuous . The shading exhibited by surfaces bound by level cut contours is related to contour orientation in a similar way to self-occluding rims . Level cut contour orientation is always equal to surface tilt , but only up to a 180° ambiguity: the adjacent surface could be convex ( i . e . the 3D surface normal points out of the shaded region ) or concave ( i . e . the 3D surface normal points into the shaded region ) . Slant is unknown along a level cut in principal , but for smooth surfaces , its rate of change along the level cut contour will be constrained and its contribution to shading ( the Acos⁡ϕs and Bsin⁡ϕs terms in Equation 2 ) is consequently likely to be dominated by the contribution of surface tilt ( the cos⁡θs-C term in Equation 2 ) . The visual system is unlikely to be deterred by any small slant-induced distortions in the shape of the cosine relationship between luminance and contour orientation , which implies that Equation 3 will still approximately hold for level cuts; the overall correlation between image intensity and orientation across large areas of the image is therefore likely to remain high .	We perceive the visual world as made of objects of different shapes , sizes and colors . Some may be smooth , shiny and reflective , whereas others are rough and uneven; some may be in shadow , while others are brightly lit . The brain must identify and distinguish all of these different features to build an accurate , three-dimensional model of the environment . Information about any visual feature originates as light bouncing off an object and entering the eye , which then captures the reflected light and focuses it onto the retina . There , cells generate electrical signals for the brain to process . However , different types of visual features can result in the same pattern of activity . The brain must rely on prior knowledge and educated guesses to disentangle the contributions made by different features , but we know little about the processes that make this possible . Here , Mooney et al . examine how the visual system can tell whether an object is blurry , or if it presents the smooth light-to-dark shading that can accompany curved shapes . The experiments show that images of shaded curved surfaces can appear blurry even when they are fully in focus . However , adding a specific type of sharp edge , called a bounding contour , eliminates this illusion . This suggests that the brain uses these sharp edges to judge whether an image is in focus . In fact , adding bounding contours can trick the visual system into perceiving a blurry image as sharp . Understanding how the human visual system interprets images could lead to advances in computer vision . Artificial vision systems – such as those used in face or license plate recognition – must determine which parts of an image are in focus before attempting to extract visual information . Identifying the cues that enable the human visual system to solve this problem could help to train computers to do the same .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2019	The perception and misperception of optical defocus, shading, and shape
Assembly of mitochondrial iron-sulfur ( Fe/S ) proteins is a key process of cells , and defects cause many rare diseases . In the first phase of this pathway , ten Fe/S cluster ( ISC ) assembly components synthesize and insert [2Fe-2S] clusters . The second phase is dedicated to the assembly of [4Fe-4S] proteins , yet this part is poorly understood . Here , we characterize the BOLA family proteins Bol1 and Bol3 as specific mitochondrial ISC assembly factors that facilitate [4Fe-4S] cluster insertion into a subset of mitochondrial proteins such as lipoate synthase and succinate dehydrogenase . Bol1-Bol3 perform largely overlapping functions , yet cannot replace the ISC protein Nfu1 that also participates in this phase of Fe/S protein biogenesis . Bol1 and Bol3 form dimeric complexes with both monothiol glutaredoxin Grx5 and Nfu1 . Complex formation differentially influences the stability of the Grx5-Bol-shared Fe/S clusters . Our findings provide the biochemical basis for explaining the pathological phenotypes of patients with mutations in BOLA3 . Mitochondria are essential organelles and are involved in numerous biological tasks including fatty acid degradation , citric acid cycle , respiration , ATP production , and the biogenesis of various protein cofactors such as heme , lipoic acid and iron-sulfur ( Fe/S ) clusters . Genetic defects in these pathways are associated with a spectrum of diverse mitochondriopathies with neurological , neurodegenerative , hematological , and metabolic phenotypes . Knowledge of the molecular deficits of the affected mitochondrial components is key for the development of a comprehensive understanding of mitochondrial diseases . A mutation in human BOLA3 encoding a mitochondrial protein is associated with diverse defects summarized as multiple mitochondrial dysfunction syndrome 2 ( MMDS2; Baker et al . , 2014; Cameron et al . , 2011; Haack et al . , 2013 ) . In particular , BOLA3 deficiency results in decreased functions of respiratory complexes I and II as well as lipoic acid-dependent enzymes such as pyruvate dehydrogenase ( PDH ) , 2-ketoglutarate dehydrogenase ( KGDH ) , and glycine cleavage system ( GCS ) ( Mayr et al . , 2014 ) . BOLA3 was suggested to play a role in mitochondrial Fe/S protein biogenesis because of similar clinical and biochemical phenotypes in patients with mutations in the known ISC factors NFU1 ( causing MMDS1; Cameron et al . , 2011; Invernizzi et al . , 2014; Navarro-Sastre et al . , 2011 ) and IBA57 ( causing MMDS3; Ajit Bolar et al . , 2013; Lossos et al . , 2015; for review see Beilschmidt and Puccio , 2014; Stehling et al . , 2014 ) . Another link for BOLA3 to mitochondrial Fe/S protein biogenesis is provided by the fact that related bacterial and plant Bol proteins interact with monothiol glutaredoxins , factors that play a critical role in Fe/S protein biogenesis [Boutigny et al . , 2013; Roret et al . , 2014; Yeung et al . , 2011] ) . However , dedicated biochemical investigations of the molecular role of BOLA3 in mitochondria have not been reported hitherto . Human BOLA3 belongs to a large protein family of bacterial origin ( Aldea et al . , 1988; Willems et al . , 2013 ) . Eukaryotes such as yeast and humans possess three BOLA family members that can be discriminated by conserved sequence elements ( Figure 1—figure supplement 1 ) . To date , the precise function of the BOLA proteins and the functional relationship between the three eukaryotic BOLA family members is unclear . Yeast Bol2 ( formerly termed Fra2; [Kumanovics et al . , 2008] ) is cytosolic , and is involved in cellular iron regulation by forming heterodimeric , [2Fe-2S] cluster-containing complexes with the cytosolic monothiol glutaredoxins Grx3-Grx4 ( Li et al . , 2009; Mühlenhoff et al . , 2010 ) . The human relative BOLA2 also forms hetero-complexes with human GRX3 ( Banci et al . , 2015; Li et al . , 2012 ) , yet the exact physiological role of the Bol2 proteins is still poorly defined . Additionally to BOLA3 , human mitochondria contain the homologous BOLA1 ( Willems et al . , 2013 ) . Ablation of BOLA1 in cultured human cells increases mitochondrial protein thiol oxidation and elicits alterations in mitochondrial morphology . BOLA1 ( but not BOLA3 ) co-purified with tagged mitochondrial monothiol glutaredoxin GLRX5 that is crucial for Fe/S protein biogenesis ( Willems et al . , 2013; Ye et al . , 2010 ) . Both BOLA1 and BOLA3 have structural counterparts in many eukaryotes including yeast Bol1 and Bol3 , respectively ( Figure 1—figure supplement 1 ) . Apart from the BOLA3 patient cell analysis , no solid evidence for a direct role of the BOLA family proteins in cellular Fe/S protein metabolism has been described . We therefore sought to better define the physiological role of the mitochondrial Bol1-Bol3 ( mBols ) and cytosolic Bol2 proteins in this essential biosynthetic pathway by using the yeast S . cerevisiae as a model organism ( Beilschmidt and Puccio , 2014; Lill , 2009; Lill et al . , 2012; Netz et al . , 2014; Rouault , 2012 ) . Mitochondrial Fe/S protein biogenesis in yeast involves 17 known ISC components which were inherited from bacteria and are conserved in eukaryotes ( Johnson et al . , 2005; Lill , 2009 ) . The pathway can be divided into two major phases . First , components of the ‘core ISC machinery’ including the cysteine desulfurase Nfs1 and the scaffold protein Isu1 synthesize a [2Fe-2S] cluster , and transfer it transiently to the monothiol glutaredoxin Grx5 for subsequent assembly of mitochondrial [2Fe-2S] proteins . Second , dedicated ISC factors use the Grx5-bound Fe/S cluster to assemble a [4Fe-4S] cluster and to facilitate its insertion into mitochondrial apoproteins such as aconitase , respiratory complexes , and the radical SAM Fe/S protein lipoic acid synthase ( LIAS ) . Even though the molecular mechanisms of this late phase of Fe/S protein biogenesis are poorly resolved , recent studies have shown that the Isa1 , Isa2 and Iba57 proteins cooperate to generate a [4Fe-4S] cluster ( Brancaccio et al . , 2014; Gelling et al . , 2008; Mühlenhoff et al . , 2011; Sheftel et al . , 2012 ) . Its insertion into specific target apoproteins is subsequently facilitated by Nfu1 and the P-loop ATPase Ind1 , both transiently binding [4Fe-4S] clusters ( Bych et al . , 2008; Sheftel et al . , 2009; Tong et al . , 2003 ) . The latter ISC protein is specific for maturation of respiratory complex I , but nothing is known about the molecular mechanisms underlying [4Fe-4S] cluster insertion by eukaryotic Nfu1 and Ind1 . While most proteins of the core ISC machinery are essential for cell viability , impairment of the ISC factors involved in the second phase of Fe/S protein biogenesis only compromises mitochondrial function of yeast cells . This striking difference for phases I and II is explained by the central , additional role of the core ISC machinery in both cytosolic Fe/S protein biogenesis and cellular iron regulation . Functional defects of core ISC factors lead to impaired assembly of essential Fe/S proteins such as nuclear DNA polymerases and helicases ( Gari et al . , 2012; Netz et al . , 2012; Stehling et al . , 2012 ) and an induction of the yeast iron regulon involving the above mentioned Bol2 ( Lill et al . , 2012; Outten and Albetel , 2013; Paul and Lill , 2015 ) . Here , we employed a combination of cell biological , biochemical and ultrastructural methods to characterize the potential role of the Bol proteins in cellular Fe/S protein biogenesis and to define their position in the complex pathway . We also investigate the involvement of the mBols in cellular iron regulation in comparison to Bol2 . We first analyzed the sub-cellular localization of S . cerevisiae Bol1 ( YAL044W-A ) and Bol3 ( formerly termed Aim1; ( Hess et al . , 2009 ) by cell fractionation . Both proteins were exclusively present in the mitochondrial fraction and absent in the cytosol ( Figure 1—figure supplement 2A ) . Upon sub-fractionation of mitochondria by hypotonic swelling or detergent lysis , the Bol proteins behaved similar to the matrix proteins Tim44 and Mge1 ( Figure 1—figure supplement 2B ) . Thus , Bol1 and Bol3 are constituents of the mitochondrial matrix . To examine the physiological function of mitochondrial Bol1-Bol3 ( mBols ) , we created single and double deletion cells , as well as combined deletions with BOL2 ( Supplementary file 1A; strain background BY4742 ) . All deletion strains grew at wild-type rates on minimal media containing the non-fermentable carbon sources glycerol or acetate with the exception of bol13Δ and bol123Δ strains ( Figure 1—figure supplement 3 ) . Apparently , simultaneous deletion of BOL1 and BOL3 created a weak respiratory defect which is slightly weaker than that seen for NFU1 deletion ( Schilke et al . , 1999 ) . We then analyzed key mitochondrial enzyme activities to detect potential defects in Fe/S protein biogenesis , and we compared these results to data for a NFU1 deletion strain which shows mild , yet specific Fe/S enzyme defects ( Navarro-Sastre et al . , 2011; Schilke et al . , 1999 ) . This analysis included the mitochondrial Fe/S proteins aconitase , succinate dehydrogenase ( SDH or complex II ) , and yeast LIAS ( Lip5 ) . The latter can be assayed indirectly either by the enzyme activities or the lipoylation extent of pyruvate dehydrogenase ( PDH ) and 2-ketoglutarate dehydrogenase ( KGDH ) ( Gelling et al . , 2008; Schonauer et al . , 2009 ) . Further , we measured the activity of cytochrome c oxidase ( COX ) . Even though this enzyme does not contain Fe/S clusters , it is frequently affected by mitochondrial Fe/S protein biogenesis defects , for unknown reasons ( see , e . g . , [Mühlenhoff et al . , 2011; Sheftel et al . , 2012] ) . All enzyme activities were normalized to malate dehydrogenase ( MDH ) that showed no significant changes in all these strains relative to wild type ( not shown ) . Single deletions of BOL1 or BOL3 did not affect any of these enzyme activities relative to wild-type cells with the exception of a slight , but significant decrease of SDH in bol3Δ cells ( Figure 1A–E , bars 1–3 ) . In contrast , double BOL1-BOL3 deletions resulted in substantial , up to threefold decreases of the Fe/S cluster-dependent enzyme activities , an effect comparable to that seen for nfu1Δ cells ( bars 4 and 6 ) ( Navarro-Sastre et al . , 2011 ) . A notable exception was aconitase which retained normal activity in bol13Δ cells ( Figure 1A ) . Thus , double but not single yeast BOL gene deletions created a similar phenotype as that observed in BOLA3 patient cells ( Cameron et al . , 2011 ) . Additional ablation of BOL2 ( strain bol123Δ , bar 5 ) did not substantially exacerbate these effects suggesting that the mitochondrial and cytosolic Bol proteins act independently . The decrease in SDH was likely due to a direct defect in Fe/S cluster assembly , because SDH protein levels remained unchanged in immunoblots ( Figure 1F ) . In keeping with the decreased PDH and KGDH activities , we also found a lipoylation defect of the E2 subunits of these enzymes by immunostaining against lipoic acid ( Figure 1F ) ( Gelling et al . , 2008; Schonauer et al . , 2009 ) . All the observed effects were rather mild in comparison to defects reported for deletion mutants of other late-acting ISC factors such as Isa1 , Isa2 , and Iba57 ( Gelling et al . , 2008; Mühlenhoff et al . , 2011 ) . This notion was also evident from the lack of any effects on COX activity in all tested BOL deletion strains ( Figure 1E ) . A severe COX diminution is observed in ISA1 , ISA2 , or IBA57 mutants , yet the exact reason is unknown ( Gelling et al . , 2008; Mühlenhoff et al . , 2011 ) . In order to verify that the loss of the enzyme activities was specific , BOL1 and BOL3 genes were reintroduced into the bol123Δ deletion strain by expression from plasmids . Both Bol1 and Bol3 were able to efficiently rescue the deficiencies in SDH activities ( Figure 1—figure supplement 4 ) . 10 . 7554/eLife . 16673 . 003Figure 1 . Deficiency of both mitochondrial Bol proteins causes defects in a subset of mitochondrial [4Fe-4S] enzymes . Wild-type ( WT; strain BY4742 ) , and the indicated BOL and NFU1 deletion yeast strains were grown in minimal medium containing 2% galactose supplemented with 50 µM ferric ammonium citrate and used for the preparation of mitochondria . Mitochondrial extracts were assayed for specific activities of ( A ) aconitase ( ACO ) , ( B ) succinate dehydrogenase ( SDH ) , ( C ) pyruvate dehydrogenase ( PDH ) , ( D ) 2-ketoglutarate dehydrogenase ( KGDH ) , and ( E ) cytochrome c oxidase ( COX ) . Values were normalized to those of malate dehydrogenase ( MDH ) . Error bars indicate the SEM ( n≥4 ) . ( F ) Mitochondrial extracts were subjected to TCA precipitation and the levels of the indicated proteins and of lipoic acid attached to E2 subunits of PDH and KGDH were determined by immunostaining . Staining for porin served as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 00310 . 7554/eLife . 16673 . 004Figure 1—figure supplement 1 . Cartoon of bacterial BolA and eukaryotic Bol1 , Bol2 and Bol3 proteins . Conserved amino acid sequence elements that characterize the various members of the BOLA protein family are shown . A conserved His ( H* ) in Bol2 serves as a Fe/S cluster ligand in a hetero-dimeric complex with Grx3 ( Li et al . , 2011 ) . MTS; mitochondrial targeting sequence . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 00410 . 7554/eLife . 16673 . 005Figure 1—figure supplement 2 . Bol1 and Bol3 are proteins of the mitochondrial matrix . ( A ) Wild-type yeast cells ( W303 ) expressing Myc-tagged Bol1 ( vector p424-MET25; Supplementary file 1B ) or HA-tagged Bol3 ( from vector p425-TDH3 ) were grown in SD medium , and mitochondria ( Mito ) and post-mitochondrial supernatant ( PMS ) fractions were isolated . Immunostaining was performed using antisera raised against the mitochondrial matrix protein Mge1 , cytosolic Grx4 ( also recognizing Grx3 ) , and monoclonal anti-Myc ( A14 , Santa Cruz ) or anti-HA . ( B ) Wild-type yeast cells ( W303 ) expressing Bol1-Myc or Bol3-HA were grown in SD medium . Isolated mitochondria were either left intact ( sorbitol ) , hypotonically swollen by tenfold dilution in HEPES buffer , or lysed with 0 . 2% Triton X-100 detergent ( Diekert et al . , 2001 ) . Samples were incubated in the presence or absence of 100 µg/ml proteinase K for 20 min on ice , and proteinase K was inactivated by PMSF . After TCA precipitation , samples were analyzed by immunostaining for the Myc-tag of Bol1 , the HA-tag of Bol3 , the intermembrane space proteins Erv1 and cytochrome b2 , and the matrix proteins Tim44 and Mge1 . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 00510 . 7554/eLife . 16673 . 006Figure 1—figure supplement 3 . The growth behavior of the various BOL gene deletion strains in comparison to NFU1 deletion cells . The indicated yeast strains were grown overnight on minimal medium with 2% glucose as carbon source . Cells were diluted to OD600 = 0 . 5 in water and spotted on rich medium ( YP ) agar plates with 2% galactose , 3% glycerol or 2% acetate as indicated . Plates were incubated at 30°C for 3 days . WT , wild-type . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 00610 . 7554/eLife . 16673 . 007Figure 1—figure supplement 4 . Specific rescue of succinate dehydrogenase activities in BOL gene deletion cells by both mitochondrial Bol proteins . Wild-type ( WT; strain BY4742 ) and the indicated BOL deletion strains were transformed with vector p416-MET25 lacking ( empty ) or overexpressing BOL1 or BOL3 as indicated . Cells were grown in minimal medium with 2% glucose without uracil and used for preparation of mitochondria . Mitochondrial extracts were assayed for the specific activity of succinate dehydrogenase ( SDH ) and normalized to malate dehydrogenase ( MDH ) activities . Error bars indicate the SEM ( n≥4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 00710 . 7554/eLife . 16673 . 008Figure 1—figure supplement 5 . Fe/S enzyme activity defects in cells lacking Bol1-Bol3 after growth in lactate medium . Wild-type ( WT; strain BY4742 ) , the indicated BOL and NFU1 deletion yeast strains were grown in lactate medium and used for the preparation of mitochondria . Mitochondrial extracts were assayed for specific activities of ( A ) aconitase ( ACO ) , ( B ) succinate dehydrogenase ( SDH ) , ( C ) pyruvate dehydrogenase ( PDH ) , and ( D ) 2-ketoglutarate dehydrogenase ( KGDH ) . Values were normalized to those of malate dehydrogenase ( MDH ) . Error bars indicate the SEM ( n≥4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 008 Previous studies on yeast ISC protein depletions , e . g . , of the Isa proteins , had shown that the Fe/S protein defects were much more pronounced under strict respiratory conditions , e . g . , upon growth in lactate medium ( Kispal et al . , 1999; Mühlenhoff et al . , 2011 ) ( see also Figure 1—figure supplement 3 ) . Both aconitase and SDH activities of the various BOL deletion strains behaved similarly after growth in lactate- compared to galactose-containing media ( Figure 1—figure supplement 5 ) . In contrast , PDH and KGDH activities were much more severely ( ten- vs . threefold ) diminished in bol13Δ , bol123Δ , and nfu1Δ cells grown in lactate medium , similar to previous reports on Isa protein depletion ( Kaut et al . , 2000; Mühlenhoff et al . , 2011 ) . Together , these data show that the mBols were crucial for efficient maturation of LIAS , yet they play only an auxiliary role in SDH maturation . The levels of Bol1 and Nfu1 did not change or rather slightly decreased in the various deletion cells making a compensatory effect unlikely ( data not shown; Bol3 levels could not be tested due to the lack of antibody detection ) . Obviously , Bol1 and Bol3 perform largely complementary roles , because only simultaneous deletion of both BOL1-BOL3 genes created a significant LIAS defect . To directly characterize the mBol involvement in mitochondrial Fe/S cluster assembly , we employed a well-established 55Fe radiolabeling and immunoprecipitation assay . This also allowed the testing of additional Fe/S proteins including such with [2Fe-2S] clusters . The various BOL deletion strains were radiolabeled with 55Fe , the Fe/S proteins were immunoprecipitated , and the precipitated radioactivity was quantified by scintillation counting ( Pierik et al . , 2009 ) . Consistent with the enzyme activity data ( Figure 1 ) , we found no significant effect on 55Fe incorporation into aconitase upon single , double or triple BOL gene deletion ( Figure 2A ) . For SDH ( assayed by precipitation of the myc-tagged Fe/S cluster subunit Sdh2 co-expressed with Sdh1 ) a 40% decrease in 55Fe binding was observed upon deletion of BOL3 , also in combination with BOL1 and BOL2 deletions ( Figure 2B ) . Compared to other ISC gene deletions ( see , e . g . , Gelling et al . , 2008; Gerber et al . , 2004; Mühlenhoff et al . , 2011; Rodriguez-Manzaneque et al . , 2002; Voisine et al . , 2001 ) this effect was rather weak . Similar , yet slightly stronger decreases in 55Fe binding ( 60% ) were detected for mitochondria-targeted HiPIP , a bacterial [4Fe-4S] cluster-containing ferredoxin ( Figure 2C ) . For all the tested Fe/S proteins , single deletion of BOL1 had no effects on 55Fe binding . The observed defects were specific , because reintroduction of Bol1 or Bol3 into the bol123Δ strain via overproducing vectors completely restored wild-type 55Fe binding activity , whereas cytosolic Bol2 or Grx5 , as a control , were ineffective ( Figure 2D ) . Taken together , these data support a specific function of the mBols in the maturation of a subset of mitochondrial Fe/S proteins , whereby Bol1 and Bol3 perform largely overlapping , non-essential functions . Interestingly , the 55Fe radiolabeling assay revealed a slightly more important role for Bol3 as compared to Bol1 . This tendency , yet less well pronounced , was also observed for the enzyme activities ( see Figure 1 ) . The ability of the mBols to mutually substitute each other explains why Fe/S protein defects are best visible when both BOL1 and BOL3 are deleted . 10 . 7554/eLife . 16673 . 009Figure 2 . Bol1-Bol3 are required for de novo Fe/S cluster incorporation into specific mitochondrial [4Fe-4S] but not [2Fe-2S] proteins . ( A–F ) Wild-type ( WT , strain BY4742 ) and the indicated BOL deletion strains were transformed with vectors overproducing ( B ) Sdh2-Myc and Sdh1 , ( C and D ) HiPIP-Myc , ( E ) human FDX2-HA , or ( F ) Schizosaccharomyces pombe ( Sp ) Grx5-Myc . In part D , bol123Δ cells were additionally transformed with p414-MET25 lacking or containing BOL1 , BOL2 , or BOL3 genes or p424-TDH3-GRX5 . Cells were grown overnight in iron-poor SD medium and radiolabeled with 10 µCi 55Fe for 2 hr . The overproduced proteins were immunoprecipitated from cell extracts with specific antibodies . The amount of co-precipitated 55Fe was quantified by scintillation counting . Error bars indicate the SEM ( n≥4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 009 All Fe/S proteins tested so far contain [4Fe-4S] clusters . To get an insight into the Fe/S cluster-type specificity of mBol function , we examined the 55Fe incorporation into two mitochondrial [2Fe-2S] proteins , namely ferredoxin FDX2-HA , the human ortholog of yeast Yah1 , and Grx5 from Schizosaccharomyces pombe ( SpGrx5-Myc ) ( Gelling et al . , 2008; Uzarska et al . , 2013 ) . The deficiency in mBols did not elicit any significant decrease in 55Fe binding to these [2Fe-2S] proteins ( Figure 2E , F ) . This behavior is strikingly different from that of members of the core ISC machinery , and shows that the mBols were not required for maturation of [2Fe-2S] proteins . We conclude that Bol1-Bol3 are dedicated ISC assembly factors specifically involved in the maturation of a subset of mitochondrial [4Fe-4S] but not of [2Fe-2S] proteins . Further , the mBols act after Grx5 , the final core ISC component ( Uzarska et al . , 2013 ) , because they were not needed for 55Fe incorporation into Grx5 . We next examined if the mitochondrial Bol proteins perform an additional role in cytosolic Fe/S protein biogenesis , if they functionally cooperate with cytosolic Bol2 , or if Bol2 alone plays a crucial role in this process . Previous studies on yeast Bol2 and human BOLA2 have suggested an involvement of these cytosolic proteins in intracellular iron metabolism ( Kumanovics et al . , 2008; Li et al . , 2012 ) . First , we measured the consequences of various BOL gene deletions on the activation of the Aft1-Aft2-dependent iron regulon , a hallmark of all cells with defects in the core ISC machinery ( Lill et al . , 2012; Outten and Albetel , 2013 ) . As expected ( Kumanovics et al . , 2008 ) , bol2Δ cells displayed a substantial activation of the Aft1-dependent FIT3 and FET3 genes in a GFP-based promoter assay ( Figure 3A–B ) . In contrast , yeast strains lacking BOL1 and/or BOL3 behaved like wild-type cells indicating that the mBols are not involved in cellular iron regulation . This fits nicely to our conclusion above that Bol1-Bol3 do not belong to the core ISC system . The same situation was observed for cells lacking Nfu1 ( Figure 3B ) , another known late-acting specific ISC assembly factor ( Navarro-Sastre et al . , 2011 ) . Surprisingly , when the BOL2 deletion was combined with genetic ablation of BOL1 and/or BOL3 , the iron regulon was activated slightly stronger than in bol2Δ cells ( Figure 3A–B ) . This amplifying effect was specific for a Bol protein deficiency because it was reversed to the FET3 induction level of bol2Δ cells by reintroducing BOL1 and/or BOL3 into bol123Δ cells , and to wild-type FET3 levels by BOL2 expression ( Figure 3B ) . 10 . 7554/eLife . 16673 . 010Figure 3 . Mitochondrial Bol1-Bol3 and cytosolic Bol2 are not involved in cytosolic Fe/S protein biogenesis . Wild-type ( WT , strain BY4742 ) and the indicated BOL deletion strains harboring vector ( A ) pFIT3-Luc2 or ( B ) pFET3-GFP were cultivated in iron-replete medium to mid-log phase and the activities of the FIT3 and FET3 promoters , respectively , were determined . In ( B ) the bol123Δ cells were also transformed with centromeric plasmids producing the indicated Bol proteins . Abbreviation: Bol13; Bol1 plus Bol3 . ( C ) Leu1 ( relative to MDH ) activities were determined in the indicated deletion stains cultivated in iron-replete medium . ( D ) The indicated BOL deletion stains were radiolabeled with 55Fe , and the Leu1-bound radioactivity was determined by immunoprecipitation with α-Leu1 antibodies followed by scintillation counting . ( E ) 55Fe incorporation into Rli1-HA was determined accordingly using WT and bol123Δ cells transformed with centromeric plasmids producing the three Bol proteins as indicated or with the empty vector ( - ) . ( F ) Representative immunoblots determining the protein levels of Leu1 ( D ) and Rli1-HA ( E ) in the indicated strains . Porin served as a loading control . Error bars indicate the SEM ( n≥4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 010 Second , we investigated the effects of the three BOL gene deletions on cytosolic Fe/S proteins by assaying the enzyme activity of isopropylmalate isomerase ( Leu1 ) . No changes were observed when BOL1-BOL3 genes were deleted either alone or in combination ( Figure 3C ) . In contrast , bol2Δ cells showed a twofold decrease in Leu1 activity , which further declined upon additional deletion of BOL1-BOL3 . Collectively , these results show that the mBols are not involved in cytosolic Fe/S protein biogenesis , yet there is a cross-talk of the mBol deficiency with the BOL2 deletion . To better understand the reason of this combinatory effect , we took advantage of the 55Fe radiolabeling assay described above . Deletion of either BOL1-BOL3 or BOL2 hardly affected 55Fe/S cluster binding to Leu1 indicating that neither mitochondrial Bol1-Bol3 nor cytosolic Bol2 perform a decisive role in Leu1 Fe/S cluster incorporation ( Figure 3D , F ) . However , combining the BOL2 deletion with that of BOL1 and/or BOL3 resulted in an up to twofold decrease in 55Fe/S cluster association with Leu1 . A similar twofold decrease ( compared to wild-type ) was seen for 55Fe incorporation into another cytosolic Fe/S protein , the ABC protein Rli1 using the triple BOL deletion strain bol123Δ ( Figure 3E–F ) . We note , however , that these decreases in 55Fe incorporation were rather weak compared to defects in core ISC or CIA components ( see , e . g . , Netz et al . , 2007 , 2010; Paul et al . , 2015 ) . The Rli1 maturation defect was completely restored when any of the BOL genes was reintroduced into the bol123Δ mutant . This supports the notion that neither the mBols nor Bol2 are directly involved in cytosolic Fe/S protein maturation . Obviously , their combined absence causes alterations in both mitochondrial metabolism and cellular iron homeostasis , thereby creating conditions that negatively affect cytosolic Fe/S cluster insertion . The physiological conditions prevailing in bol123Δ cells might also explain the slight decrease in the enzyme activity of aconitase which , like Leu1 , is a highly susceptible protein to , e . g . , oxidative stress conditions or changes in iron availability ( cf . Figures 1A and 3C ) . Together , our data indicate that the mBols execute a mitochondria-specific function in [4Fe-4S] protein assembly , whereas Bol2 plays a specific role in cellular iron regulation ( Kumanovics et al . , 2008 ) . Our current and previous results suggested functions of the mBols and of Nfu1 as specialized ISC assembly factors ( Cameron et al . , 2011; Navarro-Sastre et al . , 2011 ) . Are these functions independent or overlapping ? We first analyzed whether simultaneous inactivation of Bol1-Bol3 and Nfu1 aggravates the Fe/S defects . As observed previously ( Navarro-Sastre et al . , 2011; Schilke et al . , 1999 ) , nfu1Δ cells showed a diminution in the activities of both aconitase and SDH relative to wild type ( Figure 4A–B ) . A substantial further decrease to rather low aconitase and SDH activities was found for additional deletion of BOL3 , but not of BOL1 . The BOL3 effect was not further amplified in nfu1Δbol13Δ cells . These findings show that Bol3 plays a distinct , Nfu1-independent role that cannot be executed by Bol1 . Apparently , even though Bol1 and Bol3 play overlapping roles ( see above ) , their functions are not entirely identical . 10 . 7554/eLife . 16673 . 011Figure 4 . Bol1-Bol3 and Nfu1 cannot functionally replace each other in Fe/S protein maturation . ( A–B ) Wild-type ( WT , strain BY4742 ) and the indicated deletion strains were grown in minimal medium containing 2% glucose and used for the preparation of mitochondria . Mitochondrial extracts were assayed for the indicated specific enzyme activities as described in Figure 1 . ( C–F ) The indicated deletion strains were transformed with vector p416-MET25 lacking ( empty ) or containing BOL3 or NFU1 as indicated . Cells were grown in minimal medium containing 2% galactose and used for the preparation of mitochondria . Mitochondrial extracts were assayed for the indicated enzyme activities as described in Figure 1 . Error bars indicate the SEM ( n≥4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 01110 . 7554/eLife . 16673 . 012Figure 4—figure supplement 1 . Growth complementation test of nfu1Δ and bol13Δ cells . ( A ) nfu1Δ and ( B ) bol13Δ cells were transformed with vector p414-MET25 containing no gene ( empty ) , BOL1 , BOL3 or BOL1-BOL3 ( BOL13 ) or p416-MET25-NFU1 as indicated . Cells were grown overnight on minimal medium without uracil containing 2% glucose as a carbon source . Cells were diluted to OD600 of 0 . 1 in 500 µl of liquid minimal medium without uracil containing 2% acetate as sole carbon source , and put into a 48 well plate . Plates were incubated at 30°C for 3 days with shaking , and the OD600 was measured by a TECAN plate reader every 30 min . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 012 We next tested whether the Bol1-Bol3 and Nfu1 proteins can mutually substitute each other’s functions . To this end , we first took advantage of the growth defects of the bol13Δ and nfu1Δ strains in acetate-containing minimal medium ( see above ) . Upon complementation of bol13Δ cells with BOL1 and/or BOL3 or of nfu1Δ cells with NFU1 , wild-type growth rates were restored ( Figure 4—figure supplement 1 ) . In contrast , no significant growth defect complementation was observed for nfu1Δ cells upon BOL1 and/or BOL3 expression ( part A ) . A weak complementation was seen for bol13Δ cells by NFU1 expression ( part B ) . Notably , this slight rescue effect was not observed for growth on solid acetate medium ( not shown ) . Further , we used these cells for measuring the enzyme activities of aconitase , SDH , PDH and KGDH . In agreement with the growth behavior , Bol3 was not able to rescue any of the enzyme defects of nfu1Δ cells , while Nfu1 did so ( Figure 4C–F , left ) . Likewise , Nfu1 failed to restore the enzyme activities of bol13Δ cells , while Bol3 was complementing ( Figure 4C–F , right ) . These findings show that Nfu1 and Bol3 cannot mutually substitute each other’s biochemical function , even upon overproduction . Despite the fact that Bol3 and Nfu1 both act late in mitochondrial Fe/S protein biogenesis , they appear to fulfil different tasks that cannot be taken over by the other protein . Hence , these data suggest that Bol1-Bol3 and Nfu1 perform individual , non-overlapping functions during Fe/S cluster assembly . To extend our cell biological insights into Bol protein function , we employed NMR structural studies to examine their physical interaction with Grx5 . We used the human proteins for these in vitro studies because they were more stable than the yeast counterparts . First , the solution structures of human BOLA1 and BOLA3 were determined ( Figure 5 ) . A detailed description of the two structures and of their backbone dynamic properties is provided in Figure 5—figure supplement 1 . According to the DALI server ( Holm and Sander , 1993 ) , both BOLA1 and BOLA3 show the highest structural similarity to BolA-like proteins from Arabidopsis thaliana , Babesia bovis , and Mus musculus ( Abendroth et al . , 2011; Kasai et al . , 2004; Roret et al . , 2014 ) . In particular , the BOLA1 structure matches better with those homologues that have a longer loop between β1 and β2 , while the BOLA3 structure is more similar to homologues having a shorter β1-β2 loop ( Figure 5C and Figure 5—figure supplement 2 ) . Apparently , the structural features of the loop between β1 and β2 are key for differentiating BOLA1 from BOLA3 . Remarkably , this loop is closer to the ‘invariant’ His residue ( His102 in BOLA1 and His96 in BOLA3; Figure 5 ) , which has been shown to coordinate the [2Fe-2S] cluster in the yeast Bol2-Grx3 complex ( Li et al . , 2011 ) . This region contains further possible Fe/S cluster ligands , namely His58 , His67 in BOLA1 and Cys59 in BOLA3 . Another conserved His residue ( His86 in BOLA1 and His81 in BOLA3 ) is distant from this region and thus excluded as a cluster ligand ( Figure 5 ) . Together , structural data indicate that BOLA1 and BOLA3 have a similar fold , but show local structural differences in the regions containing the potential Fe/S cluster ligands . 10 . 7554/eLife . 16673 . 013Figure 5 . NMR solution structures of human BOLA1 and BOLA3 . The structures of ( A ) BOLA1 and ( B ) BOLA3 were solved by solution NMR . Residues His58 , His67 and His102 are conserved within the eukaryotic BOLA1 proteins , and residues Cys59 and His96 are conserved within the BOLA3 proteins . ( C ) Backbone superimposition of BOLA1 ( red ) and BOLA3 ( blue ) structures depicting conserved His and Cys residues in both proteins . Yellow sticks represent potential Fe/S cluster ligands , and green sticks indicate other conserved His residues . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 01310 . 7554/eLife . 16673 . 014Figure 5—figure supplement 1 . Structural and dynamic properties of BOLA1 and BOLA3 by solution NMR . ( A ) 2D 1H-15N HSQC spectra of 15N-labeled BOLA1 and BOLA3 . ( B ) 15N R1 , R2 relaxation rates and 15N ( 1H ) NOE values per residue of BOLA1 and BOLA3 obtained at 500 MHz and 298 K . The proteins were in buffer N with 10% ( v/v ) D2O ) . The 1H-15N HSQC spectra of the apo forms of BOLA3 and BOLA1 showed well-dispersed resonances indicative of essentially folded proteins . The overall topology of the solution structure of human BOLA3 is α1β1β2η1α2η2β3 ( h: 310-helix ) , in which β1 and β2 are antiparallel , and β3 is parallel to β2 ( Figure 5 ) . All α-helices are on the same side of the β-sheet and the 310-helix located between α2 and β3 contains the His96 residue that is invariant in all prokaryotic and eukaryotic BolA homologues ( Li and Outten , 2012 ) . A cysteine residue ( Cys59 ) , typically conserved in Bol3 homologues , is located close to the invariant His96 residue in an extended loop between β1 and β2 . The two residues are not in the proper orientation to coordinate a [2Fe-2S] cluster ( Sγ of Cys59 is at an average distance of ~8 Å from Nε2 of His96 ) . The overall topology of the solution structure of BOLA1 is similar to that of BOLA3 , with the exception of the presence , in BOLA1 , of short helices in the loop regions and at the C-terminus ( α1η1β1β2η2α2α3β3α4 topology ) ( Figure 5 ) . In addition , at both termini BOLA1 has ca . 15 residues that are unstructured and whose backbone NH resonances are clustered in the central region of the 1H-15N HSQC spectrum ( part B ) . The three His residues typically conserved in Bol1 homologues are located in two spatially close regions ( Figure 5 ) . Although they are close to each other , it is not possible to predict which ones may be involved in [2Fe-2S] cluster coordination . Information on backbone motions and on the protein oligomerization state of BOLA1 and BOLA3 were obtained through 15N R1 , R2 , heteronuclear 15N ( 1H ) -NOEs NMR experiments . The 15N backbone relaxation properties showed that the N and C termini of BOLA1 are highly flexible , as about 15 to 20 residues on each terminus have negative 15N NOE values , while the folded domain is essentially rigid . BOLA3 is a rigid molecule , with the exception of the first detected residue at the N terminus , the last two residues at the C terminus and residues around invariant His96 . All of them experience fast backbone motions in the ns-ps time scale , as indicated by their low or negative 15N NOE values . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 01410 . 7554/eLife . 16673 . 015Figure 5—figure supplement 2 . Multi-sequence alignment of the mitochondrial Bol proteins . Multalin ( Corpet , 1988 ) was used to generate a multi-sequence alignment of Bol1- and Bol3-like proteins from fungi and various higher eukaryotes . The conserved His and Cys residues are indicated . The numbers refer to the residues of the human mBols . Two His ( 58 and 67 ) in BOLA1 are candidates for the structural counterpart of Cys59 of BOLA3 . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 015 In order to characterize the interaction between BOLA1 or BOLA3 and GLRX5 at the atomic level , we first studied the apo situation . 15N labeled BOLA proteins were titrated with unlabeled apo-GLRX5 , and , vice versa , 15N labeled apo-GLRX5 with unlabeled BOLA proteins in the presence of 5 mM GSH . Meaningful chemical shift changes were observed for both titrations in 2D 1H-15N HSQC NMR spectra acquired upon addition of the unlabeled partner ( Figure 6—figure supplements 1A and 2A ) . The free and bound proteins are in fast and intermediate exchange regimes relative to NMR time scale; the observed chemical shift changes of the protein amide resonances saturated at a 1:1 protein ratio . 15N R1 and R2 NMR relaxation data performed on the final 15N labeled BOLA1-unlabelled GLRX5 or 15N labeled BOLA3-unlabelled GLRX5 mixtures showed rotational correlation times consistent with the formation of a 1:1 heterodimeric complex ( Figure 6—figure supplement 3 ) . NMR titrations performed by adding unlabeled apo-GLRX5 to a 15N-labeled BOLA1/15N-labeled BOLA3 1:1 mixture showed the formation of both hetero-complexes at comparable levels , indicating similar affinities . This data fits nicely to thermophoresis results showing that both human BOLA proteins interacted with apo-GLRX5 with Kd values of 3 µM ( Melber et al . , 2016; accompanying manuscript ) . The chemical shift changes observed upon formation of apo-GLRX5-BOLA1 and apo-GLRX5-BOLA3 complexes were mapped on the solution structures of BOLA1 , BOLA3 and apo-GLRX5 . The interaction surface on GLRX5 in the two complexes comprises the GSH binding site and its surroundings ( Figure 6A , B and Figure 6—figure supplement 2A ) . On the BOLAs , the interactions involve helix α2 , part of the β-sheet and the invariant His residue ( see above ) . The other possible Fe/S cluster ligands ( His58 , His67 in BOLA1 and Cys59 in BOLA3 ) showed no significant chemical shifts . In conclusion , NMR data indicate that apo-GLRX5-BOLA1 and apo-GLRX5-BOLA3 complexes specifically interact involving the region surrounding the invariant His in BOLAs and the GSH binding site in GLRX5 . 10 . 7554/eLife . 16673 . 016Figure 6 . Structural basis of the interaction between human apo- and holo-GLRX5 with the BOLA proteins . ( A and B ) Backbone chemical shift differences , obtained from a comparison of the 1H-15N HSQC spectra of apo-GLRX5 or BOLA proteins with that of ( A ) BOLA1-apo-GLRX5 or ( B ) BOLA3-apo-GLRX5 ( 1:1 mixture in buffer N ) , were mapped on the solution structures of the proteins . ( C and D ) Backbone chemical shift differences , obtained from a comparison of the 1H-15N HSQC spectrum of apo-GLXR5 or BOLA proteins with that of ( C ) BOLA1-apo-GLRX5 or ( D ) BOLA3-apo-GLRX5 ( 1:1 mixture in buffer N ) chemically reconstituted with a [2Fe-2S] cluster , were mapped on the solution structures of the proteins . Green regions show residues with significant chemical shift changes ( that is both in terms of chemical shift and broadening beyond detection effects ) observed upon formation of the apo- or holo-complexes . Red areas depict those residues additionally affected upon holo-complex formation . Critical residues and GLRX5-bound GSH are depicted as sticks . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 01610 . 7554/eLife . 16673 . 017Figure 6—figure supplement 1 . Chemical shift changes upon complex formation between GLRX5 and BOLA proteins monitoring backbone chemical shift changes of BOLA1 and BOLA3 . Backbone weighted average chemical shift differences Δavg ( HN ) obtained by comparing 1H-15N HSQC map of ( A ) 15N-labeled BOLA1 or BOLA3 with an equimolar mixture of 15N-labeled BOLA1 or BOLA3 with unlabeled apo-GLRX5 , and of ( B ) 15N-labeled BOLA1 or BOLA3 with an equimolar mixture of 15N-labeled BOLA1 or BOLA3 chemically reconstituted with unlabeled GLRX5 to generate a [2Fe-2S] cluster ( buffer N ) . Bars with Δavg ( HN ) = 1 indicate residues whose backbone NH signals broaden beyond detection . The indicated thresholds ( obtained by averaging Δavg ( HN ) values plus 1σ ) were used to define significant chemical shift differences . Green bars show significant chemical shift changes ( that is both chemical shift differences and broadening beyond detection effects , respectively ) in the apo interactions , and red bars those additionally occurring upon holo complex formation . The position of conserved residues is highlighted . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 01710 . 7554/eLife . 16673 . 018Figure 6—figure supplement 2 . Chemical shift changes upon complex formation between GLRX5 and BOLA proteins , monitoring backbone chemical shift changes of GLRX5 . Backbone weighted average chemical shift differences Δavg ( HN ) obtained by comparing 1H-15N HSQC map of ( A ) 15N-labeled apo-GLRX5 with that of a 1:1 mixture of 15N-labeled apo-GLRX5 and unlabeled BOLA1 or BOLA3 , and of ( B ) 15N-labeled apo-GLRX5 with that of a 1:1 mixture of 15N-labeled apo-GLRX5 with unlabeled BOLA1 or BOLA3 after chemical reconstitution of a [2Fe-2S] cluster in buffer N . The data are presented as in Figure 6—figure supplement 1 . The position of the conserved Cys67 and of the residues of the GSH binding site are highlighted . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 01810 . 7554/eLife . 16673 . 019Figure 6—figure supplement 3 . Experimental and predicted rotational correlation times ( τC ) . For the isolated proteins the τC values were obtained by the HYDRONMR program . For the apo-complexes the τC values were calculated from the sum of the τC values of the two isolated proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 01910 . 7554/eLife . 16673 . 020Figure 6—figure supplement 4 . Comparison of the apo-AtGrxS14-AtBolA2 complex with apo-GLRX5/BOLA3 . Significant chemical shift changes ( red and green for A . thaliana and human proteins , respectively ) are mapped on the structures of ( A ) apo-AtGrxS14 and apo-GLRX5 upon the interaction with AtBolA2 and human BOLA3 , respectively . Vice versa , part ( B ) shows the significant chemical shift changes of AtBolA2 and human BOLA3 upon interaction with AtGrxS14 and GLRX5 , respectively . Conserved residues and the GLRX5-bound GSH are highlighted as sticks . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 020 Recently , a heterodimeric complex formed between A . thaliana ( At ) apo-GrxS14 and AtBolA2 ( showing best similarity to BOLA3 and possessing a conserved Cys in addition to the invariant C-terminal His ) was characterized by NMR ( Roret et al . , 2014 ) . Surprisingly , the human and plant complexes share only some regions of chemical shift changes , and exhibit substantial differences ( Figure 6—figure supplement 4 ) . For instance , the invariant His66 region of AtBolA2 does not appear to be an interaction interface , at variance with what we observed for both human BOLA complexes ( Figure 6A , B ) . In addition , the GrxS14-interacting regions include the two C-terminal α4 and α5 helices , whereas helix α5 of GLRX5 of both human complexes did not show chemical shifts , and a different segment of helix α4 was involved . Moreover , helix α2 of GLRX5 showed signs of interaction only in the human BOLA-GLRX5 complexes . We conclude that there is a high degree of organismic variability in the precise interaction mode between Grx and Bol proteins . We next investigated the interaction of holo-GLRX5 with the BOLA proteins by chemically reconstituting a [2Fe-2S] cluster on equimolar GLRX5 and BOLAs . NMR 1H-15N HSQC spectra were acquired for the co-reconstituted complexes of 15N labeled BOLA3 or BOLA1 with GLRX5 . The 1H-15N HSQC spectra of these co-reconstituted complexes were similar to spectra obtained by mixing BOLA1 or BOLA3 with [2Fe-2S]-containing holo-GLRX5 ( not shown ) . Upon overlaying the 1H-15N HSQC spectra of the holo-complexes with those acquired for the corresponding apo-complexes , meaningful spectral variations ( either as chemical shift changes and/or line-broadenings beyond signal detection ) were detected on both proteins ( Figure 6—figure supplements 1B and 2B ) . The 1H-15N HSQC maps of the co-reconstituted complexes having 15N labeled GLRX5 with unlabeled BOLA1 or BOLA3 were not superimposable to the 1H-15N HSQC map of the holo-GLRX5 homodimer indicating formation of [2Fe-2S] hetero-complexes 15N R1 and R2 NMR relaxation data of BOLA1 or BOLA3 , in the [2Fe-2S] chemically reconstituted complexes with GLRX5 , indicated no significant changes in protein size with respect to the apo-complexes . Overall , the NMR data showed the presence of 1:1 heterodimeric holo-complexes between GLRX5 and BOLA1 or BOLA3 . Chemical shift perturbation and line broadening analyses were performed by comparing the 1H-15N HSQC spectrum of chemically reconstituted 15N-labeled BOLA3-GLRX5 with that of BOLA3 . The residues of the loop containing the conserved Cys59 of BOLA3 showed strong chemical shift changes only in the holo hetero-complex relative to the apoform ( Figure 6B , D and Figure 6—figure supplement 1 , right parts ) . A similar comparison for the 15N-labeled BOLA1-GLRX5 complex showed that the region around His67 of BOLA1 ( structurally close to Cys59 of BOLA3 ) exhibited similar chemical shift changes , even though fewer residues were affected in BOLA1 ( Figure 6A , C , and Figure 6—figure supplement 1 , left parts ) . The chemical shift perturbation analysis for 15N-labeled GLRX5 revealed that the residues of GLRX5 involved in the apo interaction are also affected in the holo-complex and that the conserved Cys67 of GLRX5 was additionally altered in the holo complexes only ( Figure 6—figure supplement 2 ) . In addition , two other adjacent GLRX5 residues ( Lys59 for BOLA3 and Gly60 for BOLA1 interaction ) were altered that are part of the GSH binding site . The regions of GLRX5 affected by the interaction with BOLA1 and BOLA3 are essentially the same as those in the [2Fe-2S]2+ GLRX5 homodimer ( Banci et al . , 2014 ) . This indicates that , for GLRX5 , the same interaction regions are involved in both homo- and hetero-dimer formation , and the same groups ( the conserved Cys67 and GSH ) can act as iron ligands of the [2Fe-2S]2+ cluster . We used circular dichroism ( CD ) spectroscopy to follow the characteristics of Fe/S cluster coordination upon ligand exchange from the holo-Grx5 homo-dimer to the Grx5-Bol hetero-dimer ( Banci et al . , 2015; Li et al . , 2009; Li et al . , 2012 ) . GLRX5 and stoichiometric amounts of BOLA1 or BOLA3 were chemically co-reconstituted with a [2Fe-2S] cluster and analyzed by CD ( Figure 7 ) . Similar results were obtained when GLRX5 was reconstituted first and then the BOLA proteins were added ( Figure 7—figure supplement 1 ) . The CD spectrum of GLRX5-BOLA1 characteristically differed from that of holo-GLRX5 in a positive versus negative ellipticity around 400 nm with a 6 nm blue shift and 40% decrease of the ~460 nm peak with a 6 nm red shift ( Figure 7A ) . These spectral changes saturated at an equimolar ratio of holo-GLRX5 and BOLA1 suggesting a high affinity equimolar hetero-complex formation with a shared [2Fe-2S] cluster ( Figure 7—figure supplement 1B ) . The cluster coordinated by the GLRX5-BOLA1 hetero-dimer was stable against reduction by dithionite creating a CD spectrum in which the peak at 460 nm was maintained but that at 400 nm was lost . In contrast , the cluster of holo-GLRX5 was destroyed upon dithionite treatment with no residual CD peaks ( Figure 7A and Figure 7—figure supplement 1A ) . These results demonstrate that BOLA1 stabilized the shared [2Fe-2S] cluster in the hetero-dimer with GLRX5 . 10 . 7554/eLife . 16673 . 021Figure 7 . Human BOLA1 but not BOLA3 stabilizes the [2Fe-2S] cluster of holo-GLRX5 upon heterodimer formation . ( A–B ) Chemical reconstitution of Fe/S clusters in buffer R was performed with GLRX5 in the absence or presence of stoichiometric amounts of ( A ) BOLA1 or ( B ) BOLA3 , and CD spectra were monitored under anaerobic conditions . Additional spectra were recorded after addition of 2 mM dithionite ( DT ) . ( C–D ) Chemically reconstituted GLRX5 was titrated with the indicated concentrations of BOLA3 in ( C ) buffer N or ( D ) buffer P , and CD spectra ( corrected for dilution ) were recorded under anaerobic conditions . Abbreviations for varied buffer conditions: 7 , 8: pH; N , 150 mM NaCl; G , 5 mM GSH . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 02110 . 7554/eLife . 16673 . 022Figure 7—figure supplement 1 . Stoichiometric hetero-complex formation of human holo-GLRX5 with BOLA1 results in characteristic CD spectral changes indicating shared binding of the [2Fe-2S] cluster . ( A , C ) Chemically reconstituted GLRX5 was mixed in buffer R ( lacking additional GSH ) with a stoichiometric amount of ( A ) BOLA1 or ( C ) BOLA3 , and CD spectra were recorded under anaerobic conditions . Additional spectra were recorded after addition of 2 mM dithionite ( DT ) . ( B , D ) Chemically reconstituted GLRX5 was titrated with the indicated concentrations of ( B ) BOLA1 or ( D ) BOLA3 , and CD spectra ( corrected for dilution ) were recorded under anaerobic conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 02210 . 7554/eLife . 16673 . 023Figure 7—figure supplement 2 . Gel filtration verifies the complex formation between human holo-GLRX5 and BOLA1 . ( A ) The gel filtration elution profile of co-reconstituted equimolar holo-GLRX5-BOLA3 , holo-GLRX5 ( offset by OD 0 . 015 ) , or co-reconstituted equimolar GLRX5-BOLA1 ( offset by OD 0 . 045 ) was recorded at 420 nm in buffer R . Elution positions of molecular marker proteins are indicated in kDa . ( B ) As a control , the elution behavior of the holo-GLRX5+BOLA3 mixture or BOLA3 alone were recorded at 280 nm in parallel to part A . Fractions of the holo-GLRX5+BOLA3 mixture were analyzed for GLRX5 and BOLA3 by Coomassie and immunostaining ( bottom part ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 02310 . 7554/eLife . 16673 . 024Figure 7—figure supplement 3 . Preferential binding of human BOLA3 to the holo-form of NFU1 . ( A ) NFU1 was used in either apo- or holo-form and mixed at increasing concentrations with 200 nM fluorescently labeled BOLA1 or BOLA3 . Microscale thermophoresis ( MST ) analyses were performed under anaerobic conditions , and dissociation constants ( Kd ) were determined . Control experiments were performed with human [2Fe-2S] ferredoxin FDX2 , cytochrome c , and RNase A . Error bars indicate the SD ( n = 3 ) . ( B–E ) Examples for original MST data . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 024 A strikingly different Fe/S cluster behavior was observed upon BOLA3-GLRX5 complex formation . Hardly any CD spectral changes were observed upon co-reconstituting or mixing GLRX5 and BOLA3 ( Figure 7B and Figure 7—figure supplement 1C , D ) . Further , the Fe/S cluster was not protected against destruction by dithionite treatment indicating rather different coordination situations for the [2Fe-2S] clusters in the two GLRX5-BOLA heterodimers . These results are consistent with the different NMR chemical shift changes in the putative Fe/S cluster binding regions of the two BOLAs in the hetero-complexes ( Figure 6C , D ) . They do not explain , however , why NMR still detected holo-specific changes in the putative cluster-binding region of the BOLA3 hetero-complex ( Figure 6D ) . Since the NMR data were recorded in GSH-containing buffer at lower pH , we repeated the CD analysis under NMR buffer conditions . We found a general decrease of the CD peaks upon GLRX5-BOLA3 hetero-complex formation ( Figure 7C ) . These changes mainly depended on the presence of GSH , and not salt or pH ( Figure 7D ) . In all cases , the CD signal decrease did not reach saturation even after addition of a more than tenfold molar excess of BOLA3 over holo-GLRX5 . This result indicated a rather low affinity for this interference of BOLA3 with the [2Fe-2S] cluster . Overall , the CD data indicate a rather weak and GSH-dependent influence of BOLA3 on the GLRX5 [2Fe-2S] cluster which is not stabilized by hetero-complex formation . To independently analyze the GLRX5-BOLA complex formation , we used size exclusion chromatography . Holo-GLRX5 ( recorded at 420 nm ) eluted at 44 kDa indicative of homodimer with bound Fe/S cluster ( Figure 7—figure supplement 2A ) . A significantly smaller molecular mass was observed for the heterodimeric GLRX5-BOLA1 ( 32 kDa ) , while the mixture of holo-GLRX5 and BOLA3 did not alter the elution behavior of GLRX5 at 420 nm . The elution profile at 280 nm showed no significant protein peak around 30 kDa that might represent a GLRX5-BOLA3 heterodimer ( Figure 7—figure supplement 2B ) . Further , BOLA3 from the GLRX5-BOLA3 mixture eluted exclusively at the position of BOLA3 alone . These results support the formation of a stable heterodimer between GLRX5 and BOLA1 , yet the interaction of GLRX5 with BOLA3 appears to be kinetically unstable , especially in the absence of GSH . The lack of stable interaction between GLRX5 and BOLA3 raised the question of whether the latter could interact with NFU1 . Since Fe/S cluster binding to NFU1 was CD-silent , we performed in vitro binding studies by microscale thermophoresis ( MST ) , an equilibrium method that can be performed under anaerobic conditions ( Webert et al . , 2014 ) . For these interaction experiments we used both apo-NFU1 and chemically reconstituted holo-NFU1 carrying a [4Fe-4S] cluster ( Tong et al . , 2003 ) . MST indeed showed a specific interaction between the two BOLA proteins and both apo- and holo-NFU1 ( Figure 7—figure supplement 3 ) . The dissociation constants ( Kd ) of the interactions were in the range of 3 µM , with the notable exception of a fourfold higher affinity detected for BOLA3 and holo-NFU1 ( Kd = 0 . 8 µM ) . As a control , no significant affinity was observed for human [2Fe-2S] ferredoxin FDX2 , cytochrome c or RNase A as control proteins . Collectively , these results data suggest that holo-GLRX5 preferentially cooperates with BOLA1 , and holo-NFU1 with BOLA3 . In this work , we have defined the molecular function of the mitochondrial Bol1 and Bol3 proteins ( mBols ) as specific ISC assembly factors required for the insertion of [4Fe-4S] clusters into a small subset of mitochondrial target apoproteins . The two mBols act late in the ISC pathway , and execute a largely overlapping function , because only simultaneous deletion of both BOL genes was associated with appreciable effects on mitochondrial [4Fe-4S] target proteins ( Figure 8 ) . Lipoic acid synthase ( LIAS ) with its two [4Fe-4S] clusters is the primary client of mBol function because of strong effects on lipoic acid-dependent proteins ( such as PDH and KGDH ) in bol13Δ cells . These severe effects were observed under all experimental conditions tested , unlike the comparatively weak effects on SDH ( complex II ) which typically is one of the more sensitive mitochondrial Fe/S proteins upon ISC factor defects . Therefore , we cannot exclude an indirect effect of the BOL gene deletion on SDH maturation . The [4Fe-4S] aconitase , on the other hand , was not or hardly affected in bol13Δ cells distinguishing the BOL deletion phenotype from that of all other known ISC genes , with the exception of the complex I-specific IND1 ( Bych et al . , 2008; Sheftel et al . , 2009 ) . We propose that the rather inconspicuous effects on aconitase are indirect consequences of the metabolic and respiratory alterations arising from lipoic acid-dependent enzyme defects which mainly affect the citric acid cycle ( PDH and KGDH ) and the amino acid metabolism ( BCKDH , GCS; Figure 8 ) . The functions of the mBols are largely overlapping , yet not entirely identical . This was evident from small effects on LIAS and SDH in bol3Δ cells , versus no detectable alterations upon BOL1 deletion . Moreover , double deletion of BOL3 and NFU1 , encoding another late-acting ISC factor ( Navarro-Sastre et al . , 2011 ) , ( Melber et al . , 2016; accompanying manuscript ) , exacerbated the Fe/S protein defects compared to single deletions , while double deletion of BOL1 and NFU1 behaved similarly to nfu1Δ cells ( Figure 4A , B ) . Two models may explain the largely complementary functions of the two mBols . i ) The mBols assist the same biochemical reaction . In this case , the subtle differences between bol1Δ and bol3Δ strains may be due to minor target apoprotein specificities of the two mBols or the different protein interactions with Grx5 and Nfu1 . ii ) The mBols act consecutively in two independent maturation steps that both can be bypassed to some extent . Even though we biochemically favor the first possibility , current knowledge does not allow a more precise mechanistic definition of mBol function . Taken together , the major evident physiological function of the yeast mBol proteins is in lipoate synthase maturation . A similar defect is seen in human cells obtained from BOLA3 patients ( Baker et al . , 2014; Cameron et al . , 2011; Haack et al . , 2013 ) . This phenotypical resemblance suggests that the yeast mBols and at least human BOLA3 are functionally similar . Even though yeast appears to be a suitable model system for the physiological and mechanistic investigation of mBols , the two human BOLAs now need to be studied in cell culture to gain better insights into their relative function . 10 . 7554/eLife . 16673 . 025Figure 8 . Working model for the role of mitochondrial BOLA proteins as specific ISC assembly factors in the late phase of mitochondrial Fe/S protein biogenesis . Members of the core ISC assembly machinery including the sulfur donor NFS1 , the scaffold protein ISCU , and dedicated chaperones ( Chap . ) mediate the assembly of a transiently bound , glutathione ( GS ) -coordinated [2Fe-2S] cluster on the monothiol glutaredoxin GLRX5 which is essential for [2Fe-2S] protein maturation , cytosolic Fe/S protein assembly ( CIA ) , and cellular iron regulation in yeast . With the help of ISCA1 , ISCA2 , and IBA57 , the GLRX5-bound [2Fe-2S] cluster is converted into a [4Fe-4S] type . BOLA1 and BOLA3 together with NFU1 and IND1 function in a specific assembly of mitochondrial [4Fe-4S] proteins as indicated . The central target of the BOLA proteins is the Fe/S protein lipoic acid ( LA ) synthase ( LIAS ) . Its product is used as a cofactor of five mitochondrial enzymes including pyruvate dehydrogenase ( PDH ) and 2-ketoglutarate dehydrogenase ( KGDH ) . Based on the hetero-complex formation with the BOLA proteins an additional function of GLRX5 in this late phase of mitochondrial Fe/S protein assembly is likely . Mutations in human NFU1 and BOLA3 cause multiple mitochondrial dysfunction syndromes ( MMDS; for review see [Beilschmidt and Puccio , 2014; Stehling et al . , 2014] ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16673 . 025 Several findings allowed us to define the site of action of the mBols in the multi-step ISC assembly pathway ( Figure 8 ) . First , the mBols were not required for maturation of mitochondrial [2Fe-2S] proteins or cytosolic Fe/S proteins . Second , their deletion did not affect the yeast cellular iron regulon . These criteria distinguish the mBol phenotype from that of the components of the core ISC machinery including Grx5 , that is the final core ISC factor ( Uzarska et al . , 2013 ) , and suggest that mBols act subsequently to Grx5 . This view is further corroborated by our finding that , third , 55Fe incorporation into Grx5 occurred independently of mBols . Finally , the function of the mBols must be after the involvement of Isa-Iba57 proteins in the [4Fe-4S] cluster formation , because the mBols exhibit high [4Fe-4S] target specificity . In Isa-Iba57-defective cells , all mitochondrial [4Fe-4S] proteins and the non-Fe/S protein cytochrome oxidase are severely affected , unlike in bol13Δ cells ( Mühlenhoff et al . , 2011; Sheftel et al . , 2012 ) . Overall , the site of action of the mBols is similar to that of the late-acting ISC assembly factor Nfu1 ( Navarro-Sastre et al . , 2011 ) ( Figure 8 ) . However , we found that the mBol and Nfu1 proteins cannot complement each other’s function , even after overexpression , suggesting that they both fulfil a highly specific task in the dedicated insertion of [4Fe-4S] clusters into target apoproteins . The observation that the mBols function subsequently to Grx5 is somewhat unexpected in light of our findings that the mBols form binary complexes with this core ISC component . Apparently , complex formation with the mBols is not critical for Grx5’s major function as a core ISC factor ( Figure 8 ) , as the GRX5 deficiency is associated with rather severe defects in virtually all mitochondrial and cytosolic Fe/S proteins plus a misregulation of cellular iron homeostasis ( Rodriguez-Manzaneque et al . , 2002; Uzarska et al . , 2013; Ye et al . , 2010 ) , radically distinguishing this phenotype from that of the rather inconspicuous BOL1-BOL3 deletion . It is well possible that the Grx5-mBol complexes play a role at the later stage of Fe/S cluster insertion into apoproteins . This seems likely based on interaction between BOLA1 and GLRX5 observed in human cells ( Willems et al . , 2013 ) , and similar interactions reported in the accompanying manuscript ( Melber et al . , 2016 ) . At present , it is impossible to verify such a function by in vivo studies , because the potential Grx5 involvement at this later stage of the ISC pathway is hidden by the severe GRX5 deletion phenotype . In our present work , we have characterized the Grx5-mBol interactions by a number of in vitro techniques including gel filtration , CD spectroscopy and NMR structural analyses . Human GLRX5 interacted with both mBols , stoichiometrically forming a 1:1 complex both in the apo- and chemically reconstituted holo-forms , involving similar regions on the three proteins as revealed by NMR studies . However , for the holo-forms , further residues were affected , and were clustered around the region containing the invariant histidine ( His102 in BOLA1 and His96 in BOLA3; Figure 7C , D ) , which in cytosolic Bol2 is a ligand for Fe/S cluster binding with Grx3-Grx4 ( Li et al . , 2011 ) . These Bol regions include other potential cluster ligands , which are two His residues in BOLA1 , located at the end of β2 and in the loop between β1 and β2 , and one Cys in BOLA3 , located in the loop between β1 and β2 . CD spectroscopy revealed striking differences in Fe/S cluster binding to the two holo-GLRX5-BOLA complexes . Compared to holo-GLRX5 , the holo-GLRX5-BOLA1 complex showed characteristic new spectral features including a new peak at ~400 nm . GLRX5 and BOLA1 coordinate the [2Fe-2S] cluster with high affinity . The spectral features of the complex were rather insensitive to changes in ionic conditions or to the presence of additional GSH . Most characteristically , the [2Fe-2S] cluster of GLRX5 was stabilized by BOLA1 binding against destruction by the reductant dithionite indicating a stable holo-GLRX5-BOLA1 complex . All these criteria were different for the holo-GLRX5-BOLA3 complex . First , the CD spectral features of the holo-GLRX5-BOLA3 complex differed from those of holo-GLRX5 by a general signal decrease , and were observed mainly in the presence of GSH suggesting that without added GSH the GLRX5 cluster is not shared with BOLA3 . Second , the spectral changes induced by BOLA3 addition to holo-GLRX5 were not saturable indicating a rather weak influence of BOLA3 on the GLRX5 Fe/S cluster . Third , the CD signals became rather weak at high BOLA3 or salt concentrations . Finally , dithionite readily destroyed the [2Fe-2S] cluster of GLRX5 in the presence and absence of BOLA3 . In support of this data , gel filtration did not identify a stable complex between holo-GLRX5 and BOLA3 indicating that their interaction is kinetically labile , and a hetero-dimer may dissociate during the non-equilibrium method . All these findings document radically different Fe/S cluster binding properties of the two GLXR5-BOLA complexes with BOLA1 stabilizing and BOLA3 destabilizing the [2Fe-2S] cluster . Interestingly , we found a preferential interaction of BOLA3 with the holoform of NFU1 by thermophoresis . These data suggest that BOLA3 may stabilize the [4Fe-4S] cluster bound on NFU1 . A specific genetic interaction of NFU1 and BOL3 supports our biochemical findings . Double deletion of these genes exacerbated the defects of SDH and LIAS , whereas simultaneous deletions of NFU1 and BOL1 did not ( Figure 4A and B ) . This suggests that Bol3 and Nfu1 cooperate in SDH and LIAS maturation . Future biochemical studies have to address the physiological meaning of the stabilizing or destabilizing function of the mBols for the different Fe/S clusters on Grx5 and Nfu1 . The NMR structures of the mBols share a similar fold , alike to those of other eukaryotic homologous ( Abendroth et al . , 2011; Kasai et al . , 2004; Roret et al . , 2014 ) . Further , similar interacting regions for the apo- and holo-states of the GLRX5-BOLA heterodimers were found , and a similar location of the invariant His residue , a known Fe/S cluster ligand in Bol2 ( Li et al . , 2011 ) . However , we note that its surrounding is structurally different in the two BOLAs , that is His102 of BOLA1 is the first residue following a long β-strand , while His96 of BOLA3 is in a 310 helix ( Figure 6C ) . The other potential Fe/S cluster ligands are located in structurally different positions of the two mBols . In particular , the location of the two conserved His58 , His67 residues in BOLA1 and of Cys59 in BOLA3 are different ( Figure 6C ) . These features can possibly explain the observed differences in the binding and stability of [2Fe-2S] clusters to the holo heterodimers with GLRX5 . The distinct functional roles of the mBols could be viewed as consequence of the different coordination spheres of the Fe/S cluster . The role of the mBols in a late step of the mitochondrial ISC pathway fits well with our observation that their functional deficiency ( bol13Δ cells ) does not cause any significant defects on cytosolic Fe/S proteins . Interestingly , a combination of these gene deletions with BOL2 elicited significant effects , even though these were rather weak compared to known core ISC or CIA deficiencies ( see , e . g . , Gerber et al . , 2004; Hausmann et al . , 2005; Netz et al . , 2010 ) . The defects , especially in the triple deletion mutant bol123Δ , are therefore best explained by combined detrimental effects of mitochondrial respiratory and metabolic changes arising from the BOL1-BOL3 deletion ( see above ) and the disturbed cellular iron homeostasis resulting from BOL2 ablation . Single BOL2 deletion is not associated with major phenotypes apart from iron homeostasis ( Kumanovics et al . , 2008 ) ( Figure 1—figure supplement 3 ) . This is clearly different from the usual lethality of CIA gene deletions ( Netz et al . , 2014; Paul and Lill , 2015 ) , and fits well with the rather weak or even inconspicuous effects of a BOL2 deletion on the maturation of the cytosolic Fe/S proteins Leu1 and Rli1 . Any slight effects are most likely due to the auxiliary role of Bol2 in iron metabolism ( Kumanovics et al . , 2008 ) . In this function , Bol2 cooperates with Grx3-Grx4 ( human PICOT ) which , in its role in cellular iron mobilization , is essential for cytosolic Fe/S protein maturation ( Banci et al . , 2015; Haunhorst et al . , 2013; Mühlenhoff et al . , 2010 ) . Our comprehensive physiological investigations of mBol function in yeast reveal that bol13Δ cells behave similarly to human patients deficient in BOLA3 ( Baker et al . , 2014; Cameron et al . , 2011; Haack et al . , 2013 ) . Patients with the resulting multiple mitochondrial dysfunction syndrome 2 ( MMDS 2 ) show major defects in lipoic acid-dependent enzymes , similar to the most severe effect observed in bol13Δ yeast . Moreover , the patient cells are deficient in respiratory complexes I and II , the latter also being affected in yeast . We conclude that the mBol protein function is conserved from yeast to man with a major role in [4Fe-4S] cluster insertion into LIAS and respiratory complexes I and II , whereas Fe/S proteins such as aconitase are matured independently of mBol function . The conclusion of similar mBol functions in yeast and man may be surprising based on the rather inconspicuous yeast deletion phenotypes versus the lethal effect in BOLA3-deficient individuals . This difference may simply reflect the higher complexity of a multi-cellular organism and its dependence on respiration and metabolic homeostasis , whereas yeast can tolerate substantial deviations from optimal conditions . Notably , RNAi-mediated depletion of BOLA1 was associated with rather mild effects ( Willems et al . , 2013 ) predicting that the combination with a BOLA3 depletion may create a more pathological phenotype mimicking that of human patients and of bol13Δ yeast cells . Our elucidation of the target apoprotein-specific ISC assembly function of mBols will be instrumental for future mechanistic studies on how the mBols facilitate the insertion of [4Fe-4S] clusters into LIAS and possibly other Fe/S proteins . These studies will also have to reveal why other Fe/S proteins such as aconitase are matured efficiently in the absence of these specialized ISC assembly factors . S . cerevisiae strains used in this study are listed in Supplementary file 1A . Cells were cultivated in rich medium ( YP ) or synthetic minimal medium ( SC ) supplemented with amino acids as required and 2% ( w/v ) glucose , galactose or acetate , or 3% ( w/v ) glycerol ( Sherman , 2002 ) . Iron-depleted minimal media were prepared using yeast nitrogen base without FeCl3 ( ForMedium ) . Plasmids used in this study are listed in Supplementary file 1B . Plasmid constructs were verified by DNA sequencing and/or functional complementation of a corresponding yeast mutant . Genes of human BOLA1 ( residues 30–137 ) , BOLA3 ( 25–107 ) , GLRX5 ( 32–157 ) and NFU1 ( 57–254 ) ( each lacking its mitochondrial presequence ) were inserted into the multiple cloning site I of pET-Duet1 vector ( Novagen-Merck , Darmstadt , Germany ) with a N-terminal His-tag . N-terminally tagged proteins were purified from E . coli BL21 ( DE3 ) using a HisTrap HP column ( GE Healthcare ) . The eluted proteins were treated with 5 mM DTT and isolated by gel filtration on a Superdex 200 16/60 gel filtration column ( GE Healthcare ) in reconstitution buffer R ( 50 mM Tris-HCl , pH 8 . 0 , 150 mM NaCl , 5% glycerol ) . For NMR the BOLA proteins ( BOLA1 , 21–137; BOLA3 , 27–107 ) were obtained from BL21 ( DE3 ) GOLD Escherichia coli ( Novagen ) transformed with pETG20A vector ( BOLA1 ) and pET15 vector ( BOLA3 ) . Cells were grown at 37°C in LB or minimal media with ( 15NH4 ) 2SO4 and/or 13C-glucose containing 100 μg/mL ampicillin . Both proteins contained a N-terminal His-tag , which was cleaved by tobacco etch virus protease in 50 mM Tris-HCl pH 8 , 0 . 5 mM EDTA , 5 mM GSH , and 1 mM DTT overnight at room temperature . GLRX5 protein in its apo- and holo-forms was produced as previously reported ( Banci et al . , 2014 ) . All solutions used for chemical reconstitution were prepared freshly in a COY anaerobic chamber by dissolving in degassed water . Protein samples were reduced in buffer R containing 5 mM DTT for 2–3 hr on ice in buffer R containing a two fold excess of GSH . Reconstitution was initiated at room temperature by the addition of a 2–5-fold excess of ferric ammonium citrate . After five minutes a 2–5-fold excess of lithium sulfide was slowly added . After 2 hr reconstituted proteins were desalted on a PD-10 column ( GE Healthcare ) equilibrated with buffer R . Incorporation of the Fe/S clusters into apoproteins was verified by UV-Vis ( V-550 , Jasco Inc . ) and CD spectroscopy ( J-815 , Jasco Inc . ) and the determination of bound iron and acid-labile sulfur ( Pierik et al . , 1992 ) . CD spectroscopy ( J-815 , Jasco Inc . ) of human GLRX5 reconstituted either alone ( >85% holo based on iron and acid-labile sulfur determination ) or in the presence of stoichiometric amounts of either BOLA1 or BOLA3 was performed anaerobically ( protein concentrations 200–300 µM ) in buffer R , buffer N ( 50 mM phosphate buffer pH 7 , 5 mM GSH , and 5 mM DTT ) , or buffer P ( 50 mM phosphate buffer pH 8 , 150 mM NaCl , 5 mM GSH ) . Alternatively , reconstituted holo-GLRX5 was titrated with increasing concentrations of BOLA1 or BOLA3 . CD spectra were recorded anaerobically at 21°C in 1 mm cuvettes . The bait proteins were fluorescently labeled using the Monolith NT Protein Labeling Kit RED ( NanoTemper Technologies ) with NT-647 dye as recommended by the manufacturer . Labeled bait proteins ( 200 nM ) were titrated with the indicated unlabeled proteins ( from 200 µM to 6 . 1 nM ) in buffer T ( 50 mM KPi , pH 7 . 4 , 150 mM NaCl , 5% glycerol , 0 . 05 mg/ml BSA , 0 . 05% Tween20 ) . Binding assays were performed using Monolith NT . 115 ( NanoTemper Technologies ) at 21°C ( LED power between 40% and 60% , IR laser power 75% ) in standard capillaries under anaerobic conditions at 680 nm . The results were processed by NanoTemper Analysis 1 . 2 . 009 and GraphPadPrism5 software to estimate Kd values . All NMR experiments required for resonance assignment and structure calculations of BOLA1 ( PDB ID 5LCI; BMRB 34013 ) and BOLA3 ( PDB ID 2NCL; BMRB 26031 ) were recorded on Bruker AVANCE 500 , 700 and 900 MHz spectrometers on 0 . 5–1 mM 13C , 15N-labeled or 15N-labeled BOLA1 and BOLA3 samples in 50 mM phosphate buffer , pH 7 . 0 , 5 mM DTT , containing 10% ( v/v ) D2O . All NMR spectra were collected at 298 K , processed using the standard BRUKER software ( Topspin ) and analyzed through the CARA program . The 1 H , 13C and 15N resonance assignment of BOLAs were performed following a standard triple-resonance and TOCSY-based approach . Secondary structure analysis was performed by TALOS+ . Structure calculations of BOLA3 and BOLA1 were performed with the software package UNIO ( ATNOS/CANDID/CYANA ) . The 20 conformers with the lowest residual target function values were subjected to restrained energy minimization in explicit water with the program AMBER 12 . 0 . ( D . A . Case et al . University of California , San Francisco ) . The quality of the structures was evaluated using the Protein Structure Validation Software suite ( PSVS ) and the iCING program . 15N heteronuclear relaxation experiments on 15N-labeled samples of BOLA1 and BOLA3 were recorded on Bruker AVANCE 500 MHz spectrometer at 298 K to measure 15N backbone longitudinal ( R1 ) and transverse ( R2 ) relaxation rates , as well as the heteronuclear 15N[1H] NOEs . The rotational correlation time values were estimated from the R2/R1 ratio using the program QUADRATIC_DIFFUSION . The relaxation data for those NHs having an exchange contribution to the R2 value or exhibiting large-amplitude fast internal motions were excluded from the analysis . Theoretical estimates of the rotational correlation time under the chosen experimental conditions of magnetic field and temperature were obtained using HYDRONMR program following a standard procedure ( Figure 6—figure supplement 3 ) . The interaction between apo-GLRX5 and BOLA proteins was investigated by 1H-15N HSQC NMR spectra , titrating 15N-labeled apo-GLRX5 with unlabeled BOLA1 or BOLA3 , and 15N labeled BOLA1 or BOLA3 with unlabeled apo-GLRX5 in degassed buffer N containing 10% ( v/v ) D2O ) at 298K . Spectral changes were monitored upon the addition of increasing amounts of the unlabeled partner . Protein interaction between [2Fe-2S] holo-GLRX5 and BOLA proteins was investigated comparing 1H-15N HSQC NMR spectra of 15N-labeled BOLA1 ( or BOLA3 ) with that of a 1:1 15N-labeled BOLA1 ( or BOLA3 ) -unlabeled apo GLRX5 mixture chemically reconstituted with [2Fe-2S] , or of 15N-labeled apo-GLRX5 with that of a 1:1 15N-labeled apo-GLRX5-unlabeled BOLA1 ( or BOLA3 ) mixture chemically reconstituted with [2Fe-2S] , in degassed buffer N containing 10% ( v/v ) D2O ) at 298K . NMR data were analyzed with CARA program . The following published methods were used: manipulation of DNA and PCR ( Sambrook and Russell , 2001 ) ; transformation of yeast cells ( Gietz and Woods , 2002 ) ; isolation of yeast mitochondria and post-mitochondrial supernatant ( Diekert et al . , 2001 ) ; immunostaining ( Harlow and Lane , 1998 ) ; determination of enzyme activities and of promoter strength using FET3-GFP or FIT3-GFP reporter plasmids ( Molik et al . , 2007 ) ; in vivo labeling of yeast cells with 55FeCl ( ICN ) and measurement of 55Fe-incorporation into Fe/S proteins by immunoprecipitation and scintillation counting ( Molik et al . , 2007; Pierik et al . , 2009 ) . Antibodies were raised in rabbits against recombinant purified proteins . Antibodies against c-Myc or HA were obtained from Santa-Cruz , protein A sepharose from GE Healthcare .	Proteins perform almost all the tasks necessary for cells to survive . However , some proteins , especially enzymes involved in metabolism and energy production , need to contain extra molecules called co-factors to work properly . In human , yeast and other eukaryotic cells , co-factors called iron-sulfur clusters are made in compartments called mitochondria before being packaged into target proteins . Defects that affect the assembly of proteins with iron-sulfur clusters are associated with severe diseases that affect metabolism , the nervous system and the blood . Mitochondria contain at least 17 proteins involved in making iron-sulfur proteins , but there may be others that have not yet been identified . For example , a study on patients with a rare human genetic disease suggested that a protein called BOLA3 might also play a role in this process . BOLA3 is closely related to the BOLA1 proteins . Here , Uzarska , Nasta , Weiler et al . used yeast to test how these proteins contribute to the assembly of iron-sulfur proteins . Biochemical techniques showed that the yeast equivalents of BOLA1 and BOLA3 ( known as Bol1 and Bol3 ) play specific roles in the assembly pathway . When both of these proteins were missing from yeast , some iron-sulfur proteins – including an important enzyme called lipoic acid synthase – did not assemble properly . The experiments suggest that yeast Bol1 and Bol3 play overlapping and critical roles during the last step of iron-sulfur protein assembly when the iron-sulfur cluster is inserted into the target protein . Lastly , Uzarska , Nasta , Weiler et al . used biophysical techniques to show how Bol1 and Bol3 interact with another mitochondrial protein that performs a more general role in iron-sulfur protein assembly . Defects in assembling iron-sulfur proteins are generally more harmful to human cells than yeast cells . Therefore , the next step is to investigate what exact roles BOLA1 and BOLA3 play in human cells and how similar this pathway is in different eukaryotes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cancer", "biology" ]	2016	Mitochondrial Bol1 and Bol3 function as assembly factors for specific iron-sulfur proteins
Dynamic changes in protein S-palmitoylation are critical for regulating protein localization and signaling . Only two enzymes - the acyl-protein thioesterases APT1 and APT2 – are known to catalyze palmitate removal from cytosolic cysteine residues . It is unclear if these enzymes act constitutively on all palmitoylated proteins , or if additional depalmitoylases exist . Using a dual pulse-chase strategy comparing palmitate and protein half-lives , we found knockdown or inhibition of APT1 and APT2 blocked depalmitoylation of Huntingtin , but did not affect palmitate turnover on postsynaptic density protein 95 ( PSD95 ) or N-Ras . We used activity profiling to identify novel serine hydrolase targets of the APT1/2 inhibitor Palmostatin B , and discovered that a family of uncharacterized ABHD17 proteins can accelerate palmitate turnover on PSD95 and N-Ras . ABHD17 catalytic activity is required for N-Ras depalmitoylation and re-localization to internal cellular membranes . Our findings indicate that the family of depalmitoylation enzymes may be substantially broader than previously believed . Protein S-palmitoylation involves the post-translational attachment of the 16-carbon fatty acid palmitate to cysteine residues ( Conibear and Davis , 2010; Salaun et al . , 2010 ) . While a survey of palmitoylation dynamics indicated the bulk of the palmitoyl-proteome is stably palmitoylated ( Martin et al . , 2011 ) , rapid and constitutive palmitate turnover has been shown for several proteins , including the Ras GTPases , heterotrimeric G proteins , the neuronal post-synaptic density protein PSD95 , and the Lck kinase ( Magee et al . , 1987; Degtyarev et al . , 1993; El-Husseini et al . , 2002; Zhang et al . , 2010 ) . Dynamic changes in palmitoylation modulate protein localization and trafficking and can be regulated in response to cellular signaling ( Conibear and Davis , 2010 ) . Palmitoylation is mediated by a family of DHHC ( Asp-His-His-Cys ) proteins ( Greaves and Chamberlain , 2011a ) , whereas the only enzymes identified to date that remove palmitate from cytosolic cysteines , the acyl-protein thioesterases ( APTs ) APT1 and APT2 , are related members of the metabolic serine hydrolase ( mSH ) superfamily ( Duncan and Gilman , 1998; Tomatis et al . , 2010; Long and Cravatt , 2011 ) . The β-lactone core-containing compound Palmostatin B ( PalmB ) potently inhibits these enzymes and blocks depalmitoylation of N-Ras and other proteins ( Dekker et al . , 2010; Rusch et al . , 2011 ) . Hexadecyl fluorophosphonate ( HDFP ) inhibits a subset of mSHs including APT1 and APT2 and also suppresses palmitate turnover ( Martin et al . , 2011 ) . However , it is unclear if APT1 and APT2 are the only palmitoylthioesterases responsible for the depalmitoylation of cytosolic proteins ( Davda and Martin , 2014 ) . Here , we show that APT1 and APT2 inhibition or knockdown reduces palmitate turnover on some substrates but has no effect on N-Ras and PSD95 . We identified members of the ABHD17 family as novel PalmB targets that depalmitoylate N-Ras and promote its relocalization to internal membranes . This demonstrates the enzymes responsible for protein depalmitoylation are more diverse than previously believed , which has important implications for understanding the selectivity and regulation of dynamic palmitate turnover . APT1 and APT2 were proposed to act universally and constitutively to remove mislocalized proteins from intracellular membranes and allow their re-palmitoylation at the Golgi ( Rocks et al . , 2010 ) . Reported rates of palmitate turnover on different substrates vary dramatically ( Qanbar and Bouvier , 2004; Martin et al . , 2011 ) . We used a dual-click chemistry pulse-chase scheme to simultaneously measure palmitate and protein turnover of proteins expressed in COS-7 cells and labeled with the palmitate analogue 17-octadecynoic acid ( 17-ODYA ) and the methionine surrogate L-azidohomoalanine ( L-AHA ) ( Martin and Cravatt , 2009; Zhang et al . , 2010 ) . N-Ras had a rapid palmitate turnover as previously reported ( Figure 1A; Magee et al . , 1987 ) . SNAP25 turned over slowly , whereas the glutamate decarboxylase subunit GAD65 and PSD95 had intermediate rates of depalmitoylation , demonstrating that these neuronal proteins undergo palmitate turnover at comparable rates in COS-7 cells or neuronal lines ( Greaves and Chamberlain , 2011b; El-Husseini et al . , 2002 ) . A palmitoylated N-terminal fragment of Huntingtin ( N-HTT ) implicated in the pathogenesis of Huntington’s disease ( Yanai et al . , 2006 ) also showed an intermediate palmitate turnover ( Figure 1B ) . Treatment with the APT1/2 inhibitor PalmB inhibited the depalmitoylation of these substrates without affecting protein turnover ( Figure 1A , B ) . In contrast , we found three proteins identified in a global palmitoyl-proteomics analysis ( SPRED2 , GOLIM4 , and ITM2B ) ( Martin et al . , 2011 ) did not undergo significant palmitate turnover , suggesting the apparent PalmB-resistant decline in palmitate labeling was due to protein instability ( Figure 1B ) . These results confirm that proteins have inherently distinct rates of depalmitoylation , potentially reflecting differential recognition by APTs ( Lin and Conibear , 2015 ) . In all cases examined , PalmB inhibited the palmitate turnover of dynamically palmitoylated proteins . 10 . 7554/eLife . 11306 . 003Figure 1 . Dual-click chemistry labeling reveals differences in protein depalmitoylation dynamics . ( A ) Pulse-chase analysis of established palmitoyl-proteins ( N-Ras , SNAP25 , GAD65 , PSD95 ) by dual-click chemistry in the presence of DMSO ( - ) or 10 μM PalmB ( + ) . Representative in-gel fluorescence scans illustrate dual detection of 17-ODYA ( palmitate analogue ) and L-AHA ( methionine analogue ) using Alexa Fluor 488 and Alexa Fluor 647 , respectively . Dashed line indicates cropping of a single gel . n = 2 per substrate . ( B ) Pulse-chase analysis of palmitate turnover on N-HTT , SPRED2 , GOLIM4 , and ITM2B by dual-click chemistry as described in ( A ) . Upper panels: representative in-gel fluorescence scans; Lower panels: Time course of substrate depalmitoyation in DMSO- and PalmB-treated cells after normalizing 17-ODYA to L-AHA signals at each chase time . n = 2 , mean ± SEM . 17-ODYA , 17-octadecynoic acid; L-AHA , L-azidohomoalanine; SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 11306 . 003 APT1 and APT2 are reported to have differential substrate specificity ( Tomatis et al . , 2010; Tian et al . , 2012 ) . We found that the selective inhibitors C83 and C115 , which target APT1 and APT2 respectively ( Adibekian et al . , 2012 ) , had little effect on N-HTT depalmitoylation when used individually but achieved significant inhibition when applied together ( Figure 2A , B ) . A similar effect was observed on GAD65 ( Figure 2—figure supplement 1A ) . Surprisingly , these inhibitors had no effect on PSD95 or N-Ras depalmitoylation when used alone ( Figure 2—figure supplement 1B , C ) or together ( Figure 2C , D ) . Double RNAi knockdown of APT1 and APT2 significantly inhibited N-HTT depalmitoylation ( Figure 2B ) and also reduced palmitate turnover on GAD65 ( Figure 2—figure supplement 1D ) but not PSD95 or N-Ras ( Figure 2C , D ) . These findings , which are consistent with a recent report showing APT1/2-independent depalmitoylation of R7BP ( Jia et al . , 2014 ) , strongly suggest that although APT1 and APT2 are responsible for depalmitoylating some proteins ( N-HTT , GAD65 ) , depalmitoylation of other cellular substrates , including PSD95 and N-Ras , involves other enzymes . 10 . 7554/eLife . 11306 . 004Figure 2 . Downregulation of APT1 and APT2 inhibits HTT depalmitoylation but does not affect palmitate turnover on PSD95 or N-Ras . ( A ) Pulse-chase analysis of N-HTT palmitoylation in the presence of DMSO , 10 μM PalmB , 10 μM APT1-selective inhibitor C83 , and/or 10 μM APT2-selective inhibitor C115 , as described in Figure 1 . n = 3 , mean ± SEM . ( B-D ) Pulse-chase analysis of ( B ) N-HTT , ( C ) PSD95 , and ( D ) N-Ras after APT1 and APT2 knockdown ( “APT1/2 RNAi” ) , treatment with DMSO , treatment with 10 μM C83 and 10 μM C115 , or treatment with 10 μM PalmB , as described in Figure 1 . n = 3 , mean ± SEM . p < 0 . 05; p < 0 . 01; *p < 0 . 001 . SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 11306 . 00410 . 7554/eLife . 11306 . 005Figure 2—figure supplement 1 . Downregulation of APT1 and APT2 inhibits GAD65 depalmitoylation but does not affect palmitate turnover on PSD95 or N-Ras . ( A-C ) Pulse-chase analysis of ( A ) GAD65 , ( B ) PSD95 , and ( C ) N-Ras palmitoylation in the presence of DMSO , 10 μM PalmB , 10 μM APT1-selective inhibitor C83 , and/or 10 μM APT2-selective inhibitor C115 , as described in Figure 2 . ( D ) Pulse-chase analysis of GAD65 after APT1 and APT2 knockdown ( “APT1/2 RNAi” ) , treatment with DMSO , treatment with 10 μM C83 and 10 μM C115 , or treatment with 10 μM PalmB , as described in Figure 2 . p < 0 . 05; **p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11306 . 005 Previous studies suggested that APT1 , APT2 , and PPT1 were the sole mSHs targeted by PalmB ( Rusch et al . , 2011 ) , whereas HDFP inhibited additional mSHs ( Martin et al . , 2011 ) . In pulse-chase experiments , we found HDFP robustly inhibited the depalmitoylation of N-Ras , PSD95 , and N-HTT ( Figure 3A-C ) . Because palmitate removal from N-Ras and PSD95 does not require APT1 or APT2 , their depalmitoylation may be mediated by a distinct mSH that is a common target of both PalmB and HDFP . To identify overlapping targets , we defined a set of 19 candidate mSHs that showed >25% inhibition by HDFP ( Supplementary file 1; Martin et al . , 2011 ) but excluded known proteases and mSHs with established luminal activity . We added to this list APT1L , which was previously implicated in BK channel depalmitoylation ( Tian et al . , 2012 ) but whose HDFP sensitivity was unknown . The PalmB sensitivity of each enzyme was evaluated by a competitive activity-based protein profiling ( cABPP ) assay , in which binding of an inhibitor occludes the enzyme active site and prevents labeling with the activity probe fluorophosphonate-rhodamine ( FP-rho ) ( Figure 3D; Kidd et al . , 2001 ) . As expected , PalmB significantly reduced FP-rho labeling of both APT1 and APT2 ( Figure 3E , H ) . In contrast , it had little effect on the labeling of seven candidates ( Figure 3F , H ) , highlighting the distinct substrate specificities of PalmB and HDFP . Four mSHs did not label with FP-Rho due to low activity or expression and could not be assessed ( Supplementary file 1 ) . Notably , PalmB potently inhibited seven candidates: FASN , PNPLA6 , ABHD6 , ABHD16A , and ABHD17A/B/C ( Figure 3G , H ) . Thus , PalmB has additional serine hydrolase targets beyond APT1 and APT2 that may function as protein depalmitoylases . 10 . 7554/eLife . 11306 . 006Figure 3 . Shared targets of Palmostatin B and HDFP identified by competitive activity-based protein profiling . ( A-C ) Pulse-chase analysis of ( A ) N-Ras , ( B ) PSD95 , and ( C ) N-HTT in the presence of DMSO , 10 μM PalmB or 20 μM lipase inhibitor HDFP as described in Figure 1 . n = 3 ( DMSO and PalmB ) or 2 ( HDFP ) , mean ± SEM . ( D ) Schematic diagram of the competitive ABPP assay used in this study . ( E-G ) Competitive ABPP of PalmB by in-gel fluorescence ( FP-Rho ) . 16 HDFP targets were incubated with 2 μM FP-Rho in the presence ( + ) or absence ( - ) of 10 μM PalmB . Western blots ( WB ) show reduced FP-Rho labeling is not due to protein loss . ( H ) Percent inhibition of each HDFP target by PalmB . n = 3 , mean ± SEM . Candidate depalmitoylases ( >50% inhibition by PalmB ) are highlighted in red . SEM , standard error of the mean , DOI: http://dx . doi . org/10 . 7554/eLife . 11306 . 00610 . 7554/eLife . 11306 . 007Figure 3—figure supplement 1 . Treatment with serine hydrolase inhibitors WWL70 , C75 , and RHC-80267 does not affect PSD95 palmitate turnover . ( A-B ) Competitive ABPP of 10 μM PalmB and ( A ) 10 μM WWL70 or ( B ) 20 μM RHC-80267 against candidate depalmitoylases and ACOT1 . Percent inhibition of each enzyme is relative to DMSO . ( C-D ) Pulse-chase analysis of PSD95 palmitoylation in the presence of: ( C ) 10 μM PalmB , 10 μM WWL70 , or 20 μM C75; and ( D ) 10 μM PalmB or 20 μM RHC-80267 , as described in Figure 2 . Dashed lines represent cropping of single gels . , endogenous serine hydrolase activity unaffected by PalmB . DOI: http://dx . doi . org/10 . 7554/eLife . 11306 . 007 The set of candidates inhibited by both PalmB and HDFP ( Figure 3G , H ) includes ABHD6 , which associates with PSD95-containing complexes at synapses ( Schwenk et al . , 2014 ) , and FASN , which functions in palmitoyl-CoA synthesis ( Wakil , 1989 ) . However , treatment with the ABHD6 inhibitor WWL70 ( Li et al . , 2007 ) or the FASN inhibitor C75 ( Kuhajda et al . , 2000 ) did not alter PSD95 depalmitoylation ( Figure 3—figure supplement 1A , C ) . Palmitate turnover on PSD95 was also unaffected by RHC-80267 , which moderately inhibited ABHD6 and PNPLA6 ( Figure 3—figure supplement 1B , D; Hoover et al . , 2008 ) . Thus , ABHD6 , PNPLA6 , and FASN are unlikely to play a primary role in PSD95 depalmitoylation . Selective inhibitors that target the remaining four candidates have not been identified . Therefore , we used pulse-chase click chemistry to test if increased expression of these enzymes enhances palmitate turnover . High levels of ABHD16A , ABHD6 , or APT1/2 had little effect on N-Ras ( Figure 4A ) or PSD95 ( Figure 4—figure supplement 1A ) depalmitoylation . Strikingly , however , expression of ABHD17A , ABHD17B , or ABHD17C accelerated palmitate cycling on these proteins ( Figure 4A , Figure 4—figure supplement 1A ) , strongly suggesting the uncharacterized ABHD17 family of mSHs are novel protein depalmitoylases . 10 . 7554/eLife . 11306 . 008Figure 4 . ABHD17A expression promotes N-Ras depalmitoylation and alters N-Ras subcellular localization . ( A ) Pulse-chase analysis of N-Ras co-expressed with candidate mSHs as described in Figure 1 . n = 3 , mean ± SEM . ( B ) Schematic of the ABHD17A wild type , catalytically-inactive ( S211A ) , and N-terminal truncation ( ΔN ) mutant proteins used in this study . ( C ) ABPP of ABHD17A wild type and mutant proteins by in-gel fluorescence ( FP-Rho ) . Western blot ( WB ) shows proteins expressed in each condition . Filled arrowheads: ABHD17A WT and S211A; Open arrowheads: ABHD17A ΔN . ( D ) Pulse-chase analysis of N-Ras co-expressed with ABHD17A wild type and mutant proteins as described in Figure 1 . n = 3 , mean ± SEM . ( E ) Representative live confocal images of EGFP-N-Ras-C181S and EGFP-N-Ras localization in COS-7 cells treated with 100 μM 2-bromopalmitate ( 2-BP ) or co-expressing the indicated thioesterases . Scale Bar = 10 μm . ( F ) Bar graph representing percentage of COS-7 cells with plasma membrane EGFP-N-Ras under each condition studied in ( E ) . n = 3 ( 100 cells counted per trial ) , mean ± SEM . p < 0 . 05; p < 0 . 01; **p < 0 . 0001 . mSHs , metabolic serine hydrolases; SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 11306 . 00810 . 7554/eLife . 11306 . 009Figure 4—figure supplement 1 . ABHD17 expression promotes PSD95 depalmitoylation . ( A ) Pulse-chase analysis of PSD95 co-expressed with candidate mSHs as described in Figure 4A . n = 3 , mean ± SEM . ( B ) Pulse-chase analysis of PSD95 co-expressed with ABHD17A wild type and mutant proteins as described in Figure 4D . n = 3 , mean ± SEM . p < 0 . 05 . mSHs , metabolic serine hydrolases; SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 11306 . 00910 . 7554/eLife . 11306 . 010Figure 4—figure supplement 2 . ABHD17A is localized to the plasma membrane and endosomal compartments . ( A ) Localization of ABHD17A wild-type protein with markers of early endosomes ( Rab5 ) , late endosomes ( Rab7 ) , recycling endosomes ( Rab11 ) , and the Golgi apparatus ( GM130 ) in COS-7 cells as determined by immunocytochemistry . Scale bar =10 μm . ( B ) Localization of ABHD17A ΔN in COS-7 cells relative to the Golgi marker GM130 by immunocytochemistry . Scale bar =10 μm . ( C ) Localization of mCherry-tagged ABHD17A wild type and mutant proteins co-expressed with EGFP-N-Ras in COS-7 cells by confocal microscopy . Scale bar =10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 11306 . 010 We focused on ABHD17A , which showed the strongest effect in promoting palmitate turnover on N-Ras and PSD95 . The ABHD17 proteins are targeted to membranes by a palmitoylated N-terminal cysteine cluster ( Kang et al . , 2008; Martin and Cravatt , 2009 ) . We found ABHD17A localized to the plasma membrane and to Rab5- and Rab11-positive endosomes ( Figure 4—figure supplement 2A ) . Mutation of the predicted active site serine ( S211A ) ( Figure 4B ) abolished ABHD17A activity ( Figure 4C ) but did not alter its localization ( Figure 4—figure supplement 2C ) , whereas removing the N-terminal amino acid residues 1-19 ( ∆N; Figure 4B ) shifted it to the cytosol ( Figure 4—figure supplement 2B , C ) and reduced its catalytic activity ( Figure 4C ) . Importantly , neither mutant stimulated N-Ras or PSD95 depalmitoylation ( Figure 4D , Figure 4—figure supplement 1B ) , suggesting both the catalytic activity and membrane localization of ABHD17A are functionally important . We next examined the cellular consequences of ABHD17A expression . Disrupting N-Ras palmitoylation by mutating the palmitoylated residue ( C181S ) or treating cells with the inhibitor 2-bromopalmitate ( 2-BP ) relocalized N-Ras from the plasma membrane to internal organelles , as previously described ( Choy et al . , 1999; Goodwin et al . , 2005 ) ( Figure 4E , F ) . Overexpression of APT1 or APT2 had little effect on N-Ras localization ( Figure 4E , F ) , consistent with a recent report ( Agudo-Ibáñez et al . , 2015 ) . In contrast , overexpression of ABHD17A , but not catalytically dead or cytosolic mutant forms , redistributed N-Ras from the plasma membrane to intracellular compartments consistent with its altered palmitoylation status ( Figure 4E , F ) . Taken together , these findings demonstrate the membrane-localized pool of ABHD17A depalmitoylates N-Ras and alters its subcellular targeting . To determine if the endogenous ABHD17 proteins regulate palmitate cycling in vivo , we investigated the effect of ABHD17 knockdown on N-Ras depalmitoylation in HEK293T cells . RT-qPCR ( Reverse transcription quantitative polymerase chain reaction ) showed efficient silencing of ABHD17A alone , or ABHD17A , ABHD17B , and ABHD17C in concert , after 72 hr with siRNA treatment ( Figure 5A ) . ABHD17A knockdown had a slight effect on N-Ras depalmitoylation ( p=0 . 084 ) . In contrast , N-Ras palmitate turnover was significantly inhibited when all three ABHD17 proteins were simultaneously downregulated ( p=0 . 0083 ) , and this was not further enhanced by the APT1 and APT2 inhibitors C83 and C115 ( Figure 5B ) . Knockdown was less effective than PalmB treatment , which could be due to activity of the residual ABHD17 enzymes . PalmB may also inhibit additional factors that either directly or indirectly affect N-Ras palmitate cycling . Taken together , these results demonstrate that ABHD17 proteins redundantly mediate palmitate turnover on N-Ras . 10 . 7554/eLife . 11306 . 011Figure 5 . Simultaneous knockdown of ABHD17 isoforms inhibits N-Ras palmitate turnover . ( A ) RT-qPCR of ABHD17A , ABHD17B , and ABHD17C transcript levels in HEK 293T cells treated with Non-Targeting siRNA ( ”NT” , black ) , ABHD17A siRNA alone ( ”A KD” , gray ) , or ABHD17A/ ABHD17B/ ABHD17C siRNAs ( ”Triple KD” , light gray ) for 72 hr . n = 3 , mean ± SEM . ( B ) Pulse-chase analysis of N-Ras palmitoylation in siRNA-transfected HEK 293T cells treated with vehicle ( DMSO ) , 10 μM C83 and C115 , or 10 μM PalmB as described in Figure 1 . n = 3 , mean ± SEM . p < 0 . 01; **p < 0 . 0001 . SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 11306 . 011 Our discovery that ABHD17 proteins are novel protein depalmitoylases expands the current repertoire of cellular APTs , and suggests depalmitoylation occurs in a substrate-selective and compartment-specific manner . Whereas APT1 and APT2 were proposed to act ubiquitously ( Rocks et al . , 2010; Vartak et al . , 2014 ) , ABHD17-mediated depalmitoylation of N-Ras at the plasma membrane may specifically attenuate oncogenic signaling pathways ( Song et al . , 2013 ) . ABHD17 proteins are also active in the brain ( Bachovchin et al . , 2010 ) , where palmitoylated PSD95 regulates AMPA ( α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid ) receptor nanodomain assemblies linked to synaptic plasticity ( Fukata et al . , 2013 ) . miRNA-138 targets APT1 to alter dendritic spine size ( Siegel et al . , 2009 ) , whereas the Caenorhabditis elegans ABHD17 homologue AHO-3 regulates starvation-induced thermotactic plasticity ( Nishio et al . , 2012 ) . Thus , functionally specialized APTs may prove to be critical modulators of palmitoyl-proteins in distinct cellular processes . The total number of cellular depalmitoylases is not known . We identified new PalmB targets , consistent with a recent report showing PalmB inhibits ABHD12 and monoacylglyerol lipase ( Savinainen et al . , 2014 ) . As the mSH superfamily consists of >110 members , only half of which are functionally annotated ( Simon and Cravatt , 2010 ) , a comprehensively survey the mSH proteome may uncover yet more depalmitoylases . APTs are a critical element of the dynamic palmitoylation cycle , thus it will be imperative to identify the complete set of cellular APTs and determine how they contribute to the regulation of dynamic palmitoylation . Plasmids expressing EGFP-N-Ras , PSD95-GFP , N-HTT-GFP , SNAP25-GFP were provided by Dr . Michael Hayden ( University of British Columbia ) . Plasmids expressing Myc-hAPT1 , GOLIM4-GFP , FLAG-SPRED2 , and GAD65-GFP were generous gifts from Dr . Takashi Izumi ( Gunma University ) , Dr . Adam Linstedt ( Carnegie Mellon University ) , Dr . Akihiko Yoshimura ( Keio University ) , and the late Dr . Alaa El-Husseini ( University of British Columbia ) , respectively . Venus-tagged Rab5 , Rab7 , and Rab11 plasmids were gifts from Dr . Nevin Lambert ( Georgia Regents University ) . EGFP-ITM2B was cloned by polymerase chain reaction ( PCR ) amplification of the ITM2B ORF ( open reading frame ) from MGC Fully Sequenced Human BRI3 cDNA , clone ID 3163436 ( OpenBiosystems; Mississauga , ON ) , using the forward primer 5’-ATTTAACCCGGGATGGTGAAGATTAGCTTCCAGCC-3’ and the reverse primer 5’-ATTTAAGGTACCTCACACCACCCCGCAGAT-3’ , followed by restriction digest and ligation with BspEI/KpnI-digested pEGFP-C3 vector from Clontech ( Mountain View , CA ) . EGFP-N-Ras-C181S was generated by Quikchange mutagenesis ( Stratagene; La Jolla , CA ) using the forward primer 5’-CAACAGCAGTGATGATGGTACCCAGGGTAGTATGGGATTGCCATGTGTGG-3’ and the reverse primer 5’-CCACACATGGCAATCCCATACTACCCTGGG TACCATCATCACTGCTGTTG-3’ with EGFP-N-Ras as the template . For cloning of mSHs for activity-profiling studies , plasmids containing corresponding human ORFs were purchased from DNASU ( Arizona State University , Tempe , AZ ) and OpenBiosystems , or obtained as clones from the hORFeome v8 . 1 Collection ( Yang et al . , 2011 ) . Genes of interest were amplified by PCR using oligos with flanking restriction sites ( described in Supplementary file 2 ) , and the resulting mSH-encoding PCR products were subcloned into vectors of interest ( FLAG-NT , generously provided by Dr . Stefan Taubert , University of British Columbia; or pCINeo , Promega [Madison , WI] ) . The ABHD17A-FLAG construct was used as the template to generate ABHD17A mutant and mCherry-tagged plasmids . S211A-FLAG in pCINeo was generated by Quikchange mutagenesis , and ABHD17A ∆N-FLAG was amplified by PCR then subcloned into pCINeo . ABHD17A-mCherry wild type and mutant plasmids were generated by pairing each forward oligo with the reverse ABHD17A-mCherry-Linker oligo as listed in Supplementary file 2 . The resulting ABHD17A fragments were fused with the PCR-amplified C-terminal mCherry cassette by overlapping extension PCR ( OEPCR ) and subcloned into pCINeo vector with EcoRI and XbaI . Similarly , mCherry-APT1 and mCherry-APT2 plasmids were constructed by fusing the N-terminal mCherry cassette with PCR-amplified APT1 and APT2 fragments using OEPCR and subcloning the resulting fragments into pCINeo vector with EcoRI and XbaI . The pSUPER vector and the shRNA pSUPER-APT1 plasmid used in knockdown studies was a generous gift from Dr . Gerhard Schratt ( University of Marburg ) , and ON-TARGETplus SMARTpool siRNAs targeting APT2 , ABHD17A , ABHD17B , or ABHD17C , as well as Non-Targeting control siRNA , were purchased from Dharmacon ( Lafayette , CO ) . Lipofectamine 2000 , Lipofectamine RNAiMax , sodium dedocyl sulfate ( SDS ) solution , L-azidohomoalanine ( L-AHA ) , Alexa Fluor 488-azide ( AF488-az ) , Alexa Fluor 647-alkyne ( AF647-alk ) , TRIzol reagent , and Prolong Gold Antifade Mountant with DAPI were purchased from Life Technologies ( Burlington , ON ) . X-tremeGENE 9 was purchased from Roche ( Indianapolis , IN ) . Palmostatin B was purchased from Merck Scientific ( Billerica , MA ) . Tris[ ( 1-benzyl-1H-1 , 2 , 3-triazol-4-yl ) methyl]amine ( TBTA ) , Tris ( 2-carboxyethyl ) phosphine hydrochloride ( TCEP ) , Triton-X 100 ( TX-100 ) , sodium deoxycholate , CuSO4 , palmitic acid , and 2-bromopalmitate were obtained from Sigma-Aldrich ( St . Louis , MO ) . 17-ODYA , C75 , WWL70 , and RHC-80267 were purchased from Cayman Chemical ( Ann Arbor , MI ) . HDFP , C83 , and C115 were gifts from Dr . Brent Martin ( University of Michigan ) , and FP-rhodamine was generously provided by Dr . Benjamin Cravatt ( Scripps Institute ) . COS-7 and HEK293T/17 cells from ATCC ( Manassas , VA ) were maintained and propagated in high-glucose Dulbecco’s Modified Eagle Medium ( DMEM ) supplemented with 10% fetal bovine serum ( FBS; Life Technologies ) , 4 mM L-glutamine and 1 mM sodium pyruvate , in a humidified incubator at 37°C , 5% CO2 . For pulse-chase metabolic studies and activity-based protein-profiling studies , COS-7 cells were transfected with cDNAs as indicated in each experiment using Lipofectamine 2000 as per manufacturer’s instructions . Cells were grown in six-well plates ( Corning; Corning , NY ) and transfected at 90% confluence with 1 μg of cDNA per well for pulse-chase analyses with inhibitors , or 2 μg cDNA per well for pulse-chase analyses with thioesterase overexpression . For immunofluorescence studies , COS-7 cells were grown on glass coverslips ( Fisher; Pittsburg , PA ) in 24-well plates ( Corning ) and transfected at 60–90% confluence with 0 . 5 μg of cDNA per well using Xtreme-GENE 9 according to product instructions . Experiments involving small molecules were carried out 20–24 hr following transfection , and experiments involving co-expression of candidate mSHs were carried out 24–48 hr post-transfection , as described below . For APT1 and APT2 studies , a double knockdown approach was used ( Bond et al . , 2011 ) where COS-7 cells were transfected with siRNA ( 100 nM final concentration per transfection ) on days 1 and 3 with 5 μL of Lipofectamine 2000 per transfection . One microgram of cDNA was co-transfected with the siRNA on day 3 , and pulse-chase studies were carried out on day 4 , 20 hr following the co-transfection . For ABHD17 studies , HEK293T cells were transfected on day 1 with siRNA in 9 μL Lipofectamine RNAiMax , and on day 3 with 1μg of EGFP-N-Ras in 4 μL Lipofectamine 2000 . Pulse-chase and RT-qPCR studies were performed on day 4 , 20 hr following cDNA transfection . Twenty hours following transfection , COS-7 cells or HEK293T cells were washed twice in phosphate-buffered saline ( PBS ) and starved in cysteine- and methionine-free DMEM containing 5% charcoal-filtered FBS ( Life Technologies ) for 1 hr . Cells were then labeled with 30 μM 17-ODYA and 50 μM L-AHA for 1 . 5 hr in this media . The labeling media was removed , and cells were briefly washed twice in PBS before chasing in complete DMEM supplemented with 10% FBS and 300 μM palmitic acid . Small molecule inhibitors or DMSO ( vehicle ) were added at chase time 0 . At indicated time points , cells were washed twice in PBS and lysed with 500 μL triethanolamine ( TEA ) lysis buffer ( 1% TX-100 , 150 mM NaCl , 50 mM TEA pH 7 . 4 , 2×EDTA-free Halt Protease Inhibitor [Life Technologies] ) . The lysates were transferred to 1 . 5 mL Eppendorf tubes ( Corning ) , vigorously shaken ( 3 × 20s ) while placed on ice in between each agitation . Lysates were cleared by centrifugation at 16 , 000× g for 15 min at 4°C . Solubilized proteins in the supernatant were quantified using Bicinchoninic acid ( BCA ) assay ( Life Technologies ) and subsequently used for immunoprecipitations as described below . For immunoprecipitations , Protein A or Protein G sepharose beads ( GE Healthcare; Mississauga , ON ) were washed thrice in TEA lysis buffer . Protein A beads were pre-incubated with rabbit anti-GFP antibodies ( Life Technologies ) and Protein G beads were pre-incubated with FLAG M2 antibodies ( Sigma-Aldrich ) for 2 hr at 4°C , before the addition 500 μg – 1 mg of transfected COS-7 cell lysates containing indicated proteins . Immunopreciptations were carried out for 12–16 hr on an end-to-end rotator at 4°C . Following immunoprecipitation , sepharose beads were washed thrice in modified RIPA buffer ( 150 mM NaCl , 1% sodium deoxycholate ( w/v ) , 1% TX-100 , 0 . 1% SDS , 50 mM TEA pH7 . 4 ) before proceeding to sequential on-bead CuAAC/click chemistry . Sequential on-bead click chemistry of immunoprecipitated 17-ODYA/L-AHA-labeled proteins was carried out as previously described ( Zhang et al . , 2010 ) , with minor modifications . After immunoprecipitation , sepharose beads were washed thrice in RIPA buffer , and on-bead conjugation of AF488 to 17-ODYA was carried out for 1 hr at room temperature in 50 μL of freshly mixed click chemistry reaction mixture containing 1 mM TCEP , 1 mM CuSO4·5H2O , 100 μM TBTA , and 100 μM AF488-az in PBS . After three washes in 500 μL RIPA buffer , conjugation of AF647 to L-AHA was carried out for 1 hr at room temperature in 50 μL click-chemistry reaction mixture containing 1 mM TCEP , 1 mM CuSO4·5H2O , 100 μM TBTA , and 100 μM AF647-alk in RIPA buffer . Beads were washed thrice with RIPA buffer and resuspended in 10 μL SDS buffer ( 150 mM NaCl , 4% SDS , 50 mM TEA pH7 . 4 ) , 4 . 35 μL 4× SDS-sample buffer ( 8% SDS , 4% Bromophenol Blue , 200 mM Tris-HCl pH 6 . 8 , 40% Glycerol ) , and 0 . 65 μL 2-mercaptoethanol . Samples were heated for 5 min at 95°C , and separated on 10% tris-glycine SDS-PAGE gels for subsequent in-gel fluorescence analyses . Twenty-four hours following transfection with mSH constructs , COS-7 cells were washed twice in PBS , transferred to a new vial by scraping in PBS , and lysed by gentle sonication on ice . Protein was quantified by BCA assay . Thirty micrograms of total protein was incubated either with DMSO or small molecule inhibitors at indicated concentrations at room temperature for 30 min , prior to the addition of FP-Rho ( 2 μM final concentration ) . Labeling reactions were carried out at room temperature for 1 hr and quenched with 4× SDS-sample buffer heated to 95°C for 5 min . Samples were separated on SDS–PAGE , analyzed by in-gel fluorescence , then transferred onto nitrocellulose membrane for Western blotting . A Typhoon Trio scanner ( GE Healthcare ) was used to measure in-gel fluorescence of SDS–PAGE gels: AF488 signals were acquired using the blue laser ( excitation 488 nm ) with a 520BP40 emission filter , AF647 signals were acquired using the red laser ( excitation 633 nm ) with a 670BP30 emission filter , and rhodamine signals were acquired with the green laser ( excitation 532 nm ) , with a 580BP30 emission filter . Signals were acquired in the linear range and quantified using the ImageQuant TL7 . 0 software ( GE Healthcare ) . For pulse-chase analyses , the ratio of palmitoylated substrates were calculated as AF488/AF647 values at each time point , normalized to the value at T=0 . Nitrocellulose membranes were blocked with PBS with 0 . 1% Tween-20 ( PBST ) containing 3% bovine serum albumin ( BSA , Sigma ) for 1 hr , and incubated with primary antibodies ( rabbit anti-GFP , 1:1 , 000; or mouse anti-FLAG M2 , 1:1 , 000 ) in PBST + 3% BSA for 2 hr , followed by 3x15 min washes with PBST + 0 . 3% BSA . Membranes were then incubated with secondary antibodies ( IRDye 800CW goat anti-mouse IgG , 1:10 , 000; or IRDye 680RD goat anti-rabbit IgG , 1:10 , 000 ) ( LI-COR Biosciences; Lincoln , NE ) in PBST + 0 . 3% BSA for 1 hr . After three washes in PBST , membranes were imaged using the LI-COR Odyssey Scanner ( LI-COR ) . Signals were acquired in the linear range using the 680nm and 800nm lasers and quantified using the Image Studio software ( LI-COR ) . COS-7 cells were co-transfected with EGFP-N-Ras and empty vector or indicated mCherry-tagged thioesterases at a 1:1 ratio ( total 0 . 5 μg DNA per well ) in Lab-Tek 8-well chamber slides ( Fisher ) . Twenty-four hours post-transfection , cells were imaged on a TCS SP8 confocal laser scanning microscope ( Leica Microsystems; Mannheim , Germany ) , and EGFP-N-Ras localization was quantified by counting 100 cells per experiment . Twenty hours post-transfection , cells were washed twice with PBS , and fixed in 4% paraformaldehyde ( PFA ) solution ( 4% PFA , 4% sucrose in PBS ) for 20 min . Cells were permeabilized for 1 min in PBS containing 0 . 1% TX-100 , washed thrice in PBS , and blocked with PBS +3% BSA for 60 min before incubating with primary antibodies ( mouse anti-FLAG-M2 , 1:500; rabbit anti-FLAG ( Sigma ) , 1:200; or mouse anti-GM130 ( BD Biosciences; San Jose , CA ) , 1:200 ) for 1 hr . Coverslips were washed thrice and incubated with secondary antibodies ( goat anti-mouse Alexa Flour 488 and goat anti-rabbit Alexa Fluor 594 ( Life Technologies ) , 1:1000 each ) for an hour . Coverslips were washed with PBS and mounted on glass slides with ProLong Gold Antifade Mountant containing DAPI . Cells were observed with an Axioplan 2 fluorescence microscope ( Carl Zeiss; Oberkochen , Germany ) using a Plan-Apochromat 100× 1 . 40 NA oil immersion objective lens . Images were acquired with a CoolSNAP camera ( Roper Scientific; Planegg , Germany ) using YFP , GFP , and Texas Red filters and MetaMorph 7 . 7 software ( MDS analytical Technologies; Toronto , ON ) , and adjusted using Metamorph 7 . 7 . Seventy-two hours post-transfection with siRNA pool ( s ) , HEK293T cells were collected in 1 mL TRIzol reagent . Samples were snap-frozen at -80°C until used . Total RNA extraction was carried out with PureLink RNA Mini kit ( Life Technologies ) following manufacturer instructions . For each sample , 1 μg of RNA was used to synthesize cDNA with QuantiTect Reverse Transcription Kit ( Qiagen; Hilden , Germany ) . RT-qPCR was performed in 15 μL reactions using a Rotor-Gene 6000 ( Qiagen ) and PerfeCTa SYBR Green FastMix ( Quanta Biosciences; Gaithersburg , MD ) with gene-specific primer pairs listed in Supplementary file 3 . ABHD17 mRNA levels were determined by the ΔΔCt method normalizing to β-actin mRNA levels . PCR efficiencies of primers were examined by standard curve of serial-diluted untreated whole cell samples . Statistical analyses were carried out by performing Student’s two-tailed t-tests using Prism 6 ( GraphPad Software , Inc . , La Jolla , CA ) , with DMSO-treated ( Figure 2 and Figure 3 ) , vector-co-transfected ( Figure 4 ) , or Non-targeting siRNA-transfected ( Figure 5 ) samples as the control group . All significant differences ( p< 0 . 05 ) are indicated in the figures .	Proteins play important roles in many processes in cells . Some of these proteins can be modified by the addition of a molecule called palmitate . This process , termed “palmitoylation” , helps direct these proteins to the compartments within the cell where they are needed to carry out their roles . One target of palmitoylation is N-Ras , which is a protein that can promote the development of cancer . We understand quite a lot about how palmitate is added to proteins , but much less about how it is removed . So far , researchers have only identified two enzymes – known as APT1 and APT2 – that can remove palmitate from proteins , but it is possible that there are others . Identifying other “depalmitoylase” enzymes could help us find ways to block the removal of palmitate from N-Ras , which could lead to new treatments for some cancers . Lin and Conibear used several biochemical techniques to search for depalmitoylase enzymes in human cells . The experiments reveal that although APT1 and APT2 are important for removing palmitate from some proteins , they are not needed to remove palmitate from N-Ras . Instead , Lin and Conibear found that an enzyme called ABHD17 removes palmitate from N-Ras . The next step following on from this work will be to find out what other proteins ABHD17 acts on in cells . A longer-term challenge will be to develop specific chemicals that inhibit ABHD17 activity and test if they are able to reduce the growth of cancer cells .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "short", "report", "biochemistry", "and", "chemical", "biology" ]	2015	ABHD17 proteins are novel protein depalmitoylases that regulate N-Ras palmitate turnover and subcellular localization
Notch signaling regulates cell fate selection during development in multiple organs including the lung . Previous studies on the role of Notch in the lung focused mostly on Notch pathway core components or receptor-specific functions . It is unclear , however , how Jagged or Delta-like ligands collectively or individually ( Jag1 , Jag2 , Dll1 , Dll4 ) influence differentiation of airway epithelial progenitors . Using mouse genetic models we show major differences in Jag and Dll in regulation and establishment of cell fate . Jag ligands had a major impact in balancing distinct cell populations in conducting airways , but had no role in the establishment of domains and cellular abundance in the neuroendocrine ( NE ) microenvironment . Surprisingly , Dll ligands were crucial in restricting cell fate and size of NE bodies and showed an overlapping role with Jag in differentiation of NE-associated secretory ( club ) cells . These mechanisms may potentially play a role in human conditions that result in aberrant NE differentiation , including NE hyperplasias and cancer . Notch signaling is a major regulator of progenitor cell fate and differentiation during organogenesis , repair-regeneration , and cancer . In mammals , four Notch receptors ( Notch1–4 ) and five ligands ( Delta-like: Dll1 , Dll3 and Dll4 and Jagged: Jag1 and Jag2 ) have been described . All ligands , except Dll3 , are Notch activating . Signaling is triggered by ligand-receptor binding through cell-cell interactions , which leads to sequential cleavage of the Notch receptor and binding of its intracellular domain ( NICD ) to a CSL/RBPJk-activator complex for activation of downstream target genes , such as HEY/HES-family members ( Radtke and Raj , 2003; Bray , 2006 ) . While different Notch receptors are known to act in a variety of biological processes , evidence from genetic studies suggest that the Notch effects are not necessarily dependent on the type of NICD but rather of NICD dosage ( Liu et al . , 2015 ) . Notably , specific Notch ligand-receptor binding in mammalian cells appears to be mostly non-selective or context-dependent . Interestingly , systemic deletion of Jag1 , Jag2 , or Dll4 has been shown to result in distinct phenotypes , suggesting that these ligands could mediate unique functions not entirely due to the receptor they activate ( D'Souza et al . , 2009; Choi et al . , 2009 ) . Indeed , Notch ligands were reported to activate distinct targets even through binding to the same Notch receptor and ligand-specific effects have been observed in multiple contexts ( Nandagopal et al . , 2018 ) . The Notch pathway plays a crucial role in the developing lung . When airways are still forming epithelial progenitors initiate a differentiation program that gives rise to secretory ( club , goblet ) , multiciliated , and neuroendocrine ( NE ) cells . Previous studies addressing the role of Notch in the lung focused largely on central components of this pathway ( Rbpjk , Pofut1 , and Hes1 ) . Disruption of Rbpjk or the o-fucosyl-transferase Pofut1 required for Notch signaling results in aberrant expansion of multiciliated and NE cells at the cost of secretory cells ( Tsao et al . , 2009; Tsao et al . , 2011; Morimoto et al . , 2010 ) . Subsequent studies showed that club cells are more sensitive to deficiency in Notch2 while Notch 1-3 receptors contribute to control the NE population in an additive manner ( Morimoto et al . , 2012 ) . However , it was unclear whether individual ligand families ( Delta-like and Jagged ) or specific ligands ( Dll1 , Dll4 , Jag1 , and Jag2 ) influence distinct aspects of differentiation of airway epithelial progenitors . Notably , these ligands have been reported in partially overlapping but also distinct domains in the lung ( Post et al . , 2000; Kong et al . , 2004; Tsao et al . , 2009; Xu et al . , 2010b; Zhang et al . , 2013; Mori et al . , 2015 ) . Here we explored the role of ligands using single and double conditional Jagged and Delta-like null alleles targeted to epithelial progenitors from early lung development . We show remarkably distinct roles of these ligands in the developing intra- and extrapulmonary airways and in the control of the expansion and differentiation of the NE microenvironment . Although the expression patterns of Jag and Dll have been reported in both epithelial and mesenchymal layers of the developing lung , specific information about their onset of expression and regional distribution in the epithelial compartment at early stages of differentiation has been scattered and not well integrated to functional studies ( Post et al . , 2000; Kong et al . , 2004; Xu et al . , 2010b; Morimoto et al . , 2012; Tsao et al . , 2009 ) . To gain further insights into this issue we revisited the spatial and temporal pattern of expression of Notch ligands when epithelial cells are initiating commitment to different cell fates in developing airways . By in situ hybridization ( ISH ) analysis none of these ligands were detectable in the airway epithelium prior to or at E11 . 5 ( not shown ) . However , at around E12 . 0 evidence of Jag2 epithelial signals in the developing trachea made it the first of all Notch ligands to be induced in the differentiation program of airways ( Figure 1A ) . Expression progressed in a proximal-to-distal fashion; at E12 . 5 low level signals were detected in the epithelium of extrapulmonary but not intrapulmonary airways . This contrasted with the strong Jag2 signals present in the esophageal epithelium and in neighboring vascular structures ( Figure 1B ) . Notably , the Jag2 detection in epithelial progenitors of the trachea and extrapulmonary airways coincided with the previously reported onset of Notch activation and appearance of the secretory cell marker Scgb3a2 locally ( Guha et al . , 2012 ) . No epithelial Jag1 could be detected anywhere in airways at these stages , although clearly present in vascular structures ( Figure 1C ) . These data supported the idea of a Jag2-Notch program giving rise to secretory cell precursors as one of the earliest events initiating differentiation in airways , even preceding the appearance of pulmonary NE cells ( PNEC ) reported to begin only within a day later ( Li and Linnoila , 2012; Kuo and Krasnow , 2015; Noguchi et al . , 2015; Sui et al . , 2018 ) . Indeed , expression of Ascl1 , which marks PNEC precursors , was first found ~E13-13 . 5 in large intrapulmonary airways and both Dll1 and Dll4 were then subsequently expressed in these precursors ( Figure 1E , D ) . By E13 . 5-E14 . 5 strong Jag2 epithelial signals were seen throughout the trachea and main bronchi , in contrast to Jag1 , nearly undetected at these sites ( Figure 1 , Figure 1—figure supplement 1 ) . At E14 . 5 NEBs and PNECs were sharply demarcated by Ascl1 , and Dll1 and Dll4 transcripts became prominently expressed in NEBs ( Figure 1E–F ) . This was accompanied by the appearance of clusters of cells adjacent to NEBs , collectively marked by expression of the uroplakin Upk3a , the cell surface stem cell ( secretory ) marker SSEA1 , the secretoglobin Scgb3a2 and low levels of the cytochrome gene Cyp2f2 and CC10 . The pattern was consistent with the initiation of a Notch-dependent program of secretory cells in the NEB microenvironment ( Guha et al . , 2012 and described later ) . Thus , Jag and Dll ligands appear in different domains and in a sequential proximal-distal fashion during the establishment of cell fate in airway progenitors , initiating with Jag2 in the trachea , Jag1 , and lastly Dll1 and Dll4 once NEBs form in intrapulmonary airways ( summary diagram Figure 1G ) . Given the distinct timing and spatial distribution of Jag ligands described above , we reasoned that common but also non-overlapping functions were likely to exist in the distinct domains of the respiratory tract . Inactivation of Jag1 in epithelial progenitors of intrapulmonary airways undergoing branching morphogenesis using a surfactant protein-c ( Sftpc ) -tet-O system was shown to disrupt epithelial differentiation ( Zhang et al . , 2013 ) , confirming the previously reported role of Notch signaling in this process ( Tsao et al . , 2009; Morimoto et al . , 2010 ) . Although efficient , in this targeting strategy Cre-mediated recombination was restricted to intrapulmonary airways , initiating at the onset of Sftpc expression in secondary buds ( ~E10 . 5 ) . Thus , information about a putative role of Jag1 in extrapulmonary airways ( trachea , main bronchi ) and at stages prior to E10 . 5 was missing . Moreover , little was known about how Jag2 influences lung development and whether there is any functional overlap between Jag2 and Jag1 . Jag2 systemic knockout animals die at birth ( Jiang et al . , 1998 ) . Lastly , no information was available about compensation of Jag by other ligands during epithelial differentiation . We used the Shh-cre line to inactivate Jag1 and Jag2 individually or in combination in early epithelial progenitors of both extrapulmonary and intrapulmonary at the onset of lung development ( Harfe et al . , 2004 ) . Jag1flox/flox; Shhcre/+ ( Jag1cnull ) , Jag2flox/flox; Shhcre/+ ( Jag2cnull ) and double ( Jag1cnull; Jag2cnull ) null mutants were analyzed at early ( E14 . 5 ) and late ( E18 . 5 ) stages of airway differentiation . Gross morphological analysis of the mutant lungs showed no notable macroscopic difference in size or shape ( not shown ) . We compared the effects of Jag1 and Jag2 loss in multiciliated-secretory cell fate selection at E18 . 5 , once differentiated cell profiles were largely established in extrapulmonary ( trachea ) and intrapulmonary ( lung ) airways . qPCR analysis of E18 . 5 lung homogenates showed significant changes in markers of epithelial differentiation in all mutants ( Figure 2A ) . Expression of the secretory markers Scgb3a2 and Scgb1a1 ( encoding CC10 ) were reduced by 86 . 2% ( p=5×10−12 ) and 85 . 6% ( p=0 . 0001 ) , respectively in Jag1cnull mutants , but only by 25 . 9% ( p=0 . 015 ) and 34 . 4% ( p=0 . 019 ) , respectively in Jag2cnull mutants . These changes were accompanied by a significant increase in Foxj1 expression in Jag1cnull ( 183% increase , p=0 . 0006 ) but not in Jag2cnull ( 13% reduction , p=0 . 501 ) mutants . Thus , the differentiation program of intrapulmonary airways was more severely affected in Jag1cnull than in Jag2cnull mutants . The predominant contribution of Jag1 to the program of secretory cell fate as represented by these markers could be clearly seen in double Jag1cnull; Jag2cnull mice . These mutants showed Scgb3a2 and Scgb1a1 nearly abolished and an increase in Foxj1 similar to that found in Jag1cnull . Altogether these results indicated that airway progenitors are largely dependent on Jag ligands to initiate secretory cell differentiation . Immunofluorescence of Foxj1 and CC10 in E18 . 5 lung sections confirmed the changes in gene expression in intrapulmonary airways revealed by qPCR and showed secretory cells less abundant in Jag1cnull compared to Jag2cnull mutants ( Figure 2B ) . Interestingly , multiciliated cell fate appeared to be minimally affected in Jag2cnull airways . Morphometric analysis showed no significant change in the number of Foxj1+ cells in Jag2cnull airways relative to control ( p=0 . 164 ) , in contrast to the ~2 . 5 fold increase in these cells in Jag1cnull mutants ( p=4 . 17×10−7 ) ( Figure 2C ) . To search for potential reasons contributing to the more severe ciliated cell phenotype in Jag1cnull compared to Jag2cnull mutants we further extended our analysis of Notch ligands to later developmental stages ( Figure 1—figure supplement 1 ) . Interestingly , ISH of E14 . 5- E18 . 5 WT lungs showed that Jag2 expression in the trachea and extrapulmonary airways continued to be robust at later stages while remaining weak and scattered in intrapulmonary bronchial epithelia . By contrast , Jag1 expression was progressively stronger in the intrapulmonary airway epithelium from E15 . 5 onwards and by E18 . 5 expression extended to the distal bronchioles . Double ISH-immunohistochemistry for Foxj1 confirmed our previous report of Jag1 localization in multiciliated cells ( Tsao et al . , 2009 ) . Thus , our data suggested that Jag1 is the predominant ligand in intrapulmonary airways mediating secretory vs . multiciliated cell fate choice during differentiation . Consequently , Jag1cnull mutants were expected to display the unbalanced abundance of multiciliated cells in intrapulmonary airways compared to Jag2cnull mutants . We also reasoned that in Jag2cnull mutants Notch-mediated signaling by Jag1 should compensate for the loss of Jag2 , preserving the balance of multiciliated vs . secretory cell differentiation in these mutants . Indeed , analysis of Jag2cnull mice confirmed the presence of Jag1 signals in the intrapulmonary airway epithelium of these mutants ( Figure 2D ) . IF staining of Krt5 and β-tubulin in E18 . 5 Jag1cnull; Jag2cnull double mutants revealed an expansion in the population of basal progenitors and a variable but also increased population of multiciliated cells ( Figure 2—figure supplement 1A ) . Single Jag1cnull or Jag2cnull mutants were then examined to assess the contribution of each Jag ligand to the double mutant phenotype . For morphometric analysis we first performed IF for p63 and Foxj1 , which label basal and multiciliated cells , respectively , and have the advantage of displaying nuclear signals , thus facilitating quantitation . Tracheal sections from all groups at E18 . 5 were analyzed and the %p63+/DAPI and %Foxj1+/DAPI were estimated . This analysis revealed that Jag2cnull tracheas had a significant increase in the number of prebasal cells ( %p63+/DAPI ) but a trend ( not significant ) towards an increase in the abundance of the multiciliated cell population ( Figure 2—figure supplement 1B , C ) . Notably , double ISH/immunohistochemistry for Jag2/p63 in WT confirmed the presence of Jag2 in prebasal cells of the developing trachea ( Figure 1—figure supplement 1; Mori et al . , 2015 ) . As expected from the distinct spatial distribution of these Jag ligands , single Jag1cnull tracheas had no detectable change in the number of basal cell progenitors compared to controls . Together these data suggest that Jag ligands mediate overlapping but distinct events along the respiratory tract epithelium . In extrapulmonary airways ( trachea here ) Jag2 predominantly contributes to the balance of basal versus luminal cells while Jag1 controls abundance of multiciliated cells . By contrast , in intrapulmonary airways Jag1 is the predominant ligand regulating the balance of multiciliated versus club cell fate with a lesser contribution of Jag2 . Next we investigated whether Jag ligands could influence cell fate events that ultimately regulate the PNEC pool in intrapulmonary airways , regardless of its organization as NEBs or as solitary cells . Thus , we compared levels of Ascl1 expression in homogenates of Jag-cnull mutant lungs at E18 . 5 , when NEBs are already widely distributed at branch point and internodal locations . qPCR analysis showed no difference in Ascl1 expression between controls and mutants in any of the Jag-deficient airways ( Figure 3A ) . Consistent with this , immunofluorescence for Cgrp , another established marker of PNEC fate ( Li and Linnoila , 2012 ) , did not reveal consistent differences in expression patterns , suggestive of alterations in NEBs' spatial distribution , size ( ~8–10 PNECs per NEB control , Jag1cnull , Jag2cnull , and double Jag1cnull; Jag2cnull ) or frequency in intrapulmonary airways of mutants compared to controls ( Figure 3A–D ) . Since in our mutants Jag is deleted well before epithelial progenitors differentiate , we concluded that Jag-mediated Notch signaling is unlikely to be involved in the mechanisms that initiate or restrict the domains of NEB or PNEC fate . Neither NEBs nor PNECs could be identified in extrapulmonary airways ( trachea ) of mutants or control animals . We then examined the effect of Jag deletion in the population of secretory progenitors tightly associated with the developing NEBs . Previous studies have shown that they arise around E14 . 5 immediately adjacent to Ascl1-expressing cell clusters , being distinguished from other secretory precursors collectively by their expression of Upk3a , SSEA1 , Scgb3a2 , and low levels of Cyp2f2 and CC10 ( Guha et al . , 2012; Morimoto et al . , 2012 ) . Immunofluorescence of E18 . 5 lung sections triple-labeled with SSEA1 , Scgb3a2 , and Cgrp identified the typical SSEA1+ Scgb3a2+ cells around Cgrp+ clusters similarly preserved in intrapulmonary airways of Jag1cnull and Jag2cnull mutants ( Figure 3B–D ) . This contrasted with the nearly absent expression of these markers outside the NEB microenvironment in Jag1cnull; Jag2cnull mutants ( Figure 3B–C , described above ) . Remarkably , in double Jag1cnull; Jag2cnull airways the only population of cells expressing secretory markers was that associated with NEBs ( Figure 3C , D ) . This NEB-associated cell population was heterogeneous in regards to expression of the markers above even in the same airway , as also observed in control lungs ( Figure 3C ) . Notably , double Jag1cnull; Jag2cnull showed no evidence of change in the size of this population relative to NEBs . Since these cells are crucially dependent on Notch signaling and their only source of ligand should be Dll1 and/or Dll4 from PNECs , we performed N1ICD immunofluorescence to examine the status of Notch activation locally . Strong N1ICD labeling was found selectively in the SSEA1+ NEB-associated cell populations of mutants , indistinguishable from controls ( Figure 3D ) . Lastly , qPCR analysis of Upk3a , a gene marker highly enriched in NEB-associated secretory cells but that also labels scattered secretory cells of intrapulmonary airways , showed markedly decreased levels of expression in double Jag1cnull; Jag2cnull lungs ( Figure 3E ) . This was consistent with the dependence of Upk3a on Notch signaling we previously reported ( Guha et al . , 2012 ) . The similar levels of Ascl1 and the Cgrp expression pattern we found in controls and double Jag mutants ( Figure 3A ) suggested that the NEB size and frequency in airways is not dependent on Jag ligands . Thus we reasoned that the remaining population of Upk3a-expressing cells in Jag1cnull; Jag2cnull mutants was associated with NEBs and that the significant decrease in Upk3a seen by qPCR resulted from the loss of the Upk3a-expressing scattered cell population . Indeed , ISH of Upk3a in these mutants showed signals restricted to branch points in proximal regions of intrapulmonary airways associated with Ascl1-expressing NEBs ( Figure 3F ) . This was further supported by the evidence of a preserved NEB-associated SSEA1 population ( Figure 3B–D ) and the fact that Upk3a requires Notch activation not present in the extensive areas devoided of Jag ligands . Together these results suggested that Jag1 and Jag2 have overlapping but also distinct roles in the cell fate specification of respiratory lineages in extrapulmonary and intrapulmonary airways . Jag ligands , however , appear to be dispensable for activation of Notch and induction of NEB-associated secretory cells since they can utilize Dll provided by their neighboring NE cells . Moreover , our data show no evidence that Jag ligands have any impact in regulating size or frequency of NEBs . Our analysis of Jag1cnull; Jag2cnull mutants identified seemingly self-contained units comprised of Dll-expressing NEBs and immediately adjacent cells able to activate and maintain robust Jag-independent Notch signaling for local secretory differentiation . Previous studies in Ascl1-/- mice showed that these units were strictly dependent on the presence of NEBs ( Guha et al . , 2012 ) . Questions remained whether preventing NEBs from expressing Dll ligands would have any impact on the NEB microenvironment or elsewhere if Jag ligands were still expressed . Unlike Jag1 and Jag2 , found in largely non-overlapping spatial and temporal patterns , Dll1 and Dll4 are collectively expressed in a very restricted fashion to PNEC/NEBs . Given the high probability of functional overlap , we generated mouse mutants in which both Dll ligands were deleted conditionally in the developing lung epithelium . Double deletion ( Dll1cnull; Dll4cnull ) was achieved from early stages using a similar targeting strategy with a Shhcre/+ line . IF analysis of E14 . 5 lungs from control mice showed the solitary PNECs and distinct small clusters of Ascl1+ cells in the epithelium of large intrapulmonary airways ( bronchi ) characteristic of the NEBs . By contrast , Dll1cnull; Dll4cnull E14 . 5 showed a striking expansion in the population of Ascl1+ cells ( Figure 4A ) . Although individual Ascl1+ cell clusters could still be identified , they often seem to coalesce in large patches to form a nearly continuous layer of Ascl1+ cells . In spite of the distribution in wider domains , these cells were not found ectopically in extrapulmonary airways or in the most distal airways undergoing branching morphogenesis . The large patches of Ascl1+ cells were identified at branch points in E18 . 5 lungs and by then co-expressed Cgrp , indicating their continued differentiation ( Figure 4B ) . Thus , loss of Dll ligands expanded the Ascl1+ pool of PNEC/NEB precursors and did not prevent these cells from initiating maturation . Interestingly , Ki67 staining showed no difference in labeling associated with the Ascl1-expressing cells between control and double mutants at E18 . 5 or E14 . 5 ( Figure 4C ) . The data supported the idea that the NEB expansion found in Dll1cnull; Dll4cnull mutants did not result from an increase in proliferation . To assess the contribution of each of these ligands to the Dll1cnull; Dll4cnull phenotype , we examined Dll1cnull and Dll4cnull individual mutants . E18 . 5 lungs were isolated from single and double mutants and changes in expression of Ascl1 and Cgrp were analyzed by qPCR in homogenates ( Figure 4D ) . Double Dll mutants showed a significant increase in these transcripts compared to controls ( Ascl1 p=1 . 1×10−6; Cgrp p=5 . 9×10−5 ) , consistent with the aberrant NE cell expansion . However , in single Dll1cnull or Dll4cnull lungs Ascl1 expression was modestly increased only in Dll1cnull and Cgrp mRNA was not altered in either of these mutants compared to controls . The marked difference in phenotype between double and single Dll mutants suggested functional redundancy between Dll1 and Dll4 in controlling NEB or PNEC-associated events . To better understand these events , we performed morphometric analysis of the NE compartment in E14 . 5 lungs to determine the impact of Dll in the size and frequency of NEBs and PNECs ( Figure 4E ) . Quantitation of the number of solitary PNECs in the airway epithelium showed no difference between controls and any of the single or double mutants , suggesting that Dll disruption affected primarily the NEB microenvironment . The frequency of NEBs per airway ( % ) was largely unaffected , although a small difference in Dll4 mutants reached statistical significance . However , the number of PNECs per NEB was significantly increased in both the Dll1cnull; Dll4cnull and single Dll4cnull airways , indicating that the size of NEBs was dramatically altered in these mutants . Together the data indicated that the mechanisms that restrict PNEC fate and limit expansion of NEB were severely disrupted in Dll mutants . The strikingly preserved integrity of the NEB microenvironment of double Jag1cnull; Jag2cnull mutants led us to hypothesize that Dll1 and Dll4 were not only necessary and sufficient to activate local Notch signaling but also endowed the unique features of the NEB-associated club cells that distinguish them from club cells elsewhere . The absence of Dll ligands in the expanded population of NEB from double Dll mutants provided an opportunity to examine this issue . We asked whether the robust activation of Notch signaling seen in NEB-associated club cells ( CCs ) of control and Jag double mutants was also present in Dll1cnull; Dll4cnull mice . Double IF for Ascl1 and N1ICD in E18 . 5 lung sections showed strong N1ICD signals in the NE-associated CCs of mutants indistinguishable from that of controls ( Figure 5A ) . Notably , the NEB expansion in Dll1cnull; Dll4cnull mutants was accompanied by a respective expansion of the NEB-associated CCs . The identity of these cell populations was further confirmed by expression of Cgrp ( NEB ) as well as SSEA1 and low Cyp2f2 ( NEB-associated CCs ) . Double ISH/immunohistochemistry showed the characteristic low levels of Cyp2f2 expression in NEB-associated cells in contrast to the strong signals outside the NEB microenvironment ( Figure 5B ) . The aberrant expansion of the NEB-associated cells was further demonstrated by qPCR analysis of lung homogenates , which showed a significant increase in expression of Upk3a in Dll1cnull; Dll4cnull mutants compared to controls ( Figure 6A ) . Of note , we found no change in expression of markers not directly associated with the NEB microenvironment , such as Scgb1a1 ( CC10 ) or Foxj1 , which suggested that the Jag ligands present in double Dll1cnull; Dll4cnull lungs were capable of activating Notch and mediating the balance of secretory-ciliated cell differentiation ( Figure 6A ) . Lastly , ISH of E18 . 5 lungs confirmed the marked expansion in the domain of expression of Upk3a in Dll1cnull; Dll4cnull mice and their association with NEBs ( Figure 6B–C ) . Together the data strongly suggested that , in spite of the inability to express Dll1 and Dll4 , key features of the NEB microenvironment are preserved in these mutants by Jag ligand activation of Notch signaling . Here we provide evidence for distinct roles of Notch ligands once epithelial progenitors initiate differentiation . We show that Jagged-driven Notch signaling differentially regulates cell type-specific programs of cell fate in a temporal and spatial fashion along the developing respiratory tract epithelium . In extrapulmonary airways ( largely trachea here ) we found that Jag ligands are not required to induce or maintain the fate of basal cell precursors . These precursors are known to generate luminal cells in fetal airways ( Yang et al . , 2018 ) and we now report that double Jag1 and Jag2 deletion leads to a major imbalance between the basal and luminal compartments with an expansion of basal cell precursors . By contrast , loss of both Jag ligands in intrapulmonary airways had no detectable impact on the NEB microenvironment . Unexpectedly , we found that Dll inactivation in Dll1cnull; Dll4cnull mutants resulted in marked expansion of NEBs and their associated secretory cells . Our analysis of the ontogeny of Notch ligands showed that Jag2 is expressed well before Jag1 in the epithelium and that both Dll1 and Dll4 appear only after NEBs form in intrapulmonary airways . Establishment of NE vs non-NE fate is known to be associated with induction of Ascl1 and a classic mechanism of lateral inhibition involving activation of Notch-Hes1 in neighboring cells ( Borges et al . , 1997; Ito et al . , 2000; Collins et al . , 2004 ) . Interestingly , although Ascl1-labeled NE precursors have been reported in the embryonic lung as early as E12 . 5 , we found no expression of Dll ( or Jag ) ligands by these cells prior to E13 . 5 ( Figure 1; Beckers et al . , 1999; Post et al . , 2000; Li and Linnoila , 2012; Kuo and Krasnow , 2015 ) . This was intriguing since there is evidence that Hes1 is expressed and already active early in the developing lung epithelium in spite of no evidence of ligand expression in intrapulmonary airways to activate Notch signaling nearby NE cells ( Tsao et al . , 2009; Noguchi et al . , 2015 ) . This suggests that at these initial stages NE vs . non-NE cell fate selection is mediated by Hes1 in a Notch-independent fashion . Consistent with this , Hes1 deletion in lung epithelial progenitors at the onset of lung development ( Shhcre/+; Hes1flox/flox ) results in aberrant expansion of NEB precursors as early as E13 . 5 ( Noguchi et al . , 2015 ) . Hes1-dependency on Notch is likely established at later developmental stages and could explain why genetic inactivation of the key Notch pathway components Pofut1 or Rbpjk using the ShhCre driver ( the same used to delete Hes1 , above ) , had no apparent effect in NE abundance at early ( E14 . 5 ) stages compared to the severe effects at later ( E18 . 5 ) stages ( Tsao et al . , 2009 ) and not shown ) . Our Dll1cnull; Dll4cnull mutants provided the first genetic proof that these ligands are crucially involved in regulating the size the NEBs , with no clear role in controlling other aspects such as the number of PNECs or NEBs per airway . By contrast , Jag ligands had no detectable influence in NEB size or abundance . Even at E14 . 5 , when NEBs are forming along the proximal-distal axis of intrapulmonary airways , Jag2 is still strongly expressed only at partially overlapping proximal domains such that nascent distal clusters of Ascl1+ cells arise in non-Jag2 or non-Jag1-expressing areas ( Figure 1—figure supplement 2 ) . Together our data suggest a model for the role of Notch ligands in lung development ( Figure 7 ) in which early , when epithelial progenitors start to differentiate , a wave of Jag-Notch activation initiating in the trachea progresses in a proximal-distal fashion to establish the balance of secretory vs . multiciliated cell fates in intrapulmonary airways . A program of NE cell fate also emerges in intrapulmonary airways and as NEB start to express Dll ligands , Dll-Notch signaling is turned on in adjacent cells to form NEB-associated CCs . Local activation of Notch signaling in these cells shelters the NEB microenvironment from the neighboring epithelium , preventing aberrant NEB expansion . This role is restricted to Dll1 and Dll4 , given that Jag1 , Jag2 single or double mutants showed no detectable effect in the size of NEBs , NEB-associated CCs or local Notch activation . The NEB microenvironment is maintained by Dll ligands that induce Notch to maintain the local balance of NE and non-NE cell types . The relevance of a late Notch-dependent phase is underscored by the expansion of the NE domain when Dll ligands are unavailable to induce Notch to generate NE-associated CCs locally . Our observations are not in conflict with recent reports that describe aggregation of NEB by a mechanism of NE cell migration ( slithering ) ( Kuo and Krasnow , 2015; Noguchi et al . , 2015; Branchfield et al . , 2016 ) . Rather we envision that the cell fate specification events described above precede these migratory and cluster-forming events , or overlap at least partially with mechanisms reported here . Intriguingly , in spite of the absence of Dll ligands and having Jag1 and Jag2 as the sole ligands available , the NEB-associated CC in Dll1cnull; Dll4cnull airways exhibited robust Notch activation and maintained the unique features of CCs in this microenvironment ( strong Upk3a , SSEA1 , and low Cyp2f2 ) . This suggested that the features above do not necessarily depend on the differential activation of Notch by NE-derived Dll ligands and can result from Jag-Notch activation in the cells immediately adjacent to NEBs . We speculate that a currently unidentified cell membrane-associated component of NE cells or a short-range diffusible signal ( s ) emanating from NEBs modulates Notch signaling or influences the cell fate program in adjacent CCs to endow these features ( Figure 7 ) . Lastly , the strength of receptor-ligand interactions is well known to depend on post-translational modifications of Notch receptors , particularly by the family of Fringe proteins , Lunatic ( Lfng ) , Manic ( Mfng ) and Radical ( Rfng ) ( Stanley and Okajima , 2010 ) . Mass spectrometric analysis has demonstrated Lfng to promote Notch activation by Dll1 and decrease its activation by Jag1 ( Kakuda and Haltiwanger , 2017 ) . Lfng is expressed in NEBs rather than in the NEB-associated CCs , which activates Notch signaling ( Xu et al . , 2010b; Xu et al . , 2010a ) . This is reminiscent of the developing intestinal epithelium where Fringe is expressed in the ligand-presenting cells to promote Notch activity in the neighboring cells ( Kadur Lakshminarasimha Murthy et al . , 2018 ) . There is currently no evidence that these proteins influence epithelial Notch signaling in the developing lung epithelium ( van Tuyl et al . , 2005; Xu et al . , 2010b; Xu et al . , 2010a ) . In summary our study provides novel insights into developmental mechanisms mediated by Jag/Dll/Notch in the lung . These observations could be of significance in studies of human conditions associated with aberrant expansion or differentiation of NEBs and their associated CCs . Indeed , analysis of human biopsies from normal donors and patients with pulmonary NE cell hyperplasias suggest that both the NE and NE-associated CC components are coordinately altered ( Guha et al . , 2017 ) . Further studies examining the impact of Notch ligands and downstream signals in these diseases are likely to provide important insights into their pathogenesis . Dll1flox/flox and Jag1flox/flox mice were provided by Dr . Julian Lewis ( Hozumi et al . , 2004; Brooker et al . , 2006 ) . Dll1cnull mice were generated by crossing Dll1flox/flox female mice with Dll1flox/+; Shhcre/+ males . Dll4flox/flox mice were obtained from Dr . Freddy Radtke ( Koch et al . , 2008 ) . Dll4cnull mice were generated by crossing Dll4flox/flox female mice with Dll4flox/+; Shhcre/+ males . Dll1cnull; Dll4cnull mice were generated by crossing Dll1flox/flox; Dll4flox/flox females with Dll1flox/+; Dll4flox/+; Shhcre/+ males . Jag1cnull mice were generated by crossing Jag1flox/flox females with Jag1flox/+; Shhcre/+ males . Jag2flox/flox mice were provided by Dr . Thomas Gridley ( Xu et al . , 2010b ) . Jag2cnull mice were generated by crossing Jag2flox/flox females with Jag2flox/+; ShhCre/+ males . Jag1cnull; Jag2cnull mice were generated by crossing Jag1flox/flox; Jag2flox/flox females with Jag1flox/+; Jag2flox/+; Shhcre/+ males . Embryos were harvested at E14 . 5 and E18 . 5 , where day 0 . 5 was counted as the morning when a vaginal plug was found . All experiments involving animals were performed in accordance with the protocols approved by Columbia University Medical Center . Whole lung and trachea were harvested from mice at E14 . 5 and E18 . 5 and fixed in 4% paraformaldehyde at 4°C overnight . Samples then underwent PBS washes and 15% and 30% sucrose washes before embedding in OCT . Samples were incubated with primary antibodies ( overnight at 4°C ) and secondary antibodies conjugated with Alexa488 , 568 , or 647 ( 1:300 ) with NucBlue Fixed Cell ReadyProbes Reagent ( DAPI ) ( Thermo Fisher #R37606 ) for 45 min . After washing , samples were mounted with ProLong Gold antifade reagent for analysis . When necessary , heat-induced epitope retrieval was performed using citric acid-based antigen unmasking solution ( Vector Laboratories #H-3300 ) . Ascl1 and N1ICD staining required tyramide amplification ( cyanine 3 or cyanine 5 ) used with horse radish peroxidase conjugation ( species-specific ImmPRESS kit , Vector Laboratories ) . Antibodies used were: anti-β-tubulin IV ( Abcam #ab11315 , 1:100 ) , anti-Ascl1 ( Thermo Fisher #14-5794-82 , 1:100 ) , anti-CC10 ( Santa Cruz sc9772 , 1:150 ) , anti-Cgrp ( Sigma Aldrich #C8198 , 1:2500 ) , anti-Foxj1 ( Thermo Fisher #14-9965-80 , 1:50 ) , anti-Ki67 ( Cell Signaling #9129 , 1:100 ) , anti-Krt5 ( Biolegend #905501 , 1:500 ) , anti-Krt8 ( Abcam #ab107115 , 1:500 ) ; anti-N1ICD ( Cell Signaling #4147 , 1:100 ) , anti-p63 ( Santa Cruz #sc8343 , 1:400 ) , anti-Scgb3a2 ( R and D Systems #AF3465 , 1:100 ) , anti-SSEA1 ( EMD Millipore #MAB4301 , 1:300 ) . Images were acquired using a Leica DMi8 microscope or Zeiss LSM710 confocal laser scanning microscope . To determine the percentage of Foxj1+ ciliated cells in control and Jag-cnull mutant intrapulmonary airways E18 . 5 coronal sections of whole lungs were stained with Foxj1 and DAPI . Two sections from two separate embryos for each genotype were used for counting . DAPI+ epithelial cells were counted in intrapulmonary airways and were compared to the number of Foxj1+ cells to determine Foxj1+ percentages . To determine the percentage of Foxj1+ ciliated cells in control and Jag-cnull mutant tracheas E18 . 5 coronal sections of whole tracheas were stained with Foxj1 and DAPI . Two sections of whole trachea from one embryo for each genotype were used for counting . DAPI+ epithelial cells were counted in one side of the trachea and were compared to the number of Foxj1+ cells to determine Foxj1+ percentages . Analysis of neuroendocrine cells and neuroepithelial bodies ( NEBs ) was performed on E14 . 5 Delta-cnull mutants . Sections were stained with Ascl1 and DAPI . Three sections from three separate embryos for each genotype were used for counting . For each section the number of intrapulmonary airways was counted , as well as the number of NEBs and solitary neuroendocrine cells . The ratios of NEBs/airway and solitary neuroendocrine cells/airway were calculated . Additionally , NEB size was examined . In each section , NEB size was determined by counting the number of neuroendocrine cells in contact with each other , where an NEB was determined to be a group of three or more cells . Frozen sections were processed as described for immunofluorescence . In situ hybridization was performed using digoxigenin-UTP-labeled probes as previously described ( Tsao et al . , 2008; Tsao et al . , 2009; Guha et al . , 2012 ) . Probes are listed in Table 1 . Hybridization probes were ordered from Integrated DNA Technologies at 25 nM with standard desalting and stored as 100 µM stocks in DEPC-treated water . Quantitate real-time PCR was performed as previously described ( Tsao et al . , 2009 ) . RNA was extracted using the RNeasy kit ( Qiagen ) and reverse transcribed using Oligo ( DT ) primers ( SuperScript III or IV kits , Thermo Fisher ) . The following primers ( Thermo Fisher ) were used: Ascl1 ( Mm03058063_m1 ) , Cgrp ( Mm00801463_g1 ) , Foxj1 ( Mm01267279_m1 ) , Scgb1a1/CC10 ( Mm00442046_m1 ) , Scgb3a2 ( Mm00504412_m1 ) , Upk3a ( Mm00452321_m1 ) . Reactions were performed using Taq-Man Advanced Master Mix ( Thermo Fisher #4444556 ) using β-actin as internal control and a Step-One Plus Instrument ( Applied Biosystems ) . ΔΔCT method was used to calculate changes in expression levels .	Cells communicate with each other by sending messages through a range of signaling pathways . One of the ways cells signal to each other is through a well-studied pathway known as Notch . In this pathway , cells display molecules on their surface , known as Notch ligands , that can activate Notch receptor proteins on the surface of neighboring cells . Once the Notch receptors bind to these ligands , they trigger various responses inside the cell . Notch ligands exist in two different families: Delta-like ( Dll ) ligands and Jagged ( Jag ) ligands . The layer of cells that lines the airways in the lungs consists of several different cell types . These include secretory cells that produce the fluid covering the airway surface , multiciliated cells , and neuroendocrine cells . Together these cells work as a barrier to protect the lung from environmental particles that may be breathed in . Additionally , the lung also has multipotent progenitor cells , which can become any of the other types . When Notch signaling is missing from the lung during embryonic development , not enough secretory cells are made , while other cell types are made in excess . This is because the multipotent progenitor cells need to communicate via Notch signaling to decide what type of cell to become and keep the right proportion of different cell types in the airways . In other organs , multipotent progenitors can become different types of cells depending on whether Notch signaling was activated by Dll or by Jag ligands , but it was unknown if this also happened in the lungs . Stupnikov et al . investigated the situation in the airways during development by looking at where and when Dll and Jag ligands first appeared , and by inactivating the genes that code for these ligands . They found that Jag ligands appeared well before Dll ligands , and that when the genes coding for Jag ligands were inactivated , more ciliated cells were produced . By contrast , loss of Dll ligands resulted in an increase in the neuroendocrine and their associated secretory cells , with little effect on the multiciliated cells . This increase resembled what is seen in some human diseases . The results suggest that the diversity of Notch effects in the airways depends on which Notch ligand is locally available . These observations may help to understand the mechanism of certain diseases involving neuroendocrine cells in the lung , such as small cell carcinoma or bronchial carcinoid tumors .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2019	Jagged and Delta-like ligands control distinct events during airway progenitor cell differentiation
PDZ domain scaffold proteins are molecular modules orchestrating cellular signalling in space and time . Here , we investigate assembly of PDZ scaffolds using supported cell membrane sheets , a unique experimental setup enabling direct access to the intracellular face of the cell membrane . Our data demonstrate how multivalent protein-protein and protein-lipid interactions provide critical avidity for the strong binding between the PDZ domain scaffold proteins , PICK1 and PSD-95 , and their cognate transmembrane binding partners . The kinetics of the binding were remarkably slow and binding strength two-three orders of magnitude higher than the intrinsic affinity for the isolated PDZ interaction . Interestingly , discrete changes in the intrinsic PICK1 PDZ affinity did not affect overall binding strength but instead revealed dual scaffold modes for PICK1 . Our data supported by simulations suggest that intrinsic PDZ domain affinities are finely tuned and encode specific cellular responses , enabling multiplexed cellular functions of PDZ scaffolds . It is of fundamental importance for cell function to organize signaling processes in space and time . Scaffold proteins play a key role in these efforts by operating as versatile nanoscale modules capable of bringing distinct molecular components in close proximity to shape specificity in cellular signaling networks and regulate output ( Good et al . , 2011; Zeke et al . , 2009 ) . A broad variety of different protein-protein and protein-lipid interacting domains are found in scaffold proteins , enabling them to bind and direct localization and function of their diverse interaction partners , such as receptors , transporters , ion channels and kinases ( Hung and Sheng , 2002; Zhu et al . , 2016 ) . Our current understanding of the dynamics and kinetics underlying scaffold interactions relies mostly on in vitro assays of single domains isolated form their native membrane environment ( Vincentelli et al . , 2015; Stiffler et al . , 2007; Long et al . , 2003; Ivarsson , 2012 ) . Such approaches reduce the complexity and makes binding assays simpler to both perform and analyze . Nonetheless , the presence of several interaction domains in one protein , together with possible formation of higher order structures of both scaffold proteins and transmembrane interaction partners ( Long et al . , 2003; Ivarsson , 2012 ) , strongly suggest that measurements in vitro cannot replicate the behavior of the native environment . Furthermore , scaffold proteins often function on , or in proximity , to lipid cellular membranes , and several scaffold protein domains directly interact with lipids ( Egea-Jimenez et al . , 2016; Pérez et al . , 2013 ) . PDZ ( PSD-95/Discs Large/ZO-1 ) domains constitute one of the most common interaction domains in scaffold proteins ( Feng and Zhang , 2009; Zhu et al . , 2016; Ye and Zhang , 2013 ) and are characterized by an elongated binding groove that interacts with the last three to five C-terminal residues of the target proteins ( Doyle et al . , 1996 ) . Scaffold proteins often contain several PDZ domains permitting them to serve as adaptors to assemble protein complexes ( Doyle et al . , 1996; Sheng and Sala , 2001; Feng and Zhang , 2009; Ye and Zhang , 2013 ) . PICK1 ( Protein Interacting with C Kinase 1 ) , for example , forms a dimeric structure containing two spatially separated , identical PDZ domains and a lipid binding N-BAR ( Bin/amphiphysin/Rvs ) domain ( Karlsen et al . , 2015 ) , enabling simultaneous binding of two interaction partners and tethering of the complex to the membrane ( Xu and Xia , 2006 ) . To perform its differential functions , PICK1 is believed to operate in two distinct modes of scaffolding; one required for clustering transmembrane interaction partners , such as the AMPA-type glutamate receptors ( AMPARs ) , the metabotropic glutamate receptor 7 ( mGluR7 ) , ephrin and the monoamine transporters ( Xia et al . , 1999; Torres et al . , 2001; Boudin et al . , 2000; Torres et al . , 1998 ) , and one required for recruiting protein kinase Cα ( PKCα ) to its transmembrane interaction partners to regulate their phosphorylation ( Baron et al . , 2002; Dev et al . , 2000; Perez et al . , 2001; Staudinger et al . , 1995 ) . The molecular mechanisms underlying these different scaffold modes remain nevertheless unknown . In contrast to neuronal scaffold interactions , drug-receptor or antibody interactions have been studied rigorously in tissues and cells uncovering molecular mechanisms of co-operativity and avidity in many biological systems ( Varner et al . , 2015 ) . To enable similar studies of scaffold interactions with membrane proteins embedded in their natural membrane environment , we take here advantage of a supported cell membrane sheets ( SCMS ) technique ( Perez et al . , 2006a; Perez et al . , 2006b ) . SCMS are prepared by pressing a glass cover-slip on an adherent cell culture . When the cover-slip is removed , the apical plasma membrane detaches from cells and a planar sheet exposing the inner surface of the membrane is exposed on the cover-slip . Binding of fluorescence labeled protein ligands to the membrane proteins exposed on the SCMS can be quantified by confocal laser scanning microscopy . Strikingly , our data reveal binding strengths for PICK1 , as well as for the PDZ tandem domain from PSD-95 ( Long et al . , 2003; Cho et al . , 1992 ) , that are two to three orders of magnitude higher than the intrinsic affinities measured in vitro . The binding strength for PICK1 is strongly dependent on an intact PDZ domain binding groove and the membrane-binding N-BAR domain but , surprisingly , independent of the tertiary and quaternary structure of the transmembrane PDZ binding partners . To our further surprise , the binding strength for PICK1 is insensitive to discrete changes in the intrinsic affinities of the PDZ binding partners , which instead are reflected in changes of maximal binding . Mathematical modeling of homo-bivalent binding demonstrates that the observed behavior is consistent with a change in the binding mode from a scenario where one PICK1 dimer binds two membrane ligands to a scenario where one PICK1 dimer binds one membrane ligand and has one free PDZ domain available for other interactions . Altogether , by quantitatively illuminating binding kinetics of PDZ domain protein scaffolds on a cell membrane , our results reveal novel principles for how cellular scaffold proteins can operate to ensure specificity and selectivity in cellular signaling networks , and furthermore how these principles change our current understanding of cellular binding equilibria and mechanisms underlying the function of scaffold proteins in general . To investigate the interaction between PICK1 and the GluA2 subunit of the AMPAR in a semi-native environment , N-terminally flag-tagged GluA2 ( SF-GluA2 ) was transiently expressed in HEK293 cells and labeled with Alexa-488 conjugated anti-FLAG M1 antibody before preparation of supported cell membrane sheets ( SCMS ) ( Perez et al . , 2006a ) ( Figure 1a–d ) . We incubated SCMSs , exposing the inner membrane leaflet and the intracellular parts of SF-GluA2 , with increasing concentrations of purified PICK1 , containing an N-terminal SNAP-tag fluorescently labeled with SNAP-surface 549 ( PICK1 ) ( Figure 1a , b ) , and measured the fluorescence intensity of bound PICK1 by confocal microscopy . By normalizing the intensities of the fluorescent signal from PICK1 ( red ) to the intensity of fluorescently labeled SF-GluA2 ( green ) and plotting this ratio as a function of increasing PICK1 concentration , we obtained a saturable binding curve ( Figure 1f ) . The apparent affinity ( Kd* ) calculated from the binding curve was 67 ± 6 nM ( mean ± s . e . m . , n = 3 ) , which , remarkably , is ~100 fold higher than the low micromolar intrinsic affinity ( Kdint ) determined for binding of the GluA2 C-terminus to the PICK1 PDZ domain using an in-solution based assay ( Erlendsson et al . , 2014 ) . To test if the observed binding was specific and dependent on the C-terminal PDZ binding motif ( -ESVKI ) in GluA2 , we added an alanine residue to the GluA2 C-terminus ( -ESVKI + A ) ( SF-GluA2 +A ) to compromise PDZ ligand binding ( Madsen et al . , 2005; Madsen et al . , 2008 ) . The apparent affinity and the maximal binding ( Bmax ) of PICK1 were significantly reduced compared to binding to SF-GluA2 for similar receptor expression levels ( Figure 1b , e , f ) . This supports that PICK1 binding to SCMS from GluA2 expressing cells is specific and depends on the interaction with the GluA2 C-terminal PDZ motif . The results also establish the use of SCMS as a new , robust , quantitative method for investigating membrane proximal scaffold interactions . To test if the increased binding strength measured in the SCMS assay , compared to the micromolar affinity measured by in-solution assays ( Erlendsson et al . , 2014 ) , was dependent on the tetrameric arrangement of subunits within the AMPARs , we transferred the 24 C-terminal residues of GluA2 onto the C-terminus of the single transmembrane spanning α-subunit of the IL-2 receptor ( TAC ) fused to YFP ( TAC-YFP-GluA2 ) ( Figure 1g ) . In contrast to the AMPARs , TAC is not believed to form higher order structures ( Spangler et al . , 2015 ) . Strikingly , we obtained a Kd* of 73 ± 19 nM ( mean ±s . e . m . , n = 3 ) for binding of PICK1 to TAC-YFP-GluA2 ( Figure 1h , i ) suggesting that the high apparent affinity was achieved independently of the tetrameric complex as well as of the membrane embedded segments of the receptor . To address how different PDZ-binding motifs affect the binding strength , we next measured the interaction of PICK1 with the C-terminus of the dopamine transporter DAT ( TAC-YFP-DAT C24 ) ( Figure 2 ) . According to in-solution binding assays , this peptide has a 10-fold higher intrinsic affinity for PICK1 than the GluA2 peptide ( Erlendsson et al . , 2014 ) . We observed specific binding also for this construct ( Figure 2—figure supplement 1 ) but only a minor increase in Kd* ( 47 ± 5 nM , mean ±s . e . m . , n = 7 ) , revealing that the binding strength measured in the SCMS assay correlates poorly with the intrinsic affinity ( Figure 1a , c ) . To further confirm specificity and rule out effects of the SNAP-tag on PICK1 , competition binding using a fixed concentration of labeled SNAP-PICK1 and an increasing concentration of unlabeled PICK1 was performed on SCMSs expressing TAC-YFP-DAT ( Figure 2b , d ) . The binding strength determined from competition binding ( Ki=29 ± 5 nM ( mean ±s . e . m . , n = 5 ) was close to that from the direct binding saturation assay and increased 100-fold compared to the intrinsic affinity of the PICK1 PDZ interaction with the DAT C-terminus obtained in solution using a fluorescence polarization competition assay ( 2 . 1 ± 0 . 4 μM ) ( Figure 2b , d ) . To complement the results for PICK1 , we also probed the binding strength of fluorescently labeled PSD-95 PDZ 1–2 tandem domain on SCMS expressing the ß1-adrenergic receptor ( Hu et al . , 2000 ) . Indeed , the measured binding strength ( EC50 = 3 ± 2 μM ) was two orders of magnitude increased compared to the intrinsic affinities previously measured in solution for either of the two domains ( 430 ± 47 µM and 120 ± 20 µM for PDZ1 and PDZ2 , respectively ) ( Møller et al . , 2013 ) ( Figure 2—figure supplement 2a , b ) . These findings further support that the binding strengths of PDZ domain scaffolding interactions are substantially higher than intrinsic PDZ affinities . Despite similar high binding strength , comparison of the binding curves for PICK1 binding to TAC-YFP-GluA2 and TAC-YFP-DAT revealed a surprising two-fold difference in total maximal binding ( Bmax ) ( TAC-YFP-DAT Bmax = 100%; TAC-YFP-GluA2 Bmax = 44 ± 9% ) ( Figure 2—figure supplement 3a , b ) . Note that this difference unlikely is due to different expression levels as the maximum binding was normalized to the YFP signal for the two different constructs . To address whether the difference was a consequence of the different intrinsic affinities of the DAT C-terminus compared to the GluA2 C-terminus , we exploited that the intrinsic affinity of PICK1 for the DAT C-terminus depends on the C-terminal valine and that substitution of the aliphatic side-chain of the C-terminal valine decrease its intrinsic affinity for PICK1; Val ( WT ) ( 2 . 3 ± 0 . 1 µM ) >Ile ( 9 . 5 ± 0 . 9 µM ) >Ala ( 49 ± 3 µM ) ( Madsen et al . , 2005 ) . Because DAT plasma membrane targeting is compromised by alterations in the extreme C-terminus ( Bjerggaard et al . , 2004 ) , we introduced the mutation series into a previously characterized fusion construct in which the DAT C-terminus is fused to β2 adrenergic receptor ( flagβ2-DAT8 ) ( Madsen et al . , 2012 ) yielding three constructs: LKV , LKI and LKA , respectively ( Figure 3—figure supplement 1a ) . The apparent affinity of PICK1 to LKV ( Kd=37 ± 5 nM , n = 8 ) ( mean ±s . e . m . ) ( Figure 3a , b ) was essentially the same as that seen for TAC-YFP-DAT ( Figure 2 ) . Moreover , despite a decrease in intrinsic affinity of up to >20 fold , we observed no differences in apparent affinity in the SCMS assay upon mutating the valine to isoleucine ( LKI Kd=39 ± 4 nM , n = 7 ) or alanine ( LKA Kd=59 ± 11 nM , n = 5 ) ( Figure 3a , b ) . Instead , we observed an unexpected reduction in maximal binding ( Bmax LKI: 56 ± 2% , LKA: 41 ± 4%; ( means ±s . e . m . ) relative to LKV ( Figure 3a , b and Table 1 ) for similar receptor surface expression levels ( Figure 3—figure supplement 1b ) . To obtain better insight into the molecular mechanism underlying the different maximal binding levels we turned to kinetic experiments . Association experiments did not show any striking difference between LKV and LKI ( Figure 3c–d ) . Both constructs displayed slow ( half-bound maxima; LKV: 24 ± 8 min , LKI: 14 ± 6 min; ( means ±s . e . m . , n = 3 ) ) , but saturable binding . PICK1 dissociation , on the other hand , revealed that whereas PICK dissociated very slowly when bound to LKI ( t½=431 ± 16 min; mean ±s . e . m . , n = 3 ) , a distinct fast component of dissociation was observed from LKV on top of the slow dissociation rate ( t½=21 ± 8 and 373 ± 51 min; mean ±s . e . m . , n = 3 ) , ( Figure 3e–f ) . This suggests that the unbinding of LKV consists of two kinetically distinct processes and that PICK1 therefore might engage in two different binding configurations depending on the concentration of PICK1 and the intrinsic affinity for the membrane embedded ligand . That is , when the concentration of PICK1 is low and the intrinsic affinity for the membrane embedded ligand is low , PICK1 might adopt the intuitive binding mode with both PDZ domains of the dimer bound to a membrane embedded ligand ( slow dissociation rate ) . However , when the concentration of PICK1 is high and the intrinsic affinity for the membrane embedded ligand is high , PICK1 might gradually switch to a binding mode with only one PDZ domain engaged in binding of the membrane embedded ligand ( fast dissociation rate ) . To probe the feasibility of this hypothesis , we turned to thermodynamic simulations . The principle of bivalency is well known to endow high affinity ( often denoted as ‘avidity’ ) and the two individual steps in the bivalent PDZ-binding of PICK1 ( aa ) to two identical membrane embedded ligands ( e . g . a receptor ) ( A ) can be represented as shown in the scheme in Figure 4a ( leading to formation of the ‘red complex’ , aAAa ) . Indeed , bivalency increases the overall affinity and residence time since this permits multiple partial unbinding and rebinding events to take place before the protein fully dissociates ( Vauquelin and Charlton , 2013; Vauquelin and Charlton , 2013 ) . However , when the bulk concentration of PICK1 [aa] is sufficiently high , the free bivalent PICK1 would be expected to outpace binding of the second PDZ domain , thereby preventing formation of the ‘red complex’ ( aAAa ) and instead leading to formation of the ‘blue complex’ ( aaAAaa ) ( Figure 4a ) . Under these conditions , the rate for formation of the blue ‘ternary’ complex ( V2 ) is larger than the rate for formation of the red complex , ( V3 ) , because V2 relies on the free bulk concentration of ligand [aa] , whereas V3 relies on the local concentration [L] of the second domain for binding , which in turn depends on the distance ( r ) between the individual domains ( Figure 4a ) . Moreover , V3 might be compromised by steric hindrance , restricted rotation freedom and entropic penalty jointly denoted , f ( Figure 4a ) ( Vauquelin , 2013; Vauquelin and Charlton , 2013 ) . The effect of this empiric factor is to scale the effective concentration [L] of the free end of the PICK dimer in the AAaa complex ( the ‘green complex’ ) . The proposed model is the simplest model explaining our data . However , additional reaction steps can be envisioned for the binding of PICK1 for example the binding of the amphipathic helix in PICK1 to the cell membrane ( vide infra ) or putative long-range allosteric structural changes induced by the first binding event . None of such effects are necessary explicitly to describe our data , but may be encompassed by the f value . By simultaneously solving the system of differential equations related to formation of all of the involved complexes ( Figure 4—figure supplement 1a–b ) , and using k1 and k-1 values derived of from the in-solution PDZ binding to soluble ligands ( Erlendsson and Madsen , 2015; Erlendsson et al . , 2014 ) together with an inter PDZ distance ( r ) of 180 Å determined from the structure of PICK1 based on Small-Angle X-ray Scattering ( Karlsen et al . , 2015 ) ) and an f value of 185 , we obtained a biphasic saturation binding curve ( Figure 4b ) that overall was in good agreement with the experimentally derived saturation binding curve observed for PICK1 binding to LKV . Importantly , the biphasic shape results from the population of the ternary blue complex ( aaAAaa ) outpacing the binary red complex ( aAAa ) at concentrations above 100 nM of PICK1 . Note that in the averaged experimental data shown in Figure 3a and c , the biphasic behavior of the saturation binding curve is likely masked by experimental variation , as supported by the fact that we could extract a representative data set and fit this to a biphasic curve in good agreement with the simulated curve ( Figure 4—figure supplement 2a–c ) . The modeling also rationalizes the differential binding observed for LVK and LKI ( Figure 3a–d ) . The lower intrinsic affinity for LKI reflects a change in the dissociation constant k-1 , and as k-1 is equivalent for the three rates ( V1 , V2 and V3 ) , the relative partitioning into the three different complexes will be unchanged . The absolute concentration dependence , however , will be parallel shifted when comparing LKV to LKI , as illustrated for a specific concentration [aa] by the arrow in Figure 4b . Consequently , a ligand with lower intrinsic affinity ( such a LKI ) will need a correspondingly higher concentration of bulk PICK1 to populate the ternary blue complex ( Figure 4 ) . This likely entails that what we observe when analyzing binding for the LKI ligand , that is the reduced apparent Bmax for LKI ( Figure 3b ) most likely represents a ‘plateau’ before transition to the blue complex , which would be observed if we experimentally were able to use even higher concentration of PICK1 . Finally , it should be noted that increasing or decreasing the f parameter , affecting k2 , would also have an important influence on the concentration-dependent formation of the blue complex ( Figure 4—figure supplement 3 ) . The discrimination between the different binding modes becomes even more perceptible in the simulation of the dissociation experiments; that is when free , labeled ligand molecules are removed and/or prevented to bind ( Figure 4c , d ) . The observed dissociation for LKV is first dominated by the unbinding of one of the ligands in the blue complex . Being governed by a single off-rate , k-1 , the initial component of the curve is fast , as would be expected from a monovalent binding event . Yet , since the so-obtained partially bound complexes ( green ) are more prone to form the bivalent red complex than to dissociate , the second component is governed by the slow dissociation of fully bound bivalent complexes ( Figure 4c ) . For LKI or GluA2 , which have lower intrinsic affinities ( numerically corresponding to lower PICK1 concentration on Figure 4b ) , the ternary ‘blue’ complex is not favored and only the slow second component is observed in the simulation ( Figure 4d ) . Importantly , the simulations are in very good agreement with the actual dissociation experiment ( Figure 3e , f ) . The model described above and in Figure 4 relies on two independent PDZ domain ligand-binding sites in the PICK1 dimer , and consequently it predicts that no differences in the apparent maximal binding should be observed for the different ligands if one of the PDZ domains is rendered non-functional . To experimentally test this prediction , we mutated Ala87 in the PICK1 to Leu ( A87L ) , a mutation previously described to abolish binding of ligands to the PDZ domain ( Erlendsson et al . , 2014; Madsen et al . , 2005 ) . We found that PICK1 A87L bind very weakly also to SCMSs ( Figure 5—figure supplement 1 ) . Next , we mixed monomers ( obtained in 0 . 1% TX-100 ) ( Karlsen et al . , 2015 ) labeled with Alexa-647 ( blue , WT ) and Alexa-549 ( red , WT or A87L ) to allow formation of PICK1 WT/WT homodimers as well as WT/A87L heterodimers ( upon dilution of TX-100 concentration to 0 . 01% , see Materials and methods for details ) . TX-100 ( 0 . 1% ) did not affect PDZ domain function per se ( Figure 5—figure supplement 2 ) . Binding of bivalent homodimeric PICK1 WT/WT was then compared to monovalent heterodimeric PICK1 WT/A87L on sheets from cells expressing either LKV or LKI ( Figure 5a , b ) . Importantly , PICK1 WT/WT binding to LKV revealed again a two-fold higher Bmax than for LKI with no difference in binding strength ( Figure 5c ) . For binding of PICK1 WT/A87L heterodimers , we likewise obtained signals in both the PICK1 A87L ( red ) and the PICK1 WT ( blue ) channels ( Figure 5b ) confirming formation of the heteromers . No PICK1 A87L signal was observed by incubation together with PICK1 WT without allowing subunit exchange , showing that PICK1 WT does not nucleate binding of A87L for example by oligomerization ( Figure 5—figure supplement 3a , b ) . In agreement with our model , we did no longer observe any difference in Bmax between LKV and LKI for the PICK1 WT/A87L heterodimers , demonstrating that the difference in Bmax indeed must rely on PDZ domain bivalency ( Figure 5d , e ) . Moreover , the resulting binding curves for LKV showed a binding strength similar to that of the PICK1 WT homodimers ( WT/WT Kd=37 ± 5 nM; WT/A87L Kd=24 ± 21 nM , mean ±s . e . m . , n = 3 ) ( Figure 5d , e ) , suggesting that one PDZ domain is sufficient to support the strong binding and corroboration of the ternary complex binding mode ( blue complex in Figure 4a ) . For LKI , however , we observed a significant decrease in the apparent affinity of PICK1 WT/A87L heterodimers in both channels ( WT/WT Kd* =29 ± 4 nM , WT/A87L Kd=123 ± 41 nM , mean ±s . e . m . , n = 3 ) ( Figure 5d , e ) , suggesting that the high binding strength observed for WT/WT does rely on the dual PDZ domains in agreement with the model . PICK1 A87L homomers do not interfere with PICK1 WT homomer binding if exchange between monomers has not been allowed ( Figure 5—figure supplement 3c , d ) . Interestingly , by tentatively combining the binding signal of each components of the dimer ( WT ( blue ) /A87L ( red ) the maximal binding for WT/A87L on LKI ( and LKV ) approached that of WT/WT on LKV , suggesting an overall increase in the number of binding sites of LKI when going from WT/WT to WT/A87L ( Figure 5f ) . Altogether , these findings strongly support a ‘dual scaffold mode’ as outlined in our model and simulations above; that is , one PICK1 WT/WT dimer preferentially binds one membrane embedded LKV ligand ( i . e . with 1:1 stoichiometry ) . In contrast , one PICK1 WT/WT dimer preferentially binds two membrane embedded LKI ligands ( i . e . with 1:2 stoichiometry ) , thereby preserving a high binding strength due to the avidity obtained by doubling the PDZ interactions . Upon compromising the PDZ bivalence ( PICK1WT/A87L ) , however , PICK1 is forced to obtain the 1:1 configuration , which increase Bmax at the expense of binding strength for LKI ( Figure 5f ) . To test whether the differential PICK1 scaffold modes could have a functional consequence , we tested the ability of PICK1 to inhibit recycling of its interaction partners ( Citri et al . , 2010; Madsen et al . , 2012 ) . As model system we used our flagß2-DAT8 constructs expressed in HEK293 FlpIN cells with tetracycline inducible expression of eYFP-PICK1 ( Madsen et al . , 2012 ) . Expression of eYFP-PICK1 did not affect the isoproterenol-induced internalization of any of the constructs ( Figure 5g ) , however , expression significantly inhibited the reinsertion into the plasma membrane of LKV normalized to LKV +A after alprenolol treatment ( 41 . 5 ± 12% vs 100% , mean ±s . e . m . , n = 6 ) as previously described ( Madsen et al . , 2012 ) . In contrast , eYFP-PICK1 expression did not significantly affect the reinsertion of either LKI or LKA ( 88 . 8 ± 14% and 95 . 9 ± 3 . 9% , respectively , mean ±s . e . m . , n = 6 ) ( Figure 5h ) despite the fact that these display similar affinities on SCMSs . This suggests that the ability of PICK1 to reduce recycling of an interaction partner could be dependent on formation of the ternary ( blue ) complex and hence the ability to recruit for example kinases ( Figure 4a ) . To further investigate how one PDZ domain in the PICK1 dimer is sufficient to mediate interaction with a membrane embedded binding partner with nanomolar binding strength ( LKV ) , we mutated two hydrophobic residues in the membrane binding amphipathic helix preceding the BAR domain in PICK1 ( PICK1 V121E , L125E ) ( Figure 6a ) ( Holst et al . , 2013; Herlo et al . , 2018 ) . PICK1 V121E , L125E showed markedly reduced binding strength for the LKV ( PICK1 V121E , L125E , Kd=286 ± 126 nM , mean ± s . e . m . , n = 3 ) on SCMSs compared to PICK1 WT without affecting maximal binding ( Figure 6b and Table 3 ) . This supports that the binding strength of the PICK1 WT relies in part on membrane binding . We also assessed the possible functional consequence of reduced binding strength of this PICK1 variant . Expression of eYFP-PICK1 V121E , L125E reduced the reinsertion of LKV significantly ( +Tet 50 . 9 ± 8 . 3% vs –Tet 86 . 0 ± 13% , mean ±s . e . m . , n = 6 ) and to the same extend as eYFP-PICK1 WT ( +Tet 51 . 3 ± 3 . 1 vs –Tet 81 . 7 ± 7 . 7 , mean ±s . e . m . , n = 6 ) ( Figure 6c–d ) , suggesting that this function of PICK1 is independent on the avidity provided by the membrane binding helix , but instead relies on a sufficiently high PICK1 concentration to achieve formation of 1:1 configuration ( Figure 4 , blue ) . Finally , we tested the role of the membrane binding amphipathic helix in synaptic localization of PICK1 by employing a lentiviral molecular replacement strategy in hippocampal neurons . Using a previously characterized construct ( Citri et al . , 2010 ) , we knocked down endogenous PICK1 with shRNA ( sh18 ) and replaced it with either shRNA-resistant GFP-PICK1 WT or GFP-PICK1 V121E , L125E . Similarly , to endogenous PICK1 , virally expressed GFP-PICK1 WT showed partial clustering and significant localization to dendritic spines , which was reflected in a high level of co-clustering with the synaptic marker PSD-95 ( Figure 6f , colocalization shown in white in merged picture ) . GFP-PICK1 V121E , L125E had similar levels of expression in somatic regions ( 101 ± 8% of wt , p=0 . 94 , n = 16 ) ( Figure 6g ) , but distributed much less into the dendrites and displayed a more diffuse localization than GFP-PICK1 WT . The total number of GFP-PICK1 V121E , L125E clusters relative to PSD-95 clusters was significantly reduced compared to GFP-PICK1WT ( 48 . 5 ± 7% of GFP-PICK1 WT , p<0 . 0001 , n = 24 ) ( Figure 6h ) , whereas the number of PSD-95 clusters per length of dendrite was unchanged ( 97 ± 10% , p=0 . 81 , n = 24 ) ( Figure 6i ) . Interestingly , the localization to dendritic spines and consequently the co-clustering of GFP-PICK1 V121E , L125E with PSD-95 was almost abolished compared to GFP-PICK1 WT ( 0 . 17 ± 0 . 04 , p<0 . 0001 , n = 24 ) ( Figure 6j ) , This demonstrates that the binding strength provided by the membrane binding capability of the amphipathic helix is essential for stable synaptic localization of PICK1 . Here , we utilize SCMS’s to elucidate how PDZ scaffold proteins bind transmembrane proteins in a controlled cell membrane environment . As a result of the avidity of multiple protein-protein and/or protein-lipid interactions , the two PDZ scaffold proteins PICK1 and PSD-95 both displayed binding strength several orders of magnitude higher than the intrinsic affinities for individual PDZ interactions ( see Tables 1–3 ) . For PICK1 , we sought to address the relative role of native constituents of the SCMSs to this increased binding strength . We probed the binding to the membrane and its native proteins constituents by either disrupting the PDZ binding sequence of the overexpressed membrane-protein ( Figure 1f , Figure 2 – figure supplement 1b and Figure 3a ) , which reduced the maximal binding to 20–30% of the control depending on the construct . Conversely , disrupting the PICK1 PDZ domain ( A87L ) eliminated binding almost completely ( Figure 5—figure supplement 1b ) . These findings have two implications: 1 ) PICK1 binds native membrane proteins in the SCMS using the PDZ domain and 2 ) PICK1 does not bind the membrane in absence of PDZ interactions . Given the slow non-equilibrium conditions of the system , the interplay between the overexpressed membrane proteins and the native constituents of the SCMS are not easily interpreted . Disruption of the membrane binding amphipathic helix in PICK1 reduced the affinity of the interaction ( Figure 6b ) , suggesting that the interaction with the membrane does play a role in context of the PDZ binding . Conversely , we would argue that the PDZ interaction with transmembrane proteins native to the SCMS would be effectively competed by the overexpressed transmembrane proteins ( given their comparable affinities for example Figures 1f and 3a ) and consequently play a minor role in context of overexpression . Regardless the molecular explanation of the high binding strength in the SCMSs , it implies that often scaffold proteins may be at saturating concentrations in vivo – in particular given high local concentrations of transmembrane interaction partners . Tuning of affinities/avidities may therefore represent a relative inefficient mechanism of regulation , which we importantly could demonstrate by the ability of mutant PICK1 ( PICK1V121E , L125E ) to retain cellular function despite a ten-fold drop in binding strength according to the SCMS assay . The slow binding kinetics observed for PICK1 , on the other hand , imply that the scaffold interactions that underlie biological regulation are likely to occur under non-equilibrium conditions . Consequently , factors affecting the binding kinetics ( e . g . sterical hindrance and flexibility ) may have hitherto unappreciated biological impact . Similar slow kinetics were recently demonstrated for bivalent receptor ligands by kinetic simulations ( Vauquelin and Charlton , 2013 ) . The slow off-rates observed for PICK1 moreover provide a putative biological mechanism to convert at transient signal ( such as an activity dependent exposure of the GluA2 C-terminus ) into a long-lived effect ( e . g . membrane localization of PICK1 ) . Our most striking finding was that intrinsic PICK1 PDZ affinities did not correlate directly to the binding strength determined on SCMSs , but surprisingly resulted in changes in the observed maximal binding . We initially considered , that this might involve an element of ‘kinetic proofreading’ ( Hopfield , 1974; McKeithan , 1995 ) whereby the PDZ domain would enable insertion of the amphipathic helix ( Herlo et al . , 2018 ) in the membrane given sufficient residence time , however , the difference in maximal binding was preserved after mutation of the amphipathic helix . Instead , our kinetic modeling together with the experimental data suggests that this behavior reflects an affinity dependent switch in the scaffolding mode of PICK1 – that is , using both PDZ domains for interacting with membrane associated PDZ partners with low intrinsic affinity , but only one PDZ domain when the local bulk concentration is above or close to the intrinsic affinity . This behavior was recently predicted for bivalent ligands too using kinetic simulations ( Vauquelin , 2013; Vauquelin and Charlton , 2013 ) . Thus , after binding to the first binding site penalties associated with binding of the second site of the bivalent interaction ( e . g . from reduced entropy , straining of the molecule and steric hindrance ) may rather favor binding of a second molecule from the solution ( Figure 4a ) . Importantly , this behavior is likely favored by the large distance ( ~20 nm ) between the two PDZ domains in PICK1 ( Karlsen et al . , 2015 ) , which will render the effective local concentration , [L] , of the second PDZ domain after binding of the first relatively low compared to for example the tandem domains in PSD95 . On the other hand , one might expect the steric restriction , f , of the two closely spaced and structurally aligned PDZ grooves in PSD95 PDZ1 and 2 to be more prominent than in the flexible PICK1 . We also considered modeling explicitly an allosteric effect in the binding of PICK1 to the membrane embedded ligands ( Kramer and Karpen , 1998; Lu and Ziff , 2005 ) . However , the evidence for allostery in PICK1 is controversial ( Karlsen et al . , 2015 ) and an allosteric model is not needed to explain our data . Still , it should be noted that allostery could implicitly be contained in the f value . To better understand the complex behavior between scaffold proteins and proteins embedded in biological membranes further structural and dynamic insight of the system from experiments and molecular dynamics simulations would be needed combined with for example a statistical mechanics approach . Functionally , the scaffolding mode for high intrinsic affinity ligands ( including DAT and NET ) ( Erlendsson et al . , 2014 ) would leave one PDZ domain free for recruitment of cytosolic proteins ( e . g . kinases , phosphatases , GTPases ) . Only this mode supports the PICK1 function in regulating recycling of the β2AR . Since fast recycling of β2AR has been shown to rely on the PKA phosphorylation of S345/S346 ( Vistein and Puthenveedu , 2013 ) , and the PICK1 PDZ domain binds both Calcineurin B ( Iida et al . , 2008 ) and PKA regulatory subunits ( Ammendrup-Johnsen , Gether , Madsen , Unpublished results ) , PICK1 might regulate the β2-DAT trafficking by recruiting either of these components . For ligands with lower intrinsic affinity ( including GluA2 , ASIC1a , HER2 , Glt1b , mGluR7b and Ephrin B1 ) ( Erlendsson et al . , 2014 ) , however , both PDZ domains would be engaged with membrane-associated ligands . This scaffold mode will potentially lead to clustering of the ligands and possibly changes in lateral diffusion , rotational flexibility , and molecular orientation - including positioning of the C-terminus relative to the membrane . Posttranslational modification of the C-termini or PICK1 as well as variations in the local concentration of PICK1 might complicate this simplified scheme . Finally , since the PICK1 N-BAR domain , like other N-BAR proteins , is recruited to biological membranes in a curvature sensitive manner ( Herlo et al . , 2018 ) the PICK1 concentration on membranes might be almost two orders of magnitude higher in areas of high curvature such as endocytic structures , endosomes or vesicles in the biosynthetic pathway ( Bhatia et al . , 2009 ) . This , in turn , would strongly drive interaction with membrane embedded interaction partners towards the ‘blue’ ternary configuration with one PDZ domain open for recruitment of cytosolic proteins thereby evoking spatio-temporal control on such proteins . This may indeed be the actual scenario for the modulation of the recycling of the β2-DAT constructs . Vice versa , the switch in scaffolding mode would allow stoichiometrically more binding to membrane embedded ligands in areas of membrane curvature possibly feeding into the already recursive nature of membrane curvature sensing and deformation ( Madsen and Herlo , 2017 ) . Consequently , it would be very interesting to address , whether the clustered nature of the PICK1 binding on the SCMS’s would coincide with areas of high membrane curvature . In summary , we present a novel experimental angle on scaffold processes , allowing quantitative description of the ensemble of protein-protein as well as protein-lipid interactions on a cell membrane and our results redefine several aspects of the scaffold function . We firmly believe that this approach will continue to shape our understanding of the protein ensembles that orchestrate cellular signaling and trafficking processes with the aim of developing rational therapeutic strategies for targeting these interactions in disease . pRK5 SF-GluA2 was constructed by PCR amplifying the DNA sequence encoding the signal flag sequence ( SF ) from a previous SF-β ( 2 ) -adrenergic receptor construct ( Madsen et al . , 2012 ) with sequences coding for the restriction sites XhoI and AgeI attached in the ends of the primers . Using these restriction enzymes the PCR product was inserted into a pRK5 pHluorin-GluA2 construct ( a gift from Richard Huganir , Baltimore ) thereby removing the sequence coding for the pHluorin in the process . In the pRK5 SF-GluA2 +A construct an alanine was added in the end of the construct by a one step quick change . pcDNA3 . 1 SF-β ( 2 ) -adrenergic receptor constructs LKI and LKA were made from the previous pcDNA3 . 1 SF-β ( 2 ) -adrenergic receptor-DAT-LKV ( Madsen et al . , 2012 ) using one step quick changes . pcDNA3 . 1 Tac-YFP-DAT/GluA2 constructs were constructed by PCR amplifying the sequence encoding YFP from a pEYFP-C1 vector and inserting this in a previous TAC-DAT construct ( Madsen et al . , 2012 ) using a single HindIII site between the TAC and YFP sequence . The generation of FLAG-β2 DAT LKV , with the 8 C-terminal residues of the human DAT ( -TLRHWLKV ) and LKV + A with the 8 C-terminal residues of the human DAT with an additional alanine that disrupts the PDZ binding to PICK1 ( Bjerggaard et al . , 2004 ) ( -TLRHWLKVA ) , was described previously ( Madsen et al . , 2012 ) . FLAG-β2 DAT LKI and LKA were generated similar by PCR and the resulting fragments were cleaved with KpnI and BamHI and ligated into pcDNA3 FLAG-β2ARHis6 ( kind gifts from Dr . Mark von Zastrow , Departments of Psychiatry and Cellular and Molecular Pharmacology , University of California , San Francisco , CA ) . The V121E , L125E double mutation was introduced into pcDNA5/FRT/TO eYFP-PICK1 ( rat ) by quick change PCR and used for generation of a tetracyclin inducible stable Flp-In T-REx 293 cell line ( see below ) . The V121E , L125E double mutation was subsequently subcloned into FUGWH1sh18GFPPICK1 WT ( kind gift from Dr . Malenka , using a BsrG1 fragment and checking for orientation creating FUGWH1sh18GFPPICK1 V121E , L125E . SNAP-PICK1 was made by sub-cloning SNAP into the single MfeI site between the thrombin cleavage site and PICK1 in pET41 PICK1 ( Madsen et al . , 2005 ) resulting in an N-terminal fusion of PICK1 . PICK1 A87L was introduced in pET41a SNAP-PICK1 by quick-change PCR . PICK1 L121E , V125E was sub-cloned from the FUGWH1 vector into a pET41a vector reversing the R411 to the native G411 . PSD95 PDZ1-2 was a kind gift from Kristian Strømgaard and was prepared ultimately as described in Bach et al . ( 2008 ) . SNAP-PICK1 WT , SNAP-PICK1 A87L , PICK1 WT , PICK1 A87L and PICK1 121–125 proteins are all expressed in E . Coli BL21 DE3 pLysS in the pET41 vector in LB medium . PSD-95 PDZ1-2 were expressed in E . coli One Shot BL21 Star ( DE3 ) . Induced using 75 mg/L IPTG and grown at 30°C for 4 hr . For purification the cells were lysed and centrifuged for 30 min at 18000 rpm . Subsequently glutathione sepharose 4B beads ( Life technologies ) were added to the supernatant and the suspension was incubated under slow rotation for 1 hr at 4°C . The beads were pelleted ( 5 min 3000 g at 4°C ) and washed in TBS buffer ( 50 mM Tris pH 7 . 4 , 125 mM NaCl , 0 . 01% or 0 . 1% TX-100 , 1 mM DTT ( no DTT if malemide staining ) ) . The beads were then transferred to PD-10 spin columns ( BIO-RAD ) . For SNAP staining , SNAP-surface 488 , 549 , 647 was used ( NEB ) . Labeling efficiency for PICK1 was 72 ± 24% ( average ±SD across all experiments ) . For maleimide staining PICK1 WT and PICK1 L121E , V125E Alexa C5 malemide 488 , 568 , 647 ( ThermoFisher ) was used . The beads were then washed thoroughly to remove excess dye and then cleaved from the beads using Thrombin ( 0 . 075 U/μl Novagen ) . Prior to use protein was ultra-centrifuged at 100 , 000 g for 30 min to remove aggregates . The PSD95 PDZ1-2 domain protein was labeled with Alexa Fluor 488 Carboxylic succimidylester ( NHS ) ( ThermoFisher ) The buffer of the purified protein was changed to a 0 . 1 M sodium bicarbonate solution . 50–100 µL of the reactive dye ( 1 mg/ml ) was added under constant stirring . The mixture was incubated for 1 hr at room temperature before the reaction was stopped by the addition of 100 mM Tris buffer ( pH 8 . 3 ) and incubation for 15 min . Labeled protein and free dye were separated with a Micro Bio-Spin column P-30 , and another buffer exchange was performed to get the protein in HEPES buffer ( 10 mM HEPES , 150 mM NaCl , pH 7 . 4 ) . Before use , protein concentration and degree of labelling ( DOL ) of all constructs were measured taking both the protein and the dye into accountDOL=Adye∙ε280A280-Adye∙ε280εdye∙εdyewhere A280 and Adye represent the sample’s absorbance at 280 nm and the dye excitation wavelength , respectively , ε280 and εdye the extinction coefficients of the protein at 280 nm and of the dye , respectively . HEK 293 Grip Tite cells ( kind gift from Jonathan Javitch , Columbia University , USA ) grown in standard DMEM 1965 with serum and pen-strep . and 500 µg/ml geneticin . Cells were tested negative for mycoplasma . Cells were transfected using lipofectamine 2000 and 3 µg plasmid DNA . The cells were then seeded in six well plates with approx . 400 . 000 cells pr . well in 2 ml DMEM 1965 . Cells were used no later than 24 hr later in order to avoid cell clusters . Receptor labeling was achieved by using 1 µg/ml of Alexa Fluor 488/568/647 ( ThermoFisher ) primary conjugated monoclonal ANTI-FLAG M1 antibody ( sigma ) one hour prior to use . Labelling degree is determined as above . To generate a cell line with stable tetracycline-inducible expression of YFP-tagged PICK1 V121E , L125E , we used the Flp-In T-REx system and the Flp-In T-REx 293 cell line ( Invitrogen ) as previously described for YFP-tagged PICK1 ( Madsen et al . , 2012 ) . The cells were maintained in DMEM 1965 with Glutamax ( l-alanyl-l-glutamine ) containing 10% fetal calf serum at 37°C in a humidified 5% CO2 atmosphere . Cells were tested negative for mycoplasma . Before transfection , cells were selected using 15 µg/ml blasticidin and 100 µg/ml Zeocin ( both from Invitrogen ) . Cells ( 90% confluent ) were transfected using Lipofectamine 2000 ( Invitrogen ) with a total of 3 µg of DNA in a 1:9 ratio of the pcDNA5/FRT/TO with the eYFP-PICK1 V121E , L125E insert and pOG44 vector ( Invitrogen ) in Opti-MEM ( Invitrogen ) overnight . Cells were then split to 50% confluence and grown for an additional 24 hr with no antibiotics before selection was induced using 15 µg/ml blasticidin and 150 μg/ml hygromycin . Cells were maintained until visible foci appeared after which the cells were harvested , pooled , and further maintained as a polyclonal cell line ( Flp-In T-REx 293 eYFP-PICK1 V121E , L125E ) . For transient transfections of Flp-In T-REx 293 eYFP-PICK1 cells with B2 DAT constructs , cells were seeded in 25 cm2 cell flasks ( 1 × 106 cells ) or 75 cm2 cell flasks ( 3 × 106 cells ) and grown in medium without selection for ~20 hr to reach ∼70% confluence . Cells were transfected with Lipofectamine 2000 ( Invitrogen ) for 16 hr in medium using 0 . 1 μg of DNA/75 cm2 flask . In general , we transfected >80% of the cells . The transfected cells were trypsinized and seeded on polyornithine-coated coverslips in 6-well plates ( 300 , 000 cells/well ) or 96-well ELISA plates ( 35 , 000 cells/well ) for 48–72 hr prior to experiments . After ∼20 hr , medium was changed to new medium without or with tetracycline ( 1 µg/ml ) to induce expression of eYFP-PICK1 . Supported Cell-membrane sheets were prepared following procedure presented by Perez et al . ( Perez et al . , 2006a ) In brief , cover glasses ( Round 25 mm , thickness #1 , VWR 631–1346 ) were clenched for 20 min at max . power in a Harric plasma cleaner and then coated in 0 . 3 mM poly-L-ornithine hydrobromide ( Sigma-Aldrich Product # P-8638 ) for approx . 30 min and then washed in water . Cells were allowed to swell in ddH2O for at total of 60 s before pressing a cover glass slide with poly-L-ornithine treated surface down towards cells . The supported cell membranes on the cover glass were then overlaid with sheet buffer ( 120 mM KCl , 2 mM MgCl2 , 0 . 1 mM CaCl2 , 10 mM HEPES and 30 mM glucose pH 7 . 35 ) containing 1 mg/ml BSA and kept on ice for 20 min . washed and replaced with protein containing sheet buffer and incubated 2 hr on ice in the dark . Temperature was kept low to avoid aggregation of PICK1 . Sheets were then washed in sheet buffer and PBS and then fixated for 40 min . using 4% PFA before mounting onto object glasses using Prolong gold antifade reagent ( Life technologies ) . The stained sheets were visualized using a Zeiss LSM 510 confocal laser-scanning microscope using an oil immersion numerical aperture 1 . 4 63x objective ( Zeiss , Jena , Germany ) . The Alexa Fluor 488 dye antibody and YFP were excited with the 488 nm laser line from an argon–krypton laser , and the emitted light was detected using a 505–550 nm bandpass filter . BG-547 SNAP was excited at 543 nm with a helium–neon laser , and the emitted light was detected using a 560–615 nm band pass filter . The Alexa Fluor 647 was excited at 633 nm with another helium-neon laser , and the emitted light was detected using a 650 nm long pass filter . Resulting images were analyzed using ImageJ software . All scale bars on shown images corresponds to 10 µm . Freshly purified SNAP-PICK1 labeled with malemide Alexa 488 , 568 or 647 , or SNAP surface dye , was ultra-centrifugated at 100000 g in a Beckman airfuge to remove possible aggregates . From these a concentration series were made in intracellular mimicking buffer ( see above ) , and the prepared supported membrane sheets were then overlaid with protein containing solution for 2 hours . The level of SNAP-PICK1 binding pr . Ligand molecule in the membrane was quantified by dividing the measured 585/633 nm SNAP-PICK1 signal by the measured 505-550 nm YFP/Alexa 488 signal . The average intensities in areas containing intact membrane sheets were recorded by manually defining regions of interest covering these . One ROI per cell from approximately 10 cells were imaged for each condition . In the subsequent averaging process the following selection criteria were applied: All images were qualitatively assessed , and SCMS with too high ( oversaturated pixels ) or too low ( below 2σ , compared to background ) were discarded . The intensities were subsequently corrected based on degree of labeling and laser gain settings . Correction of measured channel intensities:Icorr=Iobs∙L%∙GV-Iback∙G ( V ) . Where , Iobs is mean intensity of sheet or ligand , Iback is mean intensity of background . Labeling degree correction was made using L%=Iobslabelling degree∙[100 , 1000] . If PMT gain is adjusted this is corrected by GV=V2V1α∙n , where α is Conductance of dynodes of PMT and n the number of dynodes inside the PMT . The fractional binding is calculated using as IcorrligandIcorr ( receptor ) . All curves form independent measurements are combined and subsequently normalized . All binding curves are systematically performed in parallel , and in presence of positive and negative controls . The resulting sigmoidal binding curves are fitted using the relationship:IPICKIRec=BottomIPICKIRec+TopIPICKIRec-BottomIPICKIRec1+10logKd[PICK1] Competition binding was performed by premixing 100 nM of labeled SNAP-PICK1 with increasing concentrations of unlabeled PICK1 WT . The reaction is then allowed to equilibrate for 120 min . Because of the assumed constructed binding kinetics the resulting binding curves are fitted to the equation given above for the saturation curves , and the half maximal binding is reported as the Kivalue . The heterodimers of PICK1 WT and the PDZ binding deficient mutant A87L were made by separately purification of either construct in 0 . 1% TX-100 . After centrifugation to remove aggregates the monomeric constructs are mixed ( simultaneously ) in intracellular mimicking buffer where the TX-100 concentration was diluted at least 100-times at all concentrations thereby favoring formation of heterodimers . PICK1 A87L is added in 4 times excess at all PICK1 WT concentration . For the homodimer oligomerization experiments TX-100 concentration is kept at 0 . 01% during purification . Triton X-100 did not directly affect the intrinsic affinity of the PDZ interaction in solution . Also , under conditions that did not allow monomer exchange , we observed no significant binding of SNAP-PICK1A87L or effect on binding of SNAP-PICK1WT ( Figure 5—figure supplement 3 ) . The fluorescence polarization assay was performed essentially as described in Madsen et al . ( 2005 ) and Erlendsson et al . ( 2014 ) . For saturation binding experiments , full-length PICK1 or mutants was diluted in buffer ( 50 mM Tris , pH 7 . 4 , 125 mM NaCl , 1 mM DTT , 0 . 01% Triton X-100 ) to various concentrations and a final volume of 100 l in black low-binding 96-well microtiter plates ( Corning Glass ) . A volume of 5 µl of Oregon Green-labeled DAT C13 peptide ( OrG DAT C13 ) was added to each well to a final concentration of 20 nM . For the FP competition binding assay , increasing concentrations of unlabeled peptide was diluted in the wells , and a fixed 70% saturating concentration of PICK1 or mutants was added together with OrG DAT C13 as described above . All plates were incubated on ice for 30 min and analyzed on a PolarStar Omega FP reader ( BMG , Germany ) using a 488 nm excitation filter and a 520 nm emission filter . FP is calculated using FP= ( IV-g∙IH ) / ( IV+g∙IH ) . Where g is the g-factor , and Iv and IH are the intensities of the emission measured in the vertical and horizontal planes , respectively . For the Saturation binding the FP value is a weighted average of bound and unbound ligand and therefore the kd can be fitted using the equation: FP-FPf= ( FPb-FPf ) ∙[Rt] kd+[Rt] , where FP is the observed FP value , FPf and FPb are the FP value for free and bound ligand , respectively , and [Rt] is the total PICK1 concentration . For competition binding , ki values for the peptide ligands were determined by fitting the binding curves to the equation , FP=FPf+ ( FPb-FPf ) ∙[Rt]kd∙1+xki+[Rt] , where x is the competitor and kd is the apparent dissociation constant for OrG DAT C13 . The apparent Kd was obtained as described before from the saturation binding experiments . All peptides were purchased from Shafer , Copenhagen , Denmark . For ELISA-based trafficking experiments , FLAG-tagged β2-AR receptor variants were labeled with 1 μg/ml M1 mouse anti-FLAG antibody for 30 min at 4°C in parallel in two 96-well plates . In half of the wells on each plate , receptor internalization was stimulated with isoproterenol ( 10 µM ) at 37°C for 25 min . The wells on the other half of the plate were left at 37°C and are referred to as non-treated . Subsequently , the action of the internalizing agent ( isoproterenol ) was terminated by addition of alprenolol ( 10 μM ) . One plate was left at 4°C for 1 hr to stop further trafficking , and the other plate was left at 37°C to allow further trafficking . Subsequently , cells were washed in DMEM 1965 , fixed for 10 min . at 4°C , and washed twice in PBS before 30 min blocking in PBS + 5% goat serum and incubation with 0 . 5 µg/ml horseradish peroxidase-conjugated goat anti-mouse IgG ( Thermo Scientific ) . Finally , cells were washed twice in PBS + 5% goat serum and twice in PBS before addition of SuperSignal ELISA Femto Maximum Sensitivity Substrate ( Thermo Scientific ) . The luminescence was detected in a Wallac Victor2 plate reader after 2 min . Internalization is expressed as the ratio of the surface signal from isoproterenol treated receptors relative to the non-treated cells on the 4°C plate . Recycling is expressed as the proportion of internalized receptor that was recovered at the cell surface during 1 hr . In Figure 5h , these values were all normalized to the respective signal of β2-LKV + A with and without tetracycline induction , respectively , which was assayed on separate plates . Statistical significance was determined using ANOVA ( Figure 5 ) or Student's t test ( Figure 6 ) , multiple samples , as indicated in legend . A thermodynamic cycle model provides a detailed representation of bivalent ligand binding ( Kramer and Karpen , 1998; Müller et al . , 1998; Vauquelin and Charlton , 2013 ) and is here adapted for homobivalent ligands , ‘aa’ . ( Figure 4a and Figure 4—figure supplement 1 ) . The ligand’s identical binding domains , ‘a’ , can bind simultaneously to the proximate , identical target sites ‘A’ ( target pairs are denoted as ‘AA’ ) . Each step is a reversible bimolecular process . The initial binding event allows the second , still free binding domain to acquire a high , constant concentration , [L] , near its target site . Its association to be represented by a composite first-order rate constant , k2 , that takes account of [L] , the intrinsic association rate constant , k1 , and a penalty factor , ‘f’ ( see Figure 4 ) . The model also allows a second ligand to bind to partly occupied AA . Ligands are assumed to be in large excess over the targets , so that their concentration in the bulk of the aqueous phase remains constant over time . To simulate the concentration and/or time- wise changes in the different modes of AA occupancy ( abbreviations in Figure 4 ) , the corresponding differential equations ( Figure 4—figure supplement 1 ) adapted from Vauquelin ( 2013 ) are successively solved in parallel over very small time intervals . Hippocampal neurons were prepared from embryonic day 19 Wistar rats . The hippocampi were isolated in ice-cold dissection media ( HBSS ( Gibco 14185–0529 , 1% penStrep ( Gibco 15140122 ) , 1% pyruvate ( Gibco 11360070 ) , 1% HEPES ( Gibco 15630080 ) , 30 mM glucose ) and cleared of blood vessels and meninges . The hippocampi were kept in 2 mL ice-cold dissection media until tituration . For tituration 40 μl of papain ( Worthington LS003119 ) was added and incubated for 20 min at 37°C . The dissection media was removed , and hippocampi washed twice with culturing media ( Neurobasal ( Gibco 21103–049 ) , 5% FCS , 1% PenStrep , 2% Glutamax 1 ( Gibco 35050–038 ) , 2% B27 supplement ( Gibco 17504044 ) ) . Titration was performed by pipetting 10 times up and down with the a glass pipette rounded in the end by burning , followed by 10 times with a pipette with end closed to half diameter by burning . Media was added to a total of 5 mL and filtered through a 70 µm cell filter . The neurons were plated on glass slides ( size 25 mm , thickness No . 1 , Glaswarenfabrik Karl Hecht , Sondheim , Germany ) in 2 ml of culturing media with a density of 150 , 000 cells pr . well . Prior to culturing the glass slides were treated with concentrated nitric acid for 3–4 hr , washed in milliQ water overnight and then burned in 96% ethanol . At 1 DIV the media is replaced with culturing media without FCS . At 14 DIV 5 µl of lentivirus packed with FUGWH1sh18GFPPICK1 WT or FUGWH1- sh18eGFPPICK1 V121EL125E was added to each well . The lentiviral was produced as previously described ( Eriksen et al . , 2009 ) . At 20–22 DIV the neurons were fixed in 4% PFA for 20 min ( 10 min on ice and 10 min at room temperature ) , washed three times in PBS , permeabilised and blocked in 0 . 05% triton X100% and 5% goat serum for 20 min at room temperature . Then labelled with mouse anti-PSD95 ( Antibodies inc . #72–028 ) 1:500 and rabbit anti-GFP ( abcam #ab290 ) 1:500 for 1 hr at room temperature . Followed by staining with goat-anti mouse Alexa-488 and goat-anti rabbit Alexa-568 both 1:500 . After three washes with PBS and one wash with milliQ water the slides were mounted to coverslips . Neurons were imaged using an inverted confocal laser-scanning microscope ( LSM 510 , Carl Zeiss ) . The Alexa 488 was excited with the 488 nm laser line from an argon–krypton laser , and the emitted light was detected using a 505–550 nm bandpass filter , whereas the Alexa 568 was excited at 543 nm with a helium–neon laser , and the emitted light was detected using a 585 nm long-pass filter . A Zeiss Plan-Neofluar 63X/1 . 3 oil immersion lens was used for imaging . Images were analysed with the ImageJ software . A region of interest that contains transduced dendrites after the first branch point from soma was chosen . In the PICK1 channel the image was thresholded at two times the mean intensity , in the PSD95 channel the images were threshold at a fixed value within one experiment and the number of particles quantified . All statistics performed using One-way ANOVA using Dunnett's multiple comparisons post test or parametric students t-test , both using 95% confidence intervals . ns p>0 . 05 , p≤0 . 05 , p≤0 . 01 , p≤0 . 001 , ***p≤0 . 0001 . All in-text numbers are reported as the mean of independent experiments ± s . e . m unless stated otherwise . Graphpad Prism 6 ( La Jolla California USA , www . graphpad . com ) was used for all graphs and statistics .	Inside a cell , many different signals carry information that is essential for the cell to remain healthy and perform its role in the body . It is , therefore , very important that the signals are sent to the right places at the right times . Scaffold proteins play an essential role in organizing these signals by bringing specific proteins and other molecules into close contact at particular times and locations within the cell . Defects in scaffolding proteins can lead to cancer , psychiatric disorders and other diseases , so these proteins represent potential new targets for medicinal drugs . Many scaffolding proteins assemble groups of proteins on the surface of the membrane that surrounds the cell . Previous studies have shown that scaffolding proteins are able to bind to several other proteins as well as the membrane itself at the same time . However , the precise way in which scaffolding proteins assemble such groups is not clear because it is technically challenging to study this process in living cells . To overcome this challenge , Erlendsson , Thorsen et al . used a new experimental setup known as supported cell membrane sheets – which provides direct access to the side of the cell membrane that usually faces into the cell – to study two scaffolding proteins known as PICK1 and PSD-95 . The experiments show that PICK1 and PSD-95 bind to their partner proteins up to 100 times more strongly than previously observed using other approaches . This is due to the scaffolding proteins binding more strongly to both their partners and the membrane . Unexpectedly , the experiments show that the shape and physical characteristics of the partner protein have no effect on the increase in the strength of the binding . Further experiments suggest that altering the ability of the PDZ domain of PICK1 to bind to partner proteins changes the mode of action of the PICK1 protein so that it can activate different responses in the cell . Together these findings imply that the ability of scaffolding proteins to bind to their partner proteins is finely tuned to encode specific responses in cells in different situations – a hypothesis that Erlendsson , Thorsen et al . are planning to test in intact cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2019	Mechanisms of PDZ domain scaffold assembly illuminated by use of supported cell membrane sheets
The extent to which brain structure is influenced by sensory input during development is a critical but controversial question . A paradigmatic system for studying this is the mammalian visual cortex . Maps of orientation preference ( OP ) and ocular dominance ( OD ) in the primary visual cortex of ferrets , cats and monkeys can be individually changed by altered visual input . However , the spatial relationship between OP and OD maps has appeared immutable . Using a computational model we predicted that biasing the visual input to orthogonal orientation in the two eyes should cause a shift of OP pinwheels towards the border of OD columns . We then confirmed this prediction by rearing cats wearing orthogonally oriented cylindrical lenses over each eye . Thus , the spatial relationship between OP and OD maps can be modified by visual experience , revealing a previously unknown degree of brain plasticity in response to sensory input . In cats and monkeys neurons in the primary visual cortices are selective for both the orientation of the visual input ( orientation preference , OP ) and its eye of origin ( ocular dominance , OD ) ( Hubel and Wiesel , 1977 ) . These feature preferences are arranged spatially in the form of OP and OD maps , with stereotypical structure within each map , and strong spatial relationships between them ( Blasdel and Salama , 1986; Bonhoeffer and Grinvald , 1991; Bartfeld and Grinvald , 1992; Obermayer and Blasdel , 1993; Hübener et al . , 1997; Nauhaus et al . , 2012 ) . While some aspects of the structure of OD and OP maps individually are plastic in response to altered visual input , such as monocular deprivation ( Hubel et al . , 1977; Shatz and Stryker , 1978; Farley et al . , 2007 ) or stripe rearing ( Sengpiel et al . , 1999; Tanaka et al . , 2006 ) , none of these manipulations has succeeded in modifying the overall spatial relationships between OD and OP maps , which have appeared immune from environmental influence . In particular OP map pinwheels , where domains representing all orientations meet at a point , always tend to lie close to the center of OD regions . This is true even after manipulations of the visual input such as rearing animals with artificially induced strabismus ( Hubel and Wiesel , 1965; Löwel , 1994; Löwel et al . , 1998 ) or monocular deprivation ( Crair et al . , 1997 ) . However , whether this relationship is a fundamental aspect of map structure that is determined by innate mechanisms ( Godecke and Bonhoeffer , 1996; Crair et al . , 1998; Kaschube et al . , 2002; Katz and Crowley , 2002; Tomita et al . , 2013 ) , and thus beyond the limits of brain plasticity , is unclear . Computational models of map formation based on Hebbian plasticity principles have played an important role in understanding the mechanisms governing visual development . In particular ‘dimension reduction’ models predict negative correlations between the local gradient magnitudes of different maps ( Durbin and Mitchison , 1990; Swindale , 1996 ) , thus explaining why pinwheels , which have a high orientation gradient , normally tend to lie near the centre of OD regions , where the ocularity gradient is small . However , such models also suggest that the spatial relationship between pinwheels and OD regions might be sensitive to visual experience ( Giacomantonio et al . , 2010 ) . Here we used a computational model to predict how rearing animals with visual input biased to vertical orientations in one eye and horizontal orientations in the other eye ( cross-rearing ) would change these relationships . We then confirmed this prediction by raising cats with weak ( -10 dioptre ) cylindrical lenses placed in front of their eyes throughout the critical period . This demonstrates a form of plasticity in the relationships between visual feature maps that has not previously been observed . The elastic net algorithm ( Durbin and Mitchison , 1990 ) uses Hebbian learning to optimize a trade-off between coverage and continuity constraints , and can explain many aspects of visual map formation ( Swindale , 1996; Goodhill , 2007 ) . When simulating normal rearing ( Erwin et al . , 1995; Carreira-Perpinan and Goodhill , 2004; Carreira-Perpinan et al . , 2005 ) this reproduces the experimental observations cited above that pinwheels tend to be located near the center of OD columns . However , previous simulations of map development ( Giacomantonio et al . , 2010 ) suggested that this relationship would be disrupted when horizontal orientations were over-represented in one eye and vertical orientations were over-represented in the other ( Hirsch and Spinelli , 1970; 1971; Blakemore , 1976 ) . Here we simulated this scenario using the elastic net algorithm to quantify further the degree to which the relationship between pinwheels and OD borders would change as a function of the strength of over-representation ( see Materials and methods ) . To measure the relationship of pinwheels to OD regions we separately divided the left and right eye regions of simulated OD maps into five equally sized bins , which represented areas of the OD map from the centres of the OD columns to the border regions , similarly to Hübener et al . ( 1997 ) . As the strength of over-representation ( α ) increased , the histograms became increasingly biased away from the center of OD regions , indicating a shift in the relative location of pinwheels towards the OD borders ( Figure 1 ) . To determine whether this is due to the movement of pinwheels , OD borders , or both , we then fixed the random seed in the algorithm and explored how the spatial relationships changed within reproducible maps as a function of α . An example is shown in Figure 1g , from which it is clear that both pinwheel positions and OD borders move in the cross-reared case . The average distance that pinwheels move from their original ( α = 1 ) positions as a function of α is shown in Figure 1h . Thus , in the model , cross-rearing alters the spatial relationship between pinwheels and OD borders by causing movement of both . 10 . 7554/eLife . 13911 . 003Figure 1 . The elastic net model predicts changes in spatial map relationships under cross-rearing . ( a ) Simulated orientation preference map ( colours ) , orientation pinwheels ( black dots ) , and ocular dominance borders ( black lines ) under normal rearing ( relative strength of over-representation of horizontal and vertical contours in the input α = 1 ) . ( b ) Histogram of pinwheel locations relative to the OD borders under normal rearing , showing a preference for pinwheels located near the centre of OD regions as previously observed experimentally . Error bars show ± 1 SEM from 10 independent simulations . The dashed line shows the expected distribution if pinwheels were arranged randomly . ( c ) Simulated orientation preference map for the cross-reared condition ( α = 3 ) . ( d ) Histogram of pinwheel locations relative to OD borders for this case . ( e , f ) Simulated orientation preference maps and corresponding histogram of pinwheel locations relative to OD borders for a higher level of cross-rearing ( α = 5 ) . In the simulations of cross-rearing , pinwheels are shifted away from the centre and towards the border of OD regions . ( g ) A cropped region of a simulated OP and OD map produced with the same random seed but increasing strengths of over-representation . Circles show the location of a pinwheel and lines show the location of the adjacent OD border . Both the pinwheel and the OD border move under cross-rearing relative to their positions under normal rearing ( α = 1 ) , but the distance between them decreases . ( h ) The average distance that pinwheels move from their original positions ( measured in units of average OP map wavelength ) as a function of the strength of cross-rearing . Errors bars show ± 1 SEM across all pinwheels in the map . Scale bars in panels a , c , e and g indicate 15 pixels in the simulated feature maps . Source data for this figure are available in Figure 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13911 . 00310 . 7554/eLife . 13911 . 004Figure 1—source data 1 . This HDF5 file contains the numerical values shown in Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13911 . 004 However , plasticity of this type has not yet been observed experimentally , leaving open the possibility that the relationship between pinwheels and OD columns could instead be determined by intrinsic mechanisms and is not susceptible to environmental modification . To directly test these predictions we reared cats from 3 weeks of age with -10 dioptre cylindrical lenses mounted comfortably in front of their eyes using soft neoprene masks . The lens covering the left eye had its axis aligned vertically , so that the left eye was exposed to high contrast contours with primarily horizontal orientation . Conversely , the lens covering the right eye had its axis aligned horizontally , so that the right eye was exposed to high contrast contours with primarily vertical orientation ( Figure 2 ) . The lens strength was chosen based on preliminary observations that animals wearing -10 dioptre lenses exhibited normal behavioural activity , while animals with higher power lenses ( e . g . , as used in Tanaka et al . [2006] ) were noticeably less active . Animals wore the masks for 6 hr per day while the room was illuminated and were otherwise kept in darkness . During light periods the animals were monitored at least every 30 min to ensure the masks remained in place and to encourage active visual behaviours , such as chasing light patterns and balls . Paw striking behaviour towards objects in front of the animals appeared normal . From age 20 weeks we used intrinsic signal optical imaging followed by extracellular single unit recordings to map OP and OD preferences in cortical areas 17 and 18 of 5 cross-reared animals ( 10 hemispheres ) , and 6 normally reared control animals ( 11 hemispheres ) . 10 . 7554/eLife . 13911 . 005Figure 2 . Optical characteristics of the -10 dioptre cylindrical lenses . ( a ) Circular square wave test grating ( 1 cycle/° ) viewed normally ( no lens ) . ( b ) The same grating viewed through the -10 dioptre cylindrical lens , with the lens axis aligned horizontally , attenuating horizontal and preserving vertical contours . ( c ) The radially symmetric distribution of power over spatial frequency for the test grating viewed normally . ( d ) The radially asymmetric distribution of power over spatial frequency for the same grating when viewed through the -10 dioptre cylindrical lens with the lens axis aligned horizontally . Contours orthogonal to the axis are preserved while contours parallel to the axis are attenuated . Power spectra shown in ( c ) and ( d ) are normalized to the peak power ( black ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13911 . 005 We recorded extracellular spiking responses to quantify the tuning properties of single units in both the normal and cross-reared animals . In the five cross-reared animals we recorded from 182 units from 20 electrode tracks , 76 with a preference for input from the right eye ( which experienced predominantly vertical contours ) and 106 with a preference for input from the left eye ( which experienced predominantly horizontal contours ) . In the six control animals , we recorded from 86 units from 17 electrode tracks . Electrode tracks were positioned without reference to cortical map structure . Consistent with previous work ( Coppola et al . , 1998; Li et al . , 2003 ) and theoretical predictions ( Hunt et al . , 2013 ) , the distribution of preferred orientation for units from the control animals exhibited an over representation of the cardinal orientations ( Figure 3a ) . This distribution was well described by a sine curve with period 90º ( r2 = 0 . 69 ) . In contrast , distributions of preferred orientation for units from the cross-reared animals showed clear biases depending on the eye providing the dominant input . Units with a preference for input from the left eye showed a bias for horizontal orientations ( Figure 3b , black line ) , while those with a preference for input from the right eye showed a bias for near vertical orientations ( Figure 3b , gray line ) . These distributions were consistent with the orientation of the lenses fitted over each eye and were well described by sine curves with period 180º ( r2 = 0 . 75 , left eye; r2 = 0 . 79 , right eye ) . In contrast to the distribution from the control animals , sine curves with period 90º provided a relatively poor account of the data ( r2 = 0 . 11 , left eye; r2 = 0 . 03 , right eye ) . 10 . 7554/eLife . 13911 . 006Figure 3 . Tuning properties of single units . ( a ) The distribution of preferred orientation in control animals exhibited an over representation of cardinal orientations . The dashed line shows the expected distribution if all orientations were represented equally . The best-fitting sine curve with period 90° had peaks at 84° and 174° ( thick line , r2 = 0 . 69 ) . ( b ) Distributions of preferred orientation in the cross-reared animals , for units driven predominantly by input from the left ( black ) and right eye ( grey ) . The best-fitting sine curves with period 180° peaked at 174° for the left eye ( thick black line , r2 = 0 . 75 ) and 104° for the right eye ( thick grey line , r2 = 0 . 79 ) . Cross-rearing thus caused a systematic change in the distributions of preferred orientation of single units . Cross-rearing also caused an increase in monocularity ( c ) and an increase in preferred spatial frequency ( d ) of single units . Source data for this figure are available in Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13911 . 00610 . 7554/eLife . 13911 . 007Figure 3—source data 1 . This HDF5 file contains the numerical values shown in Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 13911 . 00710 . 7554/eLife . 13911 . 008Figure 3—figure supplement 1 . Inter-ocular difference in preferred orientation of single units is not altered by cross-rearing . The mean inter-ocular difference in preferred orientation ( ΔOP ) is significantly greater than 0 in both control ( mean ΔOP = 11 . 8° , p < 0 . 001 , two-tailed t-test ) and cross-reared animals ( mean ΔOP = 9 . 8° , p < 0 . 001 , two-tailed t-test ) . The distributions of ΔOP for single units from control and cross-reared animals are not significantly different ( p = 0 . 44 , two-tailed , two-sample t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13911 . 00810 . 7554/eLife . 13911 . 009Figure 3—figure supplement 2 . Preferred temporal frequency of single units is not altered by cross-rearing . The distributions of preferred temporal frequency for single units from control and cross-reared animals are not significantly different ( p = 0 . 2 , Kruskal-Wallis test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13911 . 00910 . 7554/eLife . 13911 . 010Figure 3—figure supplement 3 . Contrast sensitivity of single units is not altered by cross-rearing . The distributions of semi-saturation contrast ( σ , see Equation 4 ) for single units from control and cross-reared animals are not significantly different ( p = 0 . 81 , Kruskal-Wallis test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13911 . 010 We also calculated the inter-ocular difference in preferred orientation , ∆OP , for each unit as the preferred orientation for the right eye minus the preferred orientation for the left eye . Consistent with previous reports ( Nelson et al . , 1977; Cooper and Pettigrew , 1979 ) we found a torsional disparity in the preferred orientation of the two eyes in both our control ( mean ΔOP = 11 . 7° , p<0 . 001 , two-tailed t-test ) and cross-reared ( mean ΔOP = 9 . 8° , p<0 . 001 , two-tailed t-test ) animals . We found no significant difference in the distribution of ΔOP for control and cross-reared animals ( p=0 . 44 , two-tailed , two-sample t-test; Figure 3—figure supplement 1 ) . Units from cross-reared animals showed a higher level of monocularity than those from control animals ( Figure 3c ) . The median monocularity index ( MI , see Material and methods ) of units from control and cross-reared animals was 0 . 24 and 0 . 38 , respectively . This difference was significant ( p=0 . 002 , Kruskal-Wallis test ) . Cross-rearing therefore appears to induce subtle changes in the combination of input from the two eyes at the level of single neurons . In addition to each unit’s ocular dominance and orientation preference , we also quantitatively measured their tuning for spatial and temporal frequency ( See Materials and methods ) . We found no significant differences in the monocular spatial or temporal frequency tuning for dominant vs non-dominant eyes or for left vs right eyes ( regardless of dominance ) in either control or cross-reared animals ( p>0 . 05 , Kruskal-Wallis tests ) . We therefore combined the populations of left- and right-eye dominant units within the two experimental groups ( i . e . , control and cross-reared ) and compared their tuning parameters for the dominant eye . Units from cross-reared animals exhibited marginally higher preferred spatial frequencies compared to those from control animals ( median preferred spatial frequency 0 . 24 and 0 . 16 cycles/° for cross-reared and control animals respectively; Figure 3d ) . While this difference was significant ( p<0 . 001; Kruskal-Wallis test ) , the tuning curves were very broad and overlapped substantially . We found no difference in either the bandwidth ( p=0 . 19; Kruskal-Wallis test ) or skew ( p=0 . 40; Kruskal-Wallis test ) of the spatial frequency tuning curves for normal and cross-reared animals . Similarly , we found no significant difference in either the preferred temporal frequency ( p=0 . 2 , Kruskal-Wallis test; Figure 3—figure supplement 2 ) , the temporal frequency bandwidth ( p=0 . 28 , Kruskal-Wallis test ) or the skew of the temporal frequency tuning curves ( p=0 . 63 , Kruskal-Wallis test ) of units from control and cross-reared animals . Using the optimal grating parameters ( orientation , size , spatial and temporal frequency ) for each unit we also measured their response as a function of stimulus contrast . We found no difference in either the maximum spike rate ( p=0 . 56 , Kruskal-Wallis test ) or the semi-saturation contrast ( p=0 . 81 , Kruskal-Wallis test; Figure 3—figure supplement 3 ) of units from control and cross-reared animals . Selectivity for stimulus parameters , responsiveness and sensitivity to stimulus contrast therefore appeared normal in the cross-reared animals . While previous studies have usually used red light ( wavelengths >600 nm ) for intrinsic signal imaging , here we used green light ( 520 nm ) due to the much higher signal to noise ratio obtained ( Figure 4a ) . Conventional techniques for map generation from intrinsic signal imaging data are based on the calculation of difference images by either subtraction or division by a reference image that is independent of the stimulus of interest . We found that these techniques left strong blood vessel artifacts in the maps derived from green light data , particularly for ocular dominance maps . However we were able to overcome this problem by using extended spatial decorrelation ( ESD ) , a more sophisticated analysis technique ( Stetter et al . , 2000 ) ( see Materials and methods ) . This technique robustly separated the noise and mapping signals even in the green light data ( Figure 4b–d ) . Consistent with a brief earlier report ( Frostig et al . , 1990 ) , by imaging one of our control animals with both green ( 520 nm ) and red ( 609 nm ) wavelengths we found that the maps produced were very similar ( Figure 4e–h ) . We also compared measures of orientation preference from the OP maps with corresponding measures obtained from single unit recordings from the superficial layers of the cortex at each electrode track location . Where we obtained robust estimates from both the imaging and unit recordings we found a high level of correlation between the two measures ( r2 = 0 . 64 , p=0 . 003 ) with a median absolute difference of 15 . 2° . 10 . 7554/eLife . 13911 . 011Figure 4 . Extended spatial decorrelation recovers OP and OD maps from green light imaging . ( a ) Time course of the relative change in reflectance ( ∆R/R ) during a trial , averaged over all pixels and all trials , measured with red ( 609 nm ) and green ( 520 nm ) light . The shaded region shows the stimulus period . Consistent with earlier reports ( Sirotin and Das , 2009; Sirotin et al . , 2009 ) , green light produced a much stronger signal . ( b–d ) Representative source coefficient time series from the extended spatial decorrelation algorithm ( see Materials and methods ) . Sources 46–50 ( as per legend ) for ( b ) the real component of the OP map , ( c ) the imaginary component of the OP map , and ( d ) the OD map . The shaded region shows the stimulus period . The mean over the pre-stimulus period was subtracted from each source . It is clear in each case that one source ( in this case source 50 ) most strongly represents the signal of interest . ( e , f ) OP maps generated from the green ( e ) and red ( f ) light responses . ( g , h ) OD maps generated from the green ( g ) and red ( h ) light responses . Over the region shown here ( the one used for analysis ) , the correlation between red and green OP maps was r2 = 0 . 78 and between the red and green OD maps was r2 = 0 . 77 . Colour encodes the preferred orientation in the OP maps ( as per legend ) and brightness encodes eye preference in the OD maps , with black and white representing the left and right eyes , respectively . Data from a control animal . Scale bars: 1 mm . Source data for this figure are available in Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13911 . 01110 . 7554/eLife . 13911 . 012Figure 4—source data 1 . This HDF5 file contains the numerical values shown in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 13911 . 012 We measured OP and OD maps in cortical areas 17 and 18 of both normal and cross-reared animals using intrinsic signal optical imaging ( typical maps for each case are shown in Figure 5a–f ) . Cross-rearing caused a very slight reduction in orientation selectivity , as derived from the OP maps ( Figure 5—figure supplement 1 ) . Although the selectivity distributions appear very similar , the difference was statistically significant ( p<0 . 001 , two-sample Kruskal-Wallis test ) . However cross-rearing induced profound changes in map structure . In control animals there was an over-representation of cardinal ( horizontal and vertical ) orientations in the OP map . This is consistent with the distribution of orientation preferences of single units described above , and with previous reports for ferrets ( Coppola et al . , 1998 ) and cats ( Li et al . , 2003 ) . The proportion of the map devoted to different orientations was well fit by a sine curve with period 90º ( r2 = 0 . 6; Figure 5g; Figure 5—figure supplement 2 ) . However , in the cross-reared animals the OP maps calculated for each eye showed proportions that were now better fit by sine curves with period 180º ( r2 = 0 . 64 , left eye; r2 = 0 . 71 , right eye ) . In each case these curves peaked close to the orientation of the lens covering that eye ( Figure 5h; Figure 5—figure supplement 2 ) . As a comparison , the best-fitting sine curves with period 90º had r2 values of 0 . 33 ( left eye ) and 0 . 13 ( right eye ) . Thus , cross-rearing caused substantial shifts in the distribution of orientations across the cortex . 10 . 7554/eLife . 13911 . 013Figure 5 . Cross-rearing changes the distribution of orientation preferences . ( a ) OP map , ( c ) OD map and ( e ) overlay of OD and OP contours for a control cat . ( b ) OP map , ( d ) OD map and ( f ) overlay of OD and OP contours for a cross-reared cat . While qualitatively the control and cross-reared maps look similar , quantitative analysis revealed differences . ( g ) Proportion of cortical area representing different orientations from binocular stimulation for all control hemispheres ( thin line: mean ± 1 SEM , thick line: least-squares sine curve fit ) . The best-fitting sine curve with period 90° had peaks at 7° and 97° , and an r2 value of 0 . 6 . For comparison the dashed line at a frequency of 1/8 represents equal proportions . ( h ) Data from left ( thin black line ) and right ( thin grey line ) monocular stimulation for all cross-reared hemispheres . The best-fitting sine curve with period 180° peaked at 0° for the left eye ( horizontal orientations , thick black line ) and 148° for the right eye ( vertical orientations , thick grey line ) . The r2 values for the fits were 0 . 64 ( left eye ) and 0 . 71 ( right eye ) . In contrast the best-fitting sine curves with period 90° ( not shown ) had r2 values of 0 . 33 ( left eye ) and 0 . 13 ( right eye ) . Thus cross-rearing caused a systematic shift in the proportions of the maps occupied by each orientation , towards the orientation that each eye predominantly experienced . As in Figure 4 , colour encodes the preferred orientation in the OP maps and brightness encodes eye preference in the OD maps . Scale bars: 1 mm . Source data for this figure are available in Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13911 . 01310 . 7554/eLife . 13911 . 014Figure 5—source data 1 . This HDF5 file contains the numerical values shown in Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 13911 . 01410 . 7554/eLife . 13911 . 015Figure 5—figure supplement 1 . The distribution of orientation selectivity is slightly altered by cross-rearing . The mean distributions of normalised orientation selectivity for all control and cross-reared hemispheres . Normalised orientation selectivity was calculated as the absolute value of each pixel in the OP map , normalised to the maximum selectivity value in each map separately . Cross-reared hemispheres have slightly lower median orientation selectivity ( p<0 . 001 , two-sample Kruskal-Wallis test ) . Although the difference is very small , it is statistically significant because each data point is derived from many hundreds of pixels . Error bars show ± 1 SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 13911 . 01510 . 7554/eLife . 13911 . 016Figure 5—figure supplement 2 . Distributions of orientation preferences in each hemisphere . ( a ) Proportion of cortical area representing different orientations from binocular stimulation for all control hemispheres ( colored lines ) , their mean ( thin black line ) , and the best fitting sine curve with period 90° ( thick black line ) . ( b , c ) Proportion of cortical area representing different orientations from monocular stimulation of the left and right eyes , respectively , for all cross-reared hemispheres ( colored lines ) , their mean ( thin black lines ) , and the best fitting sine curve with period 180° ( thick black lines ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13911 . 016 The characteristic relationships between OP and OD maps are intersection angles and the distance of pinwheels to OD borders . There was no statistically significant change in pinwheel density ( Kaschube et al . , 2010 ) between rearing conditions ( Figure 6a ) . The distribution of intersection angles between OD and OP maps was also similar in both rearing conditions ( Figure 6—figure supplement 1 ) . There were subtle changes in the spatial distribution of orientation selectivity: in control animals orientation selectivity was slightly greater near OD borders , while in cross-reared animals selectivity was slightly greater near the centre of OD regions ( Figure 6—figure supplement 2 ) . However there was a clear effect on the distance of pinwheels to OD borders . In our control animals , the distribution of pinwheels relative to OD borders was very similar to that of Hübener et al . ( 1997 ) ( Figure 6b ) but in cross-reared animals there was a significant shift of pinwheels away from the centre of OD regions ( Figure 6c ) . Specifically , there was a significant under representation of pinwheels in the bin corresponding to the centre of OD columns ( p=0 . 01 , two-tailed t-test , power = 0 . 77 , 95% confidence interval for difference in means = [0 . 06 , 0 . 36] ) , as predicted in our model simulations ( Figure 6d ) . Thus , the altered visual input during cross-rearing changed a fundamental aspect of the spatial relationship between OP and OD maps . The similarity between the model prediction and our data provides further evidence that models based on dimension reduction , such as the elastic net , capture essential elements of the mechanisms by which cortical maps develop . 10 . 7554/eLife . 13911 . 017Figure 6 . Spatial relationship between pinwheels and ocular dominance is modified by rearing condition . ( a ) Pinwheel density relative to squared map wavelength was not significantly different between control and cross-reared animals , both being consistent with the theoretically predicted value of π ( dashed line ) ( Kaschube et al . , 2010 ) . ( b ) Pinwheel locations relative to the centres/borders of OD regions were quantised into 5 bins similarly to Hübener et al . ( 1997 ) . For control animals , pinwheels were disproportionately overrepresented at the centre of OD regions ( n = 71 pinwheels total in control hemispheres ) , consistent with previous data ( Hübener et al . , 1997 ) . ( c ) In strong contrast , for the cross-reared animals pinwheels were disproportionately underrepresented at the centre of OD regions ( n = 55 pinwheels total in cross-reared hemispheres ) . ( d ) Computational simulations using the elastic net reproduced the shift of pinwheels away from the centres of OD regions in the cross-reared compared to control condition ( data replotted from Figure 1b , f ) . For all graphs error bars show ± 1 SEM . p-values in ( a ) and ( c ) are from two-tailed , two-sample t-tests . Source data for this figure are available in Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13911 . 01710 . 7554/eLife . 13911 . 018Figure 6—source data 1 . This HDF5 file contains the numerical values shown in Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 13911 . 01810 . 7554/eLife . 13911 . 019Figure 6—figure supplement 1 . The distribution of intersection angles of the contours of the OP and OD maps is unchanged by cross-rearing . ( a ) The distributions for control and cross-reared hemispheres are not significantly different ( p = 0 . 36 , two-sample Kruskal-Wallis test ) . For the quantification we used ( see Materials and methods ) , the random distribution is a sine curve . ( b ) For comparison , the distributions for control and cross-reared hemispheres using the same quantification as in Hübener et al . ( 1997 ) . Here , the random distribution is uniform . For these data we cannot say in either case that the distributions are non-random . Errors bars show ± 1 SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 13911 . 01910 . 7554/eLife . 13911 . 020Figure 6—figure supplement 2 . The spatial layout of orientation selectivity is very slightly altered by cross-rearing . The mean , normalised orientation selectivity in each of the five ocular dominance bins used to quantify pinwheel location relative to ocular dominance borders ( see Materials and methods ) , averaged over all hemispheres in each condition . In control animals , selectivity was higher near the OD borders . but in cross-reared animals selectivity was higher near OD centres . * denotes a significant difference between conditions ( p < 0 . 001 , two-tailed , two-sample t-tests ) . Error bars show ± 1 SEM over all the pixels in each bin . DOI: http://dx . doi . org/10 . 7554/eLife . 13911 . 020 Despite the apparent robustness of the relationship between pinwheels and OD columns in earlier studies that have manipulated visual input , here we showed that this relationship could be altered by cross-rearing . We found consistent changes in cortical response properties from both our electrophysiology and optical imaging . Our findings agree with previous cell population surveys and imaging experiments on cats raised in environments with a single orientation ( stripe-rearing ) ( Blakemore and Cooper , 1970; Sengpiel et al . , 1999 ) or with orthogonal orientations presented to the left and right eye ( Hirsch and Spinelli , 1970; Blakemore , 1976 ) . These studies , like our own , revealed an over-representation of single units tuned to the orientation that matches the rearing conditions . Interestingly , we found no significant difference in ΔOP – the inter-ocular difference in preferred orientation – of units from our control and cross-reared animals . This might seem surprising given the mismatched inputs in the cross-reared animals . However , previous work using much stronger orientation biases found that cells tend to become monocular rather than maintain large ΔOP values ( Hirsch and Spinelli , 1970; 1971; Blakemore , 1976 ) . Consistent with these earlier reports , we observed an increase in the monocularity of units from our cross-reared animals compared to the controls ( median monocularity index , MI , of 0 . 38 versus 0 . 24 for the cross-reared and control animals , respectively ) . However , only about 10% of single units in our cross-reared animals were monocular , suggesting that relatively normal binocular summation developed in these animals . This is consistent with the relatively low power lenses that we used during rearing ( -10 dioptres ) , compared to the higher power lenses used by others ( 67 dioptres , cats , Tanaka et al . ( 2006 ) ; 167 dioptres , mice , Kreile et al . ( 2011 ) ) . Evidently , cortical neurons in our cross-reared animals retained access to a sufficiently broad range of orientations to form fused binocular images during development . Similarly , units from our cross-reared animals generated similar maximum spike rates and had indistinguishable contrast sensitivity to the control animals , again suggesting that the lenses had little impact on basic visual processing . A theoretical prediction based on symmetry principles was that pinwheels move during normal development and annihilate in pairs ( Wolf and Geisel , 1998 ) , leading to a final pinwheel density relative to map wavelength of π ( Kaschube et al . , 2010 ) . However chronic imaging experiments have suggested that , with normal visual input , the positions of pinwheels do not move from the positions at which they are first observed during development ( Chapman et al . , 1996; Crair et al . , 1998 ) . In adult cats , chronic imaging of OP maps in response to spatially synchronous cortical activity induced by intracortical electrical microstimulation has shown substantial rearrangement of pinwheel positions ( Godde et al . , 2002 ) . However , our results reveal for the first time that pinwheels can be repositioned relative to OD columns purely by abnormal sensory input . In simulations of normal rearing , pinwheels form near the centre of OD columns to provide good coverage ( Figure 1a , b ) : the gradient of OP is highest where the gradient of OD is lowest . However in the cross-reared case the over-represented orientations are the first to appear in the cortex , aligned with the initial formation of OD columns ( Giacomantonio et al . , 2010 ) . The representation of other orientations then forms around this initial configuration , and thus pinwheels are pushed away from the centre of the OD regions , which are dominated by the over-represented orientations ( Figure 1c–f ) . In the model , movement of both pinwheels and OD borders contribute to the change in their spatial relationship; however we cannot directly assess whether this is also true in our experimental data . The degree of plasticity of the tuning of V1 neurons depends on their position in the OP map , in particular whether they lie close to a pinwheel or in an iso-orientation domain ( i . e . depending on the spatial gradient of OP ) . However , the direction of this dependence is controversial: while Dragoi et al . ( 2001 ) reported greater plasticity of tuning curves at pinwheels in response to an adapting visual input , Schuett et al . ( 2001 ) found less plasticity at pinwheels in response to pairing a visual stimulus at a particular orientation with direct cortical electrical stimulation . It would be interesting to perform cortical microstimulation studies in cross-reared animals , to determine whether the plasticity of pinwheels that have been repositioned away from the center of OD regions is the same as unperturbed pinwheels in normally reared animals . From a methodological perspective , we found that blood vessel artefacts common in intrinsic signal imaging of cortex with green light can be overcome by using more statistically sophisticated analysis techniques than are typically employed . Changes in cortical reflectance of green wavelengths are believed to indicate changes in blood volume , while those at longer ( red ) wavelength primarily indicate changes in blood oxygenation ( Sirotin and Das , 2009; Sirotin et al . , 2009 ) . Several previous studies have employed shorter wavelengths for intrinsic signal imaging ( Spitzer et al . , 2001; Versnel et al . , 2002; Sirotin and Das , 2009; Sirotin et al . , 2009 ) and our results suggest that this does not strongly affect the resulting maps , provided appropriate analysis techniques are used ( Figure 4 ) . Notably , the strength of the lenses worn by our cross-reared animals ( -10 dioptres ) was considerably less than that in recent stripe rearing studies in cats ( 67 dioptres ) ( Tanaka et al . , 2006 ) and mice ( 167 dioptres ) ( Kreile et al . , 2011 ) . This was motivated by our preliminary observations that weak lenses were required to avoid obvious behavioural changes in the kittens . The relatively low power of the lenses in our experiment suggests sensitivity of visual cortical structure to relatively small variations in sensory input . This raises the intriguing possibility that a component of the large variability in visual map structure between individuals , seen in cats ( Kaschube et al . , 2002 ) and humans ( Adams et al . , 2007 ) , could be partly due to individual variations in visual experience ( astigmatism , the closest analogy to rearing with cylindrical lenses in humans , has been reported with powers of up to 6 dioptres [Mitchell et al . , 1973] ) . More generally , our results redefine the limits of cortical plasticity , and emphasize the importance of appropriate patterns of sensory stimulation during critical periods for normal brain development . Five animals wore -10 dioptre cylindrical plano-concave lenses mounted in soft neoprene rubber masks ( Dzioba et al . , 1986 ) . The lenses were fitted to the masks with the planar surface facing towards the cat’s eyes . The axis of the cylindrical component for the left eye was vertical and for the right eye was horizontal . Figure 2a shows a pattern of concentric circles ( spatial frequency 1 cycle/° ) . Figure 2b shows the same pattern viewed through a -10 dioptre cylindrical lens with its axis oriented horizontally . Both images were captured using a Nikon D90 DSLR camera fitted with a Nikkor 50 mm f/1 . 2 lens . The cylindrical lens compresses the image along the direction orthogonal to its cylindrical axis . Figures 2c and d show the distribution of power over spatial frequency for images of the test pattern viewed normally ( Figure 2c ) and viewed through the -10 dioptre cylindrical lens ( Figure 2d ) . Animals were housed indoors in windowless rooms measuring 2 . 5 m × 3 . 0 m . Cross-reared animals wore the masks for 6 hr each day from 3 weeks of age . For the remaining 18 hr each day ( i . e . , when not wearing the masks ) the room lights were extinguished . We took special care to eliminate any stray light entering the room around the doorframe . Six control animals ( i . e . , not cross-reared ) were housed under identical conditions , subject to the same 6:18 hr light:dark cycle . The walls of the rooms housing both control and cross-reared animals were covered with wallpaper consisting of high-contrast images containing all orientations . We performed optical intrinsic signal imaging of primary visual cortex in eleven adult cats ( 3 . 3–3 . 8 kg; 6 control , 5 cross-reared ) . Animals were prepared for acute physiological recordings as described previously ( van Kleef et al . , 2010 ) . Animals were anaesthetised by intramuscular injection of ketamine hydrochloride ( 10 mg . kg-1 ) and medetomidine ( 15 μg . kg-1 ) . Once deeply anaesthetised , as confirmed by the absence of corneal and toe withdrawal reflexes , animals were placed on a heated surgery table and intubated to ensure adequate respiration . Anaesthesia was then maintained for the remainder of the surgery by inhalation of gaseous isofluorane ( 0 . 7–1 . 0% in O2 ) . To ensure that adequate levels of oxygenation and anaesthesia were maintained at all times ( during surgery and throughout the remainder of the experiment ) , animals were immediately instrumented to allow continuous monitoring of non-invasive physiological indicators , including heart rate and saturation of peripheral oxygen ( SpO2 ) , blood pressure , the electroencephalogram ( EEG ) and end-tidal CO2 concentration . The EEG was sampled continuously at 256 Hz and Fourier analysis was performed within a sliding window 30 s in duration . The depth of anaesthesia was considered adequate when the power in the EEG was concentrated in the delta band ( <4 Hz ) and no change in power distribution was observed in response to noxious stimuli . An increase in power in the EEG at frequencies >8 Hz was interpreted as a sign the depth of anaesthesia may be inadequate . If at any stage the depth of anaesthesia was deemed to be inadequate the concentration of the anaesthetic agent was increased . The cephalic vein was cannulated to permit delivery of fluids and intravenous drugs . Animals were then transferred to a stereotaxic frame and the head fixed using ear bars , a bite bar and a screw placed on the skull at the midline 30 mm anterior to inter-aural zero . Body temperature was maintained at 37 . 7°C by means of an electric blanket under feedback control . A craniotomy ( 10 mm × 15 mm; A4 to P6 ) was made spanning the midline to expose primary visual cortex ( Areas 17 and 18 ) in both hemispheres . A stainless steel chamber was fixed to the cranium with dental acrylic . The dura mater was removed and the recording chamber filled with silicone oil ( Dow Corning 200 , 50 cSt ) and sealed with a glass cover slip . To prevent eye movements during recording , animals were subject to neuro-muscular blockade by continuous intravenous infusion of vecuronium bromide ( 0 . 1 mg . kg-1 . h-1 ) . During neuro-muscular blockade animals were mechanically ventilated to maintain end-tidal CO2 between 3 . 5 and 4% . For fluid replacement all animals received a constant intravenous infusion of Hartmann’s solution ( 25% by volume ) , 5% glucose in 0 . 9% NaCl solution ( 25% by volume ) and a 10% amino acid solution ( Synthamine-17; 50% by volume ) at a rate of 2 . 5 mL . kg-1 . h-1 . Animals also received daily injections of atropine ( 0 . 05 mg . kg-1; s . c . ) to reduce salivation , dexamethasone phosphate ( 1 . 5 mg . kg-1; i . m . ) to reduce cerebral oedema , and a broad-spectrum antibiotic ( Clavulox; 0 . 2 mL . kg-1; i . m . ) to prevent infection . To prevent desiccation of the corneas the eyes were fitted with neutral power rigid gas-permeable contact lenses . Refractive errors were assessed by reverse ophthalmoscopy and corrected as required using spherical lenses placed in front of the eyes to focus the stimulus on the retina . Eye drops ( 1% atropine sulphate; 10% phenylephrine hydrochloride ) were administered daily to cause dilation of the pupils and retraction of the nictitating membranes . After surgery and for the remainder of the experiment , anaesthesia was maintained by inhalation of gaseous halothane ( 0 . 5–1 . 0% ) in a 60:40 mixture of N2O and O2 . At the conclusion of the experiment animals were euthanized by intravenous injection of an overdose of barbiturate ( sodium pentobarbital; 150 mg . kg-1 ) . Visual stimuli were generated by a ViSaGe visual stimulus generator ( Cambridge Research Systems Ltd . , Cambridge , UK ) and displayed on a calibrated Clinton Monoray CRT monitor ( modified Richardson Electronics MR2000HB-MED CRT with fast DP104 phosphor , 100 Hz refresh , resolution 1024 × 768 pixels ) viewed binocularly or monocularly from a distance of 28 cm . For imaging , stimuli consisted of luminance defined oriented square wave gratings presented within a circular aperture ( diameter 60° ) on an isoluminant grey background matched to the mean luminance of the gratings . Orientation preference maps were obtained by recording responses to presentation of high contrast ( Michelson contrast , 100% ) gratings ( 0 . 15 cycles per degree ) drifting ( temporal frequency 2 Hz ) in one of 16 directions equally spaced between 0° and 360° . Each stimulus direction , together with a blank condition ( no grating ) , was presented at least 30 times with the order of presentation randomised across trials . For single unit recordings , stimuli consisted of patches of sine wave gratings presented within a circular aperture on an isoluminant grey background matched to the mean lumunance of the gratings . For each recorded unit we systematically identified optimal grating parameters ( orientation , direction of motion , spatial and temporal frequency , size and position ) for stimuli presented monocularly to the right and left eye . The exposed cortex was imaged using a Pantera 1M60P high-sensitivity 12-bit area scan CCD camera ( Teledyne DALSA , Waterloo , ON Canada ) fitted with a macroscope consisting of a pair of Nikkor 50 mm f/1 . 2 lenses ( Ratzlaff and Grinvald , 1991 ) . The focal plane was positioned 500 μm below the cortical surface , determined by focusing on the surface vasculature and then lowering the camera by way of a micromanipulator . The camera was configured to bin sensor pixels 2×2 producing images with a resolution of 512 × 512 pixels ( each pixel being 24 μm square ) . Image acquisition was restricted to a region of interest comprising a subset of the full imaging frame . During imaging , the cortex was epi-illuminated using a custom built LED light source with a peak wavelength of 520 nm ( Agilent Technologies; HSMQ-C150 ) . Local increases in cerebral blood flow , indicative of neural activity , caused a reduction in cortical reflectance and a corresponding reduction in the resulting image intensity . Cortical responses were imaged during presentation of an ensemble of visual stimuli ( gratings , described above ) . For each stimulus presentation images were acquired continuously at a rate of 5 Hz for a period of 10 s , the onset of which was synchronized to the phase ( maximum inspiration ) of the respirator . Visual stimuli were presented for 5 s beginning 2 s after the beginning of image acquisition . Each stimulus presentation was followed by a recovery period ( minimum duration 3 s ) during which the stimulus monitor displayed an isoluminant gray screen , the luminance of which was matched to the mean luminance of the gratings . All data processing and analyses were conducted using MATLAB . MATLAB code for map pre-processing , generation and analysis is available at https://github . com/nickjhughes/feature-map-stats . Each raw image encompassed both hemispheres , and was cropped to two rectangular regions , one covering the exposed region of areas 17 and 18 in each hemisphere . In cats , both areas 17 and 18 receive direct input from the dorsal lateral geniculate nucleus and are therefore both primary visual cortex ( Payne and Peters , 2001 ) . The remaining analyses were performed on each hemisphere separately , hereafter a ‘dataset . ’ All of the images in each dataset were then spatially aligned . The first image in the dataset was arbitrarily chosen as the reference frame , and every other image was translated to maximise the linear correlation between it and the reference frame . Trials from opposite directions of stimulus motion were pooled , and then all trials for each stimulus condition for each dataset were pixel-wise averaged , resulting in 50 imaging frames for each of 8 stimulus orientations ( 0° , 22 . 5° , … , 157 . 5° ) , for both eyes . Each of these frames was then high-pass Gaussian filtered ( σ = 20 pixels = 480 μm ) to remove large scale changes in illumination across the images , and low-pass Gaussian filtered ( σ = 2 pixels = 48 μm ) to satisfy the requirement of the extended spatial decorrelation method ( see below ) that the sources be smooth . Image frames were zero-padded prior to filtering . This had no effect on the map regions subsequently analysed since they were sufficiently far from the edges of the frames . The sign of the values in all images were then reversed , as a decrease in reflectance indicates an increase in neural activity . To generate feature maps from the data we initially tried vector-averaging methods typically used for red-light data ( Hübener et al . , 1997 ) . While these produced OP maps similar to those we subsequently produced using the extended spatial decorrelation ( ESD ) method of Stetter et al . ( 2000 ) ( see below ) , the OD maps produced were compromised by blood vessel artefacts . We therefore employed the ESD method for our analysis as follows . Each map ( i . e . , the two vector components of the orientation preference ( OP ) map and the ocular dominance ( OD ) map ) was constructed by combining the responses to the appropriate stimulus conditions . Let Rθ , L be the overall average response to orientation θ produced by the left eye , and similarly for the right eye: ( 1 ) real ( OP ) =∑θ0 . 5 ( Rθ , L+Rθ , R ) cos ( 2θ ) imag ( OP ) =∑θ0 . 5 ( Rθ , L+Rθ , R ) sin ( 2θ ) OD=∑θRθ , R−∑θRθ , L where θ values in the sum were 0° , 22 . 5° , … , 157 . 5° . Monocular OP maps were generated similarly , except that only the left or right eye response was considered , rather than the average of the two . Following the process described by Stetter et al . ( 2000 ) , these maps were generated using each of the 50 imaging frames separately , resulting in maps that consisted of 50 frames . The frames of each map were first decorrelated from each other using principal component analysis ( i . e . , each frame’s variance was made equal to 1 and the covariance , between frames , was made equal to 0 ) . Single-shift ESD was then performed on each map using a shift of Δr = ( 5 , 5 ) pixels , resulting in 50 sources . The particular shift chosen had no appreciable difference on the resulting maps . The time series of the coefficients of each source were used to choose which of the 50 sources corresponded to the feature map in each case . The source chosen was the one with the coefficient time series that most closely matched that of the typical intrinsic signal optical imaging response ( i . e . , a rise at the time of stimulus onset , up to a maximal value by the time the stimulus was turned off ) . The correct source was obvious in most cases ( for an example see Figures 4b–d ) . In some cases , the correct source was chosen by looking at which source had both a response with the same spatial extent as the corresponding OP map and an appropriate coefficient time series . Each frame of the chosen source was then multiplied by its corresponding coefficient , and the feature map was defined as the average over frames 31 to 35 ( i . e . , the final 5 stimulus frames ) . Finally , each map was low-pass Gaussian filtered ( σ = 12 pixels = 288 μm ) . These filter parameters were chosen to remove any remaining high-frequency noise ( due to remaining blood vessel artifacts ) while minimising spatial distortion of the resulting feature maps . Only areas inside masks were used in the following analyses . Each analysis mask was defined as the area within the anatomical boundaries of area 17/18 , with any areas without strong orientation preference removed . All of the above processing was performed prior to this masking . For a binocular or monocular orientation map , orientation distributions were generated as follows . Each pixel was binned into 22 . 5° wide orientation preference bins centred on 11 . 25° , 33 . 75° , … , 168 . 75° , according to its preferred orientation . These bins were used to generate a histogram , which we refer to as the orientation distribution . Sine curves were fit to the mean of the histograms for a particular condition ( binocular orientation maps from control cats , and left and right monocular orientation maps from cross-reared cats ) using least-squares estimation . The period was set to 90° for control maps , and 180° for cross-reared maps , matching the clear and expected periodicity of the data . Fits to the cross-reared maps of sine curves with period 90° were also performed for comparison . The estimated value of the phase of the sine curve was then used to determine the location of the maxima . The quality of the fits was assessed using r2 , the square of the correlation coefficient of the data and the fitted function . To calculate the crossing angle distributions , the zero-level contour of the OD map was calculated , as were the 0° , 22 . 5° , … , 157 . 5° contours of the OP map , and the angle between the OD and OP contours at all points of intersection were calculated . The angle was defined as the difference between the orientations of the tangents of the two contours at the intersection point . The distributions of these angles were calculated for all orientation maps . We compared these between rearing conditions , and also against the expected distribution for unrelated maps , which follows a sine curve ( Morton , 1966 ) . Pinwheel locations were defined in orientation maps as described previously ( Carreira-Perpinan et al . , 2005 ) : an integral of orientation preference was calculated along a closed path of radius 1 pixel around each pixel , and the pixel was defined as a pinwheel if the integral was ± 180° . For eight-connected clusters of such pixels , the cluster's centre of mass was defined as the pinwheel location . Pinwheel density was calculated as described previously ( Kaschube et al . , 2010 ) : the number of pinwheels per pixel multiplied by the square of the map wavelength . The wavelength was defined as the mean Fourier wavelength of the map averaged over all directions . To measure the relationship between the location of pinwheels and the OD domains we used the metric described by Hübener et al . ( 1997 ) with a slight modification . The positive and negative values in each OD map were separately binned into 5 equally sized regions , corresponding to values ranging from the centres to the borders of the OD columns , and the bin corresponding to each pinwheel was calculated . The bins corresponding to both the positive and negative sections were pooled , leaving 5 bins corresponding to values from the centre to the border of the OD columns . Thus , the centre of the OD column falls into the bin below the 20th percentile and the border region into the bin above the 80th percentile . This metric differs slightly from that used by Hübener et al . ( 1997 ) in that it considers the regions on either side of 0 separately , rather than binning all the values into 10 equal bins . This modification respects OD borders and takes into account contralateral bias in the OD map . Elastic net simulations were performed as described previously ( Carreira-Perpinan et al . , 2005; Giacomantonio et al . , 2010 ) . A MATLAB implementation of the elastic net algorithm is available at http://faculty . ucmerced . edu/mcarreira-perpinan/research/EN . html . Parameters for our simulations were as follows . Feature dimensions: 20×20 spatial positions in a square array in a unit square , 2 OD values of -0 . 05 and 0 . 05 and 6 OP values of radius 0 . 08 for each spatial position , giving 4800 feature points in total . Cortical array: 128×128 with non-periodic boundary conditions , α ( weighting of coverage term ) = 1 for non-overrepresented orientations , α > 1 for the overrepresented orientation in each eye , and β ( weighting of continuity term ) = 10 . Initial value of annealing parameter K = 0 . 2 , multiplied by 0 . 9925 at each iteration . Simulations ( n = 10 per condition ) were terminated at K = 0 . 0358 , shortly after the OD and OP maps had formed . The only difference in parameters between simulations of the control and cross-reared conditions were the values of α . After intrinsic signal imaging , extracellular signals from single units were acquired using gold or platinum tipped , lacquer coated tungsten microelectrodes ( FHC , Bowdoin , ME USA ) . The extracellular potential was amplified , band-pass filtered ( 300 Hz – 5 kHz ) and then sampled at 40 kHz using a 1401Plus and Spike2 software ( Cambridge Electronic Designs , Cambridge , UK ) . After isolating a single unit ( based on the size and shape of the extracellular spike waveform ) the approximate size and location of the receptive fields for the right and left eye were mapped using hand-driven bright or dark bars projected onto a tangent screen . The size and location of the receptive fields together with the unit’s tuning for direction , spatial and temporal frequency , and it’s sensitivity to stimulus contrast were then determined quantitatively using circular patches of sine wave gratings under the control of the stimulus computer . The receptive fields of all recorded units were located within 10° of the area centralis .	The structure of the brain results from a combination of nature ( genes ) and nurture ( environment ) . The brain’s ability to adapt to changes in the environment is known as plasticity , and the young brain is especially plastic . An animal’s sensory experiences in early life help to determine how its brain will process sensory input as an adult . One of the best sensory systems in which to study this process is the visual system . Within the visual system , some brain cells respond only to input from the left eye and others only to input from the right eye . Cells that respond to input from the same eye are arranged to form columns . Within each column , some cells respond only to lines with a particular orientation . Cells with different preferred orientations are grouped together in patterns that resemble pinwheels . The relative positions of the pinwheels and eye-specific columns within the brain tissue belonging to the visual system have so far been robust to changes in visual experience during development , suggesting that they are determined by an animal’s genes . However , Cloherty , Hughes et al . have now tested the unexpected predictions of a computer model . The model suggested that rearing animals so that they saw mostly vertical lines through one eye , and mostly horizontal lines through the other , would cause a form of plasticity that had never been observed before . Specifically , it would change the relative positions of the pinwheels and eye-specific columns within the visual parts of the brain . This prediction turned out to be correct . Young cats that wore special lenses – which slightly distorted what they saw but did not obviously affect their behavior – showed the predicted changes in brain structure . The results confirm that this aspect of brain structure is partly determined by nurture , as opposed to being entirely specified by nature . A key future challenge is to identify the chemical signaling that enables sensory input to have these effects on brain structure . It might then be possible to use drugs to restore normal brain activity in cases where abnormal sensory input has altered the brain , for example in the condition known as amblyopia ( or “lazy eye” ) .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Sensory experience modifies feature map relationships in visual cortex
Treatment for medulloblastoma , the most common malignant brain tumor in children , remains limited to surgical resection , radiation , and traditional chemotherapy; with long-term survival as low as 50–60% for Sonic Hedgehog ( Shh ) -type medulloblastoma . We have shown that the transcription factor Atonal homologue 1 ( Atoh1 ) is required for Shh-type medulloblastoma development in mice . To determine whether reducing either Atoh1 levels or activity in tumors after their development is beneficial , we studied Atoh1 dosage and modifications in Shh-type medulloblastoma . Heterozygosity of Atoh1 reduced tumor occurrence and prolonged survival . We discovered tyrosine 78 of Atoh1 is phosphorylated by a Jak2-mediated pathway only in tumor-initiating cells and in human SHH-type medulloblastoma . Phosphorylation of tyrosine 78 stabilizes Atoh1 , increases Atoh1’s transcriptional activity , and is independent of canonical Jak2 signaling . Importantly , inhibition of Jak2 impairs tyrosine 78 phosphorylation and tumor growth in vivo . Taken together , inhibiting Jak2-mediated tyrosine 78 phosphorylation could provide a viable therapy for medulloblastoma . Medulloblastoma ( MB ) is the most common malignant primary brain tumor in children ( Dhall , 2009 ) and is thought to originate from rhombic lip-derived precursors that have failed to differentiate into their usual progeny ( Hatten and Roussel , 2011 ) . Molecular studies have allowed us to classify MB into four major subtypes: WNT , sonic hedgehog ( SHH ) , Group C , and Group D ( Taylor et al . , 2012 ) . The cell of origin differs among the subgroups , as does the clinical prognosis ( Taylor et al . , 2012 ) . SHH-type tumors arise from precursors of the cerebellar granule neurons and can contain mutations in Smoothened , Patched , or other genes in this pathway ( Rimkus et al . , 2016 ) . Elevated expression of Atoh1 , the basic helix-loop-helix transcription factor that is crucial for the development of the cerebellar granule neurons ( Ben-Arie et al . , 1996 ) , can also drive uncontrolled cell proliferation and formation of medulloblastoma ( Flora et al . , 2009 ) . In addition to attempting to inhibit SHH signaling—a strategy that can be foiled by the development of secondary mutations ( Yauch et al . , 2009 ) —we hypothesized that controlling the levels of Atoh1 protein will provide another therapeutic entry point for the treatment SHH-driven medulloblastoma . To determine the influence of Atoh1 gene dosage on MB development and prognosis , we examined the rate of tumor formation in a mouse model of Shh-dependent MB , the Nd2::SmoA1Tg line ( Hatton et al . , 2008 ) . These mice have a constitutively active Shh pathway in all cerebellar granule cells and spontaneously develop MBs ( Hatton et al . , 2008 ) . Whereas almost a quarter ( 23 . 1% ) of Nd2::SmoA1Tg mice developed MB and succumbed to the disease within 300 days of life , fewer than one in ten ( 6 . 7% ) Nd2::SmoA1Tg mice – heterozygous for Atoh1 ( Atoh1+/- , Nd2::SmoA1Tg ) – developed signs of disease ( Figure 1a , Figure 1—source data 1 for all statistical analysis ) . Atoh1 heterozygosity dropped the overall tumor incidence by over three-fold and never reached that of Nd2::SmoA1Tg , Atoh1 wildtype mice . Importantly , while both groups expressed the tumorigenic transgene ( Nd2::SmoA1 ) at similar levels ( Figure 1—figure supplement 1a ) , levels of Gli2 , the primary Shh mediator in medulloblastoma and direct target of Atoh1 ( Read et al . , 2009; Klisch et al . , 2011 ) were reduced in the Atoh1 heterozygous mice . Hence , a 50% reduction in Atoh1 protein level has a direct , positive , effect on survival in this medulloblastoma mouse model . This suggests that Atoh1 protein dosage is critical for the development of medulloblastoma in mice . The regulation of Atoh1 expression during the development of the cerebellum is highly complex , involving not only Shh signaling but also other signaling pathways required for proper development , such as BMP and WNT signaling ( Butts et al . , 2014 ) . While this regulation is transcriptional in nature , another layer of regulation has been reported to influence Atoh1 protein levels: phosphorylation ( Forget et al . , 2014 ) . Hence , we asked whether Atoh1 was aberrantly phosphorylated in the tumors . Using tumors isolated from Nd2::SmoA1Tg mice , we performed IP-MS on tumor-initiating cells that express Atoh1 [marked by cell surface protein CD15 ( Read et al . , 2009 ) ] and CD15-negative cells ( Figure 1—figure supplement 1b ) . We tested the integrity of both cell populations by flank injections of sorted cells in immunosuppressed mice and found that Atoh1 expression in the tumor-initiating cells is required to drive secondary tumor formation ( Figure 1—figure supplement 1b ) . Using Atoh1 IP-MS , we discovered that tyrosine 78 ( Y78 , Figure 1—source data 2 ) on Atoh1 is phosphorylated exclusively in tumor-initiating cells . We confirmed phosphorylation of this Atoh1 residue in MB tissue using a phospho-specific antibody ( p-Atoh1-Y78 , Figure 1b ) . We further confirmed these findings in tumorigenic and control tissues from mice and found that Atoh1 Y78 phosphorylation is specific to the tumor-initiating cells ( Figure 1c ) . Y78 immuno-reactivity was observed mostly in the initiating cell islets of the tumor; neither the Y78-specific phospho-antibody , nor an antibody against CD15 , reacted in the normal tissue , even in the presence of strong Atoh1 expression in the external granule layer ( Figure 1c ) . Because Atoh1 is not endogenously expressed in neuronal cell lines , and reliable primary cultures of human SHH-type medulloblastoma cells are not established , we turned to modify and study medulloblastoma cell lines that have been used in the past . We engineered cell lines expressing either doxycycline-inducible wildtype ( Y78 ) , phospho-dead ( Y78F ) or phospho-mimetic ( Y78D ) Atoh1 , using three different human-derived MB parental cell lines: DAOY ( marked in blue ) , UW228 ( orange ) and ONS-76 ( purple ) ( Zanini et al . , 2013 ) . While these cell lines have been used extensively , it has been recently suggested that at least the DAOY cell line is more ‘Shh-like’ than the others ( Higdon et al . , 2017 ) . We found that Y78 phosphorylation status did not alter the interaction of Atoh1 with its E-protein binding partner or its nuclear-cytoplasmic localization ( Figure 2—figure supplement 1a–b ) . However , upon doxycycline withdrawal , Y78 phosphorylation dramatically increased Atoh1 protein stability in all three parental lines ( Figure 2a , Figure 2—figure supplement 1c–d ) . Protein half-life of phospho-mimetic Atoh1 was triple that of wildtype Atoh1 ( Figure 2—source data 1 ) , while the RNA stability of the transgenes was comparable ( Figure 2—figure supplement 1e ) . To test the transcriptional activity of the Atoh1 variants we used an Atoh1-specific luciferase reporter system ( Klisch et al . , 2011 ) . Overexpression of phospho-mimetic Atoh1 ( Y78D ) , but not the phospho-mutant ( Y78F ) , increased Atoh1 transcriptional activity ( Figure 2b , Figure 2—figure supplement 2b–c ) . In a proliferation assay using all three parental cell lines , we found that the enhanced activity of Atoh1 Y78D resulted in hyper-proliferation , whereas the phospho-dead Atoh1 resulted in a slight decrease in cell proliferation ( Figure 2—figure supplement 2d–f ) . While the increase in transcriptional activity could be due to the increased half-life , the steady-state levels of the three Atoh1 mutants were similar in the luciferase lysates ( Figure 2—figure supplement 2a ) . These data argue for two distinct effects of tyrosine 78 phosphorylation: increased Atoh1 stability as well as increased Atoh1 transcriptional activity . To test whether this phosphorylation drives tumor growth in vivo , we injected DAOY cells [known to form secondary tumors ( Jacobsen et al . , 1985 ) ] stably expressing doxycycline-inducible Atoh1-GFP WT , Atoh1-GFP Y78F , or Atoh1-GFP Y78D in flank xenografts . While these flank xenograft models lack the fully functional immune system , they provide a useful model for longitudinal tumor growth monitoring with minimal burden to the animal . The mice were fed doxycycline chow continuously to express the transgenes . Expression of wildtype or phospho-dead Atoh1 ( Y78F ) led to a similar average tumor weight , but overexpression of phospho-mimetic Atoh1 ( Y78D ) increased tumor weight by roughly 40% ( Figure 2c ) . These data suggest that tyrosine 78 phosphorylation increases Atoh1 transcriptional activity , which in turn increases cellular proliferation in vitro as well as in vivo . To identify the tyrosine kinase responsible for the aberrant phosphorylation of Atoh1 in the malignant cells , we revisited our IP-MS data . We recovered 637 Atoh1-interacting proteins from CD15-positive cells and 278 interactors from CD15-negative cells ( Figure 1—source data 2 ) . As would be expected , E-proteins were the strongest interactors present in all pull down assays , but interestingly only one tyrosine kinase , Janus kinase 2 ( Jak2 ) , was present exclusively in tumor-initiating cells . The Jak2 paralog Jak1 was present in the non-initiating cell population ( Figure 1—source data 2 ) . The Janus kinases typically function in the cytoplasm , downstream of cytokine signaling , and activate canonical cascades such as signal transducer and activator of transcription ( STAT ) , mitogen activated protein kinase ( MAPK ) , and/or the phosphotidylinositol 3-kinase ( PI3K ) –Akt pathways ( Levine et al . , 2007 ) . Although one of the most common gain-of-function mutations in Jak2 ( Jak2V617F ) is associated with increased proliferation in human myeloproliferative diseases ( Levine et al . , 2007 ) , Jak2 function in medulloblastoma has not been investigated . We tested Jak1-3 and Tyrosine Kinase 2 ( Tyk2 ) , as well as Protein Tyrosine Kinase 2 ( Ptk2 ) in a bimolecular fluorescence complementation assay ( BiFC ) for an Atoh1 interaction ( Neet and Hunter , 1996 ) . Only Jak2 displayed a strong interaction with Atoh1 as indicated by the shift in GFP signal detected by flow cytometry ( Figure 3—figure supplement 1a ) . We further investigated whether these kinases were able to phosphorylate Y78 using a luciferase-based kinase assay . Once again , only Jak2 was able to phosphorylate wildtype Atoh1 ( Y78 ) in a dose dependent manner in vitro ( Figure 3—figure supplement 1b ) . This phosphorylation was specific to Y78 , as the addition of phospho-mutant Atoh1 ( Y78F ) abolished this effect ( Figure 3—figure supplement 1b ) . Given that both Atoh1 and Jak2 are expressed during cerebellar development and both are highly overexpressed in MB tissue ( Figure 3—figure supplement 1c ) , we tested whether the Jak2-Atoh1 interaction occurs in vivo . We immunoprecipitated Atoh1 or Jak2 and found a strong interaction in malignant but not normal cerebellar tissue ( Figure 3a ) . To determine whether Jak2 is a critical kinase phosphorylating tyrosine 78 , we isolated Shh-type MB cells and cultured them in the presence or absence of a strong and highly specific Jak2 inhibitor ( AG490 ) ( Kobayashi et al . , 2015 ) . One-hour treatment with the inhibitor was sufficient to drastically reduce Jak2 and Atoh1 Y78 phosphorylation ( Figure 3b ) . These findings demonstrate that Jak2 is activated in Shh-type MB , where it interacts with Atoh1 and phosphorylates it on tyrosine 78 . We examined whether Jak2 influences Atoh1 activity through its canonical downstream cascades by using the Atoh1 transcriptional reporter assay . Co-transfection of Atoh1 and Jak2 dramatically increased Atoh1 activity in all three MB lines ( Figure 3c , Figure 3—figure supplement 1d–e ) . Inhibiting Jak2 with AG490 abolished this , but blocking STAT , PI3K or MAPK signaling in this assay had no effect ( Figure 3c , Figure 3—figure supplement 1d–e ) , arguing that Jak2 influences Atoh1 directly . Prior studies have argued for non-canonical roles of Jak2 , e . g . Jak2 phosphorylates tyrosine 41 on histone H3 ( Dawson et al . , 2009 ) . Moreover , most of the genes regulated by Jak2 do not contain the predicted STAT DNA-binding motifs ( Ghoreschi et al . , 2009 ) suggesting the existence of unidentified transcriptional mediators . Overexpression of Jak2 pheno-copied the transcriptional gain-of-function phenotype of Atoh1 Y78D , increased the stability of wildtype Atoh1 ( Y78 ) to Y78D levels ( Figure 3—figure supplement 2 , Figure 2—source data 1 ) and increased cellular proliferation in vitro , which was blocked by phospho-mutant Atoh1 ( Y78F ) and the addition of AG490 ( Figure 3—figure supplement 3 ) . Having identified Jak2 as a critical kinase phosphorylating tyrosine 78 , we asked if inhibition of Jak2 might be a viable strategy to decrease Atoh1 levels in vivo . We isolated Shh-type medulloblastoma cells , infected them with lentiviral particles harboring Jak2 shRNAs tagged with RFP , and grafted these into mice . Control shRNA-infected tumors grew rapidly , but both Jak2 shRNA-infected tumors showed dramatically reduced tumor growth ( Figure 4a ) , in accordance with their potency to inhibit Jak2 ( Figure 4b ) . The decrease in Jak2 resulted in a concomitant decrease in Y78 phosphorylation ( Figure 4b ) . Moreover , Jak1 , a close relative of Jak2 , did not compensate for the inhibition of Jak2 as it was expressed at similar levels in control and Jak2 inhibited tumors ( Figure 4b ) . Sections from control shRNA-infected tumors revealed large areas of infected cells ( Figure 4c left panel , red ) that expressed Atoh1 and had pockets of tumor-initiating cells ( CD15-positive , yellow ) that co-localized with phosphorylated Y78 ( blue ) . In contrast , tumors infected with the most potent Jak2 shRNA revealed scattered initiating cell clusters , none of which contained Y78 phosphorylated Atoh1 ( Figure 4c , right panel ) . Taken together , here we demonstrate that Atoh1 , a protein not expressed in the post-natal normal brain , is expressed specifically in SHH-type medulloblastoma malignancies ( Figure 4d ) , and that tyrosine 78 phosphorylation by Jak2 increases Atoh1 activity and stability , thus augmenting the proliferation of tumor-initiating cells in MB . Our results suggest that targeting Atoh1 by inhibiting its phosphorylation on tyrosine 78 presents a viable avenue for targeting MB cells without impinging on healthy brain tissue . Moreover , tyrosine 78 of Atoh1 can be detected in protein lysates of human samples of SHH-type medulloblastoma ( Figure 4e ) , further supporting the clinical relevance of these findings . All described assays as well as standard molecular biology techniques ( RNA extraction , qRT-PCR , general cell culture , cell lysis , co-immunoprecipitation ( co-IP ) , nuclear-cytoplasmic fractionation and immunoblotting ) can be requested in step-by-step , ‘protocol-style’ format . Plasmids used in this study are available upon request ( see Key Resource table ) . The following assays were described elsewhere: virus production and stable cell line engineering ( Rousseaux et al . , 2016 ) , Atoh1-specific dual luciferase reporter assay ( Klisch et al . , 2011 ) , immunofluorescence staining ( Flora et al . , 2009 ) , BiFC ( Pusch et al . , 2011 ) , ADP glow in vitro kinase assay ( Ohana et al . , 2011 ) . Generation of mouse strains and genotyping procedures were previously described: Atoh1 lacZ ( Ben-Arie et al . , 2000 ) ( B6 . 129S7-Atoh1tm2Hzo/J , RRID:IMSR_JAX:005970 ) , Atoh1-GFP ( Rose et al . , 2009 ) ( B6 . 129S-Atoh1tm4 . 1Hzo/J , RRID:IMSR_JAX:013593 ) , Nd2::SmoA1 ( Hatton et al . , 2008 ) ( C57BL/6-Tg ( Neurod2-SmoA1 ) 199Jols/J , RRID:IMSR_JAX:008831 ) and NOD/SCID ( Shultz et al . , 1995 ) ( NOD . CB17-Prkdcscid/J , RRID:IMSR_JAX:001303 ) recipient . All mice in this study were Nd2::SmoA1 C57/Bl6 on either Atoh1-GFP homozygous ( Nd2::SmoA1Tg Atoh1Atoh1-GFP/Atoh1-GFP , 100% Atoh1 protein ) or Atoh1 heterozygous ( Nd2::SmoA1Tg Atoh1Atoh1-GFP/LacZ , 50% Atoh1 protein ) background . All procedures were approved in advance under the guidelines of the Center for Comparative Medicine , Baylor College of Medicine and were performed in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals . Both male and female mice were used . Mice were group-housed in temperature-controlled rooms on a 14–10 hr light–dark cycle , at constant temperature ( 22–24°C ) and received food and water ad libitum . Mice of both sexes were randomly assigned to different treatment groups . All patients provided written informed consent and tissues were collected under an IRB approved protocol at Baylor College of Medicine ( BCM ) . Western blot ( overnight incubation with a 1:5000 dilution ) : anti-p-Atoh1-Y78 ( generated by GeneScript , Piscataway , NJ upon our request ) , anti-GFP ( GeneTex , Irvine , CA Cat# GTX26556 RRID:AB_371421 ) , anti-Jak2 ( Cell Signaling Technology , Danvers , MA Cat# 3230 , RRID:AB_2128522 ) , anti-p-Jak2 ( Tyr1007/1008 , Cell Signaling Technology Cat# 3771S , RRID:AB_330403 ) , anti-Vinculin ( 1:20 , 000 , Sigma-Aldrich , St . Louis , MO Cat# V9131 RRID:AB_477629 ) , anti-beta Actin-HRP ( 1:40 , 000 , Abcam , Cambridge , MA Cat# ab20272 RRID:AB_445482 ) , anti-Rabbit-HRP ( 1:20 , 000 , Bio-Rad/AbD Serotec , Hercules , CA Cat# 170–5046 RRID:AB_11125757 ) , anti-FLAG M2-HRP ( 1:10 , 000 , Sigma-Aldrich Cat# A8592 RRID:AB_439702 ) , anti-Mouse-HRP ( 1:50:000 , Jackson ImmunoResearch Labs , West Grove , PA Cat# 715-035-150 RRID:AB_2340770 ) ; Immunofluorescence: anti-p-Atoh1-Y78 ( 1:50 ) , anti-GFP ( 1:200 , Abcam Cat# ab13970 RRID:AB_300798 ) , anti-RFP ( 1:100 , RF5R , Abcam Cat# ab125244 RRID:AB_10973556 ) , anti-CD15-PE ( 1:50 , BD Biosciences , San Jose , CA Cat# 347420 RRID:AB_400298 ) , anti-chicken IgY-Alexa647 ( 1:100 , Thermo Fisher Scientific , Waltham , MA Cat# A-21449 RRID:AB_2535866 ) , anti-Mouse IgG2b-DyLight405 ( 1:100 , Jackson ImmunoResearch Labs Cat# 115-475-207 RRID:AB_2338801 ) , anti-Rabbit IgG-Alexa647 ( 1:100 , Jackson ImmunoResearch Labs Cat# 711-606-152 RRID:AB_2340625 ) ; FACS labeling: endogenous GFP signal from Atoh1-GFP KI mice , anti-CD15-Dylight649 ( 1:50 , Novus , Novus Biologicals , Littleton , CO Cat# NB100-2672C RRID:AB_1724222 ) ; Co-IP: anti-GFP ( 2 µg per IP , [N86/8] , UC Davis/NIH NeuroMab Facility Cat# 75–131 RRID:AB_10671445 ) , anti-Jak2 ( 5 µg per IP , Cell Signaling Technology Cat# 3230 RRID:AB_2128522 ) . The following cell lines were purchased from ATCC ( Manassas , VA ) : DAOY ( ATCC Cat# HTB-186 , RRID:CVCL_1167 ) , 293T ( ATCC Cat# CRL-3216 , RRID:CVCL_0063 ) . UW228 ( RRID:CVCL_4460 ) ( Huang et al . , 2005 ) and ONS-76 ( Japanese Collection of Research Bioresources Cell Bank , Japan Cat# IFO50355 , RRID:CVCL_1624 ) ( Sun et al . , 2013 ) were kindly provided by Dr . Charles G . Eberhart ( John Hopkins University , Baltimora , MD ) . All cell lines were verified using visual inspection and comparison to previously published studies . Cells were found to be negative for mycoplasma contamination . Lines were cultured as adherent cells in DMEM containing 10% FBS and antibiotics using standard cell culture practices ( Geraghty et al . , 2014 ) . For proliferation analysis cells were cultured as medullospheres ( Zanini et al . , 2013 ) for 3 days prior to Atoh1 induction by addition of doxycycline [CAS: 24390-14-5 , 0 . 2 µM] for 24 hr . Short-term culture of Shh-type MB cells was identical to cerebellar granule cell culture as described previously ( Gao et al . , 1991 ) . Inhibition of Jak2 canonical downstream cascades was as described using SH-4–54 [CAS: 1456632-40-8 , 0 . 5 µM , STAT ( Haftchenary et al . , 2013 ) ] , LY294002 [CAS: 154447-36-6 , 50 μM , PI3K ( Maira et al . , 2009 ) ] , PD184352 [CAS: 212631-79-3 , 1 µM , MAPK ( Sebolt-Leopold et al . , 1999 ) ] , or AG490 [CAS: 133550-30-8 , 5 μM , Jak2 ( Kobayashi et al . , 2015 ) ] . All chemicals were purchased from Selleckchem , Houston , TX . A total of 108 cells from Atoh1-positive/CD15-positive and Atoh1-positive/CD15-negative cell populations of Nd2::SmoA1Tg , Atoh1Atoh1-GFP/Atoh1-GFP tumors were used for immunoprecipitation . Cells were collected and lysed in lysis buffer ( 50 mM Tris , pH 8 . 1 , 137 . 5 mM NaCl , 2 mM EDTA , 0 . 5% Triton-X-100 , supplemented with fresh protease inhibitors and phosphatase inhibitors ) , homogenized using 1 ml insulin syringes . GFP-trap beads ( 20 µl per IP , ChromoTek , Germany ) were added to cleared supernatant and immunoprecipitation reactions were performed overnight at 4°C with gentle head-to-tail rotation . After three washes , immune-complexes were eluted with glycine , vacuum dried and dissolved in 50 mM ammonium bicarbonate , pH = 10 for MS . MS was performed by Proteomics and Metabolomics Core Labs at BCM using standard protocols ( Jung et al . , 2017 ) . We considered a true positive interactor as being able to be pulled down at least in two independent experiments with at least five unique peptides . Cell proliferation assay was performed on medullospheres of engineered cell lines using CellTiter 96 Non-Radioactive Cell Proliferation Assay ( Promega , Madison , WI ) as per manufacturer’s protocols . Xenografts were generated in immunocompromised mice ( NOD/SCID ) previously described ( Morton and Houghton , 2007 ) . Cells used were ( i ) Nd2::SmoA1Tg , Atoh1Atoh1-GFP/Atoh1-GFP tumors FACS sorted for Atoh1-positive/CD15-positive , Atoh1-positive/CD15 negative and Atoh1-negative cells ( Figure 1—figure supplement 1b ) ; ( ii ) the human medulloblastoma cell lines DAOY ( Atoh1-GFPTG , Atoh1-GFP Y78FTG , Atoh1-GFP Y78DTG , Figure 2c ) ; or ( iii ) Nd2::SmoA1Tg , Atoh1Atoh1-GFP/Atoh1-GFP tumors infected with shCON , shJak2#58 , shJak2#61 viral particles ( Figure 4 ) . Briefly , 105 cells were mixed in 100 μl cold matrigel ( BD Bioscience , San Jose , CA ) and injected subcutaneously into the flank region of nude mice ( n = 6 per group ) . Xenografts were left to grow for up to 1 months then harvested and weighed . When the inducible cell lines were used mice were fed doxycycline chow ( 200 mg/kg , Bio-Serv , ) ad libitum throughout the experiment . Statistics were performed using Prism 7 . 0 ( GraphPad Software , La Jolla , CA ) . Half-life was calculated based on normalized densitometric values obtained by ImageJ and solved for x when y = 0 . 5 for the following formula: y = a ( e−bx ) as previously described ( Rousseaux et al . , 2016 ) . Animal feedings , treatments and immunofluorescence analyses were performed in a single-blinded fashion . No blinding was used for the remaining analyses . The group size was determined based on previous studies in our laboratories . No animal was excluded . No data points were excluded from the statistical analyses , and variance was similar between the groups being statistically compared . For complete statistical analyses , please refer to Figure 1—source data 1 . All data presented are of mean ± standard error of the mean ( s . e . m . ) , , * and **** denote p<0 . 05 , p<0 . 01 , p<0 . 001 , and p<0 . 0001 , respectively . ns denotes p>0 . 05 . The data sets supporting the conclusions of this article are included within the article and its additional files . Plasmids are available upon reasonable request .	Medulloblastoma is the most common solid brain tumor that develops in children , with more than five hundred new cases diagnosed in the United States every year . There are four broad types of medulloblastoma . One of these is called the “Sonic Hedgehog” subtype , named after the biological pathway that becomes re-activated in these tumors . Only about half of patients with this subtype survive for more than 10 years . Moreover , medulloblastoma treatment combines surgery , chemotherapy and radiation , which can cause severe side effects including psychiatric disorders and cognitive impairment . Several drugs that treat medulloblastoma by targeting the Sonic Hedgehog pathway are currently being tested in clinical trials . However , these drugs are usually only effective for a limited time before the tumor evades the treatment . Therefore , there is a need to develop new treatment options for medulloblastoma , perhaps by targeting different signaling pathways in the cells . A protein called Atoh1 is needed for proper brain development in humans , but is not normally present after the first year of life . This protein is , however , re-expressed at high levels in medulloblastoma in mice and humans and is essential for Sonic Hedgehog-type medulloblastoma to form in mice . Klisch et al . used genetic techniques to reduce the amount of Atoh1 in mice that develop medulloblastoma . This intervention reduced the number of mice that developled tumors and increased their lifespan . Biochemical experiments showed that the tumor stem cells of the mice contain a modified version of Atoh1 where a phosphate molecule is bound to a particular region of the protein . This phosphorylation increased the amount and activity of Atoh1 in the cell , and so caused tumors to grow more quickly in mice . Phosphorylated Atoh1 was also detected in samples taken from human medulloblastoma tumors . Klisch et al . also found that an enzyme called Jak2 phosphorylates Atoh1 . Inhibiting Jak2 reduced the levels of Atoh1 in medulloblastoma cells and slowed tumor growth in mice . Future work could investigate different ways of preventing Atoh1 phosphorylation , with the hope of finding new treatments for Sonic-Hedgehog-type medulloblastomas .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "short", "report", "cancer", "biology" ]	2017	Jak2-mediated phosphorylation of Atoh1 is critical for medulloblastoma growth
Neurons regulate ionic fluxes across their plasma membrane to maintain their excitable properties under varying environmental conditions . However , the mechanisms that regulate ion channels abundance remain poorly understood . Here we show that pickpocket 29 ( ppk29 ) , a gene that encodes a Drosophila degenerin/epithelial sodium channel ( DEG/ENaC ) , regulates neuronal excitability via a protein-independent mechanism . We demonstrate that the mRNA 3′UTR of ppk29 affects neuronal firing rates and associated heat-induced seizures by acting as a natural antisense transcript ( NAT ) that regulates the neuronal mRNA levels of seizure ( sei ) , the Drosophila homolog of the human Ether-à-go-go Related Gene ( hERG ) potassium channel . We find that the regulatory impact of ppk29 mRNA on sei is independent of the sodium channel it encodes . Thus , our studies reveal a novel mRNA dependent mechanism for the regulation of neuronal excitability that is independent of protein-coding capacity . The neuronal action potential is sensitive to abrupt changes in environmental temperatures ( Peng et al . , 2007; Buzatu , 2009 ) . Thus , the failure of neurons to adjust their physiological properties in response to a fast rise in temperature can lead to neurological disorders such as febrile seizures ( Bassan et al . , 2013 ) . Previous theoretical and experimental studies suggested that one of the main mechanisms for maintaining normal neuronal excitability , circuit integrity , and behavioral robustness under varying environmental temperatures depends on changes in the abundance and membrane half-life of various voltage-dependent ion channels ( Marder and Prinz , 2003; O’Leary et al . , 2013; Rinberg et al . , 2013; Rosati and McKinnon , 2004; Tang et al . , 2010 , 2012 ) . However , the actual molecular mechanisms that mediate these processes are largely unknown . Several whole genome transcriptomics studies revealed that natural antisense non-coding transcripts ( NATs ) are prevalent in eukaryotes ( Lapidot and Pilpel , 2006; Okamura et al . , 2008 ) . Although the function of the majority of NATs is still unknown , evidence suggests that at least some cis-NATs are likely to act as regulatory RNAs of protein-coding transcripts ( Borsani et al . , 2005; Okamura et al . , 2008; Watanabe et al . , 2008 ) , including a recent report about a non-coding NAT that regulates a neuronal potassium channel ( Zhao et al . , 2013 ) . Furthermore , some NATs have been shown to play a role in the physiological response to various stresses in plants ( Borsani et al . , 2005; Katiyar-Agarwal et al . , 2006 ) . Whether NATs play a role in the post-transcriptional regulation of ion channels and neuronal excitability was unknown . The response of neurons to acute heat stress is likely to require rapid changes in ion channel functions . We hypothesized that NATs play a role in the posttranscriptional regulation of ion channel function in response to stress . Therefore , we screened the well-annotated genome of the fruit fly Drosophila melanogaster to identify known excitability-related ion channels that might be regulated by endogenous NATs . Using this approach , we found that the gene seizure ( sei ) , which encodes the sole fly homolog of the human Ether-à-go-go Related Gene ( hERG ) inward rectifying K+ channel ( Titus et al . , 1997; Wang et al . , 1997 ) , is located downstream of the degenerin/epithelial sodium channel ( DEG/ENaC ) ppk29 ( Liu et al . , 2012; Thistle et al . , 2012; Zelle et al . , 2013 ) . The two genes are convergently transcribed on opposite DNA strands , and have complementary 3′UTRs that overlap by 88 nucleotides , which we confirmed by fully sequenced cDNAs deposited in NCBI and 3’RACE analysis ( Figure 1A ) . The two opposing physiological functions of sei and ppk29 ( K+ and Na+ channels respectively ) , and the realization that their transcripts could form natural sense/antisense RNA duplexes ( Katayama et al . , 2005; Czech et al . , 2008 ) led us to hypothesize that the mRNAs of sei and ppk29 may regulate each other via the formation of natural endogenous dsRNAs . Since mRNA-dependent interaction between sei and ppk29 requires that the two genes will be co-transcribed we first analyzed expression data from the modENcode ( Cherbas et al . , 2011 ) and the FlyExpress ( Robinson et al . , 2013 ) projects . Although previous studies suggested that ppk29 function might be a sensory-specific ( Liu et al . , 2012; Thistle et al . , 2012 ) , our analysis revealed that sei and ppk29 are co-expressed in neuronal cell lines ( Figure 1—figure supplement 1A ) and are both enriched in the fly central nervous system ( Figure 1—figure supplement 1B ) . In situ hybridization in the fly brain also demonstrated neuronal co-expression ( Figure 1B–D ) . Furthermore , we used cell-specific mRNA enrichment ( Thomas et al . , 2012 ) to demonstrate that both genes are co-expressed in motor neurons in vivo ( Figure 1E ) . Together , these data support spatial co-expression of sei and ppk29 . 10 . 7554/eLife . 01849 . 003Figure 1 . sei and ppk29 are co-expressed in the nervous system . ( A ) The chromosomal architecture of sei and ppk29 ( 2R:19 , 934 , 934- 19 , 944 , 660 ) . Coding exons are in black . 3′ and 5′ untranslated regions ( UTRs ) are in gray . AY058350 , fully sequenced sei cDNA; BT029266 , fully sequenced ppk29 cDNA . Black triangles represent transposons insertion sites . Arrows represent direction of transcription . Yellow boxes , sei 3′RACE product . Green boxes , ppk29 3′RACE product . ( B ) In situ hybridization shows sei and ppk29 are co-expressed in neuronal tissues . Antisense riboprobes . Scale bar , 100 μm . ( C ) Higher magnification of white box in B . White arrowheads , optic lobe neurons . Red , ppk29 signal; Green , sei signal; Blue , DAPI nuclear stain . Scale bar , 10 μm . ( D ) Sense riboprobe controls . Scale bar , 100 μm . ( E ) Translating Ribosome Affinity Purification ( TRAP ) of mRNAs from larval motor neurons shows that sei and ppk29 are co-enriched in these cells relative to total body RNA . mRNA levels for each gene were measured with Real-Time qRT-PCR . N = 4 per gene . p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 01849 . 00310 . 7554/eLife . 01849 . 004Figure 1—figure supplement 1 . ppk29 and sei are co-expressed in Drosophila neuronal tissues . ( A ) ppk29 and sei are co-expressed in neuronal cell lines . Data are from the modEncode database . Expression levels represent average strand-specific unique RNA-seq reads . BG1 and BG2 are neuronal cell lines . Schneider 2 ( S2 ) is an undefined embryonic cell line . ( B ) Expression of ppk29 and sei in different tissues . Orange bars highlight neuronal tissues . Average data are presented as mean ± SEM ( n = 4 arrays per tissue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01849 . 004 Previous studies suggested that transcriptional changes in ion channel transcript abundance could play a role in the adaptation of neurons to changes in environmental temperatures ( Marder , 2011 ) . Thus , as a first test of our hypothesis that these two ion channels might interact antagonistically to regulate the neuronal response to heat we measured the relative expression levels of both genes in wild type animals that were adapted to variable environmental temperatures . In agreement with our hypothesis , we found that when animals adapted to high temperature ( 37°C ) the transcripts levels of sei went up and ppk29 went down relative to their levels at 25°C . In contrast , adaptation to colder temperature ( 13°C ) led to an opposite effect on the expression of both genes ( Figure 2 ) . We conclude that both sei and ppk29 are likely to play opposite roles in the regulation of neuronal activity in response to changes in ambient temperature , and that the possible interaction between these two genes is physiologically relevant . 10 . 7554/eLife . 01849 . 005Figure 2 . sei and ppk29 transcripts are inversely regulated in response to changes in ambient temperature . ( A ) Temperature adaptation protocol . Total time from 25–37°C or 25–13°C is 7 hr . ( B ) Real-time qRT-PCR data . Different letters above bars represent statistically significant post hoc analyses ( Tukey’s , p<0 . 05 , N = 4 per group ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01849 . 005 The data presented in Figure 2 , and previous reports that indicated that mutations in sei are highly sensitive to heat stress ( Titus et al . , 1997; Wang et al . , 1997 ) , led us to hypothesize that mutations in ppk29 might lead to a protection from heat stress . Based on our model presented in Figure 6 , such a protective effects for ppk29 mutations may arise from the loss of sodium currents or alternatively due to the upregulation of SEI-dependent potassium currents . As was previously reported , we found that multiple independent mutations in sei lead to rapid seizures and paralysis in response to acute heat stress ( Figure 3A ) . In contrast , flies carrying independent insertional alleles of ppk29 demonstrated protection from the effects of heat stress relative to wild type and sei mutant animals ( Figure 3A , Figure 3—figure supplement 1A , B; ppk29P1 and ppk29P2 are described in Figure 1A ) . These data confirmed our hypothesis that sei and ppk29 play opposing roles in the neuronal response to heat stress , and are likely playing an important adaptive role in environmentally induced neuronal plasticity . We also observed contrasting behavioral responses to heat stress in animals that carry single copy insertional alleles of sei or ppk29 in trans with a chromosomal deficiency that covers both loci ( Figure 3—figure supplement 1C , D ) . These data indicate that the effects of either mutation on behavior are specific and not due to other background mutations . 10 . 7554/eLife . 01849 . 006Figure 3 . RNAi-dependent knockdowns of ppk29 and sei expression lead to opposing effects on heat-induced paralysis . ( A ) The behavioral response to heat stress in sei and ppk29 mutants . Left panel , cumulative paralyzed flies over time . Right panel , same data as in left panel presented as time to total paralysis ( n = 16 , p<0 . 001 , one-way ANOVA ) . Different letters above bars represent significantly different groups ( Tukey post hoc analysis , p<0 . 05 ) . ( B ) Representative extracellular recordings from motor neurons from each genotype at 25°C and 38°C . ( C ) Summary neurophysiological data ( n = 8-10 per genotype , p<0 . 01 , *p<0 . 001 , one-way ANOVA with a Tukey post-hoc test ) . ( D ) Neuronal downregulation of sei or ppk29 with gene-specific RNAi constructs . Data presented as in A ( n = 16 , p<0 . 001 , one-way ANOVA ) . ( E ) sei and ppk29 mRNA levels in sei and ppk29 mutant lines . Analyses were by relative real-time quantitative RT-PCR analyses . Left panel , sei mRNA . Right panel , ppk29 mRNA ( n = 4 per genotype , p<0 . 05 , one-way ANOVA ) . ( F ) sei and ppk29 mRNA levels in sei and ppk29 RNAi-knockdown lines . Analyses as in E ( n = 4 per genotype , p<0 . 05 , one-way ANOVA ) . Data are presented as mean ± SEM . Different letters above bars represent significantly different groups ( Tukey post hoc analysis , p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01849 . 00610 . 7554/eLife . 01849 . 007Figure 3—figure supplement 1 . ppk29 mutations confer protection from heat-induced paralysis . ( A ) Two independent ppk29 transposon-insertional alleles do not complement each other . Data presented as cumulative paralyzed flies over time . ( B ) Same data as in A presented as total time to paralysis . Different letters above bars represent significantly different genotypes ( one-way ANOVA Tukey's post hoc test; n = 16 , p<0 . 001 ) . ( C and D ) sei or ppk29 transposon-insertional alleles in trans across a deficiency chromosome ( DfBSC136 ) that covers both loci . Control DfBSC652 has the same genetic background as DfBSC136 but does not cover the sei/ppk29 loci . Analyses and data presentations are as in panels A and B . ( E and F ) Gene knockdowns of sei or ppk29 by mutations or neuronal RNAi in larvae lead to heat sensitivity or protection phenotypes respectively that are analogous to the adult phenotypes ( G ) Acute RNAi-dependent targeting of ppk29 or sei in the adult nervous system with the GeneSwitch ( GS ) version of the pan-neuronal promoter elav was sufficient to phenocopy the mutant phenotypes . Mutant phenotypes were apparent only in the RU486 feeding group ( RU486+ ) ( n = 16 , p<0 . 001; two-way ANOVA with a Tukey's post-hoc test ) . The interaction term between genotype and drug was also significant ( p=<0 . 001 ) . Data are presented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 01849 . 00710 . 7554/eLife . 01849 . 008Figure 3—figure supplement 2 . Mutations in sei and ppk29 do not affect gross locomotion at room temperature . ( NS; n = 10 per genotype , one-way ANOVA with Tukey's post-hoc test ) . Data are presented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 01849 . 008 Previous studies indicated that the temperature-sensitive phenotype of sei mutants is associated with heat-induced neuronal hyperexcitability ( Kasbekar et al . , 1987 ) . Therefore , we hypothesized that mutations in ppk29 will lead to a hypoexcitable neuronal phenotype under heat stress . We found that the spontaneous neuronal activity of larval motor neurons is not different between ppk29 , sei and wild type animals at 25°C . In contrast , at 38°C wild type neurons show a small but significant increase in neuronal activity , while sei mutant neurons become hyperexcitable . In contrast to sei null and wild type animals , ppk29 mutants are unable to increase neuronal firing rates in response to heat stress , which is consistent with a hypoexcitability phenotype ( Figure 3B , C ) . We also confirmed that the larval excitability phenotypes of sei and ppk29 mutants are correlated with behavior . As in our neurophysiological studies , we found that at 25°C all genotypes show normal larval locomotion ( Videos 1–3 ) . However , exposure to 38°C lead to an abnormal seizure-like locomotion in wild type animals ( twitching and rolling ) . This phenotype is significantly higher in sei mutant and RNAi knockdown larvae but completely absent in ppk29 mutant larvae ( Figure 3—figure supplement 1E , F; Videos 4–6 ) . We conclude that sei and ppk29 affect the behavioral sensitivity to heat stress via the contrasting regulation of neuronal excitability in both larval and adult stages . 10 . 7554/eLife . 01849 . 009Video 1 . Wild type larva at 25°C . DOI: http://dx . doi . org/10 . 7554/eLife . 01849 . 00910 . 7554/eLife . 01849 . 010Video 2 . seiP larva at 25°C . DOI: http://dx . doi . org/10 . 7554/eLife . 01849 . 01010 . 7554/eLife . 01849 . 011Video 3 . ppk29P1 larva at 25°C . DOI: http://dx . doi . org/10 . 7554/eLife . 01849 . 01110 . 7554/eLife . 01849 . 012Video 4 . Wild type larva at 38°C . DOI: http://dx . doi . org/10 . 7554/eLife . 01849 . 01210 . 7554/eLife . 01849 . 013Video 5 . seiP larva at 38°C . DOI: http://dx . doi . org/10 . 7554/eLife . 01849 . 01310 . 7554/eLife . 01849 . 014Video 6 . ppk29P1 larva at 38°C . DOI: http://dx . doi . org/10 . 7554/eLife . 01849 . 014 Similarly to the mutant adult phenotypes , neuronal RNAi-dependent knockdown of sei and ppk29 mRNAs with the neuronal elav-GAL4 driver lead to contrasting phenotypes that are identical to the phenotypes observed in mutants ( Figure 3D ) . These data demonstrate that the observed phenotypes are neuronal-specific and suggest that quantitative changes in neuronal mRNA levels of either sei or ppk29 are sufficient to induce high-sensitivity or protective phenotypes respectively . Analyses of mRNA levels in mutants and RNAi-expressing animals support the hypothesis that downregulation of ppk29 mRNA is associated with increased sei mRNA levels , but the converse effect is not evident ( Figure 3E , F ) . Together , these data demonstrate that the regulatory interaction between the mRNAs of sei and ppk29 is not symmetric; changes in ppk29 mRNA level downregulate sei mRNA , but not the other way around . We also observed contrasting phenotypes when we expressed the same gene-specific RNAi constructs in adult neurons only by using the hormonally-induced GeneSwitch version of the elav-GAL4 ( Osterwalder et al . , 2001; Figure 3—figure supplement 1G ) . These data show that the contrasting effects of sei and ppk29 mRNA dowregulation on the neuronal response to heat stress are physiological rather than developmental . We did not observe any general locomotion defects in sei or ppk29 mutants at 25°C ( Figure 3—figure supplement 2 ) . Together , data presented in Figures 2 and 3 suggest that the protective effect of mutations in ppk29 are symptomatic in the sense that they lead to a pre-stress increase in sei transcript levels , which leads to a higher ability of the nervous system to deal with the acute heat stress even without prior adaptation to slow temperature increase . Although our data suggest that the contrasting heat-induced phenotypes of sei and ppk29 mutants are possibly mediated via mRNA-dependent interactions , they do not exclude the possibility that the two channels also interact at the protein level . Therefore , we investigated whether the protection from heat stress in ppk29 mutants is mediated by the loss of PPK29 channel activity or the up-regulation of sei mRNAs ( As shown in Figure 3E , F ) . To test this we first blocked SEI channel activity in wild type and ppk29 mutant animals by using two different hERG channel blockers ( Afrasiabi et al . , 2010 ) . These studies reveal that blocking SEI activity in wild type animals phenocopies the heat sensitivity phenotype of the sei mutation , which indicate that the drugs are successfully blocking SEI channels in the fly . Similar to wild type animals , blocking SEI activity in ppk29 mutants reduce their resistance to heat stress to a level comparable to wild type animals in a dose-dependent manner ( Figure 4A , Figure 4—figure supplement 1 ) . These data are in agreement with the expression data ( Figure 3E , F ) , and strongly indicate that the ppk29-mediated protection from heat stress is due , at least in part , to increased SEI K+ channel activity rather then the loss of ppk29-dependent Na+ currents . 10 . 7554/eLife . 01849 . 015Figure 4 . The Protective Effect of ppk29 Mutations is Mediated by SEI Channel Activity . ( A ) Blocking SEI channel activity in ppk29 mutants with the hERG channel blocker Cisapride eliminate the protective effect in a dose dependent manner ( n = 8 per genotype , p<0 . 01 , two-way ANOVA; genotype , dose , and genotype by dose showed significant effects , p=<0 . 001 ) . ( B ) Schematic representation of transgenic constructs . ( C ) Neuronal expression of ppk29-3′UTR is sufficient to rescue the majority of the protective effect of the ppk29 mutation ( n = 12 , p<0 . 01 , one-way ANOVA ) . Data are presented as mean ± SEM . Different letters above bars represent significantly different groups ( Tukey post hoc analysis , p<0 . 05 ) . ( D ) Neuronal expression of sei cDNA with or without its endogenous 3′UTR , but not the 3′UTR alone , is sufficient to rescue the sei mutation ( n = 12 , p<0 . 001 , one-way ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01849 . 01510 . 7554/eLife . 01849 . 016Figure 4—figure supplement 1 . The protective effect of ppk29 mutations depends on SEI K+ channel activity . Treating ppk29 mutants flies with hERG inhibitors cisapride ( A ) and E−4301 ( B ) lead to a significantly faster heat-induced seizures and paralyses in all tested genotypes ( n = 8 for each genotype , two-way ANOVA with a Tukey's post-hoc test; the interaction between genotype and concentration is significant for both drugs , p=<0 . 001 ) . Average data are presented as mean ± SEM . Different letters above bars represent significantly different groups . DOI: http://dx . doi . org/10 . 7554/eLife . 01849 . 016 Our hypothesis predicts that the 3’UTR of ppk29 can regulate sei function by acting as a natural antisense RNA . To test directly this hypothesis we generated transgenic fly lines that can express the cDNAs of either sei or ppk29 with or without their endogenous 3′UTR , or their 3′UTRs alone ( Figure 4B ) by using the UAS-GAL4 system . Remarkably , we found that the expression of the ppk29 endogenous 3′UTR alone or the cDNA with the 3′UTR is sufficient to rescue the ppk29 mutation . In contrast , expression of ppk29 cDNA alone is not sufficient to completely rescue the phenotype of the ppk29 mutation ( Figure 4C ) . In agreement with the pharmacological studies , these data demonstrate that the main protective effect of ppk29 mutations is mediated via 3′UTR-dependent regulation of SEI , independent of PPK29 channel functions . Nevertheless , we also found that a complete rescue of the ppk29 mutation phenotype require the expression of the ppk29 cDNA with its endogenous 3′UTR . Therefore , PPK29 channel activity may also contribute neuronal excitability independent of sei regulation . In addition , since the observed effects of ppk29 transgenes on sei function are in trans , these data show that the two genes can interact at the transcript level independent of their chromosomal proximity . Unlike for ppk29 , the neuronal expression of sei cDNA with or without its endogenous 3′UTR , but not the 3′UTR alone , is sufficient to rescue the sei mutation ( Figure 4D ) . These data further show that sei is the focal physiological element in the neuronal response to heat stress , and that the mRNA 3′UTR-dependent interaction between sei and ppk29 is not symmetric . We next investigated the role of ppk29 3′UTR in regulating sei mRNA expression and heat-induced seizures and paralysis . Consistent with our model , neuronal overexpression of sei cDNA ( with or without its endogenous 3′UTR but not the 3′UTR alone ) is sufficient to protect animals from heat stress as in ppk29 mutants ( Figure 5A ) . We also found that neuronal overexpression of a ppk29 cDNA with its endogenous 3′UTR or ppk29 3′UTR alone , but not the cDNA alone , is sufficient to induce heat sensitivity as in sei mutants ( Figure 5B ) . In agreement with the behavioral data , overexpression of the ppk29-3′UTR is sufficient to reduce endogenous sei mRNA levels but overexpression of sei-3′UTR alone does not have a similar effect on ppk29 ( Figure 5C ) . These data demonstrate that elevated levels of ppk29-3′UTR alone in trans are sufficient to affect neuronal physiology by downregulating sei mRNA levels . Expression of the ppk29 related constructs specifically in the adult nervous system by using the GeneSwitch elav-GAL4 driver demonstrate that the observed effects of ppk29-3′UTR overexpression on sei function and behavior are physiological and not developmental ( Figure 5D ) . Thus , our data prove that downregulation of sei expression leads to neuronal heat sensitivity while increased sei expression leads to a protection , and that the relative abundance of sei transcripts in neurons is affected by the expression levels of ppk29 . 10 . 7554/eLife . 01849 . 017Figure 5 . ppk29-dependent regulation of sei depends on the canonical RISC pathway . ( A ) Neuronal overexpression of sei cDNA with or without its endogenous 3′UTR in wild type animals leads to a protection from heat-induced paralysis ( n = 12 , p<0 . 001 , one-way ANOVA ) . ( B ) Neuronal overexpression of the ppk29 cDNA with its endogenous 3′UTR or the 3′UTR alone , but not the ppk29 cDNA lone , is sufficient to induce sei mutant-like heat sensitivity phenotype ( n = 12 , p<0 . 001 , one-way ANOVA ) . ( C ) Real-time qRT-PCR analyses of sei and ppk29 mRNA level . Overexpression of ppk29 cDNA with its 3′UTR or the 3′UTR alone , but not the cDNA alone , is sufficient to downregulate endogenous sei mRNA levels ( left panel ) but not conversely ( right panel ) ( n = 4 , p<0 . 05 , one-way ANOVA ) . ( D ) Adult-specific neuronal overexpression of ppk29-3′UTR with the hormone inducible GeneSwitch elav-GAL4 is sufficient to induce sei mutant-like phenotype ( n = 12 , **p<0 . 001; two-way ANOVA , genotype , RU486 , and their interaction are significant , p=<0 . 001 ) . ( E and F ) The effect of ppk29 3′UTR overexpression on heat sensitivity and sei mRNA downregulation is abolished in the Dcr-2 mutant background ( n = 12 , one-way ANOVA ) . ( F ) Real-time qRT-PCR ( n = 4 , NS , one-way ANOVA ) . Data are presented as mean ± SEM . Different letters above bars represent significantly different groups ( Tukey post hoc analysis , p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01849 . 017 The above data show that ppk29 mRNA can serve as a regulatory antisense RNA in addition to its capacity to encode for a DEG/ENaC subunit . The processing of endo-siRNAs depends on Dicer2 ( Dcr-2 ) in flies ( Czech et al . , 2008 ) . Thus , we investigated whether the regulatory impact of ppk29-3′UTR on sei mRNA levels and behavior depends on the RNAi machinery . We find that in the background of the Dcr-2 mutation neuronal overexpression of the ppk29-3′UTR has no effect on the behavioral response to heat stress ( Figure 5E ) or on the expression levels of sei ( Figure 5F ) . These data demonstrate that the regulatory function of ppk29-3′UTR depends on the endogenous siRNA pathway . The finding that ppk29 can regulate sei mRNA levels via the canonical siRNA pathway may also explain why the mRNA 3′UTR-dependent interaction between sei and ppk29 is not symmetric . Recent studies of the molecular mechanism that underly the specificity of the RNAi machinery indicate that the protein complex that mediate the recognition of the target RNA by the short dsRNA are not symmetrical . Thus , via mechanisms that are not fully understood , RISC treats only one of the strands as a guide ( Tomari et al . , 2004; Rand et al . , 2005; Betancur and Tomari , 2012; Noland and Doudna , 2013 ) . It is likely that a similar mechanism is at play here . Our data indicate that when forming siRNA duplexes , ppk29 3′UTR is the preferred guide strand during RISC loading and the subsequent mRNA target identification . Here we describe a novel mechanism for the regulation of ion channel functions and neuronal excitability via a natural antisense mRNA ( Figure 6 ) . While this is a novel mechanism , it is by no means the only known RNA-dependent mechanism for the regulation of ion channel functions . For example , the double-stranded RNA helicase maleless ( mle ) regulates the Drosophila voltage-gated sodium channel paralytic ( para ) via A-to-I RNA editing . Mutations in mle lead to aberrant editing of para , splicing errors , and subsequent low channel activity ( Reenan et al . , 2000 ) . Other examples include the putative transcription factor down and out ( dao ) , which seem to affect sei transcription levels ( Fergestad et al . , 2010 ) , the potassium-independent effects of the sei-related mammalian EAG potassium channel on cellular signaling ( Hegle et al . , 2006 ) , and other diverse mechanisms for the co-regulation of various ion channels ( MacLean et al . , 2005; Ransdell et al . , 2013 ) . 10 . 7554/eLife . 01849 . 018Figure 6 . Cartoon depicting a model for the molecular interaction between sei and ppk29 . The chromosomal organization of these two genes suggest they could generate endogenous siRNA by convergent transcription . ( I ) The complementary 3′UTRs of sei and ppk29 mRNAs form a dsRNA . ( II ) Dicer-2 cleaves dsRNAs into siRNAs . ( III ) The loaded RISC complex targets sei transcripts for degradation via the canonical siRNA pathway . DOI: http://dx . doi . org/10 . 7554/eLife . 01849 . 018 It is unlikely that the type of interaction we have identified between the mRNAs of sei and ppk29 is unique . Bioinformatic analyses of genome sequences show that at least two of the three fly and three out of the eight human eag-like ( KCNH-type ) channels are organized in a chromosomal architecture that is similar to that of sei and ppk29 ( Table 1 ) . The functional diversity of the converging genes in each of the pairs we uncovered suggest that , like in the case of sei/ppk29 , the actual protein identity is secondary to the mRNA level interactions . However , to conclusively test this hypothesis will require additional experimental molecular and biochemical analyses of these loci in the fly and mammalian systems . 10 . 7554/eLife . 01849 . 019Table 1 . Fly and human eag-like channels that are possibly regulated via convergent transcription with an unrelated mRNADOI: http://dx . doi . org/10 . 7554/eLife . 01849 . 019Specieseag-like geneConverging geneDrosophilaseippk29eaghiwHumanKCNH1HHATKCNH3MCRS1KCNH7GCAPlease note that the converging genes are functionally diverse , which suggest that their protein identities might not play a role in their regulatory functions . Our findings also indicate that the regulatory interaction between sei and ppk29 may play a role in the homeostatic response to slow changes in environmental temperature ( Figure 2 ) . However , our current genetic and transgenic tools make it impossible for us to completely disentangle the direct effects of temperature changes on sei and ppk29 transcription and the indirect effects via the interactions of their mRNAs . Thus , more studies will be needed to further establish ppk29 mRNA as a homeostatic factor , beyond its effects on the acute heat response . In contrast to the linear simplicity of the ‘central dogma of molecular biology’ ( Crick , 1970 ) , we now know that the true molecular landscape of cells is complex and far from linear . In this regard , our studies provide an additional layer of regulatory complexity , and support the idea that mRNAs , which are typically thought to solely act as the template for protein translation , can also serve as regulatory RNAs , independent of their protein-coding capacity . Thus , the abundance of convergent transcription of protein-coding genes in eukaryotic genomes suggests that many other mRNAs might serve dual functions that are not necessary associated with the same cellular or physiological processes . Furthermore , although the phenomenon of mRNA-dependent interaction between the two genes we describe here occurs in cis ( Figure 6 ) , we currently have no reason to assume that similar interactions between RNAs cannot occur in trans as well . Consequently , it is likely that some of the evolutionary changes observed in mRNAs , including those that are considered ‘neutral’ , should be re-evaluated in light of the possible regulatory function that some mRNAs might exert independently of the proteins they encode . Flies ( Drosophila melanogaster ) were raised on standard cornmeal-agar food at 25°C and 60% relative humidity with a 12 hr light: dark cycle . Unless stated differently , the w1118 strain was used a ‘wild type’ . In our hands , the heat-induced behavior and physiology of these flies were not different from the Canton-S strain . The original stocks for ppk29P1 and seiP were obtained from the Bloomington Stock Center ( Stock No . 19016 , and 21935 ) . The ppk29P2 stock ( f04205 ) was from the Exelixis collection at Harvard Medical School . All insertional alleles used in our studies were backcrossed into the w1118 background for six generation . The seits1 EMS-allele was from the Ganetzki lab ( U of Wisconsin ) . The deficiency lines Df ( 2R ) BSC136 and Df ( 2R ) BSC652 ( 9424 and 25742 ) , elav-GAL4; UAS-Dicer2 ( 25750 ) , UAS-ppk29RNAi ( 27241 ) , elav-Gal4 ( 33805 ) , elav-GeneSwitch-GAL4 ( 43642 ) and Dicer-2 mutant ( 32064 ) were from the Bloomington Stock Center ( stock no . ) . UAS-seiRNAi was from VDRC ( v3606GD ) . The transgene seiΔ3′UTR was generated by amplifying sei coding sequence ( variant RA , NP_476713 ) with primers 5′-AAAAGCGGCCGCATGTCCCACAAATCTTGCGT-3′ and 5′-AAAATCTAGACTAATTATTATTATCGAACAAGTCAAGGTG-3′ from cDNA clone GH12235 . The transgene sei-3′UTR was generated by amplifying the same sei coding sequence plus its 3′UTR ( sei-RA , length = 95 ) with primers 5′-AAAAGCGGCCGCATGT-CCCACAAATCTTGCGT-3′ and 5′-AAAATCTAGATTTTCGGTTAGGACCTTTATTGC-3′ . The transgene ppk29Δ3′UTR was generated by amplifying ppk29 coding sequence ( variant PD , NP_001097442 ) with 5′-AAAAGCGGCCGCATGTGGCGGAAGTCAGTA-ATG-3′ and 5′-AAAATCTAGACTAACCGAAAATCATGGTCTTGA-3′ from cDNA clone IP06558 . The transgene ppk29-3′UTR was generated by amplifying the same ppk29 coding sequence plus its 3′UTR ( ppk29-RD , length = 112 ) with primers 5′-AAAAGCGGC-CGCATGTGGCGGAAGTCAGTAATG-3′ and 5′-AAATCTAGATTGACTTGTTCGATAAT-AATAATTAGGGC-3′ . The transgene GFP-3′UTRsei was generated by amplifying the EGFP ORF from the pEGFP-N3 vector with primers 5′-AAAAGCGGCCGCATGGTGA-GCAAGGGCGA-3′ and 5′-CTTGTGCACAAATAAATAAGATTCACTTGTACAGCTCGT-CCATG-3′ and the sei-3′UTR from cDNA clone GH12235 with primers 5′-CATGGACG-AGCTGTACAAGTGAGGCTCACTTATGCTCGCTCAATCCGAATTATCTTATTTATTTGTGCACAAGCTGTTGCGAGGCTAAAGAG-3′ and 5′-AAAATCTAGATTTTCGGTTAGGA-CCTTTATTGCTTTTCGCTCTTTAGCCTCGCAACAGCTTGTGCACAAATAAATAAGAT-3′ followed by PCR fusion of the two DNA fragments . The transgene mCherry-3′UTRppk29 was generated by amplifying the mCherry ORF from the pCAMBIA-1300 vector with primers 5′-AAAAGCGGCCGCATGGTGAGCAAGGGCGA-3′ and 5′-TAAAGAGCGAA-AAGCAATAAAGGTCTTACTTGTACAGCTCGTCCATGC-3′ and ppk29-3′UTR from cDNA clone IP06558 with primers 5′-GCATGGACGAGCTGTACAAGTAAGACCTTTATTGCTTTTCGCTCTTTAGCCTCGCAACAGCTTGTGCACAAATAAATAAGATAATTCGGATTG-3′ and 5′-AAAATCTAGATTGACTTGTTCGATAATAATAATTAGGGCTCACT-TATGCTCGCTCAATCCGAATTATCTTATTTATTTGT-3′ followed by PCR fusion of the two DNA fragments . All transgenes were verified by sequencing and subsequently subcloned into the pUASTattB plasmid . Each of the six individual transgenes was transformed by PhiC31 integrase-based transgenesis into two different landing chromosomal landing sites ( 2L:1476459 and 3L:11837236 ) ( Bateman et al . , 2006 ) . 20–30 flies ( 2–3 days post eclosion ) were anesthetized by CO2 and transferred to standard Drosophila vials containing fresh food for 24 hr . On test day , 10 flies ( 1:1 male/female ratio ) were transferred to an empty polystyrene vial ( Genesee Scientific , San Diego , CA ) without anesthesia . Flies were allowed to recover for 10 min before vials were immersed in a 41 ± 1°C water bath ( ISOTEMP105; Fisher Scientific , Pittsburgh , PA ) . The number of cumulative paralyzed flies was counted every 15 s until all flies were paralyzed at the bottom of the vial . The proportion of paralyzed flies and the time it takes to reach total paralysis for all 10 flies were used to generate heat-induced paralysis scores . We used the negative geotaxis response as an assay for general locomotion as we previously described ( Lu et al . , 2012 ) . In short , groups of ten flies were introduced into an empty vial without anesthesia . Additional empty vial was taped on top . To assay locomotion , bottom vial was tapped down lightly and the number of flies that climbed above a marked 15 cm line in 15 s was recorded . Feeding stage 3rd-instar larvae were used . Each larva was briefly washed in distilled water to remove all food debris and then transferred to a 3% agar plate that was equilibrated to 25 ± 1°C or 38 ± 1°C . Recording of behavior started 3 min post introduction by videotaping animals for 2 min . The total numbers of larval side-twitching events were used to quantify larval ‘seizure’ like behavior ( Videos 1–6 ) . Stock solutions of hERG blockers were kept as 10 mM Cisapride ( Sigma-Aldrich , St . Louis , MO , USA ) in DMSO and 100 mM E 4301 ( Alomone labs , Jerusalem , Israel ) in distilled water . Working solution were made by diluting the stock solutions in in 2% ( wt/vol ) sucrose solution . Flies were treated in groups of 20 adults ( 1–2 days post eclosion , 1:1 mixed sex ) in a vial containing a Kimwipe tissue paper soaked with 1 ml of the drug . Flies were allowed to feed on the drug for three days at 25°C and 60% humidity . Prior to the heat stress test , treated flies were transferred to a new vial containing standard fly food without drugs for two hours . Heat-induced paralysis was assayed as above . A 10 mM stock solution of RU486 ( mifepristone , Sigma-Aldrich , St . Louis , MO , USA ) was prepared in 80% ethanol . Then , the RU486 working solution was diluted to the final concentration ( 500 μM ) in 2% sucrose . The drug was delivered to flies as described above . Flies were treated with RU-486 or 2% sucrose for 7 days at 25°C and 60% humidity . During the feeding period 200 μl RU486 working solution or a 2% sucrose solution control were added to each vial every 2 days . Prior to behavioral tests , flies were transferred into vials containing fresh standard fly food without drugs for two hours . Heat-induced paralysis was assayed as above . Extracellular recordings of larval segmental nerves were as previously reported ( Simon et al . , 2009 ) . Although these neuronal bundles include both motor and sensory fibers , previous studies demonstrated that the majority of the burst firing activity patterns observed in this preparation are generated by motor neurons alone ( Fox et al . , 2006 ) . Feeding stage 3rd-instar larvae were dissected in HL-3 solution containing 2 . 0 mM CaCl2 , 70 mM NaCl , 5 mM KCl , 4 mM MgCl2 , 10 mM NaHCO3 , 5 mM trehalose , 115 mM sucrose , 5 mM HEPES , pH 7 . 2 . Segmental nerves connecting to the ventral nerve cord were left intact . We preferentially recorded from segmental nerves that innervate the anterior segments with a polished glass electrode to suck up the nerves . Neuronal signals were filtered by a high-pass filter set at 100 Hz and a low-pass filter set at 10 kHz ( Clampex software package ) . The extracellular temperature was manipulated in the recording chamber by using a temperature-control perfusion system ( Multi Channel Systems MCS , Baden-Württemberg , Germany ) using the following protocol: ( 1 ) Recording of neuronal activity started once the perfusion system was stable at 25°C for at least 1 min . Neuronal spikes were recorded at baseline for 3 min . ( 2 ) To acutely raise the temperature , perfusion was turned off until it reached 38°C stabley for at least 1 min . ( 3 ) Recording at 38°C was initiated 1 min after perfusion was turned on again for 3 min . Total RNA from adult fly heads or whole flies was extracted with the TRIzol reagent ( Applied Biosystems , Grand Island , NY ) . First strand cDNA pool was made from total RNA ( 1 μg ) with random hexamere oligos SuperScript II reverse transcriptase ( Invitrogen , Grand Island , NY ) in 20 μl reacting volume . cDNA pool was diluted ( 1:5 ) in distilled water . Gene specific assays were used to quantify genes with the SybrGreen method using the PowerSYBR Green Super PCR Mix ( ABI Inc . , Grand Island , NY ) on an ABI7500 machine ( Applied Biosystems ) using default parameters . Gene specific assays were designed with the PrimeTime qPCR Assay design tool ( Integrated DNA Technologies ) . The housekeeping gene rp49 was used as an RNA loading control as previously described ( Lu et al . , 2012 ) . Data were transformed and analyzed according to the ΔΔCt method and are represented as relative fold differences ( Lu et al . , 2012 ) . Primer sequences used are: sei-forward: 5′-TTATTCAAAGGCTGTACTCGGG-3′; sei-reverse: 5′-GATGCCATTCGTATAGGTCCAG-3′; ppk29-forward: 5′-CCTCTCAGGTATTCTTCGTTGG-3′; ppk29-reverse: 5′-TCGGTG-GAGATGGTATAGGTC-3′; rp49-forward: 5′-CACCAAGCACTTCATCCG-3′; rp49-reverse: 5′-TCGATCCGTAACCGATGT-3′ . The double fluorescence in situ hybridization in fresh brain sections was performed following published protocols ( Jones et al . , 2007 ) . Briefly , templates for the anti-sense ( AS ) and sense ( S ) control riboprobes targeting either ppk29 or sei transcripts were synthesized by PCR reactions from pUAST-ppk29 or pUAST-sei plasmids with the following primers: sei-AS left: 5′-TAATACGACTCA-CTATAGGGCATCGATTTGATTGTGGACG-3′;sei-AS right: 5′-CAGTATTCGGTGC-CACATTG-3′; sei-S left: 5′-CATCGATTTGATTGTGGACG-3′; sei-S right: 5′-TAATAC-GACTCACTATAGGGCAGTATTCGGTGCCACATTG-3′; ppk29-AS left: 5′-TAATACG-ACTCACTATAGGGAATACGAAATGTGGCGGAAG-3′; ppk29-AS right: 5′-GCATTTCTTCGATGCTGTCA-3′; ppk29-S left: 5′-AATACGAAATGTGGCGGAAG-3′; ppk29-S right: 5′-TAATACGACTCACTATAGGGGCATTTCTTCGATGCTGTCA-3′ . The sei riboprobes were labeled by DIG ( DIG RNA Labeling Kit , Roche ) , and the ppk29 riboprobes were labeled by fluorescein ( Fluorescein RNA Labeling Kit , Roche ) . Freshly dissected female brains ( 4–5 days old ) were embedded in cryo-embedding medium ( Tisse-Tek OCT , Fisher Scientific , Pittsburgh , PA ) . Frozen tissue were cryo-sectioned at 15 μm and fixed in 4% paraformaldehyde for 5 min . Probes were used at 2 ng/μl standard ISH hybridization buffer , 65°C overnight . Post-hybridization , tissues were blocked with TNB for 30 min followed by an incubation with a peroxidase-conjugated anti-DIG antibody in TNB buffer ( 1:500; Anti-Digoxigenin-POD , Fab fragments , Roche ) for 2 hr to detect sei-specific signal . To increase signal-to-noise ratio , the Tyramide Signal Amplification system ( TSA ) with Horseradish Peroxidase ( HRP ) was used . Samples were treated for 1 hr ( 1:50 , TSA Plus Cy3 , PerkinElmer , Waltham , MA ) . Then , samples were transferred to 0 . 3% hydrogen peroxide in TNT buffer to quench HRP activity for 20 min . Subsequently , the ppk29 antisense riboprobe was detected with a peroxidase-conjugated anti-Fluorescein antibody in TNB buffer ( 1:500; Anti-Fluorescein-POD , Fab fragments , Roche ) for 2 hr . To amplify ppk29 signal , samples were treated with the primary antibody in TSA signal amplification buffer ( 1:50 , TSA Plus Fluorescein , PerkinElmer ) for 1 hr . Tissue sections were mounted with Vectashield mounting medium with DAPI and imaged with a confocal microscope . The FirstChoice RLM-RACE Kit ( Life Technologies , Grand Island , NY ) was used to characterize the 3′UTRs of sei and ppk29 by following manufacturer′s instructions . Total RNA was isolated from mixed adults and 5 μg total RNA was used for first strand cDNA synthesis . Gene specific primers for PCRs were: ppk29: 5′-ACTTGCGACTGCTCTCTATTC-3′; sei: 5′-AAACTGCACAGGGACGATTT-3′; 3′RACEOuterPrimer: 5′-GCGAGCACAGAATTAATACGACT-3′ . Positive PCR products were sequenced from both ends with the PCR primers . Translating Ribosome Affinity Purification ( TRAP ) was used to isolate mRNAs specifically from larval motor neurons according to a recently published protocol ( Thomas et al . , 2012 ) . In short , a GFP tagged version of the ribosomal protein RpL10A was specifically expressed in larval motor neurons with the motor-neuron specific driver OK6-GAL4 ( Aberle et al . , 2002; Xiong et al . , 2010 ) . Total RNA was extracted using the TRizol reagent ( Life Sciences , Grand Island , NY ) from 35 3rd instar larvae . Enrichment for sei and ppk29 transcripts in motor neurons was measured with Real-Time qRT-PCR in TRAPped mRNAs by comparing enriched vs total RNA from the OK6-Gal4>UAS-GFP::RpL10A genotype . Real-time qRT-PCR was performed as described above . All quantitative behavioral , molecular , and neurophysiological data were analyzed using the most recent version of the SAS package ( SAS Inc . ) . One-way and Two-way ANOVAs were used to analyze parametric data followed by a Tukey post hoc analyses ( p<0 . 05 ) when comparisons between individual groups were required . Data distributions are presented as error bars that denote Standard Error of the Mean .	Neurons communicate with one another via electrical signals known as action potentials . These signals are generated when a stimulus causes sodium and potassium ion channels in the cell membrane to open , leading to an influx of sodium ions , followed by an efflux of potassium ions . Changes in temperature affect the rate at which ion channels open and close , and thus affect how easy it is for a stimulus to trigger an action potential . In response to a sudden rise in temperature , neurons must adjust the number of ion channels in their membranes to ensure that they do not become hyperexcitable , which could result in epilepsy . Now , Zheng et al . have revealed one possible mechanism for how neurons do this . In the fruit fly , Drosophila , a gene for a potassium channel is found on the same chromosomal location as a gene for a sodium channel , and some of the genetic elements that regulate the expression of these two genes even overlap . However , the genes are on opposite strands of the DNA double helix . This means that when the genes are transcribed to produce molecules of messenger RNA ( mRNA ) , which is usually single stranded , some of the mRNA molecules will pair up to form double-stranded mRNA molecules . This is significant because such RNA ‘duplexes’ have been shown to inhibit the translation of conventional single-stranded mRNA molecules into proteins , or to lead to their complete degradation . Zheng et al . found that flies with mutations in the potassium channel gene display seizures in response to sudden changes in temperature . However , insects with mutations in the sodium channel gene are not affected because , surprisingly , they have a higher than expected number of potassium channels . It turns out that the mutant sodium channel mRNA molecules are unable to form RNA duplexes with potassium channel mRNA molecules: these duplexes would normally limit the number of potassium channels so , in their absence , the number of potassium channels increases , and this protects the flies from seizures . Zheng et al . also uncovered a novel mechanism by which mRNA molecules can regulate gene expression independent of their role as templates for proteins . Further work is required to determine whether this mechanism is also present in other organisms , including humans .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "neuroscience" ]	2014	Natural antisense transcripts regulate the neuronal stress response and excitability
Out of millions of ejaculated sperm , a few reach the fertilization site in mammals . Flagellar Ca2+ signaling nanodomains , organized by multi-subunit CatSper calcium channel complexes , are pivotal for sperm migration in the female tract , implicating CatSper-dependent mechanisms in sperm selection . Here using biochemical and pharmacological studies , we demonstrate that CatSper1 is an O-linked glycosylated protein , undergoing capacitation-induced processing dependent on Ca2+ and phosphorylation cascades . CatSper1 processing correlates with protein tyrosine phosphorylation ( pY ) development in sperm cells capacitated in vitro and in vivo . Using 3D in situ molecular imaging and ANN-based automatic detection of sperm distributed along the cleared female tract , we demonstrate that spermatozoa past the utero-tubal junction possess the intact CatSper1 signals . Together , we reveal that fertilizing mouse spermatozoa in situ are characterized by intact CatSper channel , lack of pY , and reacted acrosomes . These findings provide molecular insight into sperm selection for successful fertilization in the female reproductive tract . In most mammals , millions or billions of spermatozoa are deposited into the cervix upon coitus . Yet less than 100 spermatozoa are found at the fertilization site , called the ampulla , and only 10–12 spermatozoa are observed around an oocyte ( Kölle , 2015; Suarez , 2002 ) . This implies the presence of mechanisms to select spermatozoa as they travel through the female reproductive tract and to eliminate non-fertilizing , surplus spermatozoa once the egg is fertilized ( Sakkas et al . , 2015 ) . Recent ex vivo imaging studies combined with mouse genetics have shown that surface molecules on the sperm plasma membranes such as ADAM family proteins are essential for the sperm to pass through the utero-tubal junction ( UTJ ) ( Fujihara et al . , 2018 ) . By contrast , whether such selection and elimination within the oviduct requires specific molecular signatures and cellular signaling of spermatozoa is not fully understood . Mammalian sperm undergo capacitation , a physiological process to obtain the ability to fertilize the egg , naturally inside the oviduct ( Austin , 1951; Chang , 1951 ) . The emulation of sperm capacitation in vitro led to the development of in vitro fertilization ( IVF ) techniques ( Steptoe and Edwards , 1976; Wang and Sauer , 2006 ) . Since then , most studies on sperm capacitation and gamete interaction have been carried out under in vitro conditions . However , mounting evidence suggests that in vitro sperm capacitation does not precisely reproduce the time- and space-dependent in vivo events in the oviduct . Protein tyrosine phosphorylation ( pY ) , which has been utilized as a hallmark of sperm capacitation over decades , showed different patterns in boar sperm capacitated in vitro from ex vivo and in vivo ( Luño et al . , 2013 ) . In mice , pY is not required for sperm hyperactivation or fertility ( Alvau et al . , 2016; Tateno et al . , 2013 ) . Previous in vitro studies that represent the population average at a given time may or may not have observed molecular details of a small number of the most fertilizing sperm cells . Capacitation involves extensive sperm remodeling that triggers cellular signaling cascades . Cholesterol shedding and protein modifications occur within the plasma membrane ( Visconti et al . , 1999; Vyklicka and Lishko , 2020 ) . Cleavage and/or degradation of intracellular proteins by individual proteases and the ubiquitin-proteasome system ( UPS ) also participate in the capacitation process ( Honda et al . , 2002; Kerns et al . , 2016 ) . Various capacitation-associated cellular signaling pathways that include cAMP/PKA activation followed by pY increase and rise in intracellular pH and calcium result in physiological outcomes such as acrosome reaction and motility changes ( Balbach et al . , 2018; Puga Molina et al . , 2018 ) . The sperm-specific CatSper Ca2+ channel forms multi-linear nanodomains on the flagellar membrane , functioning as a signaling hub that links these events and motility regulation during capacitation ( Chung et al . , 2014 ) . Sperm from mice lacking the CatSper genes are unable to control pY development and fail to migrate past the UTJ ( Chung et al . , 2014; Ho et al . , 2009 ) . The presence and integrity of CatSper nanodomains , probed by CatSper1 , correlate with sperm ability to develop hyperactivated motility ( Chung et al . , 2017; Chung et al . , 2014; Hwang et al . , 2019 ) . It is not known how these molecular and functional events are coordinated in the individual sperm cells within the physiological context . The central hypothesis of this study is that specific modifications and processing of CatSper1 , a pore subunit of CatSper channel , govern the channel biogenesis and Ca2+ signaling state , endowing differential sperm motility and fertility during the fertilization journey in the female reproductive tract . Here we reveal that most fertilizing mouse spermatozoa in situ are molecularly and functionally characterized by intact CatSper channel , lack of pY , and reacted acrosomes . Using biochemical and pharmacological analyses , we show that CatSper1 undergoes O-linked glycosylation during sperm differentiation and maturation . Capacitation induces CatSper1 cleavage and degradation dependent on Ca2+ influx and protein phosphorylation cascades . We find that CatSper1 processing correlates with pY development in the flagella among heterogenous sperm cells capacitated in vitro and in vivo . We use ex vivo imaging and microdissection to show that the intact CatSper channel is indispensable for sperm to successfully reach the ampulla and for the acrosome to react . Finally , we use newly developed 3D in situ molecular imaging strategies and ANN approach to determine and quantify the molecular characteristics of sperm distributed along the female reproductive tract . We demonstrate that spermatozoa past the UTJ are recognized by intact CatSper1 signals which are graded along the oviduct . These findings provide molecular insight into dynamic regulation of Ca2+ signaling in the selection , maintenance of the fertilizing capacity , and elimination of sperm in the female reproductive tract . We previously found that the CatSper channel complex is compartmentalized within the flagellar membrane , creating linear Ca2+ signaling nanodomains along the sperm tail ( Chung et al . , 2017; Chung et al . , 2014; Hwang et al . , 2019 ) . Caveolin-1 , a scaffolding protein in cholesterol-rich microdomains , colocalizes with the CatSper channel complex but does not scaffold the nanodomain ( Chung et al . , 2014 ) . The molecular weight and amount of intact CatSper1 , among all CatSper subunits , specifically declines during sperm capacitation ( Chung et al . , 2014; Figure 1B , D , H , I , K ) . To better understand the molecular basis and functional implications of the unique processing of CatSper1 , we first examined CatSper1 protein expression in the testis and epididymis . Interestingly , the molecular weight of CatSper1 increases gradually during sperm development and epidydimal maturation ( Figure 1A; top ) , indicating that CatSper1 undergoes post-translational modifications . Next , we examined the nature of the modifications . Blocking tyrosine phosphatases by sodium orthovanadate or adding specific protein phosphatases , such as protein serine/threonine phosphatase 1 ( PP1 ) or protein tyrosine phosphatase 1B ( PTP1B ) to solubilized sperm membrane fraction , does not change the molecular weight of CatSper1 ( Figure 1—figure supplement 1A ) . By contrast , when the sperm membrane was subjected to enzymatic deglycosylation , O-glycosidase , but not PNGase F , shifts the apparent molecular weight of CatSper1 to closest to the CatSper1 band that corresponds to the smallest molecular weight observed in the testis ( Figure 1A , B , Figure 1—figure supplement 1B ) . These data suggest that CatSper1 in sperm is not a phosphoprotein but an O-linked glycosylated protein . During sperm capacitation , cholesterol depletion destabilizes the plasma membrane and reorganizes the lipid raft ( Nixon et al . , 2007 ) . One simple hypothesis is that the capacitation-associated changes in raft stability and distribution render CatSper1 accessible to protease activity . To test whether CatSper nanodomains are raft-associated , we performed sucrose density gradient centrifugation , which identified CatSper1 in lipid raft subdomains in mature sperm ( Figure 1C ) . Before inducing capacitation , CatSper1 is not processed in sperm cells , probably because the CatSper1-targeting protease activity is normally not in the immediate vicinity to the CatSper nanodomains in the flagellar membrane ( Figure 1—figure supplement 1C , F ) . Supporting this notion , the protease activity readily cleaves CatSper1 by solubilizing the sperm membrane fraction with Triton X-100 ( Figure 1—figure supplement 1C ) . Next , we investigated the location of CatSper1 cleavage and degradation using recombinant CatSper1 proteins and sperm lysates . The CatSper1 antibody used in this study is raised against the first N-terminal 150 amino acids of recombinant CatSper1 ( Ren et al . , 2001 ) . C-terminal HA-tagged full-length ( FL ) or N-terminal deleted ( ND ) recombinant CatSper1 are expressed in HEK 293 T cells for pull-down and detection by western blot ( Figure 1E , F ) . Solubilized sperm lysates degrade FL-CatSper1 and result in increased detection of cleaved CatSper1 by HA antibody ( Figure 1G ) . By contrast , protein levels of recombinant ND-CatSper1 are not affected by incubation with sperm lysate ( Figure 1G ) . These results demonstrate that the cytoplasmic N-terminal domain of CatSper1 is the target region for proteolytic activity in sperm cells . How is the CatSper proteolytic activity regulated ? At the molecular level , capacitation is initiated by HCO3- uptake , which activates soluble adenylyl cyclase ( sAC ) , resulting in increased cAMP levels . HCO3- also stimulates CatSper-mediated Ca2+ entry into sperm cells by raising intracellular pH ( Kirichok et al . , 2006; Figure 1—figure supplement 1F ) . We thus examined whether the proteolytic activity requires cAMP/PKA and/or Ca2+ signaling pathways . Adding a PKA inhibitor H89 or the St-Ht31 peptide , which abolishes PKA anchoring to AKAP , accelerated CatSper1 degradation during sperm capacitation ( Figure 1H , Figure 1—figure supplement 1D ) . Consistently , calyculin A , a serine/threonine protein phosphatase inhibitor , suppresses the capacitation-associated CatSper1 degradation ( Figure 1H ) . These data suggest that regulation of the proteolytic activity targeting CatSper1 involves protein phosphorylation cascades . Interestingly , adding Ca2+ ionophore A23187 to the sperm suspension induces CatSper1 processing even under non-capacitating conditions that either do not support PKA activation or outright inhibit PKA activity ( Figure 1I , J , Figure 1—figure supplement 1E , top ) . This effect of Ca2+ influx by A23187 is not simply due to a rise in the intracellular Ca2+ but presumably also requires membrane events because loading sperm with BAPTA-AM cannot prevent the proteolytic activity under capacitating conditions ( Figure 1—figure supplement 1E , bottom ) . Thus , we hypothesize a Ca2+ dependent protease that is indirectly regulated by protein phosphorylation such as calpain ( Ono et al . , 2016 ) might process CatSper1 . We observed that all three classes of calpain inhibitors prevent CatSper1 from capacitation-associated degradation ( Figure 1K ) . All these results corroborate our hypothesis that the responsible protease ( s ) is associated with the flagellar membrane but is not localized inside the CatSper nanodomains ( Figure 1—figure supplement 1F ) ; capacitation-associated membrane reorganization and the increase in local Ca2+ via the CatSper channel activates the protease ( s ) . We speculate that this pathway can be indirectly modulated by PKA such as via regulation of the protease activity by PKA phosphorylation of PP1/PP2A . Inducing sperm capacitation in vitro results in a functionally heterogeneous sperm population in which no more than ~15% of cells are hyperactivated ( Neill and Olds-Clarke , 1987 ) . This is because individual sperm cells undergo time-dependent changes . Accordingly , the extent to which CatSper1 degrades varies with individual sperm cells capacitated in vitro ( Figure 1J , Figure 2A–C ) . The presence and integrity of the CatSper nanodomains , probed by the CatSper1 antibody , is an indicator of sperm capability to hyperactivate ( Chung et al . , 2017; Chung et al . , 2014 ) . Next , we examined the functional relevance of pY to sperm that can hyperactivate . Notably , we find that sperm cells that maintain intact CatSper1 develop capacitation-associated pY to a lesser degree in vitro ( Figure 2B , C ) . This finding is consistent with the reported phenotype of Catsper1 knockout sperm that exhibit potentiated pY during capacitation ( Chung et al . , 2014 ) . Thus far , our results suggest that in vitro capacitation generates a heterogeneous sperm population in which intact CatSper1 and pY development are inversely correlated at the single cell level . These heterogeneous sperm cells in vitro might reflect a collection of the time- and space-dependent changes that sperm undergo in the oviduct ( Chang and Suarez , 2012; Demott and Suarez , 1992 ) . To assess molecular changes of CatSper1 and pY in the context of sperm capacitation in vivo , we utilized Su9-DsRed2/Acr-EGFP male mice ( Hasuwa et al . , 2010 ) . Sperm from these transgenic mice have green acrosomes ( EGFP ) and red mitochondria ( DsRed2 ) ; loss of GFP indicates a reacted acrosome and RFP allows detection of sperm regardless of the acrosome state . We performed microdissection to obtain the spatially distributed sperm populations along the female reproductive tract mated at 8 hr post-coitus and flushed out sperm cells from different regions . By subsequent immunostaining , we found that CatSper1 in the spermatozoa that passed the utero-tubal junction ( UTJ ) are arranged normally along the tail , mostly protected from degradation , but in decreasing intensity and continuity more towards UTJ ( Figure 2D ) . In striking contrast , pY is not readily detected in the spermatozoa from the ampulla but appears in the oviductal sperm increasingly towards UTJ ( Figure 2E , F ) . Absence of EGFP reveals that spermatozoa from the ampulla are fully capacitated and acrosome reacted ( AR ) but those in the isthmus are undergoing AR ( Figure 2D , G ) . Ex vivo imaging of Su9-DsRed2/Acr-EGFP sperm in the reproductive tract removed from mated female mice reveals segment-specific patterns of the acrosome status ( Figure 2G ) . This result is consistent with the previous observations that AR initiates in the mid-isthmus ( Hino et al . , 2016; Muro et al . , 2016 ) and reacted spermatozoa are able to penetrate the zona in vivo ( Jin et al . , 2011 ) . Interestingly , we found that a few Catsper1-/- sperm cells that managed to arrive at the ampulla are all not acrosome reacted ( Figure 2H ) , supporting the notion that CatSper-mediated Ca2+ signaling is required for sperm acrosome reaction ( Stival et al . , 2018 ) . These results suggest that escape of CatSper1 from the cleavage and subsequent degradation suppresses pY development , enabling sperm to maintain hyperactivation capability , prime AR , and achieve the fertilization in vivo . The physiological importance of tracing a small number of spermatozoa progressing to the fertilization site prompted us to seek a method that enables direct molecular assessment of single cells inside the intact female tract . We have adapted tissue clearing technologies to establish three-dimensional ( 3D ) in situ molecular imaging systems for fertilization studies ( Figure 3 , Figure 3—figure supplement 1 , Figure 3—videos 1–6 ) . We found that various tissue clearing methods ( Chung et al . , 2013; Murray et al . , 2015; Yang et al . , 2014 ) are applicable to the reproductive organs from both male and female mice to preserve gross morphology , and fine cellular and subcellular structures . The cleared tissues preserved protein-based fluorescence and were compatible with labeling with dyes and antibodies; growing follicles ( WGA ) inside the ovary ( phalloidin ) , oviductal folds ( WGA ) and multi-ciliated ( anti-Ac-Tub ) epithelium ( PNA ) , different stages of male germs cells ( Acr-EGFP ) in the seminiferous tubules of the testis and the epididymis are readily detected ( Figure 3A–C , Figure 3—figure supplement 1 , Figure 3—videos 1–6 ) . 3D volume imaging of the whole cleared female tract labeled by WGA well illustrates the uterine and isthmic mucus and the labyrinths of passages sperm must navigate ( Figure 3A , B , Figure 3—figure supplement 1B , C , Figure 3—videos 3 , 5 , 6 ) . Moreover , 3D rendering of the images and digital reconstruction of oviductal surface and central lumen depicts continuous and non-disrupted morphology ( Figure 3D , E ) consistent with reported dimensions and parameters ( Stewart and Behringer , 2012 ) , validating the integrity of the processed oviduct . Next , we combined tissue clearing with an in vivo sperm migration assay ( Chung et al . , 2014; Yamaguchi et al . , 2009 ) to molecularly analyze different sperm populations during the fertilization process . Among tested clearing methods , we found that passive clearing of CLARITY-processed reproductive tracts from time-mated females retains the location and stability of gametes within the track past UTJ ( Figure 3F–I , Figure 3—videos 4 , 7 and Figure 4—video 1 ) ; whole-animal fixation by trans-cardiac perfusion perturbs minimally and rapidly arrests all cellular function while tissue-hydrogel matrix fills the lumen of the oviduct and provides supportive meshwork to prevent gamete loss during subsequent labeling steps ( Figure 3—figure supplement 2 ) . This new in situ imaging platform enables capturing a moment of sperm-egg interaction; a spermatozoon that approaches a fertilized egg protruding the 2nd polar body in the ampulla is detected in a cleared female tract 8 hr post-coitus immunostained by acetylated tubulin antibody ( Figure 3 , Figure 3—video 7 ) . CatSper1 antibody specifically recognizes sperm cells transfixed inside the ampulla in cleared female tract ( Figure 3H , I ) . Tissue clearing allows 3D volume imaging of the female tract but does not compromise the resolution . Two linear CatSper1 domains typically observed by confocal imaging are easily observed in the sperm cells inside an ampullar region of the whole cleared female tract ( Figure 3I ) . Thus , the integrity of CatSper1 in sperm cells at different locations along the female tract can be subjected to quantitative analysis . With this new imaging strategy to detect sperm cells that remain transfixed in the female reproductive tract ( Figure 3 ) , we investigated acrosome state and CatSper1 integrity in sperm populations directly from the cleared tract of females 8 hr after mating , focusing on a few anatomically defined regions ( Figure 4 ) . Based on the earlier results from micro-dissection or ex vivo imaging ( Figure 2D–H ) , we anticipated that sperm cells that successfully reach the ampulla would be CatSper1-intact and acrosome reacted . As expected , most sperm cells located in the ampulla exhibit linearly arranged intact CatSper1 and reacted acrosomes ( Figure 4A , top , Figure 4—video 1 ) . In the middle isthmus , both CatSper1 and acrosome remain intact in most sperm cells , but mixed patterns are observed in some cells ( Figure 4A , middle , Figure 4—video 1 ) . Interestingly , acrosome is largely intact in the sperm clusters in the proximal isthmus close to UTJ whereas CatSper1 is barely detected ( Figure 4A , bottom , Figure 4—video 1 ) . This contrasts with the reduced but readily visible CatSper1 in the sperm from the same region by microdissection ( Figure 2D ) . It is possible that the relatively longer tissue processing time and subsequent labeling could have contributed to lower the signal to noise ratio to a certain degree . Notably , 3D volume imaging of this mid isthmus regions reveals sperm cells aligned in one direction towards the ampulla , providing unprecedented insight into sperm taxis in the fertilization process ( Figure 4—video 1 ) . Our qualitative but semi-quantitative analyses suggest that CatSper1 is largely protected from degradation once in the oviduct; acrosome reaction initiates in the mid-isthmus and is completed in the ampulla before interacting with the oocytes ( Figure 4B ) . These results are consistent with our initial observations from microdissection and ex vivo imaging studies ( Figure 2D , G ) , validating the information obtained by our in situ molecular imaging platform . Taken together , we conclude that intact CatSper1 , lack of pY , and reacted acrosome are molecular and functional signatures of most fertilizing spermatozoa in the physiological context . Processing 3D volumetric fluorescent data presents a significant challenge; analyses of sperm in the female tract includes object identification in the voluminous specimen , object separation from background noise , and object alignment in three dimensions . To address these logistics problems , we took an advantage of the artificial neural network ( ANN ) approach for automatic localization and signal isolation . We performed a proof-of-principle investigation utilizing CatSper1 distributions in sperm cells from our 3D in situ molecular imaging ( Figure 5 ) . First , we manually annotated 3D fluorescent signatures of sperm , somatic nuclei and background noise from the original images . These signatures were placed in different abundance models in the ANN 3D training environments ( Figure 5A , Figure 5—figure supplement 1 , Figure 5—video 1 ) for subsequent ANN training using MatLab ANN module . We performed a supervised iteration process where the sperm locations were predefined in the training environments ( Figure 5—figure supplement 2A ) . We evaluated the performance of individual ANN according to their sensitivity and specificity in detecting sperm cells and somatic nuclei , and the abundance ( voxel occupancy ) of noise ( Figure 5B , Figure 5—figure supplement 2B , C ) . Detection sensitivity is chosen as a major parameter used to evaluate the ANN performance in the training environment simulated with the values similar to those in real samples . The specificity required for sperm detection is lower than the sensitivity , thus provides mainly empty analytical frames that are easily removed manually . After iteration and performance evaluation , we selected the best performing ANN and analyzed images from our experimental samples for which we manually counted sperm number ( Figure 5C ) . The selected ANN is able to recognize all the sperm detected manually and the ANN sensitivity varies around 90% in individual samples ( Figure 5—figure supplement 2C ) , validating the ANN performance . Furthermore , the false-negative detection of sperm all comes from the sperm with dubious signals in the antecedent human eye evaluation; the 90% of the ANN detected sperm expresses well recognizable CatSper1 fluorescent staining patterns ( Figure 5C , Figure 5—figure supplement 3A ) . In order to pair each CatSper1 signal containing tail with the head from the same cell in the subsequent analysis , we took the reverse approach to the environment production by removing the detected noise and somatic cell nuclei from the analytical frames ( Figure 5—figure supplement 3B ) . The pre-processed CatSper1 fluorescent signal were then subjected to subsequent alignment , pattern linearization , and intensity detection ( Figure 5D , Figure 5—figure supplement 3C ) . These steps make possible calculation and visual representation of the fluorescent intensity parameters along the sperm tail related to their CatSper1 integrity status ( Figure 5E–G ) . The quadrilateral and linear organization of the Ca2+ signaling nanodomains discovered by super-resolution imaging of CatSper1 ( Figure 6A ) is an indicator of a sperm cell’s ability to hyperactivate and fertilize the egg in vitro ( Chung et al . , 2017; Chung et al . , 2014 ) . The present study demonstrates that incubating sperm cell under capacitating conditions in vitro induces CatSper1 cleavage and degradation , leading to a heterogeneous sperm population ( Figures 1 and 2 ) . Building on our observations of sperm cells from microdissection , ex vivo imaging , and CLARITY-based in situ molecular imaging ( Figures 2 , 3 and 4 ) , we hypothesize that CatSper1 is a built-in countdown timer for sperm death and elimination in the female tract; CatSper1 cleavage and degradation , triggered in a time- and space-dependent manner along the female tract , signals to end sperm motility , and ultimately sets sperm lifetime in vivo . With our newly developed automated ANN method to obtain high-quality 3D fluorescent images of CatSper1 in the sperm cells from cleared female tract samples , we further tested this idea by quantitatively analyzing the CatSper1 signals in situ . Our in situ imaging platform offers the typical resolution that a confocal microscopy can provide; two separated CatSper1 arrangement along the sperm tail ( Chung et al . , 2017 ) are detected without any computational processing ( Figures 3I and 4A ) . This encouraged us to develop an analytical procedure to assess the status of CatSper1 quadrilateral and linear distributions . We isolated the fluorescent signal from a proximal region of the principal piece close to the annulus where the immunolabeled CatSper1 signal is the most intense ( Figure 6B ) . To superpose the individual cross-sectional images according to the expected four intensity peaks , we aligned randomly oriented transversal-projection images by placing the quadrant with the highest fluorescent intensity to top right corner ( Figure 6B , inset ) . The aligned images were then superposed ( Figure 6C ) and used for statistical purposes to represent quadrilateral arrangement of CatSper1 in individual sperm cells ( Figure 6D ) . The individually processed images of sperm cells from the oviductal regions close to UTJ , middle isthmus , and ampulla were again superposed to create cumulative diagrams and heat maps corresponding to these regions ( Figure 6E ) . They show quadrilateral distribution of enriched CatSper1 signal more clearly from the sperm population in the ampulla compared to the population in the oviduct close to UTJ ( Figure 6E , G , H ) . To further quantify and statistically analyze our outputs , we divided the pre-processed images of individual sperm cells on 80 round areas ( Figure 6—figure supplement 1A ) and calculated fluorescent intensities among them . The quantified intensity from the 80 areas were plotted; the observed four peaks ( highest intensity ) and valleys ( lowest intensity ) were used to calculate the delta value among them to represent the quality of CatSper1 quadrilateral structure ( Figure 6F ) . Our quantitative analysis ( Figure 6G , Figure 6—figure supplement 1B ) shows consistent results with our previous semi-quantitative analysis by manual assignment of the CatSper1 patterns ( Figure 4 ) . Together with the whole tissue image processing ( Figure 3E ) , the quantitative analysis clearly visualizes that sperm populations located along the cleared oviduct have statistically different CatSper1 quadrilateral intensity delta values ( Figure 6H ) . Testicular spermatozoa undergo maturation and biochemical alterations in the intraluminal environment of the epididymis ( Cornwall , 2009 ) . Glycan-modifying enzymes such as glycosidases and glycosyltransferases are present in the epididymal luminal fluid ( Tulsiani , 2003 ) . Here we have shown that CatSper1 in mouse sperm is an O-linked glycosylated protein with gradually increasing molecular weight from the testis to the epididymis during male germ-cell development . The different forms of native CatSper1 may represent different degrees of glycosylation . Heterologously expressed CatSper1 cannot reach the plasma membrane , remaining instead at the ER/Golgi ( Chung et al . , 2017; Chung et al . , 2011; Ren et al . , 2001 ) . It is intriguing that the molecular weight of recombinant CatSper1 is similar to one of the testicular forms of CatSper1 but bigger than that of the enzymatically deglycosylated and naked polypeptide . O-linked glycosylation takes place in the cis-Golgi for secreted and transmembrane proteins after the protein is folded ( Röttger et al . , 1998 ) , suggesting that additional modification is required for native CatSper1 to exit the Golgi . Determining the precise identity and modification site may help to clarify the long-sought functional expression of the CatSper channel in heterologous systems . In rodents , sialyltransferase displays maturation-associated quantitative changes ( Ram et al . , 1989; Scully and Shur , 1988 ) and sperm lose sialic acid from the surface during capacitation ( Ma et al . , 2012 ) . Sperm glycoproteins promote sperm migration and survival in the female reproductive tract ( Ma et al . , 2016 ) . We speculate that mature CatSper1 in sperm contains terminal sialic acid residues , consistent with the small drop in Catsper1 molecular weight during capacitation . The dynamic sugar modifications on CatSper1 may serve as a binding site for decapacitation factors and/or a recognition site during capacitation . Supporting this idea , it was previously shown that mouse sperm lacking the CatSper channel cannot pass through the UTJ ( Chung et al . , 2014; Ho et al . , 2009 ) . Capacitation-associated CatSper1 degradation is blocked by incubation with a 26S proteasome inhibitor , MG-132 ( Chung et al . , 2014 ) . Now we show that solubilized sperm membrane fraction contains additional proteolytic activities that cleave within CatSper1 NTD . The proteolysis involves two distinct pathways: Ca2+ entry and phosphorylation cascades . We hypothesize that a member of calpains , the Ca2+ dependent modulatory protease family , might cleave CatSper1 , as their proteolytic activity can be regulated by PKA ( Du et al . , 2018 ) . Among 15 calpain proteins identified in mammals ( Ono et al . , 2016 ) , calpain1 and calpain11 were previously detected in our sperm proteome ( Hwang et al . , 2019 ) . Intriguingly , we observed that CatSper1 processing requires not a simple rise in intracellular Ca2+ , but rather Ca2+ influx mediated by the CatSper channel which normally accompanies membrane reorganization during capacitation . Increased Ca2+ level overrides the phosphorylation effect on calpain1 activity ( Du et al . , 2018 ) . We speculate that calpain11 might be similarly regulated to calpain1 , as their domain structures and catalytic residues are conserved ( Ono et al . , 2016 ) . Since recombinant CatSper1 is cleaved more specifically by sperm lysates , we propose that the testis-specific calpain11 ( Ben-Aharon et al . , 2006 ) may target CatSper1 . The effect of CatSper1 truncation on channel activity and sperm motility remains to be determined in future studies . CatSper1 truncation may be coordinated with molecular changes of other CatSper subunits . For example , the protein level of CatSper2 , but not CatSper3 or 4 , also decreases after capacitation when probed with the antibody recognizing its C-terminal domain ( CTD ) ( Figure 1; Chung et al . , 2014 ) . Since the cytoplasmic modulatory subunits , CatSperζ and Efcab9 , mainly interact with the channel pore ( Hwang et al . , 2019 ) , specific processing of the intracellular domains of pore subunits could alter the interactions and subsequent channel activity . Spermatozoa that successfully navigate to the fertilization site in the female reproductive tract and interact with the egg are recognized by intact CatSper1 . CatSper1 processing may lead to a loss of control in hyperactivation and eventually end sperm life . An increase in pY is one of the various capacitation-associated parameters observed from in vitro capacitated sperm cells ( Visconti et al . , 1995 ) . Subsequently , pY was observed in the flagellum of mouse and human sperm interacting with the oocyte in the medium that supports sperm capacitation and fertilization in vitro ( Sakkas et al . , 2003; Urner et al . , 2001 ) . This correlation of pY and the zona binding previously established pY as an indicator of successful sperm capacitation . More recently , however , different observations have been made with in vivo approaches . In sows inseminated close to ovulation , spermatozoa found in the UTJ exhibited more phosphorylation in the flagella than those bound to oviductal epithelial cells ( OEC ) , where pY was limited to the equatorial region in the sperm head or no pY was observed ( Luño et al . , 2013 ) . In mice , the testis-specific tyrosine kinase , Fer , is demonstrated as a master kinase for capacitation-associated pY ( Alvau et al . , 2016 ) . Surprisingly , homozygous Fer-mutant male mice are fertile even though their sperm do not develop pY . All together , these results lead to a new interpretation of the physiological significance of pY: successful sperm capacitation does not require pY development . Determining the precise time and place of pY development in sperm in situ would help to elucidate its function in sperm capacitation and fertilization . Here we have shown that sperm cells , which have capacitated in vivo and successfully migrated to the ampulla , are characterized , not only by intact CatSper1 , but also by relative lack of pY development and reacted acrosome . These results coincide with our observations from in vitro capacitated sperm cells and other previous studies; pY development inversely correlates with CatSper1 integrity at the single cell level ( Figure 2 ) ; genetic and pharmacological ablation of Ca2+ entry potentiates pY ( Chung et al . , 2014; Navarrete et al . , 2015 ) ; AR occurs in mid-isthmus before contacting an oocyte ZP ( Hino et al . , 2016; Jin et al . , 2011; Muro et al . , 2016 ) . Sperm remaining in the female reproductive tract need to be eliminated after fertilization . They may undergo apoptosis and phagocytosis in the female reproductive tract ( Aitken and Baker , 2013; Chakraborty and Nelson , 1975 ) and/or become lost in the peritoneal cavity ( Mortimer and Templeton , 1982 ) . pY is reported to mediate apoptosis in immune cells ( Yousefi et al . , 1994 ) and cancer cells ( Liu et al . , 1994 ) . We propose that capacitation-associated global pY development represents degenerating sperm which might concomitantly lose motility . It is intriguing that capacitation-associated reactive oxygen species ( ROS ) generation activates intrinsic apoptotic cascade and compromises sperm motility ( Koppers et al . , 2011 ) . Consistent with this idea , ROS inactivates protein tyrosine phosphatase ( Tonks , 2005 ) and enhances pY development in sperm ( Aitken et al . , 1998 ) . Inhibition of PKA anchoring to AKAPs , which induces CatSper1 truncation and degradation , also suppresses acrosome reaction in capacitating sperm cells in vitro ( Stival et al . , 2018 ) . Thus , CatSper-mediated Ca2+ signaling directly or indirectly contributes to sperm acrosome reaction in the female tract . Future work will determine molecular mechanisms by which CatSper channel activity fine-tunes Ca2+ signaling to regulate hyperactivated motility , as well as how the Ca2+ signaling is linked to coordinate acrosome reaction . The successful development of in vitro capacitation and fertilization systems provided fundamental insights into sperm capacitation , fertilization , and early embryogenesis . On the other hand , it is evident that the in vitro systems have limitations . Sperm numbers required for IVF are much higher than those observed at the fertilization site in vivo ( Suarez , 2007 ) . Sperm capacitated in vitro do not encounter the anatomically and spatially distinct environment of the female reproductive tract , for example , missing their interaction with the oviductal epithelial cells . In vitro capacitation also lacks secretory factors from the male and female reproductive tracts that can affect the surface protein dynamics during the capacitation process ( Flesch and Gadella , 2000 ) . Mouse models that typically use epidydimal sperm for in vitro studies do not contain secretions from male glands . This is in contrast with ejaculated sperm from human and domestic animals . Recent studies have observed sperm behavior in the physiological context through ex vivo imaging of sperm in the mouse and bovine oviducts under transillumination ( Hino and Yanagimachi , 2019; Ishikawa et al . , 2016; Kölle et al . , 2009; Muro et al . , 2016 ) . Yet this technique is limited in providing molecular information at a single cell level , as live imaging is not easily amenable to direct molecular labeling and 3D volume imaging . Here , we report new systems to molecularly examine individual sperm cells capacitated in vivo . Polymerization of the hydrogel-embedded time-mated female reproductive tract followed by passive clearing provides a stable meshwork to minimally disturb the original location of sperm cells inside the female tract . This approach allowed us to assess the fine organization of CatSper nanodomains in the sperm cells distributed along the female reproductive tract . We showed that both the intensity and the quadrilateral detection of the domains probed by CatSper1 appear as the common pattern of sperm reaching the ampulla and potentially fertilizing the oocyte . The experimental outputs complement the molecular and functional information of sperm released from micro-dissected female tracts and ex vivo imaging , identifying molecular and functional signatures of fertilizing sperm in the physiological context . Furthermore , we demonstrate the efficacy of topological heat-map representations of cumulative results by automatic sperm detection and image post-processing and averaging; this method provides statistically robust presentation and interpretation of the volumetric image data . The present study opens up new horizons to microscopically visualize and analyze molecular events in single sperm cells that achieve fertilization . This will allow us to better understand physiologically relevant cellular signaling pathways directly involved in fertilization . We also have illustrated that the same approach of tissue-clearing based 3D in situ molecular imaging is applicable to study gametogenesis in situ . Future areas for investigations as natural extensions of the current study are gameto-maternal interaction , development , transport , and implantation of early embryos and maternal-fetal communication . Developing gamete-specific antibodies and/or knockout validated antibodies to probe molecular abundancy and dynamics in situ and post-processing tools for various parameters will be critical to this end . Catsper1-null ( Ren et al . , 2001 ) and Su9-DsRed2/Acr-EGFP ( Hasuwa et al . , 2010 ) mice were generated in the previous study and maintained on a C57BL/6 background . Su9-DsRed2/Acr-EGFP mice were crossbred with Catsper1-null mice to generate Su9-DsRed2/Acr-EGFP Catsper1-null mice . WT C57BL/6 and B6D2F1 male and CD1 female mice were purchased from Charles River Laboratories ( Wilmington , MA ) . Mice were cared for in compliance with the guidelines approved by the Yale Animal Care and Use Committees . HEK293T and COS-7 cells were purchased from ATCC and authenticated by positive detection of SV40 DNA and STR-based analysis using the Cell ID System ( Promega ) . They were cultured in DMEM ( GIBCO ) supplemented with 10% FBS ( Thermofisher ) and 1× Pen/Strep ( GIBCO ) at 37°C , 5% CO2 condition . Cultured cells were used to express recombinant proteins ( HEK293T cells ) or make total cell lysates ( COS-7 cells ) . In-house rabbit polyclonal CatSper1 ( Ren et al . , 2001 ) , CatSper3 ( Qi et al . , 2007 ) , CatSperε ( Chung et al . , 2017 ) antibodies were described previously . Polyclonal CA-IV antibody ( M-50 ) was purchased from Santacruz . Monoclonal antibodies were purchased from BD Biosciences: anti-caveolin1 ( clone 2297 ) ; EMD Milipore: anti-phosphotyrosine ( clon4G10 ) , anti-acetylated tubulin ( clone 6-11B-1 ) , anti-HA agarose ( clone HA-7 ) ; Thermo Scientific: anti-HA magnetic beads; and Cell Signaling Technology: β-actin ( clone 13E5 ) and HA ( clone C29F4 ) . HRP-conjugated goat anti-rabbit IgG and goat anti-mouse IgG were from Jackson Immunoresearch . PNA-Alexa 568 , WGA-Alexa 555 , WGA-Alexa 647 , goat anti-mouse IgG ( Alexa 488 or 647 ) , and goat anti-rabbit IgG ( Alexa 568 or Alexa 647 ) were from Invitrogen . H89 , calyculin A , and calpain inhibitor I were purchased from Calbiochem . ST-Ht31 was from Promega . Calpain inhibitor II and III were from Enzo life science . All other chemicals were from Sigma-Aldrich unless indicated . Sperm cells were released from caput , corpus , or cauda regions of the epididymis in M2 medium ( EMD Millipore ) . To induce capacitation , sperm from caudal epididymis were incubated in human tubular fluid ( HTF ) medium or M16 ( EMD Milipore ) containing 25 mM sodium bicarbonate at 37°C , 5% CO2 condition at 2 × 106 cells/mL concentration for the indicated time . Sperm cells were incubated under capacitating conditions with or without the following chemicals: H89 ( 50 μM ) , ST-Ht31 ( 10 μM ) , Calyculin A ( 100 nM ) , calpain inhibitor I ( 20 μM ) , calpain inhibitor II ( 20 μM ) , or calpain inhibitor III ( 20 μM ) . Sperm cells suspended in M2 medium ( 2 × 106 cells/mL ) were incubated with A23187 ( 10 μM ) to induce Ca2+ influx under non-capacitating conditions . NEB10β bacterial strain ( NEB ) was used for molecular cloning . Genomic regions encoding full-length ( FL , 1–686 aa ) and N-terminal domain deleted ( ND , 345–686 aa ) mouse CatSper1 were amplified from mouse CatSper1 expression vector ( Hwang et al . , 2019 ) . The PCR products were subcloned into pcDNA3 . 1 ( - ) vector using NEBuilder HiFi DNA Assembly ( NEB ) to express the recombinant proteins tagged with HA at C-terminus ( pcDNA3 . 1 ( - ) -FL-Catsper1-HA and pcDNA3 . 1 ( - ) -ND-Catsper1-HA ) . HEK293T cells were transfected with constructs encoding FL-CatSper1 or ND-CatSper1 to express the recombinant proteins transiently . Polyethyleneimine was used for the transfection following the manufacturer’s instruction as previously . Sperm migration assay was performed as previously described ( Chung et al . , 2014 ) . Briefly , female mice were introduced to single-caged Su9-DsRed2/Acr-EGFP males for 30 min and checked for the vaginal plug . Whole female reproductive tracts were collected 8 hr post-coitus and subjected to ex vivo imaging to examine spermatozoa expressing reporter genes in the tract ( Eclipse TE2000-U , Nikon ) . Female reproductive tracts from timed-mated females to Su9-DsRed2/Acr-EGFP or Su9-DsRed2/Acr-EGFP Catsper1-null males were collected 8 hr post-coitus . Sperm cells were released by micro-dissection of the female reproductive tract followed by lumen flushing of each tubal segment ( cut into ~1–2 mm pieces ) . Each piece was placed in 50 μL of PBS on glass coverslips and the intraluminal materials were fixed immediately by air-dry followed by 4% PFA in PBS . Ampulla and uterine tissue close to UTJ were placed in 100 μL of PBS and vortexed briefly to release the sperm within the tissues . Fixed sperm cells were subjected to immunostaining . Non-capacitated or in vitro capacitated sperm cells on glass coverslips were washed with PBS and fixed with 4% paraformaldehyde ( PFA ) in PBS at RT for 10 min . Fixed samples were permeabilization with PBS-T for 10 min and blocked with 10% normal goat serum in PBS for 1 hr at RT . Blocked sperm cells were stained with primary antibodies , anti-CatSper1 ( 10 μg/mL ) , and anti-phosphotyrosine ( 1:1000 ) , at 4°C for overnight , followed by staining with secondary antibodies for 1 hr at RT . Hoechst was used for counterstaining sperm head . Sperm cells were mounted ( Vectashield , Vector Laboratories ) and imaged with confocal microscopes ( Zeiss LSM710 Elyra P1 and Olympus Fluoview 1000 ) . The overall strategy for the artificial neural network ( ANN ) image processing is described in Figure 5—figure supplement 2A . The individual signal patterns ( sperm , somatic cell nuclei , and noise ) were isolated from the original volume images using Zen Blue ( Carl Zeiss ) and IMARIS software ( Oxford instruments ) and exported as . obj/ . fbx files . The isolated signal patterns were used to generate 3D training environments for ANN by importing different abundancies of the individual components ( Figure 5—figure supplement 1 ) to the 3D environment operating system , Blender 2 . 79 ( https://www . blender . org/ ) ; the individual 3D training environment ( ~104 ) generated together with the exactly defined coordinates of individual components were exported as . obj/ . fbx/Notepad++ files . The ANN training environments were used to develop the ANN detecting the sperm in situ . The ANN training was carried out using MATLAB 9 . 3 ( R2017b ) software ANN toolbox . The input to the ANN would be virtual z-stacks of the produced training environments . The isolated sperm signal patterns were used as a target signature . The supervised training process was performed by comparing the vector coordinates of the individual sperm signatures in the output with the pre-defined vector coordinates of the signatures in the input . This approach also enabled us to evaluate the ANN performance and to quantify signature detection sensitivity and specificity ( Figure 5—figure supplement 2A ) , panel 6 ‘ANN performance evaluation’ . The detection sensitivity and specificity of ANNs were the major performance indicators used to select ANNs . ANNs with the best performance in detecting the sperm signature were subsequently applied to detect the sperm fluorescent signatures and their post-processing in real volumetric data . In the real environments , selected ANNs showed both sensitivity and specificity around 90% ( Figure 5—figure supplement 2C ) . See the related source data . Statistical analyses were carried out with a one-way analysis of variance ( ANOVA ) with the Tukey post hoc test . Differences were considered significant at p<0 . 05 . For ANN analysis , both parametric ( ANOVA; Tukey post hoc ) and non-parametric ( KW-ANOVA ) tests were carried out to evaluate the presented differences; both tests resulted in the same significance output with differences considered significant at p<0 . 05 .	When mammals mate , males ejaculate millions of sperm cells into the females’ reproductive tract . But as the sperm travel up the tract , only a handful of the ‘fittest’ sperm will actually manage to reach the egg . This process of elimination prevents the egg from being fertilized by multiple sperm cells and stops the eggs from being fertilized outside of the womb . A lot of what is known about fertilization in mammals has come from studying how sperm and eggs cells interact in a Petri dish . However , this approach cannot explain how sperm are selected and removed as they journey towards the egg . Previous work suggests that a calcium channel , which sits in the membrane surrounding the sperm tail , may provide some answers . The core of this channel , known as CatSper , is made up of four proteins arranged into a unique pattern similar to racing stripes . Without this specific arrangement , sperm cells cannot move forward and fertilize the egg in time . To investigate the role of this protein in more depth , Ded et al . established a new way to image the minute structures of sperm cells , such as CatSper , in the reproductive tract of female mice . Experiments in a Petri dish revealed that sperm cells that have been primed to fertilize the egg are a diverse population: in some cells one of the proteins that make up the calcium channel , known as CatSper1 , is cleaved , while in other cells this protein remains intact . Visualizing this protein in the female reproductive tract showed that sperm cells close to the site of fertilization contain non-cleaved CatSper1 . Whereas sperm cells further away from the egg – and thus closer to the uterus – are more likely to contain broken down CatSper1 . Taken together , these findings suggest that the state of the CatSper1 protein may be used to select sperm that are most likely to reach and fertilize the egg . Future studies should address what happens to the calcium channel once the CatSper1 protein is cleaved , and how this channel controls the movements and lifespan of sperm . This could help identify new targets for contraception and improve current strategies for assisted reproduction .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2020	3D in situ imaging of the female reproductive tract reveals molecular signatures of fertilizing spermatozoa in mice
Lipid droplets ( LDs ) are ubiquitous organelles that store neutral lipids , such as triacylglycerol ( TG ) , as reservoirs of metabolic energy and membrane precursors . The Arf1/COPI protein machinery , known for its role in vesicle trafficking , regulates LD morphology , targeting of specific proteins to LDs and lipolysis through unclear mechanisms . Recent evidence shows that Arf1/COPI can bud nano-LDs ( ∼60 nm diameter ) from phospholipid-covered oil/water interfaces in vitro . We show that Arf1/COPI proteins localize to cellular LDs , are sufficient to bud nano-LDs from cellular LDs , and are required for targeting specific TG-synthesis enzymes to LD surfaces . Cells lacking Arf1/COPI function have increased amounts of phospholipids on LDs , resulting in decreased LD surface tension and impairment to form bridges to the ER . Our findings uncover a function for Arf1/COPI proteins at LDs and suggest a model in which Arf1/COPI machinery acts to control ER-LD connections for localization of key enzymes of TG storage and catabolism . Nearly all organisms balance fluctuations in the availability of energy sources with the need for energy expenditure . With its high energy content , triacylglycerol ( TG ) stored in lipid droplets ( LDs ) is the primary means of storing energy for many organisms ( Thiele and Spandl , 2008; Fujimoto and Parton , 2011; Brasaemle and Wolins , 2012; Walther and Farese , 2012 ) . LDs also store lipid precursors for membrane synthesis ( e . g . , cholesterol and glycerophospholipids ) needed , for example , when cells exit quiescence and expand membranes for cell division ( Kurat et al . , 2009 ) . Due to their function in lipid storage , LDs are central to the development of pathologies associated with excess lipid accumulation , ranging from atherosclerosis and cardiovascular disease to obesity and metabolic syndrome ( Krahmer et al . , 2013 ) . Unlike most organelles , LDs are not delimited by a bilayer membrane but instead are covered with a monolayer of phospholipid surfactant , which is important for their stability in cells ( Tauchi-Sato et al . , 2002; Krahmer et al . , 2011; Yang et al . , 2012a ) . In this sense , LDs constitute the dispersed phase of a cellular emulsion , with the phospholipid monolayer acting as a surfactant at the interface of the oil core with the aqueous cytosol ( for review , see Thiam et al . , 2013b ) . Proteins specifically located at the LD surface execute many of the reactions of lipid storage or mobilization . For example , enzymes mediating TG synthesis and hydrolysis localize to LDs , where they mediate LD expansion and shrinkage , respectively ( Kuerschner et al . , 2008; Schweiger et al . , 2008; Stone et al . , 2009; Murugesan et al . , 2013; Wilfling et al . , 2013 ) . How such enzymes are specifically targeted to LDs is a poorly understood , yet fundamental question . Unbiased genome-wide screens in model systems , such as Drosophila cells , revealed factors that are required for LD targeting of proteins ( Beller et al . , 2008; Guo et al . , 2008 ) . Specifically , members of the Arf1/COPI machinery , but not other proteins involved in secretory trafficking ( e . g . , COPII or clathrin ) , are necessary for normal LD morphology and for the targeting of some proteins to LDs ( Beller et al . , 2008; Guo et al . , 2008; Soni et al . , 2009 ) . Depletion of Arf1/COPI proteins from cells leads to the formation of relatively uniform LDs of a characteristic size that exhibit impaired lipolysis ( Beller et al . , 2008; Guo et al . , 2008 ) . Consistent with this , Arf1/COPI proteins are required for LD localization of the major TG lipase ATGL ( brummer in Drosophila ) ( Beller et al . , 2008; Soni et al . , 2009; Ellong et al . , 2011 ) . ATGL was shown to behave , biochemically , as an integral membrane protein ( Soni et al . , 2009 ) , and it was suggested that this lipase is transported to LDs from the ER by vesicular trafficking . In vesicular trafficking , the best-characterized function of Arf1/COPI proteins is in retrograde transport , that is , retrieving ER resident proteins from the Golgi apparatus ( Nickel et al . , 2002 ) . In this pathway , Arf1 is loaded with GTP by a nucleotide exchange factor , such as GBF1 [gartenzwerg ( garz ) in Drosophila] . The activated Arf1-GTP then recruits the coatomer , a heptameric protein complex , leading to the formation of a coated transport vesicle . Subsequent uncoating of the vesicle allows its transport and fusion to the target membrane ( e . g . , the ER ) . It is unknown how Arf1/COPI proteins function in LD biology . Although one possibility is that Arf1/COPI proteins target proteins to LDs via bilayer vesicles , a variety of studies suggest a function directly at LDs . First , Arf1 and its GEF , GBF1 , as well as other members of the COPI machinery , have been found on LDs in proteomic and cell biological studies ( Nakamura et al . , 2005; Bartz et al . , 2007; Ellong et al . , 2011; Bouvet et al . , 2013 ) . Second , the expression of dominant-negative Arf1T31N , which binds its exchange factor tightly , localizes to LDs ( Guo et al . , 2008 ) . Third , Arf1Q71L that cannot hydrolyze GTP ( and hence acts as a dominant-negative mutant in vesicular trafficking ) activates lipolysis from LDs ( Guo et al . , 2008 ) . Most recently , GTP-bound Arf1 and COPI proteins were shown to bud nano-LDs of ∼60 nm diameter from a phospholipid covered oil-water interface in vitro ( Thiam et al . , 2013a ) , indicating that this machinery can interact with monolayer interfaces such as what is found at LD surfaces . Collectively , these data suggest an alternative , so far untested model , in which Arf1/COPI proteins function in cells directly at LDs in a way that enables protein targeting . Besides ATGL , other enzymes involved in TG metabolism also localize to LD surfaces . For example , at least one isoenzyme catalyzing each step of de novo TG synthesis from glycerol-3-phosphate ( e . g . , GPAT4 , AGPAT3 , and DGAT2 ) localizes to a subset of LDs . Each of these enzymes contains two hydrophobic segments likely forming a hairpin in the ER membrane or the LD monolayer ( Wilfling et al . , 2013 ) . LD localization of these enzymes enables LDs to synthesize TG locally and expand their neutral lipid cores under conditions of excess energy ( fatty acid ) availability . Recent evidence indicates that these enzymes re-localize to a subset of LDs from the ER via abundant membrane bridges that form between the organelles ( Wilfling et al . , 2013; Thiam et al . , 2013a ) . Intriguingly , this targeting reaction can occur rapidly at a particular LD , from which TG synthesis enzymes were absent for a long time ( Wilfling et al . , 2013 ) . How the targeting process is initiated and how bridges between LDs and the ER are established is unknown . Here we investigate the mechanism of Arf1/COP protein function in cellular LD protein targeting by using a combination of cell biological and biochemical approaches . In contrast to the canonical role of these proteins in vesicular trafficking , we uncover a mechanism of action that relies on altering the surface lipid composition of LDs . Based on the presence of the Arf1/COPI machinery at LDs , we propose a newly identified function of Arf1/COPI proteins in modulating LD surfaces to enable protein targeting . Many cell types , including Drosophila S2 cells , contain two populations of LDs: a few rather large , expanding LDs , several microns in diameter , and many smaller ( less than a micron diameter ) LDs ( Wilfling et al . , 2013 ) . Depletion of the Drosophila Arf1 homologue Arf79F or βCOP results in a relatively uniform LD population ( Beller et al . , 2008; Guo et al . , 2008 ) . We quantified this phenotype and found that depletion of either Arf79F or βCOP results in a relatively narrow , monodisperse distribution of LDs that lies intermediate in size ( mean ∼1 . 3 μm ) between small and larger expanding LDs ( Figure 1A ) . 10 . 7554/eLife . 01607 . 003Figure 1 . The COPI machinery is required for LD targeting of specific proteins . ( A ) The bimodal size distribution of control cells ( black line ) with few large LDs and many small LDs shifts a monodisperse size in Arf1/COPI-depleted cells ( green and red line ) . The figure shows the density function of the LD size distribution . ( B ) Endogenous GPAT4 detected by immunofluorescence localizes to LDs ( stained by BODIPY ) in control treated cells , but not in the absence of any of the COPI machinery components , except εCOP . ( C ) The amount of GPAT4 fractionating with LDs ( detected by thin layer chromatography of TG ) is reduced in cells depleted of βCOP . ( D ) Arf1/COPI effects on LD protein targeting are protein specific , as Lsd1 targeting to LDs is not affected in cells depleted of Arf1/COPI . Cherry-Lsd1 localizes to LDs stained with BODIPY in the absence of Arf79F ( middle panel ) or βCOP ( bottom panel ) . Scale bars are 10 μm ( overview ) or 1 μm ( inlay ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01607 . 00310 . 7554/eLife . 01607 . 004Figure 1—figure supplement 1 . The COPI machinery is required for LD targeting of specific proteins . ( A ) Targeting of GPAT4 from the ER to LDs depends on Arf1/COPI . GFP-GPAT4 localizes to LDs stained with lipidtox in control-treated cells , but not in the absence of COPI machinery subunits , Arf79F ( middle panel ) or βCOP ( bottom panel ) . ( B ) Depletion of 'COP does not significantly affect the secretion of HRP . ( C ) The number of GPAT4 positive LDs per cell remains unaffected when comparing control-treated cells with cells in which 'COP has been depleted . ( D ) DGAT2 , catalyzing the formation of TG from DAG and fatty acid–CoA depends on Arf1/COPI for its targeting to LDs . Cherry-DGAT2 was expressed in S2 cells , and localized in respect to LDs ( stained with BODIPY ) by fluorescence microscopy in control cells and cells depleted for Arf79F ( middle panel ) or βCOP ( bottom panel ) . Scale bars are 10 μm ( overview ) or 1 μm ( inlay ) . ( E ) Brummer LD targeting in S2 cells is dependent on Arf1/COPI . GFP-Brummer localizes to LDs stained with lipidtox ( top panel ) . GFP-Brummer LD targeting is abolished in the absence of either Arf79F ( middle panel ) or βCOP ( bottom panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01607 . 004 Since the Arf1/COPI-depleted cells lacked large expanding LDs , we tested whether Arf1/COPI depletion affected the LD localization of enzymes catalyzing LD expansion by examining LD localization of fluorescent GFP-tagged or endogenous GPAT4 ( detected by immunofluorescence ) . We found that depletion of Arf79F , garz , or any of the coatomer subunits , with the exception of εCOP , impaired the LD localization of GPAT4 ( Figure 1B , Figure 1—figure supplement 1B , C ) . Similarly , depletion of Arf79F or βCOP compromised LD targeting of the triglyceride-synthesis enzyme DGAT2 ( Figure 1—figure supplement 1D ) . Defective GPAT4 localization to LDs with Arf1/COPI depletion was also evident in subcellular fractionation experiments , where the amount of GPAT4 in the LD fraction was greatly diminished in the absence of βCOP ( Figure 1C ) . Consistent with previous reports ( Beller et al . , 2008; Soni et al . , 2009 ) , we found that brummer was also missing from LDs in Arf1/COPI-depleted cells ( Figure 1—figure supplement 1E ) . The targeting defect was apparently specific to proteins targeting LDs from bilayer membranes , as at least some proteins that localize to LDs from the cytoplasm , such as the Drosophila perilipin Lsd1 , were not affected by Arf1/COPI depletion ( Figure 1D ) . The absence of TG synthesis enzymes likely explains the absence of large LDs , and the defect in lipase targeting and the associated defect in lipolysis , likely contribute to the increase in size of small LDs to an intermediate size . Some of the proteins requiring Arf1/COPI for LD localization , such as specific isoenzymes of TG synthesis ( including GPAT4 ) , access LDs from the ER through membrane bridges ( Wilfling et al . , 2013 ) . We hypothesized that Arf1/COPI activity on LDs is required for establishing the junctions between the ER and LDs . To test this hypothesis , we performed add-back experiments with Arf1/COPI in GPAT4 localization assays . We fused LD-containing cells depleted for βCOP and expressing GFP-tagged GPAT4 localized in the ER , with wild-type cells that provide Arf1/COPI proteins in trans ( Figure 2A ) . After cell–cell fusion , the COPI pool from wild-type cells rapidly equilibrates in the mixed cytoplasm . This led to rapid targeting of GFP-GPAT4 to some of the pre-existing LDs ( Figure 2B ) , with a variable lag phase of 1–25 min ( Figure 2C ) . LD targeting of GPAT4 invariably occurred directly from the ER through a number of junctions between the two organelles ( Figure 2C; Video 1 ) . After the initial lag phase , GPAT4 targeting was rapid , with a characteristic time τ of 3 . 6 ± 1 min ( Figure 2D , Figure 2—figure supplement 1A–C ) . A mathematical model using the ( experimentally determined ) apparent diffusion constant of GPAT4 in the ER ( 0 . 035 ± 0 . 005 μm2/sec ) revealed that roughly 5–9 connections between a LD and the ER are required to obtain the observed speed of GFP-GPAT4 targeting to LDs ( Figure 2—figure supplement 1B , ‘Materials and methods’ ) . This is consistent with the observed number of connections to large LDs in fluorescence and EM images ( Figure 2C , FW , MJO , and TCW , unpublished observations ) . 10 . 7554/eLife . 01607 . 005Figure 2 . Arf1/COPI mediate LD protein targeting by establishing connections between the ER and LDs . ( A ) Schematic representation of cell–cell fusion experiments . ( B ) Fusion of βCOP depleted cells expressing GFP-GPAT4 and induced LDs with WT cells rapidly rescues GFP-GPAT4 targeting to LDs . Soluble cherry fluorescent protein is expressed as a marker for content mixing of fused cells . Scale bars are 10 μm ( overview ) or 1 μm ( inlay ) . ( C ) Time lapse analysis of GFP-GPAT4 targeting to LDs . Upper panels shows representative images from time-lapse imaging of cell–cell fusion experiments . Arrows point to apparent connections between LDs and the ER . Scale bar = 1 μm . Lower panel shows quantitation of GFP-GPAT4 localization to LDs in nine independent cell–cell fusion experiments . Time = 0 indicates fusion and content mixing of cells . ( D ) Rate of GFP-GPAT4 recruitment to LDs after cell–cell fusion . Insert shows the histogram of characteristic recovery time τ . DOI: http://dx . doi . org/10 . 7554/eLife . 01607 . 00510 . 7554/eLife . 01607 . 006Figure 2—figure supplement 1 . Mathematical model for GPAT4 targeting to LDs through bridges with the ER . ( A ) Measurement of GPAT4-GFP diffusion in the ER . D = 0 . 035μm2/sec denotes for the apparent diffusion coefficient of GPAT4 in the ER , determined by FRAP recovery experiments . Normalized fluorescence intensity of recovery in a bleached ER tubule from three independent experiments are shown . ( B ) Mathematical Model Estimating the Number of Connections , nc , Necessary to Target GPAT4 . The image and cartoon of a connection between ER and LD , with the different connection parameters , are shown . The gradient of concentration of GPAT4 between ER and LD follows: ( 1 ) C0−N/S = J/k , where C0 is the GPAT4 concentration in the ER , considered constant; N is the molar number of GPAT4 molecules on the LD at time t; S is the surface of the LD , J is the surface flux of GPAT4 molecules from ER to LD; k , homogeneous to a permeability , reflects the speed at which GPAT4 molecules travels along the connections to reach the LD , and writes as D/I . l is the characteristic length of the ER-LD connection . Within a time dt , the amount of proteins , dN , targeting the LD , via nc connections , writes as: ( 2 ) dN = nc × JLdt . L is the characteristic length of the section of ER-LD connections . Combining ( 1 ) and ( 2 ) yields: ( 3 ) dN/dt = nc × kL ( C0−N/S ) , with the solution: ( 4 ) N = C0S ( 1−exp ( −t/τ ) ) , where τ = S/ ( nc × kL ) and equals: ( 5 ) τ = lS/ ( nc × DL ) . τ is the characteristic time of GPAT4 targeting to LDs and experimentally found to be τ = 3 . 6min in Figure 2D . From ( 5 ) , the required number of connections to target GPAT4 during that time τ is nc = lS/ ( τ×DL ) . The surface S of the LD is 4πR2 and l/L varies between 1 . 5 and 3 depending on the experiment; we take l/L ∼2 , and define τ as τ = lS/ ( DL ) = 8πR2/D . τ varies between 21–31min ( depending on the diffusion coefficient D ) . Therefore nc , writing as nc = τ/τ , varies between 5 to 9 connections . ( C ) Time Lapse Analysis of GFP-GPAT4 Targeting to LDs . Rate of GFP-GPAT4 recruitment to LDs after cell-cell fusion for 12 different experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01607 . 00610 . 7554/eLife . 01607 . 007Video 1 . Time lapse analysis of GFP-GPAT4 targeting to LDs in cell–cell fusion experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01607 . 007 We next asked how Arf1/COPI proteins trigger the formation of LD-ER connections . If Arf1/COPI proteins act directly on LDs in this process , then a portion of these proteins should localize to LDs . To test this , we determined the localizations of the Arf1 exchange factor garz and αCOP in Drosophila S2 cells . For each protein , we observed foci localizing to the surface of some LDs in addition to signal likely reflecting the Golgi pool of the proteins ( Figure 3A ) . Importantly , LD co-localization occurred more frequently than would be expected by overlaying a random pattern of foci onto the LD signals ( Figure 3A , Figure 3—figure supplement 1A ) . We also observed abundant co-localization of GFP-tagged Arf79F with the LD marker CGI-58 on LDs ( Figure 3—figure supplement 1B ) , and the signal was distinct from signals marking the Golgi apparatus ( Figure 3—figure supplement 1C ) . 10 . 7554/eLife . 01607 . 008Figure 3 . The COPI machinery localizes to the LD surface . ( A ) The endogenous COPI machinery stained with αCOP or garz antibodies ( red ) localizes to LDs in S2 cells . Frequencies of colocalization of αCOP and garz spots with LDs from experiments are higher than expected from a random distribution . ( B ) The endogenous COPI machinery localizes to LDs in NRK cells . NRK cells stained for βCOP or GBF1 by immunofluorescence ( red ) show partial colocalization with LDs stained with BODIPY ( green ) . Colocalization of βCOP with LDs in NRK cells is not random . Relative frequencies of βCOP , KDEL receptor and clathrin spots colocalizing with LDs determined in experiments are respectively compared to the frequencies of colocalization from a binomial random distribution . From the two frequencies ( experiment vs simulation ) , a significant overrepresentation of βCOP on LDs is observed , whereas clathrin and KDEL receptor ( KDELR ) are not found on LDs . For ( A ) and ( B ) scale bars are 10 μm ( overview ) or 1 μm ( first inlay ) or 250 nm ( second inlay ) . Statistical significance was tested by a student t test with p<0 . 01 ( n = 30 ) . ( C ) Localization of β’COP ( green ) to the LD surface ( perilipin3 , red ) using confocal ( upper panel ) and super-resolution STED microscopy ( lower panel ) . Scale bar = 500 nm ( overview ) or 100 nm ( inlay ) . ( D ) Localization of β’COP to LDs is efficiently blocked by treatment of cells with the Arf1 GEF inhibitors brefeldin A or golgicide A . Scale bar = 10 μm ( overview ) or 1 μm ( inlay ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01607 . 00810 . 7554/eLife . 01607 . 009Figure 3—figure supplement 1 . COPI machinery localizes to the surface of LDs . ( A ) The probability P of having n colocalization events from nA type A dots ( A = βCOP , αCOP , clathrin , KDELR ) follows a binomial distribution . ns is the fraction with the highest probability of colocalization from the random distribution . nexp is the fraction of colocalization observed in the experiment . If nexp is larger than ns , A is overrepresented on LDs . A representative plot of P ( n ) obtained for an experiment of βCOP colocalization with LDs is depicted showing that β COP is overrepresented on LDs ( nexp >> ns ) . ( B ) Arf79F-cherry ( red ) colocalizes with the LD marker protein CGI-58 ( green ) . LDs are visualized by MDH ( blue ) . ( C ) GFP-GM130 ( green ) and Arf79F-cherry ( red ) show colocalization in the perinuclear region of S2 cells , representing the Golgi apparatus ( lower panel ) but not on LDs ( upper panel ) . ( D ) Arf79F-GFP binding to LDs is stabilized by treatment with brefeldin A . Arf79F-GFP localization was followed after the addition of 100 µM brefeldin A over time . LDs are stained with LipidTOX . Left panels show representative images of such an experiment . Right panel shows a quantification of at least three independent experiments . Error bars represent the s . d . of the mean . Scale bars are 10 μm ( overview ) or 1 μm ( inlay ) . ( E ) NRK cells stained for β’COP by immunofluorescence ( red ) show partial colocalization with LDs stained with BODIPY ( green ) . ( F ) NRK cells costained for endogenous βCOP ( red ) and β’COP ( green ) show colocalization in the same structures on LDs visualized by MDH ( blue ) . ( G ) NRK cells co-stained for endogenous clathrin ( green ) and βCOP ( red ) ( upper panel ) or β′COP ( green ) and KDELR ( red ) ( lower panel ) show no colocalization on LDs ( blue ) . ( H ) Endogenous β’COP ( green ) colocalizes with the LD marker protein perilipin3 ( red ) . LDs are visualized by MDH ( blue ) . ( I ) NRK cells co-stained for endogenous GM130 ( green ) and βCOP ( red ) show colocalization in the perinuclear region of cells , representing the Golgi apparatus ( lower panel ) but not on LDs ( upper panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01607 . 009 To test whether Arf1/COPI localization to LDs was conserved between different species , we localized GBF1 , β′COP , and βCOP in mammalian NRK cells . Similar to findings in Drosophila cells , some signal from each protein localized in foci to LDs ( Figure 3B , Figure 3—figure supplement 1E ) . Also in this case , LD colocalization of the coatomer protein βCOP was overrepresented , compared with the expectation for a random pattern . Moreover , βCOP and β′COP were localized to the same foci ( Figure 3—figure supplement 1F ) . In contrast , clathrin or KDEL receptor were either underrepresented or only occasionally showed a focus on LDs , consistent with a randomly distributed pattern ( Figure 3B , Figure 3—figure supplement 1G ) . In NRK cells , the colocalizing foci completely overlapped with the LD marker perilipin3 in confocal and super-resolution stimulated emission depletion ( STED ) images ( Figure 3C , Figure 3—figure supplement 1H ) , but were distinct from signals marking the Golgi apparatus ( GM130 , Figure 3—figure supplement 1I ) . As expected by its interaction with GTP-loaded Arf1 , LD localization of COPI coat was blocked completely when cells were incubated with brefeldin A ( Figure 3D , Figure 3—figure supplement 1D ) , which inhibits Arf1 nucleotide exchange factors . This inhibition by brefeldin A leads to the formation of a stable , abortive complex of the compound with Arf1 ( Walker et al . , 2011 ) . Similar results were obtained with Golgicide A , a specific inhibitor of GBF1 Arf1 exchange factors ( Figure 3D; Dobrosotskaya et al . , 2002 ) . The presence of Arf1/COPI on LDs prompted us to test whether this machinery can bud nano-LDs from cellular LDs , similar to the way it forms COPI-coated nano-LDs from artificial oil-water interfaces ( Thiam et al . , 2013a ) . We isolated LDs from Drosophila S2 cells and incubated them with purified Arf1/COPI proteins . Electron microscopy revealed that specifically in the presence of the Arf1/COPI machinery and a non-hydrolyzable GTP analogue ( GTPγS ) , abundant protein-covered nano-LDs were formed ( Figure 4A ) . The nano-LDs had an average diameter of 65 nm ± 10 nm ( Figure 4A ) , consistent with the size range of COPI-coated vesicles ( Simon et al . , 1996 ) or the size of COPI-coated nano-LDs formed at artificial oil-water interfaces ( Thiam et al . , 2013a ) . For vesicle formation by COPI , dimerization of Arf1 is required ( Beck et al . , 2008 ) . Interestingly , the formation of nano-LDs was unaffected in reactions performed with the Arf1Y35A mutant , which is deficient in dimer formation ( Beck et al . , 2008; Figure 4—figure supplement 1A , B ) . This lack of requirement for dimerization of Arf1 in nano-LD formation might reflect a lower energy barrier in the scission step of budding off a nano oil droplet compared with a vesicle . 10 . 7554/eLife . 01607 . 010Figure 4 . Arf1/COPI bud nano-LDs from purified , cellular LDs . ( A ) Purified LDs from S2 cells were incubated with components of the Arf1/COPI machinery in the presence or absence of GTPγS . Representative electron micrographs reveal abundant nano-LDs formed in the presence of activated Arf1/COPI . Scale bars are 500 nm ( overview ) or 100 nm ( inset ) . Histograms show the size distribution of nano-LDs formed . ( B ) Purified LDs have the ability to activate Arf1 by GTP loading . Purified LDs incubated with Arf1 , GTPγS , and fluorescently labeled COPI , but without the addition of a nucleotide exchange factor , are able to recruit COPI ( green ) in a GTP-dependent manner ( top left panel and bottom right panel ) . COPI binding is abolished by blocking the exchange factor garz with an antibody ( bottom left panel ) , or by digesting LD proteins with trypsin prior to the experiment ( top right panel ) . Recruitment is partially restored by addition of a soluble Arf1-GEF , ARNO . Adding a secondary antibody ( red ) that recognizes the αgarz antibody labeled LDs dependent on the presence of the primary antibody . Scale bars are 5 μm ( overview ) or 1 μm ( inlay ) . ( C ) Quantification of the recruitment of COPI to purified LDs . For each experiment in ( B ) the average intensity of 15 LDs was determined . ( D ) Nano-LDs formed from cellular LDs into the buffer visualized by fluorescence microscopy detecting Arf1 ( red ) , COPI ( green ) and LDs ( MDH labeled , blue ) . Scale bar is 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01607 . 01010 . 7554/eLife . 01607 . 011Figure 4—figure supplement 1 . Purified LDs were incubated with Arf1-Y35A , coatomer , ARNO and GTPγS , upon budding conditions shown in Figure 4A . The left panel shows an electron micrograph reaction products with budded nano-LDs as shown in the inlay . Dimerization deficient Arf1-Y35A is able to induce nano-LDs formation similarly to Arf1 . Quantifications were done using the approach of the paper Thiam et al . , 2013a . The right panel shows an electron micrograph without the COPI machinery . DOI: http://dx . doi . org/10 . 7554/eLife . 01607 . 011 We next tested whether purified LDs can activate Arf1 by GTP loading and , as a consequence , form COPI nano-LDs . Addition of ARNO , a soluble Arf1 nucleotide exchange factor to the in vitro budding reaction increased the number of nano-LDs observed , demonstrating that exchange activity was limiting ( Figure 4A ) . Using fluorescently labeled coatomer , we observed recruitment to LDs in such reactions in a GTP-dependent manner ( Figure 4B ) . COPI binding to LDs was abolished efficiently by blocking the exchange factor garz with an antibody or by digesting LD proteins with trypsin before the experiment ( Figure 4B , C ) . In either case , recruitment could be partially restored by adding a soluble Arf1-GEF , ARNO . When we added a secondary antibody against the αgarz antibody , LDs were labeled if the primary antibody was present , further indicating that Arf1-GEF was on the LDs ( Figure 4B ) . In addition to COPI labeling of the LD surface in these reactions , we observed nano-LDs ( stained by BODIPY ) in the supernatant from reactions containing fluorescently labeled Arf1 ( Cy3 ) and COPI ( Alexa647 ) , as well as GTPγS ( Figure 4D; Video 2 ) , directly demonstrating nano-LD formation from isolated LDs . 10 . 7554/eLife . 01607 . 012Video 2 . Time lapse video of Arf1/COPI coated nano-LDs formed from cellular LDs . Nano-LDs are visualized by fluorescence microscopy detecting Arf1 ( red ) , COPI ( green ) and LDs ( MDH labeled , blue ) . Shifts between channels are due to short time delays between channel acquisitions , during which nano-LDs diffuse in solution . DOI: http://dx . doi . org/10 . 7554/eLife . 01607 . 012 The budding of nano-LDs , with a very high surface to volume ratio , from the surface of donor LDs is predicted to remove primarily phospholipids . Therefore LDs from Arf1/COPI-depleted cells should contain more phospholipids than LDs from control cells . Indeed , when we compared lipids of purified LDs from cells depleted of βCOP with those from control cells , we found the levels of phosphatidylcholine ( PC ) and phosphatidylethanolamine ( PE ) , but not TG , increased ( Figure 5A ) . 10 . 7554/eLife . 01607 . 013Figure 5 . Lack of Arf1/COPI increases phospholipids on LDs , abolishing GPAT4 LD targeting . ( A ) PC and PE , but not TG levels are increased in LDs from βCOP depleted cells compared with WT cells . ( B ) LD ( green ) targeting of endogenous CCT1 ( red ) is delayed in cells depleted of βCOP . Time = 0 indicates the addition of oleate to the cells . Ratios between nuclear and LD targeted CCT1 signals are shown . Error bars represent the SD of the mean ratio from 100 cells . Western blot analysis shows decreased targeting of CCT1 to LDs when cells are depleted for βCOP . ( C ) Efficient co-depletion of CCT1 and Arf1 or βCOP restores GPAT4 targeting to LDs even in the absence of a functional COPI machinery . ( D ) Arf1/CCT1 or βCOP/CCT1 co-depletion blocks HRP secretion . Error bars represent the SD of triplicate measurements . DOI: http://dx . doi . org/10 . 7554/eLife . 01607 . 01310 . 7554/eLife . 01607 . 014Figure 5—figure supplement 1 . Depletion of COPI machinery components is efficient . Expression of indicated subunits were measured by quantitative real-time CR . Primers used are listed in Table 1 . Means ± SD of three experiments are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01607 . 014 We previously discovered that the enzyme CCT1 , catalyzing the rate-limiting step of PC synthesis , binds to LDs deficient in PC , effectively acting as a biosensor for PC on expanding LDs ( Krahmer et al . , 2011 ) . We therefore reasoned that Arf1/COPI depletion , by causing increased PC levels on LDs , would affect the time course of CCT1 recruitment to LDs . Indeed , CCT1 localized to LDs at later times during LD expansion ( Figure 5B ) . The model of Arf1/COPI removing primarily phospholipids from donor LDs predicts that the effects of Arf1/COPI depletion might be overcome by alternative treatments limiting the availability of phospholipids for LDs . To test this prediction , we decreased PC synthesis by depleting CCT1 , either alone or in combination with Arf79F or βCOP . Depletion efficiency was equally efficient in single and double depletions ( Figure 5—figure supplement 1 ) . As expected from previous studies , CCT1 depletion resulted in coalescence of LDs into giant LDs , due to limiting availability of phospholipids on LDs ( Figure 5C; Guo et al . , 2008; Krahmer et al . , 2011 ) . Also , as predicted , CCT1 depletion did not abolish GPAT4 targeting to LDs . Strikingly , when CCT1 was depleted concomitantly with Arf79F or βCOP , GPAT4 targeting to LDs was efficiently restored ( Figure 5C ) . The ability of CCT depletion to complement deficient Arf1/COPI function was specific to the GPAT4 targeting to LDs , as CCT1 depletion did not restore the defect in protein secretion due to Arf79F or βCOP depletion ( Figure 5D ) . If Arf1/COPI proteins function to remove phospholipids from LDs and thus allow membrane bridges to be established between the ER and LDs , then modulating the LD surface properties by other means should similarly alter protein targeting to LDs . To test this prediction , we added PC to cells . In agreement with the hypothesis , adding excess PC prevented GPAT4 targeting to the LD surface ( Figure 6A ) . We suspect that , in this experiment , PC accumulates on LD surfaces and shields their TG cores , thereby lowering surface tension , and thus might prevent the establishment of membrane bridges with the ER . 10 . 7554/eLife . 01607 . 015Figure 6 . LD surface properties modulate GPAT4 LD targeting . ( A ) Addition of exogenous PC to S2 cells inhibited GPAT4 LD targeting in βCOP or control RNAi-treated cells . Cholesterol ( chol ) addition to cells restored GPAT4 LD targeting in βCOP-depleted cells . Targeting efficiency depends on the ratio of added cholesterol and PC in βCOP or control RNAi-treated cells . ( B ) The artificial compounds SR59230A or stearylamine rescued GPAT4 LD targeting in βCOP depleted cells . The numbers of GPAT4-targeted LDs per cell are shown . Error bars represent the SD from the mean number of GPAT4-targeted LDs in 40 cells . Statistical significance was calculated using ANOVA , followed by a Dunnett test with a 99% confidence interval ( p=0 . 01 ) . ( C ) GPAT4 targeting to phospholipid monolayers depends on the surface tension . Buffer drops containing GPAT4-GFP-labeled microsomes are formed in a microfluidics device by flow focusing . The buffer micro-reactors are surrounded by oil of different composition ( TG containing PC/PE ( 0 . 25% ea . ) or PC/PE ( 0 . 25% ea . ) + 2% cholesterol , or cholesterol only ( 0 . 5% ) ; concentrations are w/w compared to TG ) . Each formed buffer drop pass through a zigzag region where microsomes inside the buffer drop are constantly brought into contact with the monolayer at the oil interface . Drops are arrested in a network of trapping chambers . In the presence of PC/PE , little GPAT4-GFP is targeted to the monolayer but stays in microsomes . Addition of 2% cholesterol or cholesterol alone significantly increased GPAT4-GFP signal on the monolayer . Quantification of the relocalization efficiency of GPAT4 from microsomes to the monolayer interface . Bar = 100 μm ( device ) or 25 µm ( drop ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01607 . 01510 . 7554/eLife . 01607 . 016Figure 6—figure supplement 1 . Cholesterol leads to an increase of surface tension at a TG/buffer interface . ( A ) Cholesterol increases the surface tension of PC/PE monolayer at a TG/buffer interface . Surface tension was measured by a drop weight method for the indicated phospholipid/cholesterol ratios . Error bars represent the SD of the mean from a minimum of 15 experiments . ( B ) Cholesterol decreases the stability of artificial oil microdroplets in buffer . Time course of the optical density evolution of TG droplets with PC/PE and increasing amounts of cholesterol is shown . Error bars represent the s . d . of the mean from six independent experiments . ( C and D ) Exogenously added cholesterol is incorporated into cellular membranes and LDs . Total lipids from control or cholesterol-treated cells normalized to 200 μg of protein ( C ) or isolated LD fraction normalized to 50 μg of protein ( D ) , where separate by TLC and stained using Hanessian's stain . The band intensity for PE , PC and cholesterol was measured from three biological replicates . Western blots show the purity of the LD fraction . ( E ) The initial oil stream ( containing TG , cholesterol and/or phospholipids ) is split into two streams that flow perpendicularly to a buffer stream with GPAT4-GFP containing microsomes . ‘Flow focusing’ the oil streams continuously pinches small buffer micro-reactors with defined size into the oil stream . After formation , the micro-reactor enters a zigzag pattern that creates a chaotic flow inside the micro-reactor . This increases the contact of the encapsulated microsomes with the interfacial monolayer formed at the buffer micro-reactor phase and the surrounding oil . Micro-reactors are arrested in a network of trapping chambers after the zigzag , allowing imaging and analysis of the monolayer targeted GPAT4 population . DOI: http://dx . doi . org/10 . 7554/eLife . 01607 . 016 Conversely , we predicted that a surfactant with a low potential to shield TG , therefore generating higher LD surface tension , might restore GPAT4 targeting to LDs even in the setting of Arf1/COPI depletion . We hypothesized that cholesterol ( which in Drosophila cells is normally only present in very low amounts ) , with its small head-group and pronounced cone shape , might act in this manner . As predicted , in vitro measurements confirmed that cholesterol addition increased the surface tension of a TG-buffer interface when added in the presence of phospholipids ( PC and PE ) mimicking the LD surface composition ( Figure 6—figure supplement 1A ) . Additionally , emulsion stability was reduced by cholesterol ( Figure 6—figure supplement 1B ) . When cholesterol was added to Arf1/COPI-depleted cells , the cholesterol content increased at LDs ( Figure 6—figure supplement 1C , D ) . Importantly , adding cholesterol to cells was sufficient to restore targeting of GPAT4 to LDs in Arf1/COPI-depleted cells ( Figure 6A ) , and the number of GPAT4-positive LDs depended on the ratio of cholesterol and PC added to cells ( Figure 6A ) . To test whether the effect is due to cholesterol’s physical properties , or alternatively , to some physiological change in the cells induced by cholesterol , we repeated these experiments with SR59230A and stearylamine . Both of these agents are surface active , amphiphilic molecules that normally do not occur in cells , but which induce LD destabilization in vivo ( Murphy et al . , 2010 ) , likely by increasing LD surface tension or by decreasing line tension of coalescence intermediates . In agreement with the findings with added cholesterol , adding SR59230A or stearylamine efficiently restored GPAT4 targeting to Arf1/COPI-depleted LDs ( Figure 6B ) . To further test whether changes of LD surface properties , introduced by the action of Arf1/COPI , controls GPAT4 targeting to LDs through membrane bridges , we reconstituted this reaction in vitro with a microfluidic device ( Figure 6—figure supplement 1E ) . We introduced microsomes harboring GFP-GPAT4 into buffer-in-oil micro-reactors ( Figure 6C ) . Mixing the content of the micro-reactors by flow through zig-zagging micro-channels led to localization of some GPAT4 to the monolayer delimiting the TG phase . The amount of GFP-GPAT4 at the monolayer depended on its composition and varied according to the surface tension . Similar to the situation in cells , monolayers rich in cholesterol and having higher surface tension , bound GFP-GPAT4 more efficiently than control monolayers of PC and PE ( Figure 6C ) . The Arf1/COPI machinery is important for governing LD morphology , protein targeting , and consequently lipolysis ( Beller et al . , 2008; Guo et al . , 2008; Soni et al . , 2009; Ellong et al . , 2011 ) . However , how Arf1/COPI proteins act to affect LDs has been unknown . Recent evidence from in vitro experiments using artificially generated oil-water interfaces show that GTP-bound Arf1 and COPI proteins are sufficient to bud nano-LDs ( Thiam et al . , 2013a ) , suggesting that the Arf1/COPI machinery might perform a similar function at the oil-water interfaces of LDs in cells . The current studies show that Arf1/COPI machinery has an additional function other than its canonical function in forming bilayer vesicles namely that these proteins can control the formation of membrane bridges between LDs and the ER to mediate targeting of specific proteins ( such as ATGL , GPAT4 , and DGAT2 ) from the ER to LDs . Taken together , our data are most consistent with a model for the function of Arf1/COPI in which these proteins act directly on LDs to remove phospholipids from the LD surfaces through the formation of nano-LDs . Budding of nano-LDs in turn , increases surface tension of the donor LD and allows membrane bridges to be established between this LD and the ER . These membrane bridges provide a pathway for the localization of membrane-associated proteins , such as ATGL and GPAT4 , and it allows them to diffuse to the LD surface where they perform key steps in TG metabolism . Without functional Arf1/COPI , TG synthesis enzymes fail to target LDs , which as a consequence cannot expand to form large LDs . In addition , as reported ( Beller et al . , 2008; Soni et al . , 2009 ) , ATGL fails to target LDs leading to a defect in lipolysis and a mild increase in the size of small LDs . Consistent with this model , incubation of cells depleted for components the Arf1/COPI machinery with chemicals that increase LD surface tension , such as cholesterol , stearylamine or SR59320A is sufficient to restore GPAT4 targeting . Various proteomic , biochemical , and cell biological studies showing that components of the Arf1/COPI machinery are present on LDs ( Nakamura et al . , 2005; Bartz et al . , 2007; Guo et al . , 2008; Ellong et al . , 2011; Bouvet et al . , 2013 ) are also consistent with this model . Calculations based on the size of the targeted LDs and the formed nano-LDs suggest that only a few nano-LD budding events are required to significantly increase the surface tension of the donor LD ( Thiam et al . , 2013a ) . Thus , Arf1/COPI activity that results in the budding of nano-LDs will cause a significant change in the surface properties of existing LDs , and these changes are required to enable interactions of the LD monolayer surface with bilayer membranes . Specifically , we posit that the increase in the surface tension of LDs allows for the formation of bridges with the ER , whereas the densely packed phospholipid shell on LDs with low surface tension , where Arf1/COPI have not acted , are refractive to forming a bridge with the ER . In an alternative and possibly complementary model , Arf1/COPI might also function to maintain ER lipid composition or structure in a manner that allows for the formation of bridges with LDs . In other systems inhibition of the Arf1 guanine-nucleotide exchange factor led to collapse of the Golgi apparatus into the ER and ectopic cleavage and activation of the transcription factor SREBP ( Walker et al . , 2011 ) . In Drosophila , SREBP up-regulates phospholipid synthesis ( Dobrosotskaya et al . , 2002 ) , which could indirectly affect LD surface properties ( Krahmer et al . , 2011 ) . However , in our experimental system , we did not detect increased SREBP cleavage , up-regulation of the SREBP target genes ( such as CCT1 , acetyl-CoA synthase , acetyl-CoA carboxylase and fatty acid synthase ) or changes in cellular PC or PE levels after Arf1 depletion ( Figure 5—figure supplement 1 and MJO and TCW , unpublished observations ) . Once ER-LD bridges are established , GPAT4 , and presumably other enzymes ( e . g . , AGPAT3 , DGAT2 , or ATGL/brummer ) rapidly migrate to LDs . The time course of enzyme relocalization , in our experiments triggered at some point during oleate loading or after adding back COPI by cell–cell fusion , suggests that , once LD-to-ER bridges are established , targeting is limited by diffusion across the bridges . It is unclear how cargo that migrates from the ER to LDs is selected . Intriguingly , the Arf1/COPI mechanism appears to operate specifically for proteins that are embedded in the membrane , such as GPAT4 and ATGL , which behaves similarly to GPAT4 as an integral membrane protein ( Soni et al . , 2009 , and FW , MJO , RVF , and TCW , unpublished observations ) . It is also unclear why these LD-targeted proteins accumulate on LD surfaces . Accumulation could be mediated by partitioning into the oil phase , but the mechanism providing energy for the reaction is not yet known . Our findings provide a number of new questions for investigation . It is unknown if the Arf1/COPI machinery is constitutively active stochastically on some LDs or if is regulated . Interestingly , data from in vitro budding reactions from oil-water interfaces indicate that Arf1/COPI can act only on membranes sufficiently covered by phospholipids ( Thiam et al . , 2013a ) . Therefore , Arf1/COPI might be part of a system that detects LDs that are sufficiently coated by PC ( i . e . , have reached a sufficiently low surface tension ) and thus are suitable for further expansion . It is also unclear how Arf1/COPI-mediated protein targeting is affected by lipolysis . Generation of surface-active lipids during lipolysis , such as fatty acids or diacylglycerol , might increase LD surface tension and subsequently augment the triggering of ER-LD bridge formation , thereby allowing more lipases to migrate to LDs . Also unclear is how the specificity of membrane bridge formation of LDs to the ER is controlled . Finally , it will be of interest to determine the fate of the nano-LDs formed by Arf1/COPI actions . Nano-LDs are similar in size to typical COPI vesicles . However , in contrast to vesicles , they are made up of a small oil core that is likely coated with a monolayer of phospholipids and may contain specific proteins . The model emerging from our studies highlights how cells solved a fundamental problem—how to deal with LDs , which are essentially emulsified oils in the aqueous cytosol . Through the actions of the Arf1/COPI machinery , the surface properties of LDs can be altered such that proteins are able to access them . Among all coat complexes known to function in vesicular trafficking , the Arf1/COPI system has unique properties that make it ideally suited to function in this process . All other vesicular coat complexes require exchange factors that contain trans-membrane spanning protein segments , which are unlikely to be found on LDs . Arf1/COPI does not require such a factor . By this unique mechanism , cells can alter the surface properties of LD emulsions and enable them to interact with membranes , so that specific enzymes can gain access to LDs and facilitate dynamic changes in lipid storage or utilization . Our findings additionally provide evidence for a previously unrecognized cellular mechanism by which Arf1/COPI proteins can control protein trafficking . Rabbit polyclonal antibodies used: anti-GPAT4 ( Wilfling et al . , 2013 ) , anti-CCT1 ( Wilfling et al . , 2013 ) , anti-GBF1 ( BD Biosciences , San Jose , CA ) , anti-KDEL-receptor ( KDELR; gift from Dr JE Rothman; Yale University ) , anti-βCOP ( gift from Dr JE Rothman; Yale University ) , anti-perilipin3 ( TIP47; Novus Biologicals , Littleton , CO ) , anti-αCOP ( Abcam , Cambridge , MA ) , anti-GRP78/BiP ( ET-21 ) ( Sigma–Aldrich , St . Louis , MO ) and anti-garz ( Wang et al . , 2012 ) ( gift from Dr A Paululat; University of Osnabrück ) . Mouse monoclonal antibodies used: anti-GM130 ( BD Biosciences ) , anti-tubulin ( Sigma–Aldrich ) , anti-β’COP ( gift from Dr JE Rothman; Yale University ) , and anti-clathrin heavy chain ( x22 ) ( Thermo Scientific , Waltham , MA ) antibody . The following secondary antibodies were used: Alexa Fluor 568 goat anti-rabbit ( Invitrogen , Grand Island , NY ) , Alexa Fluor 488 goat anti-mouse ( Invitrogen ) , ATTO 647N ( STED ) goat anti-rabbit ( Active Motif , Carlsbad , CA ) , and goat anti-mouse STAR470SX ( Abberior , Göttingen , Germany ) . Full-length cDNA encoding CG5295 ( brummer ) and CG10374 ( Lsd1 ) were obtained from the DGRC ( https://dgrc . cgb . indiana . edu/ ) and subcloned into the pENTR/SD/DTOPO vector ( Invitrogen ) and indicated destination expression vectors ( actin promoter ) . The destination vectors used in this study are part of the Drosophila Gateway Vector Collection and are available from the DGRC ( https://dgrc . cgb . indiana . edu/ ) . WT Drosophila S2 or stably transfected cells ( pAGW-brummer or pACherryW-Lsd1 ) were cultured , treated with oleate , transfected and depleted by RNAi as described ( Krahmer et al . , 2011 ) . Cells were analyzed 4 days after RNAi treatment . Table 1 contains a list of primers to generate dsRNAs for RNAi . A segment of pBluescript backbone was used as the template for control RNAi . Expression of the ss-HRP construct was induced and the secretion assay was performed as previously described ( Bard et al . , 2006 ) . If not otherwise indicated , cells were treated after RNAi treatment with 1 mM oleate for 8 hr . Exogenous lipids ( PC , or cholesterol , or PC/cholesterol ) were added to Drosophila S2 cells at the second day of the RNAi treatment . The final concentration of these lipids in the growth medium was 5 mM . On the fourth day , medium was replaced by fresh medium containing 1 mM oleate and LD formation was induced for 8 hr before cells were fixed . The artificial lipid SR59230A was added to RNAi treated cell during the last hour of oleate treatment to a final concentration of 100 µM . NRK cells were cultured in DMEM with 10% FBS and antibiotics ( 100 units of penicillin and 100 µg of streptomycin per ml ) . Cells were split onto glass bottom plates and incubated in the culture media the day before imaging . LDs were induced by treatment with 0 . 5 mM oleate for 2 hr before fixation and imaging . 10 . 7554/eLife . 01607 . 017Table 1 . Sequences of primers used for RNAi experimentsDOI: http://dx . doi . org/10 . 7554/eLife . 01607 . 017GeneGene IDForwardReversegarzCG8487TTGCACAAACTTTGATTCCTGCATATCGGCGCACTATAATCArf79FCG8385TAGCGATTAGCGTTCTTCACTGCCAAATGCAATGAACGαCOPCG7961AGGAAGCTAAGCTTGTCAAAGGACGAGTCTGGAGTGTTTβCOPCG6223CCAGTCAGTTGGGTGACCTTCCTAGCAAGCCCATAACCAAβ’COPCG6699ATCTTGCTTCCCACAACGTCCCGAAGGACAACAACACCTTγCOPCG1528ATTACGTTCACAGCACGCAGCAGAGGAGGGCTATGACGACζCOPCG3948CCGTCGCAGATCTCGTCGCATCCTGGCCAAGTACTAεCOPCG9543AGGTGCCAGATGTTGGTCTCCCAACTCGGTGCTATTCGATδCOPCG14813AAGCTGTCTGCGCCATAAATTCCAGTGGCACATTCCAATACCT1CG1049ACATCTATGCTCCT1CTCAAGGCCTCTGCAGACTCTGGTAACTGCpBluescriptAATTCGATATCAAGCTTATCGATTAAATTGTAAGCGTTAATATTTTG Fluorescently labeled Arf1 was generated using an Arf1-variant in which the single cysteine residue of Arf1 was replaced with serine , and the C-terminal lysine was replaced with cysteine , yielding Arf1-C159S-K181C . Published work has demonstrated that exchanging the C-terminal lysine of the small GTPase with a Cys- residue , and subsequent fluorescent labeling ( using thiol-reactive dyes on Cys181 ) , does not inhibit Arf1-function ( Beck et al . , 2008; Manneville et al . , 2008 ) . In short , human Arf1-C159S-K181C and yeast N-myristoyltransferase were coexpressed in Escherichia coli supplied with BSA-loaded myristate . Cell lysates were subjected to 35% ammonium sulfate , and the precipitate , enriched in myristoylated Arf1 , was further purified by DEAE-ion exchange . Eluted fractions of interest were concentrated in spin-column filters with a 10-kD cutoff ( Millipore ) , and fluorescently labeled using Cy3-maleimide ( GE Healtcare ) according to the manufacturer’s protocol . To remove excess dye , samples were purified by gel filtration using a Superdex 75 column ( GE Healthcare ) . Recombinant coatomer protein was expressed and purified , as described in Sahlmuller et al . , ( 2011 ) . In short , Sf9 insect cells were infected with baculovirus encoding for heptameric coatomer . Coatomer complexes were isolated from the soluble protein fraction by nickel-affinity purification , concentrated in spin-column filters with a 250-kD cutoff ( Millipore ) , and fluorescently labeled using Alexa-Fluor-647-NHS ( Molecular Probes ) according to the manufacturer’s protocol . Excess imidazole and dye was removed by gel filtration using a Superose 6 column ( GE Healthcare ) . Cells were treated with 1 mM oleate , stained with BODIPY , and subsequently imaged and measured as described ( Wilfling et al . , 2013 ) . Density plots were computed using R ( http://www . r-project . org/ ) . For live-cell imaging and immunostaining , cells were prepared and imaged as described ( Wilfling et al . , 2013 ) . The antibody dilution buffer used for immunostaining of perilipin3 in NRK cells did not contain detergent . The permeabilization buffer used for immunostaining of CCT1 in Drosophila S2 cells had a final concentration of 0 . 1% NP-40 . Also , the buffer for first and secondary antibody dilution was detergent free . LDs were stained with 1 µg/ml BODIPY ( Invitrogen ) or LipidTOX ( Invitrogen ) or 10 mM of MDH ( Yang et al . , 2012b ) . STED microscopy ( Hell and Wichmann , 1994 ) was performed on a custom-built system featuring an 80 MHz mode-locked Ti:Sapphire laser ( Chameleon Ultra II , Coherent ) tuned to either 760 nm or 770 nm as the depletion beam . The 140 fs pulses output from this laser were stretched to several hundred picoseconds using a glass block and a 100 m polarization-maintaining optical fiber ( Thorlabs ) to prevent multiphoton excitation of the fluorophores . A spatial light modulator in the depletion beam path allowed phase modulation for generating a toroidal depletion focus in the sample and for correction of system induced optical aberrations ( Gould et al . , 2012 ) . For fluorescence excitation , 510 nm and 640 nm pulsed diode lasers ( PicoQuant ) were electronically synchronized to the depletion beam and an electronic delay ( Colby Instruments ) allowed adjustment of the relative arrival time of the laser pulses at the sample . Excitation and STED beams were combined using dichroic mirrors and focused into the sample through a 100×/1 . 4NA oil immersion objective lens ( UPLSAPO 100XO/PSF , Olympus ) . Imaging was preformed via beam scanning . A 16 kHz resonant scanner and a galvanometer mirror ( EOPC ) were placed in the beam path and imaged into the pupil plane of the objective lens to scan the beams through the sample . Fluorescence from the sample was collected by the objective lens , de-scanned by the scan mirrors , separated from laser light using dichroic mirrors , bandpass filtered ( FF01-685/40 for ATTO647N or FF01-593/46 for STAR470SX; both from Semrock ) , and focused onto 105 μm core diameter ( ATTO647N: ∼0 . 7 Airy units; STAR470SX: ∼0 . 8 Airy units ) multimode fibers ( Thorlabs ) connected to single-photon counting avalanche photodiodes ( APD; ARQ-13-FC , Perkin Elmer ) . APD counts were acquired using a FPGA-based data collection board ( PCIe-7852R , National Instruments ) and custom acquisition software programmed in LabView ( National Instruments ) . Recorded pixel values were linearized ( on the DAQ card ) to account for the sinusoidal velocity profile of the resonant mirror and normalized according to the pixel dwell times such that the center pixel was divided by unity . Dual-color imaging of ATTO647N and STAR470SX were performed using sequential frame acquistions similar to previously published reports using a long Stoke’s shift fluorophore as the second color channel ( Schmidt et al . , 2008 ) . Laser powers ( measured at the objective back aperture ) were ∼16 μW of 510 nm excitation light and ∼130 mW of 760 nm STED light for STAR470SX and ∼17 μW of 640 nm excitation light and ∼130 mW of 770 nm STED light for ATTO647N . Images were acquired with a 20 nm pixel size in a 1024 by 1024 image format with 500 accumulations per line , resulting in a frame rate of 0 . 032 Hz . To assess whether the overlapping signals of βCOP , clathrin , KDELR ( in NRK cells ) and αCOP , garz ( for S2 cell ) with BODIPY was erratic a Matlab script was written . The population of the immunostained foci was denoted A and BODIPY stained LDs were denoted LD . For each colocalization experiment , a minimum of 15 snapshots was taken . Each image was first analyzed to assess the frequency of colocalization between A and LD from the experiment ( 1 ) , from a random situation where A-type particles were randomly distributed ( 2 ) ; both situations were then compared ( 3 ) . 1 . The brightness contrast was adjusted for each channel of the picture using ImageJ . After applying a threshold binary images were generated for each channel . The total number of A-type particles , nA , and their corresponding radius , rA , were determined . For each LD particle , the distance of the first A-type neighbor was determined . Negative distances corresponded to overlapping of A and LD . The colocalized fraction of A-type particles with LD population was given by nexp/nA , where nexp was the number of colocalized A-type spots ( number of negative distances ) . 2 . The random distribution of A particles was based on an analytical model following a binomial distribution hypothesis . The choice of a binomial distribution model was adequate to assess overdispersion of A-type particles ( Rosner , 2011 ) . From the binary mask of the LD population , each LD of radius rLD was given a new radius rLD + rA . The probability of colocalization of A and LD can be formulated by the probability of having an A-type dot colocalizing to a LD of the new defined radius . In the total field occupied by the cells ( areaf denoting for the area of the field ) , the total area fraction occupied by LD is given by:s= ∑LDπ ( rLD+rA ) 2areaf The probability to colocalize n A-type dots out of nA is:P ( n ) = ⊂nAn sn ( 1−s ) nA−n Colocalization of ns ( ns = snA ) from the nA particles has the highest occurrence ( P ( n ) < P ( ns ) ) . Therefore the most likely situation from a random distribution was the colocalization of ns A-type particles with LD . Likewise we observed that a random simulation based on a normal distribution results in similar values for ns ( data not shown ) . 3 . If ns >> nexp , particles A are excluded from LDs; if ns << nexp , they are enriched on LDs . Expression levels were measured by quantitative Real-Time PCR . Total RNA was prepared with the RNeasy Mini Kit ( Qiagen ) ; 1 μg was used for first-stand cDNA synthesis with the iscript cDNA synthesis kit ( BioRad ) . Real-time quantitative PCR was performed on a LightCycler 480II ( BioRad ) using the Power SYBR green mix ( Applied Biosystems ) . Pimers used are listed in Table 2 . 10 . 7554/eLife . 01607 . 018Table 2 . Sequences of primers used for RT-PCRDOI: http://dx . doi . org/10 . 7554/eLife . 01607 . 018GeneGene IDForwardReverseGAPDH1CG12055TTGTGGATCTTACCGTCCGACCTTAGCCTTGATTTCGTCArf79FCG8385TTACAGTGTGGGATGTGGGGAAGATAAGACCTTGTGTATTCTGGβCOPCG6223GACTTCTGCAATATCAAGGCCGGTTTCGTAAACAATATTGCCGCCT1CG1049GATACGGAGTGCGTCAAATTCATCGGACAGAGTCCA Purfication of lipid droplets was done as previously reported ( Wilfling et al . , 2013 ) . Lipids were extracted ( Folch et al . , 1957 ) , separated on silica TLC plates ( Merck ) with chloroform/methanol/acetic acid/formic acid/water ( vol/vol ) ( 70:30:12:4:1 ) for phospholipids or n-heptane/isopropyl ether/acetic acid ( 60:40:4 ) for neutral lipids , and detected by Hanessian’s Purified LDs incubated with ARF/COPI at various conditions were absorbed to continuous carbon-coated grids ( glow discharged ) at room temperature for 5 min , rinsed briefly with HKM buffer ( 25 mM HEPES-KOH at pH 7 . 4 , 100 mM KCl , 10 mM MgCl2 ) , and stained with 1% uranyl formate for 20 s . Negatively stained samples were imaged under low-dose conditions in an FEI Tecnai12 microscope ( 120 kV ) . Micrographs were collected at 26 , 000 × magnification using Gatan 4K × 4K CCD camera , giving a pixel size of 4 . 5 Å . The diameters of nano LDs were manually measured on digital micrographs . The surface tension of different lipids or lipid mixtures was measured using a drop weight method . HKM buffer containing different concentrations of phospholipids and/or cholesterol was formed in a TG oil phase . Buffer drops were slowly formed in the oil ( at a flow rate of 20 µl/hr ) to allow dynamic interfacial equilibrium . At a critical size the drop detaches . For each concentration , videos of this process were taken using a 1394 Unibrain camera . From the inner diameter d of the injection tube ( d = 250 µm ) , the surface tension is determined by mg/ ( πdf ) where f is a Wilkinson geometric parameter correction that depends on the ratio between d and the radius of the detached drop and g is the gravity constant . The mass m of the drop was calculated according to m = vΔρ ( v is the volume of the drop and Δρ is the volume mass difference between oil and buffer phases ) . 2 . 5 mg of DOPE and 2 . 5 mg of DOPC ( Avanti Polar Lipids ) were solubilized in 250 mg TG ( Sigma–Aldrich ) by sonication . Lipids were then mixed with buffer ( 25 mM HEPES-KOH at pH 7 . 4 , 100 mM KCl , 10 mM MgCl2 ) in a ratio of 1/16 ( oil/buffer ) by vortexing and sonication for 5 min using a Branson 3510 sonicator water bath . The emulsion was added to the indicated amounts of cholesterol and sonicated for 2 min . The optical density of the emulsion was monitored over a time course of 2 hr in 1-min intervals by a TECAN infinite M200 . Drosophila S2 cells were co-transfected with pAW-VSVG and pAW-cherry . After 24 hr cells were mixed 1:1 with a stable transfected cell line of GFP-GPAT4 depleted for β-COP for 4 days and treated for 8h with 1 mM oleate . The cell mixture was prepared for live cell imaging as described ( Wilfling et al . , 2013 ) . Fusion of cells was initiated by addition of a low pH buffer ( 10 mM Na2HPO4 , 10 mM NaH2PO4 , 150 mM NaCl , 10 mM MES , 10 mM HEPES , pH 5 ) for 30 s . After incubation with the fusion buffer cells were immediately shifted to regular growth medium . The microfluidics device was fabricated by well-established soft lithography techniques . A wafer mold was made by lithography with a negative resin ( SU8-2035 ) . The device was made of a poly-dimethylsiloxane polymer , used to replicate the pattern on the mold and stuck on a glass cover slip . The height of the device is 58 ± 5 µm . A buffer and oil stream was generated using a syringe pump . By flow focusing , defined buffer drops were generated in the oil stream . These buffer micro-reactors contained GFP-GPAT4 microsomes . The oil used was a mixture of TG with phospholipids and/or cholesterol ( PC/PE each 0 . 25% ( wt/wt ) ; PC/PE/cholesterol 1/1/4 with 2% ( wt/wt ) cholesterol; cholesterol 0 . 5% ( wt/wt ) ; indicated lipid concentrations are compared to TG ) . To ensure the same frequency of interaction of the microsomes with the monolayer interface , the flow rate of the buffer and oil stream ( 150 and 30 µl/hr ) was kept constant for all experiments .	Just like the body contains organs that perform different jobs , the cells within the body contain organelles that carry out different tasks . The endoplasmic reticulum , for example , makes proteins that are sent to other organelles or to destinations outside the cell . Each organelle is typically sealed within a membrane made from a double layer of phospholipids—molecules that have a phosphate ‘head’ group at one end , and two fatty acid ‘tails’ at the other . Proteins are shuttled between the organelles inside membrane-bound packages called vesicles . There is , however , an exception to this rule . Lipid droplets are organelles that store fats and oils inside a single layer of phospholipids . This layer can include enzymes that break down the contents of the droplet , or make new fat molecules , depending on the needs of the cell and the organism . However , it is not clear how these enzymes get from the endoplasmic reticulum to the lipid droplet . Previous work had suggested that a protein complex called Arf1/COP—which is also involved in the movement of vesicles around the cell—might recruit the enzymes to the lipid droplets . However , none of the other proteins known to be involved in vesicle transport were needed to transport the enzymes to the droplets , which suggested that the Arf1/COPI complex was using a previously unknown mechanism to move the enzymes . Now Wilfling , Thiam et al . have shown that Arf1/COPI complexes trigger the establishment of membrane bridges between the endoplasmic reticulum and the droplets , which means that vesicles are not needed to get the enyzmes to the lipid droplets . It was also shown that the Arf1/COPI complexes could pinch off tiny droplets from full-size lipid droplets taken from living cells . Wilfling , Thiam et al . suggest that this ‘budding’ process changes the composition of the phospholipid layer around the larger droplet in a way that allows it to interact directly with the membrane of the endoplasmic reticulum . By providing new insights into the trafficking of proteins between organelles , the work of Wilfling , Thiam et al . reveals mechanisms that govern the composition of lipid droplets . In the future , these pathways could be manipulated to treat conditions that result from excessive storage of fat , such as obesity or cardiovascular diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2014	Arf1/COPI machinery acts directly on lipid droplets and enables their connection to the ER for protein targeting
Eukaryotes have evolved various quality control mechanisms to promote proteostasis in the endoplasmic reticulum ( ER ) . Selective removal of certain ER domains via autophagy ( termed as ER-phagy ) has emerged as a major quality control mechanism . However , the degree to which ER-phagy is employed by other branches of ER-quality control remains largely elusive . Here , we identify a cytosolic protein , C53 , that is specifically recruited to autophagosomes during ER-stress , in both plant and mammalian cells . C53 interacts with ATG8 via a distinct binding epitope , featuring a shuffled ATG8 interacting motif ( sAIM ) . C53 senses proteotoxic stress in the ER lumen by forming a tripartite receptor complex with the ER-associated ufmylation ligase UFL1 and its membrane adaptor DDRGK1 . The C53/UFL1/DDRGK1 receptor complex is activated by stalled ribosomes and induces the degradation of internal or passenger proteins in the ER . Consistently , the C53 receptor complex and ufmylation mutants are highly susceptible to ER stress . Thus , C53 forms an ancient quality control pathway that bridges selective autophagy with ribosome-associated quality control in the ER . Autophagy is an intracellular degradation process where eukaryotic cells remove harmful or unwanted cytoplasmic contents to maintain cellular homeostasis ( Dikic and Elazar , 2018; Klionsky and Deretic , 2011; Marshall and Vierstra , 2018 ) . Recent studies have shown that autophagy is highly selective ( Johansen and Lamark , 2020; Stolz et al . , 2014 ) and is mediated by receptors that recruit specific cargo , such as damaged organelles or protein aggregates . Autophagy receptors and their cargo are incorporated into the growing phagophore through interaction with ATG8 , a ubiquitin-like protein that is conjugated to the phagophore upon activation of autophagy ( Stolz et al . , 2014; Zaffagnini and Martens , 2016 ) . The phagophore grows and eventually forms a double-membrane vesicle termed the autophagosome . Autophagosomes then carry the autophagic cargo to lytic compartments for degradation and recycling . Selective autophagy receptors interact with ATG8 via conserved motifs called the ATG8 interacting motif ( AIM ) or LC3-interacting region ( LIR ) ( Birgisdottir et al . , 2013 ) . In contrast to mammals and yeast , cargo receptors that mediate organelle recycling remains mostly elusive in plants ( Stephani and Dagdas , 2020 ) . The endoplasmic reticulum ( ER ) is a highly dynamic heterogeneous cellular network that mediates folding and maturation of ~40% of the proteome ( Walter and Ron , 2011; Sun and Brodsky , 2019 ) . Proteins that pass through the ER include all secreted and plasma membrane proteins and majority of the organellar proteins . This implies , ER could handle up to a million client proteins in a cell every minute ( Karagöz et al . , 2019 ) . Unfortunately , the folding process is inherently error prone and misfolded proteins are toxic to the cell ( Sun and Brodsky , 2019; Karagöz et al . , 2019; Fregno and Molinari , 2019 ) . To maintain the proteostasis in the ER , eukaryotes have evolved dedicated quality control mechanisms that closely monitor , and if necessary , trigger the removal of terminally misfolded proteins . Degradation of the faulty proteins is mediated by proteasomal and vacuolar degradation pathways ( Fregno and Molinari , 2019 ) . One of the main vacuolar/lysosomal degradation processes is ER-phagy . It has emerged as a major quality control pathway , and defects in ER-phagy is linked to various diseases ( Chino and Mizushima , 2020; Hübner and Dikic , 2020; Stolz and Grumati , 2019; Wilkinson , 2020 ) . ER-phagy involves cargo receptors that mediate removal of certain regions of the ER via autophagy . Several ER-resident ER-phagy receptors have been identified . These include Fam134B , RTN3L , ATL3 , Sec62 , CCPG1 , and TEX264 in mammals and ATG39 and ATG40 in yeast ( Khaminets et al . , 2015; Grumati et al . , 2017; Chen et al . , 2019; Fumagalli et al . , 2016; Smith et al . , 2018; An et al . , 2019; Chino et al . , 2019; Mochida et al . , 2015 ) . A recent study showed reticulon proteins could also function as ER-phagy receptors in plants ( Zhang et al . , 2020 ) . In addition , CALCOCO1 and Epr1 have been recently identified as cytosolic ER-phagy receptors that associate with ER-resident VAP proteins to recycle ER tubules ( Nthiga et al . , 2020; Zhao et al . , 2020 ) . Altogether , these receptors are activated during starvation or stress conditions and work together to remodel the highly heterogeneous and dynamic ER network to maintain proteostasis . Despite the emerging links , how ER-phagy cross-talks with the core ER quality control pathways remains largely unknown ( Chino and Mizushima , 2020; Dikic , 2018 ) . Here , using a peptide-competition coupled affinity proteomics screen , we identified a highly conserved cytosolic protein , C53 , that is specifically recruited into autophagosomes during ER stress . C53 interacts with plant and mammalian ATG8 isoforms via a non-canonical ATG8 interacting motif ( AIM ) , termed shuffled AIM ( sAIM ) . C53 is recruited to the ER by forming a ternary receptor complex with the UFL1 , the E3 ligase that mediates ufmylation , and its ER membrane adaptor DDRGK1 ( Gerakis et al . , 2019 ) . C53-mediated autophagy is activated upon ribosome stalling during co-translational protein translocation and results in the degradation of specific ER proteins . To identify specific cargo receptors that mediate selective removal of ER compartments during proteotoxic stress , we performed an immunoprecipitation coupled to mass spectrometry ( IP-MS ) screen to identify AIM-dependent ATG8 interactions triggered by ER stress . We hypothesized that a synthetic AIM peptide that has higher affinity for ATG8 can outcompete , and thus reveal , AIM-dependent ATG8 interactors . To identify this synthetic peptide , we performed a peptide array analysis that revealed the AIM wt peptide ( Figure 1—figure supplement 1A , B; Supplementary file 1 ) . Using isothermal titration calorimetry ( ITC ) , we showed that the AIM wt binds ATG8 with nanomolar affinity ( KD = ~ 700 nM ) , in contrast to the AIM mutant peptide ( AIM mut ) , which does not show any binding ( Figure 1—figure supplement 1C–D ) or the low micromolar-range affinities measured for most cargo receptors ( Zaffagnini and Martens , 2016 ) . As plants have an expanded set of ATG8 proteins , we first tested if any of the ATG8 isoforms specifically responded to ER stress induced by tunicamycin ( Kellner et al . , 2017 ) . Tunicamycin inhibits glycosylation and leads to proteotoxic stress at the ER ( Bernales et al . , 2006 ) . Quantification of ATG8 puncta in transgenic seedlings expressing GFP-ATG8A-I revealed that tunicamycin treatment significantly induced all nine ATG8 isoforms ( Figure 1—figure supplement 2 ) . Since all ATG8 isoforms were induced and ATG8E has a broad expression pattern , we chose ATG8E , and performed peptide competition coupled IP-MS analysis ( See Materials and methods for detailed description ) . In addition to well-known AIM dependent ATG8 interactors such as ATG4 ( Autophagy related gene 4 ) and NBR1 ( Neighbour of BRCA1 ) ( Wild et al . , 2014 ) , our analyses revealed that the highly conserved cytosolic protein C53 ( aliases: CDK5RAP3 , LZAP , IC53 , HSF-27 ) is an AIM-dependent ATG8 interactor ( Figure 1A , Supplementary file 2 , Figure 1—figure supplement 3 ) . To confirm our IP-MS results , we performed in vitro pull-down experiments . Arabidopsis thaliana ( At ) C53 specifically interacted with GST-ATG8A , and this interaction was outcompeted with the AIM wt , but not AIM mut peptide . Consistently , ATG8 receptor accommodating site mutations ( LDS − LIR Docking Site ) prevented C53 binding ( Figure 1B ) . We extended our analysis to all Arabidopsis ATG8 isoforms and showed that AtC53 interacts with eight of nine Arabidopsis isoforms ( Figure 1C ) . To probe for evolutionary conservation of C53-ATG8 interaction , we tested the orthologous proteins from the basal land plant Marchantia polymorpha ( Mp ) and showed that MpC53 interacts with one of two Marchantia ATG8 isoforms ( Figure 1—figure supplement 4 ) . As C53 is highly conserved in multicellular eukaryotes and has not been characterized as an ATG8 interactor in mammals , we tested whether human C53 ( HsC53 ) interacts with human ATG8 isoforms ( LC3A-C , GABARAP , -L1 , -L2 ) . HsC53 interacted with GABARAP and GABARAPL1 in an AIM-dependent manner via the LIR docking site , similar to plant C53 homologs ( Figure 1D , E ) . Of note , we have also tested other modes of ATG8 binding such as the recently identified UDS or the hydrophobic pocket accommodating the atypical LIR motif found in ufmylation enzyme UBA5 ( Marshall et al . , 2019; Huber et al . , 2019 ) . The UDS mutation rendered ATG8A unstable ( Figure 1B ) , whereas mutating the atypical LIR accommodating site did not affect C53 binding ( Figure 1F ) . Altogether , these data suggest that C53-ATG8 interaction is conserved across kingdoms and mediated via the LIR Docking Site . In order to examine the in vivo link between C53 and ATG8 function , we generated transgenic AtC53-mCherry Arabidopsis lines and measured autophagic flux during ER stress . Without stress , AtC53 displayed a diffuse pattern in the cell , partially overlapping with the ER marker GFP-HDEL ( Figure 2—figure supplement 1A ) . Similarly , upon carbon starvation ( -C , Figure 2A ) , which is commonly used to activate bulk autophagy , AtC53-mCherry remained mostly diffuse ( Marshall and Vierstra , 2018 ) . However , tunicamycin ( Tm ) treatment led to a rapid increase in AtC53 puncta as observed in both native promoter driven and ubiquitin 10 promoter driven transgenic lines . The C53 puncta did not colocalize with HDEL-GFP puncta formed during ER stress , suggesting C53 puncta are highly specific ( Figure 2—figure supplement 1A , B ) . The number of puncta was further increased upon concanamycin A ( ConA ) treatment that inhibits vacuolar degradation , suggesting that AtC53 puncta are destined for vacuoles ( Figure 2A ) . The AtC53 puncta disappeared when AtC53-mCherry lines were crossed into core autophagy mutants atg5 and atg2 , confirming that formation of AtC53 puncta is dependent on macroautophagy ( Figure 2A ) . Consistent with this , other ER-stressors such as phosphate starvation , cyclopiazonic acid ( CPA ) , and dithiothreitol ( DTT ) treatments also induced AtC53 puncta ( Figure 2—figure supplement 1C; Fumagalli et al . , 2016; Smith et al . , 2018; Naumann et al . , 2019 ) . The AtC53 puncta co-localized with GFP-ATG8A and GFP-ATG11 , indicating that they are autophagosomes ( Figure 2B , Figure 2—figure supplement 2A ) . Moreover , as recently shown for other selective autophagy receptors , AtC53 and HsC53 directly interacted with the mammalian ATG11 homolog FIP200 ( PTK2/FAK family-interacting protein of 200 kDa ) ( Figure 2—figure supplement 2B; Lahiri and Klionsky , 2018; Turco et al . , 2019; Ravenhill et al . , 2019; Vargas et al . , 2019 ) . Ultrastructural analysis using immunogold labelling showed that C53 is associated with ER under non-stress conditions , consistent with previous findings showing C53 associates with ER membrane proteins ( Yang et al . , 2019 ) . Electron micrographs also showed that AtC53 is recruited to autophagosomes during ER stress , consistent with our live cell imaging results ( Figure 2C , Figure 2—figure supplement 3 ) . Similar to plant proteins , transfected HsC53-GFP co-localized with mCherry-GABARAP upon tunicamycin treatment in HeLa cells . The number of HsC53 puncta increased upon bafilomycin ( BAF ) treatment , which inhibits lysosomal degradation; suggesting that HsC53 puncta eventually fuse with lysosomes ( Figure 2D ) . To support our imaging-based autophagic flux assays , we also performed western blot based autophagic flux analyses , using antibodies raised against AtC53 and HsC53 . These autophagic flux assays further demonstrated ER-stress-specific autophagic degradation of AtC53 and HsC53 ( Figure 3 ) . Having validated C53 as an autophagy substrate , we next sought to identify its ATG8-interacting motif ( AIM ) . For this purpose , we reconstituted the binary complex in vitro and determined the stoichiometry of the C53-ATG8 interaction by native mass spectrometry ( nMS ) . Both HsC53 and AtC53 formed 1:1 and 1:2 complexes with GABARAP and ATG8A , respectively; pointing to the existence of multiple binding epitopes ( Figure 4A ) . Initially , we tested all predicted canonical AIMs in AtC53 . However , even the pentuple AIM mutant bound at similar levels to ATG8 , suggesting non-canonical AIMs mediate the C53-ATG8 interaction ( Figure 4—figure supplement 1 ) . To narrow down the ATG8-binding region of C53 , we performed in vitro pull downs using truncated proteins . C53 contains an intrinsically disordered region ( IDR ) that bridges two α-helical domains located at the N and C termini . In vitro pull downs revealed that the IDR is necessary and sufficient to mediate ATG8 binding , as also confirmed with ITC and nMS experiments ( Figure 4B–D , Figure 4—figure supplement 2 ) . Multiple sequence alignment of the C53-IDR uncovered three highly conserved sites with the consensus sequence ‘IDWG’ , representing a shuffled version of the canonical AIM sequence ( W/F/Y-X-X-L/I/V ) ( Figure 4C , Figure 4—figure supplement 3 ) . Mutational analysis of the three shuffled AIM sites in HsC53 and AtC53 revealed the importance of the sAIM epitopes for binding to GABARAP and ATG8 , respectively; though in AtC53 , an additional canonical AIM had to be mutated to fully abrogate the binding ( Figure 5A ) . ITC experiments with the purified IDRs , as well as nMS and surface plasmon resonance ( SPR ) experiments with full-length proteins , also supported sAIM-mediated ATG8-binding for both HsC53 and AtC53 ( Figure 5B , C , Figure 5—figure supplement 1 ) . Circular dichroism spectroscopy showed that sAIM mutants had very similar secondary structures to the wild-type proteins , suggesting that lack of ATG8 binding is not due to misfolding ( Figure 5—figure supplement 1C ) . To verify our in vitro results in vivo , we analyzed the subcellular distribution of sAIM mutants in transgenic Arabidopsis lines and transfected HeLa cells . Confocal microscopy analyses showed that C53sAIM mutants were not recruited into autophagosomes and had diffuse localization patterns upon ER stress induction ( Figure 5B , C ) . Altogether these biochemical and cell biological analyses show that C53 is recruited to the autophagosomes by interacting with ATG8 via the non-canonical sAIMs . Next , we looked for client proteins subject to C53-mediated autophagy . Quantitative proteomics analyses of wild type and AtC53 mutant lines revealed that AtC53 mediates degradation of ER resident proteins as well as proteins passaging the ER to the cell wall , apoplast , and lipid droplets ( Figure 6 , Supplementary file 3 , 4 ) . These data are consistent with a recent study , showing that ER-resident proteins accumulate in a conditional mutant of mouse C53 ( Yang et al . , 2019 ) . Since C53 is a cytosolic protein , we then explored how it senses proteotoxic stress in the ER lumen , considering four likely scenarios: C53 may collaborate with ( i ) a sensor of the unfolded protein response ( UPR ) ( Karagöz et al . , 2019 ) or ( ii ) a component of the ER-associated degradation pathway ( ERAD ) ( Sun and Brodsky , 2019 ) . Alternatively , it may sense clogged translocons caused by ( iii ) ribosome stalling triggered during co-translational protein translocation ( Wang et al . , 2020 ) or ( iv ) aberrant signal recognition particle ( SRP ) independent post-translational protein translocation events ( Ast et al . , 2016; Figure 7A ) . In plants , there are two major UPR branches: the Ire1 pathway and bZIP17/28 pathway ( Pastor-Cantizano et al . , 2020 ) . To test the connection with the UPR system , we performed autophagic flux assays . AtC53 flux was already higher than wild type in Arabidopsis UPR sensor mutants ire1a/b and bzip17/28 , consistent with elevated ER stress levels in these mutants ( Figure 7B , C; Koizumi et al . , 2001; Kim et al . , 2018 ) . Furthermore , C53 puncta induced by tunicamycin treatment did not colocalize with Ire1b-YFP oligomers ( Figure 7D ) . Finally , inhibition of Ire1 activity in HeLa cells using chemical inhibitors 4μ8c or KIRA6 also increased HsC53 puncta ( Figure 7E ) . Together these data indicate that recruitment of C53 to the autophagosomes does not depend on UPR sensors ( Maly and Papa , 2014 ) . Next , we performed colocalization analyses using model ERAD substrates . In transgenic plant lines expressing model ERAD substrates , the client proteins did not colocalize with AtC53 puncta ( Figure 7—figure supplement 1A; Shin et al . , 2018 ) . Likewise , the model mammalian ERAD substrates GFP-CFTRΔF508 ( ERAD-C ) , A1ATNHK-GFP ( ERAD-L ) , and INSIG1-GFP ( ERAD-M ) only partially colocalized with HsC53 puncta in HeLa cells ( Figure 7—figure supplement 1B ) , suggesting C53-mediated autophagy may cross-talk with the ERAD pathway ( Leto et al . , 2019 ) . Next , we tested the effect of clogged translocons on C53 function . Remarkably , HsC53 significantly colocalized with the ER-targeted poly-lysine construct ER-K20 that leads to ribosome stalling ( Wang et al . , 2020 ) , but not with an SRP-independent translocon clogger ( Ast et al . , 2016 ) , despite both leading to a blockage at the Sec61 translocon ( Figure 7—figure supplement 2A ) . To further corroborate these findings , we tested a suite of translation inhibitors that block different steps in translation . Consistent with C53 responding to ribosome stalling , elongation inhibitors such as Anisomycin , Emetine or Puromycin induced AtC53 puncta , whereas initiation inhibitors Harringtonine or Hygromycin B did not have any effect . All inhibitors triggered mCherry-ATG8A puncta formation , suggesting the effect caused by elongation inhibitors is specific to C53 ( Figure 7—figure supplement 2B ) . HsC53 puncta were also induced by anisomycin treatment ( Figure 7—figure supplement 2C ) . Consistently , silencing of HsC53 using shRNA significantly reduced lysosomal trafficking of ER-K20 ( Figure 7—figure supplement 2D; Wamsley et al . , 2017 ) . These data suggest that C53 is activated upon ribosome stalling during co-translational protein translocation and mediates autophagic degradation of the stalled nascent chain . How is C53 recruited to the ER during ribosome stalling ? Notably , C53 has been previously linked to UFL1 , an E3 ligase that mediates ufmylation of stalled , ER-bound ribosomes , modifying ribosomal protein RPL26 ( Wang et al . , 2020; Walczak et al . , 2019 ) . To test if C53 is a part of a higher order receptor complex , we analysed the interaction of C53 with UFL1 and its ER membrane adaptor DDRGK1 ( Gerakis et al . , 2019 ) . We were able to observe both DDRGK1 and HsC53 in a single UFL1 pull down experiment ( Figure 8—figure supplement 1A ) . Further in vitro pull-down assays and yeast two hybrid analyses with the plant proteins showed that AtUFL1 directly interacts with AtC53 and AtDDRGK1 ( Figure 8A , Figure 8—figure supplement 1B ) . Consistently , AtC53 associates with DDRGK1 and UFL1 in in vivo coimmunoprecipitations and affinity purification mass spectrometry experiments ( Figure 8B , Supplementary file 5 ) . Furthermore , co-localization of UFL1 and DDRGK1 with AtC53 in punctate structures increases upon ER stress and these puncta are delivered to the vacuole ( Figure 8C , D , Figure 8—figure supplement 1C , D ) . Strikingly , AtC53 autophagic flux requires functional UFL1 and DDRGK1 , as the number of AtC53 puncta was significantly lower in ufl1 and ddrgk1 mutants ( Figure 8E , Figure 8—figure supplement 1E ) . Ultimately , autophagic flux assays using the ufmylation machinery mutants confirmed that AtC53 flux requires a functional ufmylation machinery ( Figure 8E , Figure 8—figure supplement 1F , G ) . Taken together , our data indicate that C53 is recruited to the ER by forming a heteromeric receptor complex with UFL1 and DDRGK1 . Since , DDRGK1 is an ER-membrane protein and physically linked to C53 , we analyzed the degradation of DDRGK1 during ER stress . Transgenic lines expressing DDRGK1-GFP in c53 and atg5 mutant revealed that recruitment of DDRGK1 from ER membrane to punctate structures during ER stress required both C53 and ATG5 ( Figure 8—figure supplement 2A ) . Furthermore , DDRGK1 puncta colocalized with mCherry-ATG8A in a C53-dependent manner ( Figure 8—figure supplement 2B ) . Western-blot-based autophagic flux assays further confirmed AtC53-dependent degradation of DDRGK1 ( Figure 8—figure supplement 2C ) . Interestingly , abundant ER proteins such as the Calnexin , BIP or SMT1 are not degraded by AtC53-dependent ER-phagy . Likewise , small and large ribosomal subunits are not degraded by AtC53 ( Figure 8—figure supplement 2C ) . These results are consistent with the C53 cargo clientele defined by quantitative proteomics , and point toward a highly selective , yet unknown cargo selection mechanism of C53 . We then explored how C53 is kept inactive under normal conditions . We hypothesized that the Ubiquitin like modifier UFM1 may safeguard the C53 receptor complex under normal conditions and keep ATG8 at bay . Upon ER stress , UFM1 would be transferred to RPL26 , exposing sAIMs on C53 . To test this , we first analyzed the UFM1-C53 interaction by in vitro pull-down assays and could show that AtC53 can interact with AtUFM1 ( Figure 9A ) . Furthermore , in vitro competition experiments revealed a competition between UFM1 and ATG8 for C53 binding ( Figure 9A ) . This result is reminiscent of the mutually exclusive UFM1 and GABARAP binding of UBA5 , the E1 enzyme in the ufmylation cascade ( Huber et al . , 2019 ) . We then performed in vivo co-immunoprecipitation experiments during ER stress . Consistent with our hypothesis and in vitro data , ER stress led to depletion of UFM1 and enhanced AtC53-ATG8 interaction ( Figure 9B , C , supplement 1 ) . Altogether , these data suggest that the two ubiquitin-like proteins UFM1 and ATG8 compete with each other for association with the C53 receptor complex ( Figure 9D ) . Finally , we examined if C53 is physiologically important for ER stress tolerance . First , we tested if C53 plays a general role in autophagy using carbon and nitrogen starvation assays . Carbon and nitrogen starvation are typically used to characterize defects in bulk autophagy responses ( Marshall and Vierstra , 2018 ) . In contrast to the core autophagy mutants atg5 and atg2 , CRISPR-generated Atc53 mutants did not show any phenotype under carbon or nitrogen starvation conditions ( Figure 10A , B ) . However , consistent with increased flux , Atc53 mutants were highly sensitive to phosphate starvation , which has been shown to trigger an ER stress response ( Naumann et al . , 2019; Figure 10C , Figure 10—figure supplement 1A ) . Similarly , in both root length and survival assays , Atc53 mutants were sensitive to tunicamycin treatment ( Figure 10D , Figure 10—figure supplement 1B , C ) . In addition , ufmylation machinery mutants ( Figure 10E ) , including ufl1 and ddrgk1 , were also sensitive to tunicamycin treatment but insensitive to carbon and nitrogen starvation ( Figure 10F , Figure 10—figure supplement 1D , E ) . Lastly , the Marchantia polymorpha c53 mutant was also sensitive to tunicamycin , suggesting C53 function is conserved across the plant kingdom ( Figure 10—figure supplement 1F ) . We then performed complementation assays using wild-type AtC53 and the AtC53sAIM mutant . AtC53 expressing lines behaved like wild-type plants in tunicamycin supplemented plates ( Figure 10G ) . However , AtC53sAIM mutant did not complement the tunicamycin sensitivity phenotype , and had significantly shorter roots ( Figure 10G , Figure 10—figure supplement 1G ) . Parallel to analyzing C53-mediated ER homeostasis in plants , stress tolerance assays in HeLa cells showed that silencing of HsC53 led to an induction of Bip3 chaperone protein levels ( Figure 10—figure supplement 1H ) , indicating increased ER stress . Complementation of Hsc53 silenced lines with HsC53-GFP dampened Bip3 expression ( Figure 10—figure supplement 1H ) . Altogether , these results demonstrate that C53 coordinated ER-phagy is crucial for ER stress tolerance in plant and mammalian cells . The endoplasmic reticulum is a highly heterogeneous and dynamic network that handles folding and maturation of up to a million proteins per minute in a cell ( Karagöz et al . , 2019 ) . It constantly tailors the proteome in a cell-type and physiological state dependent manner . Unfortunately , protein synthesis , folding , and maturation events are all error prone , and faulty proteins have to be efficiently recycled to prevent accumulation of toxic by-products . Since , selective autophagy is a highly efficient quality control pathway that could very quickly recycle large amounts of proteins and membranous compartments , it is not surprising to have various ER-phagy receptors focusing on re-shaping the ER during stress ( Chino and Mizushima , 2020; Wilkinson , 2020 ) . With C53 , eight ER-phagy receptors have now been identified in metazoans , working together to maintain ER homeostasis under changing cellular conditions . However , since most of ER-phagy pathways were studied during nutrient starvation , which supposedly activates bulk recycling mechanisms , selective cargo recruitment triggered upon quality control defects is still poorly understood . It is thus a major challenge to elucidate the coordination of different ER-phagy receptors and their cross-talk with the core ER-quality control pathways ( Chino and Mizushima , 2020 ) . Our findings reveal C53-mediated ER-phagy to be a central mechanism operating at the interface of key quality control pathways , controlling ER homeostasis across different kingdoms of life . Using various model systems including divergent model plant species and human cell lines , we show that C53 forms an ancient autophagy receptor complex that is closely connected to the ER quality control system via the ufmylation pathway . Unlike other ER-phagy receptors studied so far , C53 seem to be highly specific in resolving ribosome stalling triggered during SRP-dependent co-translational protein translocation . However , it remains to be shown how C53 recruit specific cargo into the autophagosomes . Interestingly , recent genome wide CRISPR screens identified ufmylation as a major regulator of ER-phagy , the ERAD pathway , and viral infection ( Leto et al . , 2019; Liang et al . , 2020; Kulsuptrakul et al . , 2019 ) . Using fluorescent reporter lines and genome wide CRISPRi screens , Liang et al . , showed that ufmylation plays a major role in regulating starvation induced ER-phagy . They showed that both DDRGK1 and UFL1 are critical for starvation-induced ER-phagy , whereas C53 mutants did not show any ER-phagy defects ( Liang et al . , 2020 ) . Our results using stable transgenic organisms show that C53-mediated autophagy is not activated by carbon or nitrogen starvation that are typically used to activate bulk autophagy ( Figure 2 and Figure 2—figure supplement 1 ) . C53 is activated by ER stress caused by phosphate depletion ( Naumann et al . , 2019 ) . Consistently , phenotyping experiments revealed that C53 and the ufmylation machinery mutants are asymptomatic during carbon or nitrogen starvation but are highly sensitive to ER-stress treatments ( Figure 10 , Figure 10—figure supplement 1 ) . Together , the two complementary studies indicate that the ufmylation machinery is tightly associated with ER-phagy in multicellular eukaryotes and plays a crucial role in ER stress tolerance . It should be noted that C53 and ufmylation proteins are essential for mammalian development ( Gerakis et al . , 2019 ) . Defects in C53 receptor complex have been associated with various diseases including liver cancer , pancreatitis , and cardiomyopathy ( Gerakis et al . , 2019 ) . Our results suggest C53 and ufmylation is also critical for stress tolerance in plants , but they are not essential for development; suggesting plants have evolved compensatory mechanisms during adaptation to sessile life . Future comparative studies could reveal these mechanisms and help us develop sustainable strategies for promoting ER proteostasis during stress in mammals and plants . Constructs for Arabidopsis thaliana and E . coli transformation were generated using the GreenGate ( GG ) cloning method ( Lampropoulos et al . , 2013 ) . Plasmids used are listed in materials section . The coding sequence of genes of interest were either ordered from Twist Biosciences or Genewiz or amplified from Col-0 or HeLa cDNA using the primers listed in the materials section . The internal BsaI sites were mutated by site-directed-mutagenesis without affecting the amino acid sequence . For Marchantia polymorpha Gateway Cloning ( Ishizaki et al . , 2015 ) was used to generate all constructs . For HeLa expression experiments , plasmids used are listed in the materials section . The constructs were made by conventional restriction enzyme-based cloning . The CRISPR/Cas9 constructs for mutating c53 , DDRGK1 and UFM1 in Arabidopsis thaliana were prepared according to the protocol described by Xing et al . , 2014 and Ma et al . , 2015 . The pHEE401E and pCBC-DT1T2 vectors for expressing two sgRNAs were provided by Youssef Belkhadir and Jixiang Kong , GMI Vienna . sgRNA target sites were chosen using the website http://crispr . hzau . edu . cn/CRISPR2/ . Each gene was targeted by two sgRNAs to remove a fragment of the gene . The CRISPR cassettes of each gene were generated by PCR amplification using pCBC‐DT1T2 as template with the primer pairs BsF/F0 and BsR/R0 , using adaptors containing the BsaI-restriction sites , respectively ( see materials section ) . The PCR products were digested with BsaI , ligated into the pHEE401E plasmid , and transformed into DH5α E . coli . Floral dipping proceeded as described previously ( Clough and Bent , 1998 ) . Genotyping primers P1 5′‐xxx‐3′ and P2 5′‐xxx‐3′ flanking each target site were used to select T1 plants that carried deletions . Sanger sequencing was performed to define the deletion . Through backcrossing with Col-0 plants and genotyping , Cas9-free plants were achieved . In Marchantia polymorpha , CRISPR/Cas9 constructs were generated by selecting two target sequences in c53 and ire1 . Synthetic oligonucleotides were annealed and inserted at the BsaI site of the entry vector pMpGE_En03 flanked by attL1 and attL2 sequences ( Sugano et al . , 2018 ) . The resultant cassettes were inserted to the destination vector pMpGE011 by LR Clonase II Enzyme Mix . The vectors were introduced into thalli of TAK1 via A . tumefaciens GV3101+pSoup , and the transformants were selected with 0 . 5 µM chlorsulfuron ( Kubota et al . , 2013 ) . Genomic DNA from transformants was amplified by PCR and sent for sequencing to verify mutations . All Arabidopsis thaliana lines used originate from the Columbia ( Col-0 ) ecotype background . Mutant lines used in this study are listed in the materials section . All transgenic plants were generated by the floral dipping method ( Clough and Bent , 1998 ) for which the plasmid constructs were prepared using the green gate cloning method ( Lampropoulos et al . , 2013 ) . Seeds were then spread on plates or liquid culture with half-strength MS media ( Murashige and Skoog salt + Gamborg B5 vitamin mixture ) with 1% sucrose , 0 . 5 g/L MES and 1% plant agar . pH was adjusted to 5 . 7 with NaOH . Seeds were imbibed for 4 days at 4°C in darkness . Plants were grown at 21°C at 60% humidity under LEDs with 50 µM/m2s and 12 hr:12 hr photoperiod . For the autophagy flux assay , TMT and in vivo immunoprecipitation , seedlings were grown in liquid culture under continuous light . Male Marchantia polymorpha accession Takaragaike-1 ( Tak-1 ) was maintained asexually and cultured through gemma using half-strength Gamborg’s B5 medium containing 1% agar under 50–60 mmol photons m−2s−1 continuous white light at 22° C unless otherwise defined ( Kubota et al . , 2013 ) . Carbon starvation: Seedlings were grown on half-strength MS media with 1% sucrose for 7 days . They were then transferred to media without sucrose , followed by wrapping the plates in aluminium foil and placing them under the same growth conditions as before for 9 days . Nitrogen starvation: Seedlings grew on half-strength MS media with 0 . 5% sucrose for 7 days . They were then transferred to media without nitrogen and put under the same growth conditions as before for 14 days . Seedlings were arranged in a similar fashion to Jia et al . , 2019 . Phosphate starvation: The method was previously described by Naumann et al . , 2019 . Seeds were surface-sterilized and germinated 5 days on +Pi medium prior to transfer to 1% ( w/v ) Phyto-Agar ( Duchefa ) containing 2 . 5 mM KH2PO4 , pH 5 . 6 ( high Pi or +Pi medium ) or no Pi supplement ( low Pi or –Pi medium ) , 5 mM KNO3 , 0 . 025 mM Fe-EDTA , 2 mM MgSO4 , 2 mM Ca ( NO3 ) 2 , 2 . 5 mM MES-KOH , 0 . 07 mM H3BO3 , 0 . 014 mM MnCl2 , 0 . 01 mM NaCl , 0 . 5 µM CuSO4 , 1 µM ZnSO4 , 0 . 2 µM Na2MoO4 , 0 . 01 µM CoCl2 , 5 g/L sucrose . The agar was routinely purified by repeated washing in deionized water and underwent subsequent dialysis using DOWEX G-55 anion exchanger ( Ticconi et al . , 2009 ) . ICP-MS analysis of the treated agar ( 7 . 3 µg/g Fe and 5 . 9 µg/g P ) indicated a contribution of 1 . 25 µM Fe and 1 . 875 µM P to the solid 1% agar medium . Images were analyzed using ImageJ software . Seedlings were grown on 9 cm round plates supplemented with the indicated drug at the indicated concentration . Seedling survival was quantified after 14 days . Differentiation between live and dead seedlings was carried out similar to Yang et al . , 2016 . Surviving seedlings were defined as seedlings which had two green cotyledons and two green true leaves . Plants with yellow leaves or cotyledons were defined as dead . Tunicamycin sensitivity: 14 days old plants were transformed to half-strength Gamborg’s B5 medium containing indicated concentration of tunicamycin and grown in continues light at 22°C to determine survival rates . 20–30 seedlings for western blot or 0 . 5–1 g seedlings for immunoprecipitation and mass spectrometry were grown in liquid culture for 5 days under continuous light with shaking at 80 rpm . Media was supplemented with different drugs ( 3 µM Torin , 10 µg/ml Tunicamycin or other drugs dissolved in DMSO ) as indicated . 1 µM of concanamycin was added , if indicated in figures , to track the contribution of vacuolar degradation . For nutrient starvation , seedlings were transferred to phosphate , nitrogen- or sucrose-depleted media ( –C , –P , -N ) . The plants were kept in the dark to reduce sucrose production by photosynthesis or to provide drug stability . Pure DMSO was added to control samples . For analyzing total protein degradation such as TMT , seedlings were flash frozen in liquid nitrogen after 24 hr treatment . For interaction analysis such as Co-immunoprecipitation , seedling treatment was stopped after 8 hr of treatment . Samples were homogenized in a bead mill ( RetschMM300 , Haan , Germany; 30 Hz , 90 s ) at 4°C with zirconium oxide grinding beads or ground by mortar and pestle for bigger sample volumes . For western blotting , SDS loading buffer was added and the sample boiled at 95°C for 10 min . Lysates were cleared by centrifugation at 16 , 000 g for 10 min and protein concentration was normalized by Amidoblack staining ( Sigma ) . Western blotting was performed following standard protocols as described below . 5 µg of lysate was loaded per lane . HeLa-Kyoto and HEK293T cells maintained in Dulbecco’s modified Eagle’s Medium ( DMEM ) with 10% FBS , 1% L-Glutamine and 1% Penicillin/Streptomycin . Transfection was performed with GeneJuice transfection reagent according to manufacturer’s instructions . 100 µl of empty media was mixed with 3 µL of GeneJuice and after 5 min of incubation a total of 1 µg of DNA mixture per transfection was added . After 20 min of incubation , transfection mixture was added dropwise to the cells . Cells were incubated with DNA for 24 hr . DNA containing media was removed and replaced with media . Both cell lines were authenticated using STR profiling and repeatedly tested negative for mycoplasma contamination . Testing and authentication were performed using the in house core facilities . Lentiviral transduced shRNA-mediated knockdown of c53 in HeLa cells: The knockdown was performed in S2 conditions . HEK293T cells were seeded 24 hr prior to transfection in DMEM without antibiotics . At 50–60% confluency , cells were transfected with 1 µg shRNA , 750 ng psPAX2 and 250 ng pMD2 . G utilizing 6 µL of GeneJuice in 250 µL of empty DMEM . After 48 hr of incubation , the virus containing media was harvested and mixed 1:1 with full media . This mixture was applied to HeLa cells that were seeded 24 hr prior . Polybrene was added to a final concentration of 4 µg/ml . After 24 hr of incubation , the medium on target cells was exchanged with full media . After 24 hr , selection with 2 µg/ml Puromycin was started . No living cells were observed in a control plate after 24 hr . After splitting cells in S2 conditions , cells were transferred into S1 conditions . Cells were seeded 24 hr prior to treatment . At 50–60% confluency treatments were started by replacing media containing the indicated drugs or full media ( untreated ) . Tunicamycin was added with a final concentration of 2 . 5 µg/ml and Torin with a final concentration of 3 µM . The treatments were stopped after 16 hr by removing the media and washing the cells with 1xPBS . A 2 hr recovery period was started by adding either media containing 100 nM Bafilomycin A1 or full media . Cells were put on ice and lysed with 100 µL of Lysis buffer ( 50 mM HEPES , 150 mM NaCl , 1 mM EDTA , 1 mM EGTA , 25 mM NaF , 10 µM ZnCl2 , 1% Triton X-100% and 10% Glycerol ) per well . After centrifugation , supernatant was mixed 1:1 with 2x Laemmli Buffer and denatured by heating to 95°C for 5 min . Each sample was loaded onto a 4–20% SDS-PAGE gradient gel ( BioRad ) and electrophoresis was run at 100V for 1 . 5 hr . SDS-PAGE was performed using gradient 4–20% Mini-PROTEAN TGX Precast Protein Gels ( BioRad ) . Blotting on nitrocellulose membranes was performed using a semi-dry Turbo transfer blot system ( BioRad ) . For images of human LC3B , a wet transfer to PVDF membranes was performed at 200 mA for 70 min . Membranes were blocked with 5% skimmed milk or BSA in TBS and 0 . 1% Tween 20 ( TBS-T ) for 1 hr at room temperature or at 4°C overnight . This was followed by incubation with primary and subsequent secondary antibody conjugated to horseradish peroxidase . After three times 10 min washes with TBS-T , the immune-reaction was developed using ECL Super-Pico Plus ( Thermo ) and detected with ChemiDoc Touch Imaging System ( BioRad ) . Protein bands intensities were quantified with Image Lab 6 ( BioRad ) . Equal rectangles were drawn around the total protein gel lane and the band of interest . The lane profile was obtained by subtracting the mean intensity of the background . The adjusted volume of the peak in the profile was taken as a measure of the band intensity . The protein band of interest was normalized for the total protein level of the whole lane . Average relative intensities and a standard error of at least three independent experiments were calculated . To generate AtC53 antibody , purified protein was sent to Eurogentec for immunization of rabbits via their 28 day program . The final bleed was purified on column conjugated with the purified protein . For pulldown experiments , 10 µl of glutathione magnetic agarose beads ( Pierce Glutathione Magnetic Agarose Beads , Thermo Scientific ) were equilibrated by washing them two times with wash buffer ( 100 mM Sodium Phosphate pH 7 . 2 , 300 mM NaCl , 1 mM DTT , 0 . 01% ( v/v ) IGEPAL ) . Normalized E . coli clarified lysates or purified proteins were mixed , according to the experiment , added to the washed beads and incubated on an end-over-end rotator for 1 hr at 4°C . Beads were washed five times in 1 ml wash buffer . Bound proteins were eluted by adding 100 µl Laemmli buffer . Samples were analysed by western blotting or Coomassie staining . Yeast two hybrid assay ( Y2H ) was performed according to the Mathmaker GAL4 Two hybrid system ( Clonetech ) following the protocol from the manufacture . Different genes were fused in frame to GAL4 activation domain of the prey vector pGADT7 and GAL4 binding domain from the bait vector pGBKT7 . Split-GFP was used as positive control . Combinations of pGADT7 and pGBKT7 vectors carrying the different genes were transformed in the yeast strains Y187 ( MAT α ) and AH109 ( MAT a ) , respectively . After mating between bait and prey strains , the diploid yeast was selected for growth on ( SD ) -Leu /- Trp , ( SD ) -Leu /- Trp /- His and ( SD ) -Leu /- Trp /- His/-Ade plates at 28°C for 2 to 4 days . 0 . 5–1 g seedlings were grown in liquid and treated as described under section Autophagy Flux Assay . After homogenization of frozen samples by bead-mill , G-TEN buffer ( 10% Glycerole , 50 mM Tris/HCl pH 7 . 5 , 1 mM EDTA , 300 mM NaCl , 1 mM DTT , 0 . 1% [v/v] Nonidet P-40/Igepal , Complete protease inhibitor tablet ) was added , vortexed , and lysates were cleared by centrifugation at 16 , 000 g for 10 min at 4°C . Protein concentration was equally adjusted using Bradford protein assay ( Sigma ) . 25 µl of RFP or GFP-Trap_A beads ( Chromotek ) were equilibrated and added to each lysate and incubated for 2 hr at 4°C on a turning wheel . Beads were washed three times with 1 mL G-TEN buffer . For western Blot analysis , beads were resuspended in 30 µl SDS-loading buffer ( 116 mM Tris-HCl pH 6 . 8 , 4 . 9% glycerol , 10 mM DTT , 8% SDS ) . On-bead bound proteins were eluted by boiling the beads for 10 min at 70°C and analysed by western blotting with indicated antibodies . For mass spectrometry experiments , the beads were further washed five times with mass spectrometry compatible buffer ( 50 mM Tris/HCl pH 7 . 5 , 1 mM EDTA ) . Buffer resuspended beads were then submitted for trypsin digestion . Bead-bound bait proteins were incubated with fluorescently labelled prey protein as described previously by Turco et al . , 2019 . 10 µl of Glutathione Sepharose 4B beads ( GE Healthcare , average diameter 90 mm ) were incubated for 30 min at 4°C ( 16 rpm horizontal rotation ) with GST-tagged bait proteins ( 4 mg/mL for GST and GST-FIP200 CTR ) . The beads were washed two times in 10x bead volume of washing buffer ( 25 mM HEPES pH 7 . 5 , 150 mM NaCl , 1 mM DTT ) . The buffer was removed , and the beads were resuspended 1:1 in washing buffer . 10 µL of a 2–5 µM dilution of fluorescently labeled binding partners ( GFP , C53-GFP and GFP-p62 ) were added to the bead suspension and incubated for 30–60 min at room temperature before imaging with a Zeiss LSM700 confocal microscope with 20 X magnification . For quantification , the maximum gray value along the diameter of each bead ( n ≥ 15 ) was measured . MS/MS Data analysis: Raw files were processed with Proteome Discoverer ( version 2 . 3 , Thermo Fisher Scientific , Bremen , Germany ) . Database searches were performed using MS Amanda ( version 2 . 3 . 0 . 14114 ) ( Dorfer et al . , 2014 ) against the TAIR10 database ( 32785 sequences ) . The raw files were loaded as fractions into the processing workflow . Carbamidomethylation of cysteine and TMT on peptide N-termini were specified as fixed modifications , phosphorylation on serine , threonine and tyrosine , oxidation of methionine , deamidation of asparagine and glutamine , TMT on lysine , carbamylation on peptide N-termini and acetylation on protein N-termini were set as dynamic modifications . Trypsin was defined as the proteolytic enzyme , cleaving after lysine or arginine . Up to two missed cleavages were allowed . Precursor and fragment ion tolerance were set to 5 ppm and 15 ppm , respectively . Identified spectra were rescored using Percolator ( Käll et al . , 2007 ) , and filtered to 0 . 5% FDR at the peptide spectrum match level . Protein grouping was performed in Proteome Discoverer applying strict parsimony principle . Proteins were subsequently filtered to a false discovery rate of 1% at protein level . Phosphorylation sites were localized using IMP-ptmRS implemented in Proteome Discoverer using a probability cut-off of >75% for unambiguous site localization . TMT-quantification: TMT reporter ion S/N values were extracted from the most confident centroid mass within an integration tolerance of 20 ppm . PSMs with average TMT reporter S/N values below 10 as well as PSMs showing more than 50% co-isolation were removed . Protein quantification was determined based on unique peptides only . Samples were sum normalized and missing values were imputed by the 5% quantile of the reporter intensity in the respective sample . Statistical significance of differentially abundant proteins was determined using limma ( Smyth , 2004 ) . Gene Ontology ( Ashburner et al . , 2000 ) enrichment was determined using DAVID ( Dennis et al . , 2003 ) ( version 6 . 8 ) . Cross species comparison of regulated proteins was performed by mapping proteins to ortholog clusters available in eggnog ( Huerta-Cepas et al . , 2016 ) . Proteins containing signal peptides were predicted using SignalP 5 . 0 ( Almagro Armenteros et al . , 2019 ) . High-density peptide array analysis was performed commercially by PEPperPRINT . This comprised a full substitution scan of wild-type peptide GVSEWDPILEELQEM , with exchange of all amino acid positions with 23 amino acids including citrulline ( Z ) , methyl-alanine ( O ) and D-alanine ( U ) . The analysis also included an N- and C-terminal deletion series of wild-type peptide GVSEWDPILEELQEM; an additional 32 spots of custom control peptide KPLDFDWEIVLEEQ , and acidic variants of this control peptide involving exchanges of selected amino acid positions with glutamic acid € . The resulting peptide microarrays contained 416 different linear peptides printed at least in triplicate ( 1412 peptide spots; wild-type peptides were printed with a higher frequency ) , and were framed by HA ( YPYDVPDYAG , 88 spots ) control peptides ( See Supplementary file 1 for the array map ) . Peptide microarrays were pre-stained with rabbit anti-GST Dylight680 at a dilution of 1:2000 to investigate background interactions with the variants of wild-type peptides GVSEWDPILEELQEM and KPLDFDWEIVLEEQ that could interfere with the main assays . Subsequent incubation of other peptide microarrays with proteins GST-ATG8A and GST at a concentration of 10 µg/ml in incubation buffer was followed by staining with secondary antibody rabbit anti-GST Dylight680 and read-out at a scanning intensity of 7 ( red ) . The control staining of the HA epitopes with control antibody mouse monoclonal anti-HA ( 12CA5 ) DyLight800 was finally done as an internal quality control to confirm the assay quality and the peptide microarray integrity . Read-out of the control staining was performed at a scanning intensity of 7/7 ( red/green ) . Quantification of spot intensities and peptide annotation were based on the 16-bit grey scale tiff files at a scanning intensity of 7 that exhibit a higher dynamic range than the 24-bit colorized tiff files; microarray image analysis was done with PepSlide Analyzer . A software algorithm breaks down fluorescence intensities of each spot into raw , foreground and background signal ( see ‘Raw Data’ tabs ) , and calculates averaged median foreground intensities and spot-to-spot deviations of spot duplicates . Statistical analyses were performed with GraphPad Prism eight software . For all the quantifications described above , statistical analysis was performed . Statistical significance of differences between two experimental groups was assessed wherever applicable by either a two-tailed Student’s t-test if the variances were not significantly different according to the F test , or using a non-parametric test ( Mann-Whitney or Kruskal-Wallis with Dunn’s post-hoc test for multiple comparisons ) if the variances were significantly different ( p<0 . 05 ) . Differences between two data sets were considered significant at p<0 . 05 ( * ) ; p<0 . 01 ( ) ; p<0 . 001 ( * ) ; p<0 . 0001 ( **** ) ; n . s . , not significant . CD spectroscopy experiments were performed using a Chirascan-Plus CD spectrophotometer ( Applied Photophysics ) . Purified proteins in 50mM sodium phosphate pH 7 . 0 , 100mM NaCl were diluted to approximately 0 . 2 mg/ml and spin-filtered with an 0 . 1µm filter . CD measurements were carried out in a quartz glass cuvette with 0 . 5 mm path length . To obtain overall CD spectra , wavelength scans between 180 nm and 260 nm were collected at 25°C using a 1 . 0 nm bandwidth , 0 . 5 nm step size , and time per point of 0 . 5 s . Both CD and absorbance data were collected at the same time over three accumulations and averaged . CD data at wavelengths where the absorptivity was above 2 . 5 are not shown ( data below 194nm ) . The raw data in millidegree units were corrected for background and drift ( Θdcorr ) . Subsequently , the differential molar extinction coefficient per peptide bond ( Δε ) was calculated , taking into account the absorptivity measured at 205 nm ( A205 ) and the calculated protein extinction coefficient at 205 nm ( ε205 ) using the equation∆ε=Θdcorr∙ε20510∙A205∙ ( N-1 ) ∙3298	For cells to survive they need to be able to remove faulty or damaged components . The ability to recycle faulty parts is so crucial that some of the molecular machinery responsible is the same across the plant and animal kingdoms . One of the major recycling pathways cells use is autophagy , which labels damaged proteins with molecular tags that say 'eat-me' . Proteins called receptors then recognize these tags and move the faulty component into vesicles that transport the cargo to a specialized compartment that recycles broken parts . Cells make and fold around 40% of their proteins at a site called the endoplasmic reticulum , or ER for short . However , the process of folding and synthesizing proteins is prone to errors . For example , when a cell is under stress this can cause a ‘stall’ in production , creating a build-up of faulty , partially constructed proteins that are toxic to the cell . There are several quality control systems which help recognize and correct these errors in production . Yet , it remained unclear how autophagy and these quality control mechanisms are linked together . Here , Stephani , Picchianti et al . screened for receptors that regulate the recycling of faulty proteins by binding to the ‘eat-me’ tags . This led to the identification of a protein called C53 , which is found in both plant and animal cells . Microscopy and protein-protein interaction tests showed that C53 moves into transport vesicles when the ER is under stress and faulty proteins start to build-up . Once there , C53 interacts with two proteins embedded in the wall of the endoplasmic reticulum . These proteins form part of the quality control system that senses stalled protein production , labelling the stuck proteins with ‘eat-me’ tags . Together with C53 , they identify and remove half-finished proteins before they can harm the cell . The fact that C53 works in the same way in both plant and human cells suggests that many species might use this receptor to recycle stalled proteins . This has implications for a wide range of research areas , from agriculture to human health . A better understanding of C53 could be beneficial for developing stress-resilient crops . It could also aid research into human diseases , such as cancer and viral infections , that have been linked to C53 and its associated proteins .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology", "cell", "biology" ]	2020	A cross-kingdom conserved ER-phagy receptor maintains endoplasmic reticulum homeostasis during stress
Homeostatic signaling stabilizes synaptic transmission at the neuromuscular junction ( NMJ ) of Drosophila , mice , and human . It is believed that homeostatic signaling at the NMJ is bi-directional and considerable progress has been made identifying mechanisms underlying the homeostatic potentiation of neurotransmitter release . However , very little is understood mechanistically about the opposing process , homeostatic depression , and how bi-directional plasticity is achieved . Here , we show that homeostatic potentiation and depression can be simultaneously induced , demonstrating true bi-directional plasticity . Next , we show that mutations that block homeostatic potentiation do not alter homeostatic depression , demonstrating that these are genetically separable processes . Finally , we show that homeostatic depression is achieved by decreased presynaptic calcium channel abundance and calcium influx , changes that are independent of the presynaptic action potential waveform . Thus , we identify a novel mechanism of homeostatic synaptic plasticity and propose a model that can account for the observed bi-directional , homeostatic control of presynaptic neurotransmitter release . Homeostatic signaling systems are believed to interface with the mechanisms of neural plasticity to stabilize neural function throughout the life of an organism ( Marder and Goaillard , 2006; Turrigiano , 2011; Davis , 2013 ) . To do so , homeostatic signaling systems require bi-directional control , being able to accurately offset perturbations that persistently increase or decrease the excitation of neurons or muscle . An evolutionarily conserved , bi-directional form of homeostatic signaling is observed at the neuromuscular junction of organisms ranging from Drosophila to human ( see Davis , 2013 for review ) . At the Drosophila NMJ , inhibition of postsynaptic glutamate receptor function leads to a compensatory increase in presynaptic neurotransmitter release that precisely restores normal muscle depolarization ( Petersen et al . , 1997; Davis and Goodman , 1998; Davis et al . , 1998; Frank et al . , 2006; Davis , 2013 ) . This is referred to as presynaptic homeostatic potentiation ( PHP ) . An opposing process , presynaptic homeostatic depression ( PHD ) , can be induced by presynaptic overexpression of the vesicular glutamate transporter ( vGlut2 ) , which causes an increase in the amount of glutamate packaged within individual synaptic vesicles and a corresponding increase in synaptic vesicle diameter ( Daniels et al . , 2004 ) . These larger vesicles produce larger average spontaneous miniature excitatory post-synaptic potentials ( mEPSPs ) . A homeostatic decrease in presynaptic vesicle release , however , maintains evoked EPSP amplitudes at wild type levels despite increased mEPSP amplitude ( Daniels et al . , 2004 ) . It remains unknown whether these opposing forms of presynaptic , homeostatic modulation , PHP , and PHD , are coordinately controlled to achieve robust , bi-directional control of muscle excitation . Can PHP and PHD be simultaneously induced to adjust presynaptic release ? If so , can the expression of PHP and PHD be coordinated to produce an accurate homeostatic response , precisely offsetting the magnitude of the perturbation ? Are PHP and PHD controlled by the same signaling system ( s ) or are they independently controlled and coordinated ? Ultimately , understanding how these signaling systems interface requires the identification and characterization of the cellular and molecular mechanisms that drive PHP and PHD . Although significant progress has been made in identifying the mechanisms that participate in PHP ( Frank et al . , 2006; Dickman and Davis , 2009; Tsurudome et al . , 2010; Müller et al . , 2011; Müller and Davis , 2012; Younger et al . , 2013 ) , nothing is known regarding the cellular or molecular mechanisms responsible for PHD ( Daniels et al . , 2004 ) . Here , we demonstrate that PHP and PHD are driven by distinct molecular mechanisms . Further , we demonstrate that PHD is achieved by a regulated decrease in presynaptic calcium channel abundance and concomitant decrease in presynaptic calcium influx . The regulated control of active-zone associated calcium channel abundance represents a novel mechanism of homeostatic synaptic plasticity . Importantly , this mechanism is consistent with the simultaneous induction of both PHP and PHD . As such , bi-directional , rheostat-like adjustment of presynaptic release could reasonably be achieved by the coordinated expression of two independent homeostatic signaling systems at the NMJ . We first confirmed that presynaptic overexpression of the vGlut2 glutamate transporter ( UAS-vGlut2 ) caused an increase in average mEPSP amplitude and a corresponding decrease in presynaptic vesicle release ( Daniels et al . , 2004; Figure 1A–E ) . Animals overexpressing UAS-vGlut2 ( referred to hereafter as vGlut-OE animals ) displayed a shift of the entire mEPSP amplitude distribution , corresponding to larger miniature release events ( Figure 1E ) . As previously shown ( Daniels et al . , 2004 ) , a decrease in presynaptic release is able to precisely offset the increase in mEPSP amplitude , thereby restoring EPSP amplitudes to wild type levels ( Figure 1A , C ) . Since EPSP amplitudes are maintained at wild type levels , these data support the conclusion that a homeostatic signaling system detects the change in average mEPSP amplitude and drives a compensatory reduction in presynaptic neurotransmitter release that precisely offsets the magnitude of increased mEPSP amplitude . This compensatory response will be referred to as PHD to distinguish this process from PHP . 10 . 7554/eLife . 05473 . 003Figure 1 . vGlut-OE animals display increased miniature EPSP amplitude and a compensatory decrease in vesicle release . ( A ) Representative EPSP traces following stimulation of a single action potential in wild type and vGlut-OE animals; representative mEPSP traces in inset . ( B ) Representative EPSC traces following stimulation of a single action potential in wild type and vGlut-OE animals; representative mEPSC traces in inset . ( C ) vGlut-OE animals have increased mEPSP amplitudes ( p < 0 . 0001 ) with no change in evoked EPSP amplitudes ( p = 0 . 08 ) , resulting in a decrease in calculated quantal content ( p < 0 . 0001 , n > 12 NMJ per genotype ) . ( D ) vGlut overexpression results in increased mEPSC amplitudes ( p < 0 . 0001 ) , but no significant change in evoked EPSC amplitudes , resulting in a decrease in calculated quantal content ( p < 0 . 01 , n > 8 per genotype , error bars = SEM ) . ( E ) The cumulative frequency of mEPSPs is shifted to greater amplitudes in vGlut-OE animals . ( F ) Quantal content is able to scale across an order of magnitude in response to perturbations that increase ( vGlut2 overexpression ) or decrease ( PhTx treatment ) quantal amplitude . An exponential function was fit to these data points with R2 = 0 . 7491 . [Ca]e = 3 mM for ( B ) and ( D ) , 0 . 3 mM for all others . Error bars = SEM for all . DOI: http://dx . doi . org/10 . 7554/eLife . 05473 . 003 Following the induction of PHP , there is a rightward shift in the calcium cooperativity of neurotransmitter release , indicating that the homeostatic potentiation of neurotransmitter release is robust over a range of extracellular Ca2+ concentrations and likely does not involve a change in the calcium sensor for synaptic vesicle fusion ( Frank et al . , 2006; Dickman and Davis , 2009; Müller et al . , 2012 ) . We performed a similar analysis for PHD in vGlut-OE animals . We observed that PHD accurately offset an increase in average spontaneous release event amplitude at elevated extracellular calcium ( 3 mM; Figure 1B , D ) , demonstrating that PHD proceeds accurately at extracellular calcium concentrations [Ca2+]e that span what is considered to be physiological in the intact organism ( compare quantal content in Figure 1C , D ) . Experiments at elevated extracellular calcium , 3 mM [Ca2+]e , were performed using two-electrode voltage clamp ( see ‘Materials and methods’ ) . Similar changes were observed at a range of intervening calcium concentrations ( 0 . 3–1 . 5 mM ) with no change in the slope of the calcium cooperativity curve , consistent with previous experiments examining calcium cooperativity in vGlut-OE animals over a narrower range of extracellular calcium ( data not shown; Daniels et al . , 2004 ) . These data demonstrate that PHD is robust across a range of extracellular calcium concentrations and is likely to be independent of any change in the calcium sensor for presynaptic release . We next addressed the accuracy of homeostatic signaling during PHD . A system that is under true homeostatic control should be able to accurately offset a wide range of perturbation . This was previously shown for PHP by plotting the average mEPSP amplitude against average quantal content for each recording made from an individual NMJ . In this analysis , the data lie along a line associated with near perfect compensation ( Frank et al . , 2006 ) . We constructed a similar plot , including recordings from vGlut-OE NMJ as well as wild type NMJ and NMJ that were incubated in sub-blocking concentrations of the glutamate receptor antagonist philanthotoxin-433 ( PhTx ) , and observed a wide range of mEPSP amplitudes , ranging from >2 mV to less than 0 . 4 mV . These data produced points lying along a clear line , providing evidence for continuous adjustment of presynaptic release , effectively maintaining a constant EPSP amplitude over a wide range of average mEPSP amplitude values ( Figure 1F ) . Moreover , each segment of the data in Figure 1F ( shown by different colors ) , when fit independently , showed a negative relationship between mEPSP amplitude and presynaptic release . This indicates that , as with PHP , the mechanisms of PHD can respond to small changes in the average mEPSP amplitude and adjust presynaptic release accordingly . In conclusion , these data suggest that PHD could be part of an integrated , bi-directional , homeostatic signaling system at the NMJ . We next asked whether the NMJ can coordinate simultaneous induction of homeostatic potentiation and depression to precisely control release . Since vGlut2 overexpression is a presynaptic perturbation and PhTx is a postsynaptic perturbation , both PHD and PHP can be induced concurrently . First , we found that PhTx application to vGlut-OE terminals significantly reduced mEPSP amplitudes in a manner quantitatively similar to that observed when PhTx is applied to wild type NMJ ( Figure 2A , B ) . At both wild type and vGlut-OE NMJ , the 50–60% drop in average mEPSP amplitude was offset by an approximate doubling of presynaptic neurotransmitter release ( Figure 2A , B ) . As a result , in both wild type and vGlut-OE NMJ , EPSP amplitudes were maintained at or near wild type levels after PhTx application ( a slight decrease in EPSP amplitude was observed in vGlut-OE animals , p < 0 . 05 ) . Furthermore , there was no difference in EPSP amplitude comparing wild type ( +PhTx ) and vGlut-OE ( +PhTx; p > 0 . 05 ) . Thus , despite the fact that vGlut-OE NMJ chronically express PHD during larval development , application of PhTx is able to induce expression of PHP that appears unaffected by prior induction of PHD . From these results , several conclusions can be made . First , these data demonstrate that the mechanisms of PHP and PHD can co-exist at a single NMJ and can be sequentially induced , at least in the sequence that we performed in this study . The reagents to perform the reverse sequence do not currently exist as there is no pharmacological or acute genetic manipulation that can rapidly increase mEPSP amplitude . Second , the data argue that the reduced presynaptic release caused by vGlut2 overexpression is not a secondary effect whereby vGlut2 overabundance disrupts the vesicle fusion apparatus , since release can be precisely potentiated at this NMJ . Finally , these data demonstrate that the mechanisms of PHP and PHD can be effectively combined to fine tune presynaptic release . 10 . 7554/eLife . 05473 . 004Figure 2 . Homeostatic potentiation can be induced at NMJ already expressing presynaptic homeostatic depression ( PHD ) . ( A ) Representative evoked EPSP traces from wild type and vGlut-OE NMJ without and with the application of PhTx . Corresponding mEPSP traces in inset for each condition . ( B ) The addition of PhTx to either wild type ( top ) or vGlut-OE ( bottom ) animals results in decreased mEPSP amplitude ( p < 0 . 0001 for both ) with a minimal change in evoked EPSP amplitude ( p = 0 . 5 for wt , p = 0 . 03 for vGlut-OE ) , resulting in a decrease in calculated quantal content ( p < 0 . 0001 for both , n = 11 for wt -PhTx , n = 8 for wt +PhTx , n = 6 for vGlut-OE–PhTx , n = 9 for vGlut-OE +PhTx . Error bars = SEM ) . [Ca]e = 0 . 3 mM for all . DOI: http://dx . doi . org/10 . 7554/eLife . 05473 . 004 Since PHP and PHD can be combined to produce an accurate homeostatic response , and since both PHP and PHD act upon the presynaptic release of neurotransmitter , it raises the question as to whether PHP and PHD are controlled by the same molecular mechanism . In one scenario , the same signaling system ( s ) could drive both PHP and PHD . Alternatively , distinct molecular mechanisms might underlie PHP and PHD and the two processes might somehow be combined or coordinated to produce an accurate homeostatic response . Considerable progress has been made identifying genes that , when mutated , block PHP . Therefore , we tested several of these gene mutations for a role in PHD . It was recently demonstrated that the potentiation of neurotransmitter release during PHP requires a presynaptic ENaC channel ( Younger et al . , 2013 ) . Briefly , in either a glutamate receptor mutant background , or following acute application of PhTx to the NMJ , subsequent pharmacological inhibition of the presynaptic ENaC channel , or genetic mutation of the gene encoding an essential subunit of this channel , eliminates the expression of homeostatic potentiation ( Younger et al . , 2013 ) . A current model suggests that sodium leak through newly membrane-inserted ENaC channels drives presynaptic membrane depolarization , a subsequent potentiation of presynaptic calcium influx , and enhanced neurotransmitter release . Importantly , ENaC channels do not appear to be present on the membrane under baseline conditions since deletion or pharmacological blockade of these channels does not decrease baseline neurotransmitter release ( Younger et al . , 2013 ) . If ENaC channels are not present on the membrane at rest , then these channels should not have a role in PHD . To test this possibility , we applied the ENaC channel inhibitor benzamil ( 25 µM ) to wild type and vGlut-OE NMJ . We found no significant effect of benzamil on presynaptic release at vGlut-OE NMJ , consistent with the expectation that ENaC channels do not participate in baseline release in wild type animals and , therefore , cannot be removed as a mechanism underlying PHD . ( Figure 3A , B ) . As a control , we tested whether application of Benzamil still prevents PHP expression in the vGlut-OE background . We found that this is , indeed , the case ( vGlut-OE +Benzamil: mEPSP = 0 . 83 ± 0 . 05 mV , QC = 39 . 2 ± 2 . 3; vGlut-OE +Benzamil +PhTx: mEPSP = 0 . 50 ± 0 . 02 mV , QC = 43 . 6 ± 3 . 0; quantal content is statistically unchanged; p = 0 . 29 , n > 10 ) . Taken together , these data imply that the mechanisms responsible for PHD are molecularly separable from those that drive PHP . 10 . 7554/eLife . 05473 . 005Figure 3 . Presynaptic homeostatic potentiation and depression operate through distinct molecular mechanisms . ( A ) Representative single action potential stimulated EPSP traces from wild type and vGlut-OE animals in the presence of the drug Benzamil with corresponding mEPSP traces in the inset . ( B ) vGlut-OE animals have increased mEPSP amplitudes in the presence of Benzamil as compared to similarly treated wild type animals ( p < 0 . 0001 ) but have unchanged evoked EPSP amplitudes , resulting in a decrease in calculated quantal content ( p = 0 . 05 , n = 9 for wt , n = 5 for vGlut-OE ) . ( C ) Representative EPSP traces as in ( A ) from rim103 homozygotes in a wild type ( left ) and vGlut2 overexpressing ( right ) background . ( D ) vGlut2 overexpression in rim103 mutants results in increased mEPSP amplitudes ( p = 0 . 0002 ) and decreased quantal content ( p = 0 . 003 , n = 12 for rim103 and n = 10 for rim103 +vGlut-OE ) . ( E ) Representative EPSP traces as in ( A ) from CacS homozygotes in a wild type ( left ) and vGlut2 overexpressing ( right ) background . ( F ) vGlut2 overexpression in CacS mutants results in increased mEPSP amplitudes ( p < 0 . 0001 ) and decreased quantal content ( p < 0 . 0001 , n = 22 for CacS , n = 11 for CacS +vGlut-OE ) . Error bars = SEM for all . DOI: http://dx . doi . org/10 . 7554/eLife . 05473 . 005 To further investigate molecular distinctions between PHP and PHD , we examined additional gene mutations shown to block PHP . We first tested whether mutations in the gene encoding Rab3 interacting molecule ( RIM ) affect homeostatic synaptic depression . We previously demonstrated that multiple loss of function alleles of the rim gene , including a deletion that removes a large portion of the coding sequence ( rim103 ) , all block PHP , independent of whether they also impair baseline presynaptic release ( Müller et al . , 2012 ) . By contrast , vGlut2 overexpression in the rim103 mutant background still induced robust expression of PHD ( Figure 3C , D ) . This experiment was performed in 0 . 4 mM extracellular calcium to normalize presynaptic calcium influx and vesicle release to that observed in wild type under standard recording conditions ( Müller et al . , 2012 ) . These data demonstrate that RIM is not required for PHD and further separate the mechanisms that promote PHD vs PHP . Finally , we tested a point mutation in the CaV2 . 1 calcium channel previously shown to block PHP ( cacS; Frank et al . , 2006 ) . The cacS mutation is a single amino acid substitution in the sixth transmembrane domain of the third repeat of the pore forming subunit of the CaV2 . 1 channel ( Brooks et al . , 2003 ) . In a current model of PHP , the cacS mutation prevents low-voltage modulation of CaV2 . 1 calcium channels following insertion of ENaC channels in the presynaptic membrane ( Younger et al . , 2013 ) . Because the cacS mutation causes a ∼30% decrease in presynaptic calcium influx and a corresponding decrease in presynaptic release at baseline ( Müller and Davis , 2012 ) , we performed our analysis at elevated ( 1 mM ) extracellular calcium . By elevating extracellular calcium , we restored EPSP amplitudes and calcium influx to levels observed in wild type animals under standard experimental conditions ( 0 . 3 mM extracellular calcium ) . Although the cacS mutation completely blocks PHP , we observed no effect on PHD at vGlut-OE NMJ ( Figure 3E , F ) . Taken together with the observation that PHD is insensitive to benzamil treatment and occurs normally in the presence of a RIM deletion , our data demonstrate that the molecular mechanisms underlying PHD are distinct from those that drive PHP . The expression of PHP is achieved , in part , by an increase in presynaptic calcium influx through CaV2 . 1 calcium channels ( Frank et al . , 2006; Müller and Davis , 2012 ) . In principle , this could be achieved by broadening the waveform of the presynaptic action potential . Conversely , PHD could be achieved by narrowing the presynaptic action potential waveform . A direct test of these possibilities has been lacking because it has not been possible to record from the type 1 presynaptic boutons of the Drosophila NMJ , which are embedded within the postsynaptic muscle cell . However , it was recently demonstrated that Archaerhodopsin can be expressed in motoneurons without affecting baseline neurotransmission . The voltage sensitive properties of Archaerhodopsin can be used to visualize the presynaptic action potential waveform with high temporal resolution ( Ford and Davis , 2014 ) . Importantly , this technique has sufficient temporal resolution to quantify changes in action potential half-width that correlate with either a 50% increase or decrease in neurotransmitter release ( Ford and Davis , 2014 ) . Therefore , this tool has sufficient sensitivity to resolve changes in action potential waveform that might be responsible for PHD or PHP . To determine whether PHP and/or PHD are achieved by a change in action potential waveform , we expressed Archaerhodopsin ( Arch-GFP ) in motoneurons of wild type animals with and without PhTx treatment , in GluRIIA mutant animals , and in vGlut-OE animals . This level of Arch-GFP expression does not alter baseline synapse function or anatomy ( Ford and Davis , 2014 ) . To image voltage , we performed confocal spot measurements , using GFP to localize the spot to the edges of individual synaptic boutons . Importantly , spot confocal imaging does not alter action potential propagation past the imaging site to a distal extracellular recording site within an NMJ and spot excitation at 643 nm has no effect on baseline neurotransmission ( Ford and Davis , 2014 ) . It is worth noting that measurements of action potential waveform at the NMJ are nearly identical to measurements made from somatic patch clamp recordings at the motoneuron cell body ( Marie et al . , 2010; Ford and Davis , 2014 ) . Using Arch-GFP to image the presynaptic AP waveform , we observed that the induction of PHP following application of PhTx to the NMJ or in animals containing a mutation in the muscle specific glutamate receptor GluRIIA did not cause an increase in AP width or WHM ( Figure 4A–C ) . We found a small , but significant , decrease in WHM following PhTx application , but this is in the opposite direction expected for the potentiation of neurotransmitter release ( Figure 4C ) . Next , we observed the AP waveform in vGlut-OE animals . This experiment was performed at two different extracellular calcium concentrations: low extracellular calcium ( 0 . 4 mM ) typically used to examine synaptic transmission in current clamp configuration , and at physiological calcium ( 1 . 5 mM ) to control for the recent observation that action potential repolarization is faster at elevated extracellular calcium ( Figure 4A , far right , compare black and gray traces; Ford and Davis , 2014 ) . In both cases , there was no change in AP waveform comparing wild type with vGlut-OE animals ( Figure 4A–C ) . We also controlled for the possibility that Arch-GFP imaging might alter the expression of PHP or PHD by recording neuromuscular transmission from the same genotypes in which we quantified AP waveforms . We found that both PHP , using either PhTx or the GluRIIA mutation , and PHD occur normally in the presence of Arch-GFP ( Figure 4D , E ) . From these data , we conclude that the homeostatic modulation of presynaptic neurotransmitter release , either PHP or PHD , is not correlated with a change in AP waveform measured at the presynaptic nerve terminal . 10 . 7554/eLife . 05473 . 006Figure 4 . Action potential waveforms do not change during presynaptic homeostasis . ( A ) Normalized average AP traces comparing PhTx-treated , GluRIIA mutant , and vGlut-OE animals to WT animals . ( B ) AP width does not change in animals in which presynaptic homeostasis has been induced ( n labeled within bars for all ) . ( C ) AP width at half maximum amplitude does not change in animals where presynaptic homeostasis has been induced , except for a small decrease in PhTx treated animals; p = 0 . 05; n for all is same as ( B ) . ( D ) Quantal content homeostatically changes in response to perturbed mEPSP amplitude in animals that express Arch ( p < 0 . 001 , n labeled within bars for all ) . ( E ) Example traces of the data summarized in ( D ) . [Ca]e = 0 . 4 mM , errors bars = SEM for all . DOI: http://dx . doi . org/10 . 7554/eLife . 05473 . 006 PHP requires two parallel changes within the presynaptic nerve terminal; ( 1 ) an increase in calcium influx through CaV2 . 1 calcium channels and ( 2 ) an increase in the RRP . If homeostatic depression is controlled by similar presynaptic mechanisms , then we would expect the process to be correlated with decreased presynaptic calcium influx and decreased RRP size . We therefore measured these parameters in vGlut-OE animals . First , we assayed the RRP in vGlut-OE animals by quantifying EPSC amplitudes during high frequency stimulation ( 30 stimuli at 60 Hz , 3 mM extracellular calcium ) and estimating the cumulative EPSC and RRP according to published methods ( see ‘Materials and methods’ ) . Estimates of the RRP using this method in wild type animals agree with estimates made according to variance mean analysis , a technique based on the measurement of EPSC amplitude variance delivered at low ( 0 . 2 Hz ) stimulus frequencies that do not induce short term synaptic modulation ( Müller et al . , 2012 ) . We found that the RRP was not significantly different when comparing wild type and vGlut-OE animals ( Figure 5A , B ) . Based on these data , we conclude that homeostatic depression is not correlated with a drop in the size of the RRP , highlighting a mechanistic difference in the expression of PHP and PHD . 10 . 7554/eLife . 05473 . 007Figure 5 . Effects of PHD on RRP size and Pr . ( A ) Top: representative EPSC trains from wild type ( left ) and vGlut-OE ( right ) animals in response to 60 Hz stimulation ( 30 stimuli ) . Bottom: cumulative EPSC amplitude for the traces shown . The line fit to the cumulative EPSC data and back extrapolated to time 0 is shown in red ( see ‘Materials and methods’ ) . ( B ) vGlut-OE animals had increased mEPSC amplitudes ( p < 0 . 0001 ) , and cumulative EPSC amplitudes that trended upward ( p = 0 . 16 ) . The calculated RRP size was unchanged ( p = 0 . 49 ) . ( C ) The relative probability of release as calculated by the train method was decreased in vGlut-OE animals at 3 mM external Ca2+ ( p = 0 . 008 ) . ( D ) vGlut-OE animals were not significantly more resistant to depression over the first 10 stimuli of the 60 Hz trains described ( a single phase decay function was fitted to points 2–10 for each muscle , generating a decay constant which was averaged across each genotype; p = 0 . 24 ) . ( E ) The probability of release was unchanged in vGlut-OE animals as measured by the paired pulse ratio at 0 . 3 mM external Ca2+ ( n = 5 for vGlut-OE , n = 6 for wt ) . For ( A ) through ( D ) : n = 9 for WT , n = 10 for vGlut-OE and [Ca]e = 3 mM . Error bars = SEM for all . DOI: http://dx . doi . org/10 . 7554/eLife . 05473 . 007 Having observed no change in the RRP size of vGlut-OE animals , we assessed whether PHD is associated with a drop in presynaptic release probability ( Pr ) by examining short-term synapse modulation . As a measure of Pr , we calculated the fraction of the total RRP released following the first action potential in a stimulus train , a value referred to as Ptrain ( Schneggenburger et al . , 1999 ) . Using this method , we found Ptrain to be significantly lower in vGlut-OE animals as compared to wild type , indicating reduced presynaptic release probability ( Figure 5C ) . Examining accumulated synaptic depression during a short stimulus train , however , we observed only a mild , not statistically significant , difference in the rate of depression ( Figure 5D ) . Finally , we assessed paired pulse stimulation at low extracellular calcium ( 0 . 3 mM ) at a range of inter-stimulus intervals and found no change in paired-pulse depression ( Figure 5E ) . In summary , we observed a significant reduction in Ptrain , but this change was not reflected by a change in short-term release dynamics as measured with paired-pulse stimulation at low extracellular calcium . Thus , while pool usage is altered under conditions of high release during a stimulus train , the pool of vesicles is apparently large enough to obscure a change in paired-pulse release dynamics under conditions of low release . We conclude that vGlut-OE is associated with a decrease in presynaptic release probability . The mechanisms driving PHP include a significant increase in presynaptic calcium influx ( Müller and Davis , 2012 ) . Therefore , we asked whether there is an opposing decrease in presynaptic calcium during PHD . We estimated presynaptic calcium influx in wild type and vGlut-OE animals according to previously published methods ( Müller and Davis , 2012 ) . In brief , nerve terminals are loaded with the calcium indicator Oregon Green Bapta 1 ( OGB1 ) and a calcium-insensitive reference dye ( Alexa 568 ) . Line scans across a single synaptic bouton are then made prior to and following single action potential stimulation . Following single action potential stimulation , we observed a ∼50% reduction in calcium influx in vGlut-OE animals compared to wild type ( Figure 6A–C ) . There was no significant difference in the basal calcium signal , indicating that this effect was not caused by differences in dye loading ( Figure 6C , left ) . This decrease in spatially averaged presynaptic calcium influx is more than sufficient to account for the depression of presynaptic neurotransmitter release caused by vGlut2 overexpression . Indeed , although the precise relationship between the spatially averaged calcium signal and release probability remains to be defined , the drop in the presynaptic calcium signal is larger than might be expected for the observed decrease in presynaptic release observed during PHD ( see ‘Discussion’ ) . These data indicate that PHD is correlated with a large decrease in presynaptic calcium influx and suggests that this is a primary mechanism driving the process . 10 . 7554/eLife . 05473 . 008Figure 6 . Animals expressing PHD have reduced presynaptic calcium influx . ( A ) Representative line-scans of WT and vGlut-OE boutons . Red arrowhead: moment of stimulation . ( B ) Example Ca2+ transients from WT and vGlut-OE animals . vGlut-OE transients are identical to WT when normalized by amplitude ( right ) . ( C ) vGlut-OE animals displayed a ∼50% drop in ΔF/F ( right ) with no change in baseline OGB-1 fluorescence ( left ) . n = 12 for WT , n = 19 for vGlut-OE , p < 0 . 0001 , see ‘Materials and methods’ for details . DOI: http://dx . doi . org/10 . 7554/eLife . 05473 . 008 A drop in presynaptic calcium influx could be achieved by a change in calcium channel number or function . Currently , no antibodies are available that robustly label the presynaptic calcium channel in vivo . Therefore , we overexpressed a GFP-tagged CaV2 . 1 channel ( referred to , hereafter , as CaV2 . 1-GFP or Cac-GFP ) presynaptically to visualize active-zone associated calcium channels . Overexpression of CaV2 . 1-GFP does not influence baseline neurotransmission and serves as a reliable reporter of active-zone localized presynaptic channels ( Kawasaki et al . , 2004; Wang et al . , 2014 ) . Previous reports demonstrate that overexpressed CaV2 . 1-GFP can report changes in presynaptic calcium channel abundance ( Wang et al . , 2014 ) . Consistent with prior reports , we found that overexpression of CaV2 . 1-GFP did not affect transmission in wild type animals and , furthermore , did not affect PHD induction in vGlut-OE animals ( Figure 7A , B ) . Imaging of overexpressed presynaptic CaV2 . 1-GFP revealed discrete puncta residing within active zones in both wild type and vGlut-OE animals ( Figure 7C , D ) . In order to measure CaV2 . 1 abundance , we first labeled presynaptic active zones with an antibody against the active zone-associated protein Bruchpilot ( Brp ) , which has previously been shown to surround presynaptic calcium channels based on super-resolution microscopy ( Liu et al . , 2011 ) . We then labeled CaV2 . 1-GFP with an antibody against GFP and measured the fluorescence intensity of the signal at active zones , using Brp puncta to define the ROI . Strikingly , we observed a 30% and 50% decrease in CaV2 . 1-GFP signal at type 1B and type 1S boutons , respectively , in vGlut-OE animals compared to wild type ( Figure 7E–H ) . There was no change in the levels of Brp , assayed by measurement of anti-BRP fluorescence intensity ( Figure 7E–H ) . We also quantified the number of Brp puncta per bouton as an estimate of active zone number per bouton and found no significant change ( vGlut-OE: 12 . 1 ± 2 . 0 and 6 . 4 ± 1 . 4 puncta per type 1B and 1S bouton , respectively , N = 11 NMJ; WT: 13 . 3 ± 1 . 5 and 7 . 5 ± 2 . 1 . p > 0 . 05 for both , N = 12 NMJ ) . These data indicate that the observed decrease in CaV2 . 1-GFP signal is not a secondary consequence of grossly altered active zone organization , consistent with previous electron microscopic analysis of the vGlut-OE NMJ ( Daniels et al . , 2004 ) . Since UAS-CaV2 . 1-GFP expression is independent of the normal transcriptional regulation of the CaV2 . 1 gene locus , these data additionally suggest that CaV2 . 1 levels are not modulated at the level of transcription during PHD , but rather that CaV2 . 1 protein is selectively restricted from entering the active zone or is selectively removed from the active zone . We conclude that reduced presynaptic CaV2 . 1 abundance underlies the decreased neurotransmitter release observed during homeostatic depression . 10 . 7554/eLife . 05473 . 009Figure 7 . Animals expressing PHD have reduced CaV2 . 1 channel levels . ( A ) Example EPSP traces of vGlut-OE animals with ( right ) and without ( left ) a Cac-GFP transgene; example mEPSPs in inset . ( B ) vGlut-OE animals expressing a Cac-GFP transgene had unchanged synaptic transmission ( n = 15 for vGlut-OE , n = 6 for vGlut-OE +Cac-GFP , [Ca]e = 0 . 3 mM ) . ( C ) Representative image of a wild type NMJ labeled with antibodies against the active zone component Brp and a transgenically expressed GFP-tagged Cac channel . ( D ) Representative vGlut-OE NMJ labeled as in ( C ) . ( E ) vGlut-OE animals displayed no change in synaptic Brp signal as compared to wild type , but had greatly reduced synaptic Cac-GFP signal at type 1B boutons ( p = 0 . 0002 , n > 1000 active zones across >10 NMJ for each ) . ( F ) The cumulative frequency of fluorescence intensity at 1B boutons for the antibodies described in ( C ) . The distribution of vGlut-OE animals was shifted to the left as compared to wild type . ( G ) Synaptic Cac-GFP signal was also reduced at type 1S boutons in vGlut-OE animals ( p = 0 . 0007 , n > 400 active zones across >10 NMJ for each . ) ( H ) The cumulative frequency for 1S boutons as described in ( F ) . Error bars = SEM for all . DOI: http://dx . doi . org/10 . 7554/eLife . 05473 . 009 The regulated trafficking of calcium channels remains a poorly understood process . This is due , in part , to the limited number of reagents for visualizing presynaptic calcium channels and the fact that active zones in most areas of the nervous system are small structures , at or near the diffraction limit of light . Although auxiliary calcium channel subunits such as α2δ have mutant phenotypes that affect presynaptic calcium influx and calcium channel abundance ( Dickman et al . , 2008; Ly et al . , 2008; Hoppa et al . , 2012 ) , it remains unknown whether these auxiliary subunits acutely modulate calcium channel abundance in response to activity , as opposed to simply establishing and maintaining baseline expression levels during development and throughout the life of a synapse . As such , exploration of how calcium channel abundance is modulated during PHD should be of considerable interest . It is interesting to note that while our measurements of presynaptic calcium channel abundance are restricted to the active zone , defined by the presence of Brp labeling , we did not observe an increase in non-synaptic CaV2 . 1-GFP signal . Furthermore , because we measured CaV2 . 1-GFP driven from a transgene under control of the GAL4-UAS expression system , whatever mechanisms are at play during PHD likely act downstream of calcium channel transcription . Therefore , we hypothesize that PHD is driven either by restricting CaV2 . 1 abundance at the level of translation , or by the regulated internalization and degradation of presynaptic CaV2 . 1 calcium channels . We have documented a dramatic decrease in presynaptic calcium channel abundance during PHD , approaching a 50% decrease in total active-zone associated CaV2 . 1-GFP levels . We acknowledge that we are visualizing transgenically over-expressed channels , tagged with GFP , rather than the endogenous protein . However , the documented ∼50% decrease in CaV2 . 1-GFP abundance is mirrored , quantitatively , by a ∼50% drop in the spatially averaged presynaptic calcium signal in response to single action potential stimulation . Furthermore , baseline neurotransmitter release is unaltered by CaV2 . 1-GFP overexpression , consistent with prior observations ( Kawasaki et al . , 2004; Wang et al . , 2014 ) , suggesting that the CaV2 . 1-GFP containing channels replace endogenous , untagged channels with little or no change in total calcium channel number at the active zone . Based upon these considerations , we argue that our measurements of decreased calcium channel number and decreased calcium influx are an accurate reflection of what is happening at the active zone during PHD . We are also confident of our estimation of presynaptic quantal content in vGlut-OE animals . There is a ∼50–60% increase in quantal size that is precisely offset by a ∼30% decrease in presynaptic vesicle release ( quantal content , Figure 1 ) . These measurements are consistent with the originally described phenotype of vGlut-OE animals ( Daniels et al . , 2004 ) . Furthermore , we have replicated and extended the finding that PHD is robustly expressed over a range of extracellular calcium concentrations ( 0 . 3–3 mM ) . Even though our measurements of calcium channel number and calcium influx are internally consistent , and our measurements of quantal content during PHD corroborate prior publications , there remains a discrepancy . The magnitude of the decrease in calcium channel abundance and presynaptic calcium influx should cause a much larger decrease in quantal content that what we observe during PHD . More specifically , the relationship between release and external calcium is a power function with an exponent of approximately three at the Drosophila NMJ ( Dickman and Davis , 2009; Müller et al . , 2011 ) . Therefore , a 50% drop in the spatially averaged calcium signal should decrease release by more than the observed 30% . The reason for this discrepancy remains unknown . We recently documented a linear relationship between the RRP and extracellular calcium at this synapse ( Müller et al . , 2015 ) , consistent with similar observations at the mammalian calyx of held ( Thanawala and Regehr , 2013 ) . One possibility , therefore , is that the mechanisms of PHD include two inter-dependent processes . First , there is a drop in calcium channel abundance . Second , there is a mechanism that potentiates the RRP to achieve precise homeostatic compensation . This hypothetical , calcium-independent , potentiation of the RRP would offset the expected reduction in the RRP attributable to decreased calcium influx and result in an RRP measurement that does not appear to change during PHD . If this model is correct , it should be possible to isolate mutations that interfere with the latter mechanism and cause presynaptic release to drop below the levels normally seen for precise PHD . This is not observed when a rim mutation is placed in the background of vGlut-OE animals , indicating that any such mechanism must be independent of the activity of RIM . Alternatively , recent evidence indicates that increasing vesicular size may directly increase Pr at hippocampal synapses ( Herman et al . , 2014 ) , suggesting another model . In this model , the overexpression of vGlut2 , which causes increased vesicle diameter ( Daniels et al . , 2004 ) , would cause a concomitant increase in Pr . Because this increase in Pr would occur in addition to increased quantal size , one would expect a much larger compensatory response as compared to perturbing quantal size alone , therefore explaining the apparent outsized reduction in calcium influx . This model would also explain why only a mild reduction in Ptrain is observed electrophysiologically at vGlut-OE synapses , despite the large decrease in presynaptic calcium influx . Further exploration of PHD should clarify whether vGlut2 overexpression itself increases Pr and/or whether additional mechanisms exist to oppose CaV2 . 1 down-regulation . Application of PhTx to vGlut-OE animals results in an NMJ that releases enlarged vesicles , but with resulting mEPSPs that are smaller than normal . In the presence of both perturbations , the homeostatic signaling system is able to restore EPSP amplitudes to precisely wild type levels . Because PhTx induces homeostatic potentiation in 10 min , and because there does not appear to be a reservoir of CaV2 . 1 channels present extra-synaptically in vGlut-OE animals , it is unlikely that there is sufficient time to repopulate all of the active zones at the NMJ with calcium channels , thereby erasing PHD and allowing the system to function solely according to PHP . Thus , the correct restoration of EPSP amplitude likely occurs through the summation of PHP and PHD . In this model , the overexpression of vGlut induces PHD and establishes a new steady state marked by increased mEPSP amplitude and decreased calcium channel abundance , with subsequent PhTx application independently inducing PHP through trafficking of ENaC channels to the presynaptic membrane . This model predicts that the opposite order of perturbation , early induction of PHP with a subsequent acute induction of PHD , would have a similar outcome , though the reagents to test this possibility do not yet exist . Both PHP and PHD presumably require retrograde signaling from the muscle to the motor neuron to drive the observed changes in presynaptic release . Diverse retrograde , trans-synaptic signaling systems have been observed including endocannabinoids at mammalian synapses ( Kreitzer and Regehr , 2001 ) and other forms of target-derived signaling at Aplysia sensory-motor synapses ( Bao et al . , 1998; Antonov et al . , 2003; Hu et al . , 2006; Cai et al . , 2008 ) . A unique aspect of homeostatic plasticity is that retrograde signaling accurately and persistently adjusts presynaptic release to offset a perturbation . As such , the retrograde signaling system ( s ) seem to have the capacity to convey quantitative information regarding the magnitude of the postsynaptic perturbation ( Davis , 2013 ) . At the Drosophila NMJ , recent evidence implicates Endostatin as a trans-synaptic signaling molecule necessary for homeostatic plasticity ( Wang et al . , 2014 ) , but little else is known . Ultimately , identifying the retrograde signals responsible for PHP and PHD and defining how these signals interact , will be necessary to understand how homeostatic , rheostat-like control of presynaptic release is achieved . In conclusion , our data define a previously unknown layer of complexity driving presynaptic homeostatic plasticity , with parallel pathways acting to correct bi-directional perturbation . Moreover , we identify a novel mechanism of synaptic plasticity that is driven by tight regulation of presynaptic calcium channel abundance . Given the importance of voltage-gated calcium channels in controlling synaptic transmission , as well as the numerous human diseases linked to their dysfunction ( Ophoff et al . , 1996; Zhuchenko et al . , 1997; Striessnig et al . , 2010 ) , further investigations into the mechanisms of CaV2 . 1 regulation during PHD should be of broad relevance . Drosophila stocks were maintained at 22–25°C on normal food . The w1118 strain was used as a wild-type control , matching the genetic background of all transgenic lines used in this study . UAS-vGlut2 flies were obtained from Aaron DiAntonio . In all cases , vGlut2 overexpression was driven by OK371-Gal4 . rim103 mutant animals were previously generated ( Müller et al . , 2012 ) . A plasmid containing Archaerhodopsin3 ( Arch ) with a C terminal GFP tag was obtained ( Addgene plasmid 22217 ) . The coding sequence for Arch was cloned into the pUAST destination vector to generate UAS-Arch . This construct was confirmed by sequencing . Transgenic flies were generated using standard injection methods by BestGene Inc . Stocks containing UAS-Arch insertions on chromosome two and three were used for experiments . All imaging and electrophysiology experiments on Arch expressing flies were performed using two copies of UAS-Arch ( 2 on chromosome II , or 1 on II and 1 on III ) and one copy of the motor neuron driver OK371-GAL4 . For Arch imaging experiments , crosses were set up and allowed to lay for 2–3 days on food containing 1 mM all-trans retinal ( ATR ) . Crosses containing ATR food were wrapped in foil and kept at 25°C . All other flies were obtained from the Bloomington Drosophila Stock Center . Sharp-electrode recordings were made from muscle six in abdominal segments two and three of third instar larvae using an Axopatch 200B , or a Multiclamp 700B amplifier ( Axon Instruments ) , as previously described ( Davis and Goodman , 1998 ) . Two-electrode voltage clamp recordings were performed with an Axoclamp 2B amplifier . The extracellular HL3 saline contained ( in mM ) : 70 NaCl , 5 KCl , 10 MgCl2 , 10 NaHCO3 , 115 sucrose , 4 . 2 trehalose , 5 HEPES , and various concentrations of CaCl2 ( see ‘Results’ and Figures ) . To assess PHP , larvae were incubated in Philanthotoxin-433 ( PhTX; 20 µM; Sigma–Aldrich , St . Louis , MO ) for 10 min ( Frank et al . , 2006 ) . The average single AP-evoked EPSP amplitude ( stimulus duration , 3 ms ) , or EPSC amplitude ( stimulus duration , 1 ms ) of each recording is based on 30 presynaptic stimuli . Quantal content is given by the ratio of the average EPSP amplitude over average mEPSP amplitude of a recording , and then averaging recordings across all NMJ for a given genotype . The apparent size of the readily-releasable vesicle pool ( RRP ) was probed by the method of cumulative EPSC amplitudes ( Schneggenburger et al . , 1999 ) , which was recently applied to the Drosophila NMJ ( Hallermann et al . , 2010; Miśkiewicz et al . , 2011; Weyhersmüller et al . , 2011; Müller et al . , 2012 ) . Muscles were clamped to their resting membrane potential ( Vm ) , or clamped to −65 mV if the Vm was more positive than −60 mV . Muscles with a Vm more depolarized than −55 mV were discarded . Synapses were stimulated with 60-Hz trains ( 30 stimuli , at least five trains per synapse ) . EPSC amplitudes during a stimulus train were calculated as the difference between peak and baseline before stimulus onset of a given EPSC . The cumulative EPSC amplitude was obtained by back-extrapolating a line fit to the last 10 cumulative EPSC amplitude values of the 60 Hz train to time 0 . Electrophysiology data were acquired with Clampex ( Axon Instruments , Foster City , CA ) , and imaging data were recorded with Prairie View ( Prairie Technologies , Middleton , WI ) . Data were analyzed with custom-written routines using Igor Pro 6 . 2 . 2 . ( Wavemetrics; custom script submitted with this manuscript ) , and mEPSPs were analyzed with Mini Analysis 6 . 0 . 3 . ( Synaptosoft , Decatur , GA ) . Statistical significance was assessed with a Student's t-test , and all error bars are SEM . Ca2+ imaging experiments were done as described in Müller and Davis ( 2012 ) . Third instar larvae were dissected and incubated in ice cold , Ca2+-free HL3 containing 5 mM Oregon-Green 488 BAPTA-1 ( OGB-1; hexapotassium salt , Invitrogen ) and 1 mM Alexa 568 ( Invitrogen ) . After incubation for 10 min , the preparation was washed with ice cold HL3 for 10–15 min . This leads to an intraterminal OGB-1 concentration of approximately 50 µM ( Müller and Davis , 2012 ) . Single action-potential evoked spatially-averaged Ca2+ transients were measured in type-1b boutons synapsing onto muscle 6/7 of abdominal segments A2/A3 at an extracellular [Ca2+] of 1 mM using a confocal laser-scanning system ( Ultima , Prairie Technologies ) at room temperature . Excitation light ( 488 nm ) from an air-cooled krypton-argon laser was focused onto the specimen using a 60× objective ( 1 . 0 NA , Olympus ) , and emitted light was detected with a gallium arsenide phosphide-based photocathode photomultiplier tube ( Hamamatsu ) . Line scans across single boutons were made at a frequency of 313 Hz . Fluorescence changes were quantified as ΔF/F = ( F ( t ) − Fbaseline ) / ( Fbaseline-Fbackground ) , where F ( t ) is the fluorescence in a region of interest ( ROI ) containing a bouton at any given time , Fbaseline is the mean fluorescence from a 300-ms period preceding the stimulus , and Fbackground is the background fluorescence from an adjacent ROI without any indicator-containing cellular structures . One synapse ( 4–12 boutons ) was imaged per preparation . The average Ca2+ transient of a single bouton is based on 8–12 line scans . Experiments in which the resting fluorescence decreased by > 15% , and/or which had a Fbaseline > 650 a . u . were excluded from analysis . Data of experimental and control groups were collected side by side . The Ca2+ indicator was not saturated by single AP stimulation because repetitive stimulation induced a further increase in the peak ΔF/F amplitude . Confocal spot imaging was made from type 1b boutons on muscle 6/7 of abdominal segments 2–4 of third instar larvae using a confocal laser-scanning microscope ( Ultima , Prairie Technologies ) . Excitation light ( 643 nm , 4 mW at back of objective ) from an air-cooled solid-state laser was focused onto the specimen using a 60× objective ( 1 . 0 NA , Olympus ) . The emission path consisted of a quad band 405 , 488 , 515 , 643 nm dichroic , 500 nm long pass filter , 600 nm dichroic , and 700/75 nm and 525/40 nm band pass emission filters for Arch and GFP emissions , respectively . Arch emission was detected with a gallium arsenide phosphide-based photocathode photomultiplier tube ( Hamamatsu ) and GFP emission was detected by a second PMT . Spot imaging from edges of single boutons was made at a sampling frequency of 4 kHz . A 1 ms stimulation of the nerve fiber was used to evoke action potentials after 60 ms of baseline image acquisition . 30–50 events were collected for each bouton at an inter-stimulus rate of 0 . 5 Hz for single action potentials and an inter-train interval of 0 . 1 Hz for trains of five stimuli . 1–5 boutons per synapse were imaged . For experiments using 1 . 5 mM calcium , 10 μM Philanthotoxin-433 ( Sigma ) or 100 μM 1-Naphthylacetyl spermine ( Sigma ) was included in the saline to prevent contraction during stimulation . Imaging data were analyzed using custom-written routines in Matlab ( Mathworks; custom script submitted with this manuscript ) and digitally filtered at 2 kHz . Fluorescence signals consisted of an exponentially decaying signal that derives from tissue fluorescence and is independent of Arch fluorescence , a rapid increase in Arch fluorescence that occurs during the first 10 ms of the photocycle ( Maclaurin et al . , 2013 ) , and the voltage dependent change in Arch fluorescence . To isolate voltage dependent changes in fluorescence , we fit a single exponential from 10 ms after the start of imaging to 5 ms before stimulus onset . This fit , which approximates the tissue fluorescence as well as the baseline Arch fluorescence , was extrapolated and subtracted from the fluorescence . Traces containing extra action potentials were removed . The resulting fluorescence measurements were averaged for each bouton and the average waveform was used to determine peak amplitude , full width , and width at half maximum ( WHM ) . To obtain precise width and WHM measurements , we used linear interpolation between data sampling points . Due to the inability to accurately measure baseline Arch fluorescence in the presence of tissue fluorescence and Arch localized to internal membranes , amplitude measurements are not reported . Average AP waveforms for each experimental condition were peak-aligned and normalized before averaging . Standard immunocytochemistry was performed as previously described . Dissected third instar larvae were fixed with ice-cold ethanol for 5′ . The following primary antibodies were used: mouse anti-Brp ( 1:100 ) ( Kittel et al . , 2006 ) , and rabbit anti-GFP ( 1:1 , 000 , Invitrogen clone 3E6 ) . Alexa-conjugated secondary antibodies were used at 1:300 ( Jackson Immuno-research Laboratories ) .	Neurons are organised into networks via junctions called synapses . The arrival of an electrical signal , called an action potential , at a neuronal synapse causes an influx of calcium ions into the cell . This in turn causes packages of chemicals called neurotransmitters that are stored inside the cell to fuse with the neuron's membrane , which releases their contents from the neuron . These molecules bind to the cell on the other side of the synapse ( another neuron or a muscle cell ) , and the action potential can be regenerated . Most biological systems are maintained within an optimal range , even in the context of a constantly changing external environment . The nervous system is no exception . The communication across synapses can be carefully controlled to ensure that action potentials pass through a neural network in a reliable way . This means that deviations away from an optimal amount of synaptic transmission trigger changes that compensate for the deviation and aim to restore the optimum . When these compensatory changes increase synaptic transmission , the process is referred to as ‘homeostatic potentiation’; when they reduce it , this is known as ‘homeostatic depression’ . Gaviño et al . studied the synapses between neurons and muscle cells in the larvae of fruit flies that had been genetically modified to produce abnormally large packages of neurotransmitters . In a clear demonstration of homeostatic depression , the larvae compensate for their altered condition by releasing fewer of their extra large packages in response to each action potential . Mutations that disrupt homeostatic potentiation have no effect on this process , suggesting that the two processes work via separate mechanisms . Indeed , further experiments reveal that synapses achieve homeostatic depression by reducing the number of calcium channels in the membrane . This limits the entry of calcium—and thus the release of neurotransmitters—in response to each action potential . Homeostatic potentiation and depression are independent processes with distinct mechanisms , and which work in combination to allow fine-grained control of communication across synapses . Future work is now needed to determine how the mechanisms of homeostatic depression and potentiation are coordinated to precisely control neural activity , and if these mechanisms are also commonly employed at synapses in the mammalian brain .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Homeostatic synaptic depression is achieved through a regulated decrease in presynaptic calcium channel abundance
HIV encodes Tat , a small protein that facilitates viral transcription by binding an RNA structure ( trans-activating RNA [TAR] ) formed on nascent viral pre-messenger RNAs . Besides this well-characterized mechanism , Tat appears to modulate cellular transcription , but the target genes and molecular mechanisms remain poorly understood . We report here that Tat uses unexpected regulatory mechanisms to reprogram target immune cells to promote viral replication and rewire pathways beneficial for the virus . Tat functions through master transcriptional regulators bound at promoters and enhancers , rather than through cellular ‘TAR-like’ motifs , to both activate and repress gene sets sharing common functional annotations . Despite the complexity of transcriptional regulatory mechanisms in the cell , Tat precisely controls RNA polymerase II recruitment and pause release to fine-tune the initiation and elongation steps in target genes . We propose that a virus with a limited coding capacity has optimized its genome by evolving a small but ‘multitasking’ protein to simultaneously control viral and cellular transcription . Transcription of protein coding genes by RNA Polymerase ( Pol ) II is regulated at several steps including initiation , elongation and termination ( Adelman and Lis , 2012; Fuda et al . , 2009; Grunberg and Hahn , 2013; Sims et al . , 2004; Zhou et al . , 2012 ) . Transcription factors coordinate the activation and/or repression of key regulatory programs by acting at one or multiple steps in this regulatory circuitry . While one group of transcription factors recruit Pol II to their target genes to induce transcription initiation , another class functions by promoting the transcriptional pause release from the promoter-proximal state to allow Pol II transition to the productive elongation phase ( Feinberg et al . , 1991; Gomes et al . , 2006; Peterlin and Price , 2006; Rahl et al . , 2010; Zhou et al . , 2012 ) . Thus , promoter-proximal pausing has been identified as a general feature of transcription control by Pol II in metazoan cells and a key regulatory step during differentiation , cell development and induction of stem cell pluripotency ( Adelman and Lis , 2012; Core and Lis , 2008; Smith and Shilatifard , 2013; Zeitlinger et al . , 2007; Zhou et al . , 2012 ) . In addition to Pol II and the basal transcription machinery , the epigenetic landscape plays another critical role in transcription regulation , mediated by covalent modifications to the N-terminal tails of histones . The study of chromatin modifications has revealed fundamental concepts in the regulation of transcription . Histone marks , in general , associate with different genomic domains ( promoters , coding units and enhancers ) and provide evidence of their transcriptional status ( Barski et al . , 2007; Creyghton et al . , 2010; Guenther et al . , 2007; Li et al . , 2007; Tessarz and Kouzarides , 2014; Zhou et al . , 2011 ) . While active promoters are marked with histone 3 lysine 4 trimethylation ( H3K4me3 ) , active transcription units are marked with H3K79me3 and H3K36me3 ( Kouzarides , 2007; Li et al . , 2007 ) . Thus , most Pol II regulated genes ( protein-coding and long non-coding RNAs ) contain a K4/K36 signature that positively correlates with active gene expression ( Guttman and Rinn , 2012; Martin and Zhang , 2005 ) . In addition to promoters , distal genomic elements referred to as enhancers modulate gene activity through gene looping or long-range chromatin interactions and further regulate the location , timing , and levels of gene expression ( Bulger and Groudine , 2011 ) . The genomic locations of these enhancers ( inter- or intra-genic ) usually correlate with high H3K4me1 and low H3K4me3 content and their activity is proportional to the levels of H3K27Ac ( Bulger and Groudine , 2011; Creyghton et al . , 2010; Kim et al . , 2010b; Kowalczyk et al . , 2012; Smallwood and Ren , 2013 ) . Additionally , H3K27Ac also appears to modulate two temporally separate events: it enhances the search kinetics of transcriptional activators and accelerates the transition from initiation into elongation leading to a robust and potentially tunable transcriptional response ( Stasevich et al . , 2014 ) . Viruses have evolved strategies to precisely orchestrate shifts in regulatory programs . In addition to sustaining transcription of their own genomes , certain viral transcription factors directly or indirectly alter existing cellular programs in ways that promote viral processes ( replication and spread ) ( Ferrari et al . , 2008; Horwitz et al . , 2008 ) and/or modulate programs in the infected target cell . It is well established that the HIV Tat protein relieves promoter-proximal pausing at the viral promoter by recruiting the positive transcription elongation factor b ( P-TEFb ) to the trans-activating RNA ( TAR ) stem-loop formed at the 5’-end of viral nascent pre-mRNAs ( D'Orso and Frankel , 2010; Mancebo et al . , 1997; Ott et al . , 2011; Zhou et al . , 2012; Zhu et al . , 1997 ) . Recently , it was discovered that Tat recruits P-TEFb as part of a larger complex referred to as super elongation complex ( SEC ) , which is composed by the MLL-fusion partners involved in leukemia ( AF9 , AFF4 , AFF1 , ENL , and ELL ) , and PAF1 . Although SEC formation relies on P-TEFb , optimal P-TEFb kinase activity towards the Pol II C-terminal domain ( CTD ) is AF9 dependent , and the MLL-fusion partners and PAF1 are required for Tat transactivation ( He et al . , 2010; Luo et al . , 2012; Sobhian et al . , 2010 ) . Moreover , Tat stimulates transcription complex assembly through recruitment of TATA-binding protein ( TBP ) in the absence of TBP-associated factors ( TAFs ) ( Raha et al . , 2005 ) , implying that Tat controls both the initiation and elongation steps of transcription , in agreement with early proposals of Tat increasing transcription initiation , stabilizing elongation and precluding anti-termination ( Kao et al . , 1987; Laspia et al . , 1989; Rice and Mathews , 1988 ) . Thus , Tat has the ability to control multiple stages in the HIV transcriptional cycle to robustly increase transcript synthesis to promote viral replication . In addition to controlling HIV transcription , Tat appears to modulate cellular gene expression to generate a permissive environment for viral replication and spread , and alter or evade immune responses ( Izmailova et al . , 2003; Kim et al . , 2010a; Kim et al . , 2013; Lopez-Huertas et al . , 2010; Marban et al . , 2011 ) . One example of Tat-mediated down-regulation is the mannose receptor in macrophages and immature dendritic cells , which plays a key role in host defense against pathogens by mediating their internalization ( Caldwell et al . , 2000 ) . A similar case is Tat’s repression of the MHC class I gene promoter , which depends on Tat binding to complexes containing the TBP-associated factor TAFII250/TAF1 ( Weissman et al . , 1998 ) . The interaction of the C-terminal domain of Tat and TAF1 suggests that Tat mediates repression functions through the transcription initiation complex . These , and many other examples , suggest that Tat is capable of altering cellular gene expression via association with factors bound to promoters . More recently , chip-on-chip and chromatin immunoprecipitation sequencing ( ChIP-seq ) approaches have revealed that Tat has the ability to bind target genes in the human genome . While these previous studies have provided early glimpses about Tat recruitment to host cell chromatin , a comprehensive description of the direct target genes and the molecular mechanisms remain poorly understood . To address those gaps in knowledge , we aimed at: ( i ) identifying Tat target genes in the human genome , ( ii ) delineating the molecular mechanisms by which Tat reprograms cellular transcription , and ( iii ) defining how Tat is recruited to host cell chromatin . One of the main challenges in the field is to obtain a high-quality , comprehensive ChIP-seq that will provide functional insights . To address this challenge , we analyzed the genome-wide distribution of Tat in the human genome using a technically improved ChIP-seq compared to the previous studies . Moreover , we investigated the molecular mechanisms using a global analysis of chromatin signatures that demarcate the position and activity of genomic domains ( enhancers , promoters and coding units ) , as well as Pol II recruitment and activity ( Barski et al . , 2007; Creyghton et al . , 2010 ) . We provide , for the first time , evidence that the direct Tat target genes share functional annotations and are regulated by common master transcriptional regulators such as T-cell identity factors . While the Tat stimulated genes ( TSG ) show a positive role in activating T cells , favoring cell proliferation to promote viral replication and spread , the Tat down-regulated genes ( TDG ) show critical roles in blunting immune system responses , nucleic acid biogenesis ( splicing and translation ) , and proteasome control . Strikingly , Tat functions as both activator and repressor by modulating Pol II recruitment and/or pause release as well as controlling the activity of chromatin-modifying enzymes to reprogram the epigenetic landscape of the host cell . Taken together , we propose that a virus with a limited coding capacity has optimized its genome by evolving a small protein ( Tat ) to perform multiple functions throughout the viral life cycle . Beyond controlling HIV transcription , Tat has evolved unique properties to occupy precise genomic domains ( promoters and enhancers ) to reprogram cellular transcription using unexpected regulatory mechanisms . We provide the molecular basis of an unprecedented paradigm with critical roles in host-pathogen interactions . Previous studies have proposed that Tat modulates cellular gene expression to generate an environment hospitable for viral replication and spread ( Kim et al . , 2010a , 2013; Li et al . , 1997; Lopez-Huertas et al . , 2010; Marban et al . , 2011 ) . However , the molecular mechanisms remain poorly understood . To elucidate how Tat performs these functions we aimed at: ( i ) identifying genomic domains occupied and regulated by Tat in the human genome , and ( ii ) defining the mechanisms of transcriptional control . We generated high-quality chromatin immunoprecipitation sequencing ( ChIP ) -seq datasets in two Jurkat CD4+ T cell lines , inducibly expressing FLAG-tagged Tat or green fluorescent protein ( GFP ) used as negative control ( Figure 1A ) . Importantly , the cell line used expresses low and physiologic levels of Tat , which mimic those detected during HIV infection ( Figure 1—figure supplement 1 ) . We utilized this minimalistic system in order to more precisely examine the effects of Tat alone rather than in an infection setting . The latter approach may compromise the investigation due to the introduction of other viral products that may also affect cellular behavior and/or alter Tat activity ( Frankel and Young , 1998 ) . 10 . 7554/eLife . 08955 . 003Figure 1 . Genomic domains occupied and regulated by Tat in CD4+ T cells . ( A ) Western blot of Jurkat-GFP and -Tat cell lines treated ( + ) or not ( – ) with DOX using the indicated antibodies . ( B ) Genome-wide distribution of Tat across the human genome . ( C ) Integration of the FLAG ChIP-seq and RNA-seq datasets defines a set of genes directly regulated by Tat . ( D ) Validation of the RNA-seq dataset using qRT-PCR on the indicated TSG , TDG or non-target genes as negative controls ( mean ± SEM; n = 3 ) . ( E ) Individual tracks showing FLAG ChIP-seq and the corresponding RNA-seq dataset in the GFP and Tat cell lines . ( F ) Functional annotation of biological processes enriched at TSG and TDG . This figure is associated with Figure 1—figure supplements 1–10 . Direct targets , genes directly bound and regulated by Tat; ChIP-seq , chromatin immunoprecipitation sequencing; DOX , doxycycline; GFP , green fluorescent protein; RNA-seq , RNA sequencing; qRT-PCR , quantitative real time polymerase chain reaction; TDG , Tat downregulated genes; TSG , Tat stimulated genes; TSS , transcription start site . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 00310 . 7554/eLife . 08955 . 004Figure 1—figure supplement 1 . Tat protein expression levels in the Jurkat Tat-SF model matches the levels of Tat detected during HIV infection . Western blot ( FLAG and Tat ) of Jurkat Tat-SF cell line treated with ( + ) or without ( – ) DOX , and Jurkat cells latently infected with HIV ( clone E4 ) induced with ( + ) or without ( – ) TNF-α for 24 hr . A β-actin western blot is shown as loading control . DOX , doxycycline; TNF-α , tumor necrosis factor . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 00410 . 7554/eLife . 08955 . 005Figure 1—figure supplement 2 . Technical improvement of Tat ChIP-seq in CD4+ T cells . ( A ) Comparison of ChIP-seq Tat peak numbers in the Jurkat-GFP and -Tat cell lines using the Tat and FLAG antibodies . The number of Tat peaks in a similar Jurkat-Tat cell line from the study of Marban et al . , ( 2011 ) is indicated . ( B ) Overlay of Tat peaks from this study and those identified by Marban et al . , ( 2011 ) using ChIP-seq . The number of common Tat peaks between our ChIP-seq data and Marban’s at a distance of <0 . 1 kb or <1 kb from the Tat sites is shown . ( C–E ) Genome browser views of FLAG ChIP-seq tracks in the GFP and Tat cell lines along with the FLAG peaks called by MACS in the Tat cell line and the Tat track by Marban et al . , ( 2011 ) . ChIP-seq , chromatin immunoprecipitation sequencing; GFP , green fluorescent protein; MACS , model-based analysis of ChIP-seq . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 00510 . 7554/eLife . 08955 . 006Figure 1—figure supplement 3 . Non-functional Tat mutants have compromised chromatin interaction and modulation of cellular gene expression . ( A ) Scheme of Tat showing the position of its domains ( AD , RBD and Ct ) along with the location of the mutated residues . ( B ) Western blots of Jurkat CD4+ T cell lines expressing GFP , wild-type Tat or non-functional mutants ( C22A , K50Q , R52R53K ) ( D'Orso et al . , 2012 ) . Cells from panel ( B ) were used in ChIP assays to analyze the occupancy of GFP , Tat or the non-functional mutants at class I TSG promoters ( C ) , class II TSG promoters ( D ) , class I TDG promoters ( E ) and class II TDG promoters ( F ) . Values representing the average of three independent experiments ( mean ± SEM; n = 3 ) . Cells from panel ( B ) were used to isolate total RNA and the expression of class I TSG ( G ) , class II TSG ( H ) , class I TDG ( I ) and class II TDG ( J ) was measured by qRT-PCR , normalized to RPL19 , and plotted as fold RNA change over the GFP line arbitrarily set at 1 ( mean ± SEM; n = 3 ) . AD , activation domain; ChIP-seq , chromatin immunoprecipitation sequencing; Ct , C-terminal domain; GFP , green fluorescent protein; qRT-PCR , quantitative real time polymerase chain reaction; RBD , RNA-binding domain; TDG , Tat downregulated genes; TSG , Tat stimulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 00610 . 7554/eLife . 08955 . 007Figure 1—figure supplement 4 . Tat-induced transcriptome changes are also observed at the protein level . Flow cytometry analysis of Jurkat-GFP and -Tat cell lines induced with doxycycline for 24 hr and stained with CD4-APC and CD69-PE antibodies or unstained ( negative control ) to monitor levels of CD4 and CD69 proteins expressed at the cell surface . Note the increase in the CD69+ population in the presence of Tat . GFP , green fluorescent protein . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 00710 . 7554/eLife . 08955 . 008Figure 1—figure supplement 5 . Distribution of Tat occupancy at promoter and/or intragenic domains in TSG and TDG . The percentage of sequences from the total is indicated . TDG , Tat downregulated genes; TSG , Tat stimulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 00810 . 7554/eLife . 08955 . 009Figure 1—figure supplement 6 . FLAG ChIP-qPCR analysis on the indicated genomic loci . ChIP assay to analyze the distribution of Tat or GFP at the CD69 ( A ) , ADCYAP1 ( B ) , CD1E ( C ) and RAG1 ( D ) locus in the GFP ( green ) and Tat ( black ) cell lines . The position of the amplicons used in ChIP-qPCR and their distance to the TSS ( arrow ) is shown with the schematic of each locus . The schemes are not in real scale . Values represent the average of three independent experiments ( mean ± SEM; n = 3 ) . ChIP , chromatin immunoprecipitation; GFP , green fluorescent protein; qPCR , quantitative polymerase chain reaction; SEM , standard error of the mean; TSS , transcription start site . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 00910 . 7554/eLife . 08955 . 010Figure 1—figure supplement 7 . The genes modulated by ectopic expression of Tat are also detected during a time-course HIV infection experiment . ( A ) Jurkat T cells were infected with HIV ( NL4-3 ) and levels of p24/Capsid protein was quantified using ELISA at different time points post-infection ( 0 , 3 and 7 hr; 1 , 2 , 4 , 6 , 8 , 10 , and 12 days ) . Values represent the average of three independent experiments ( mean ± SEM; n = 3 ) . Cells from panel ( A ) were used to isolate total RNA and the expression of three TSG: CD69 ( B ) , FAM46C ( C ) , and PPM1H ( D ) ; and three TDG: CD1E ( E ) , EOMES ( F ) and FBLN2 ( G ) normalized to RPL19 was measured by qRT-PCR and plotted as fold RNA change over the GFP cell line arbitrarily set at 1 ( mean ± SEM; n = 3 ) . The points in the curve were fitted to a non-linear regression in GraphPad Prism . ELISA , enzyme-linked immunosorbent assay; GFP , green fluorescent protein; HIV , human immunodeficiency virus; qRT-PCR , quantitative real time polymerase chain reaction; SEM , standard error of the mean; TSG , Tat stimulated genes; TDG , Tat downregulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 01010 . 7554/eLife . 08955 . 011Figure 1—figure supplement 8 . HIV infection of central memory CD4+ T cells triggers deregulation of TSG and TDG detected in the genome-wide approaches . ( A ) Scheme of the pipeline used to generate primary central memory T cells ( TCM ) and infect with replication competent HIV to identify differentially expressed genes . ( B ) qRT-PCR analysis on the indicated class I and II TSG , TDG and non-target genes ( mean ± SEM; n = 3 ) . Cells from panel ( A ) were used to isolate total RNA and the expression of initiating ( In ) and elongating ( El ) transcripts for class I TSG ( C ) , class II TSG ( D ) , class I TDG ( E ) and class II TDG ( F ) was measured by qRT-PCR , normalized to RPL19 , and plotted as fold RNA change: HIV infection/mock infection ( mean ± SEM; n = 3 ) . HIV , human immunodeficiency virus; qRT-PCR , quantitative real time polymerase chain reaction; TDG , Tat downregulated genes; TSG , Tat stimulated genes; SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 01110 . 7554/eLife . 08955 . 012Figure 1—figure supplement 9 . Response network of TSG and TDG . A response network was constructed based on RNA-seq data using a p-value cut off p = 0 . 01 . The network consists of 138 nodes and 161 edges . Multiple edges between nodes indicate multiple evidence from different protein-protein interaction datasets ( [Human Protein Reference Database] [Goel et al . , 2011]; IntAct [Kerrien et al . , 2012]; NetworKin , BHMRSS , CORUM , Hynet [Konig et al . , 2010] and NCBIs HIV-1 interaction databases ) . Green and red nodes denote TSG and TDG , respectively . Edges are directional according to the respective databases . Edges with circles as ‘arrow tips’ denote phosphorylation reactions , ‘diamond shaped tips’ refer to general activation , and ‘arrows’ describe other reactions . Non-directional edges indicate binding . HIV , human immunodeficiency virus; RNA-seq , RNA sequencing; TDG , Tat downregulated genes; TSG , Tat stimulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 01210 . 7554/eLife . 08955 . 013Figure 1—figure supplement 10 . Enrichment of TSG and TDG in publicly available datasets of differentially expressed genes identified in HIV infection and replication experiments . ( A ) The enrichment analysis with 48 publicly available differentially expressed gene datasets identified in HIV and replication experiments and 14 HIV relevant pathways from MSigDB are shown . ( B ) The functional annotation by the MSigDB canonical pathways ( set c2 . cp ) is depicted . Twelve significantly TSG enriched gene sets and 43 significantly TDG enriched gene-sets have been identified . P-values are Bonferroni adjusted for multiple testing . The vertical line indicates FDR = 0 . 05 . FDR , false discovery rate; HIV , human immunodeficiency virus; MSigDB , Molecular Signatures Database; TDG , Tat downregulated genes; TSG , Tat stimulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 013 We established a ChIP-seq pipeline using both a Tat antibody ( Ab ) previously used in ChIP-seq ( Marban et al . , 2011 ) and a FLAG Ab . We identified genomic domains enriched with statistically significant occupancy relative to input DNA using the model-based analysis of ChIP-seq ( MACS ) algorithm with a stringent cutoff ( FDR<0 . 05 ) ( Zhang et al . , 2008 ) . Unexpectedly , we did not find a significant number of overlapping peaks between the Tat and FLAG ChIP-seq datasets , suggesting that the Tat Ab performs poorly in ChIP , providing a molecular explanation for the low-quality dataset generated by Marban et al . ( Marban et al . , 2011 ) ( Figure 1—figure supplement 2A–E ) . We thus focused on FLAG , which enabled us to generate a better quality , more comprehensive ChIP-seq dataset compared with previous studies ( Kim et al . , 2010a; Marban et al . , 2011 ) , which served as the key basis to elucidate novel Tat functions in the control of cellular transcription . To precisely map Tat binding sites in the human genome , we called FLAG peaks in both the Tat and GFP cell lines , then discarded those peaks in the intersection of both datasets , considering only the FLAG peaks unique to the Tat cell line to be true Tat binding events . We found that the majority of FLAG peaks in the Tat cell line represent bona fide Tat binding sites because they are not detected in the GFP cell line ( Figure 1—figure supplement 2A , C–E ) . Importantly , the ChIP signal for FLAG was barely detectable in the GFP cell line or in cells expressing inactivating Tat mutations in the activation domain ( such as C22A ) that abolish Tat recruitment to target genes and/or impair gene expression activation ( D'Orso et al . , 2012 ) ( Figure 1—figure supplement 3 ) . In contrast to the C22A non-functional Tat mutant , synonymous mutations within the RNA-binding domain ( such as K50Q and R52R53K ) have less pronounced effects on Tat binding to chromatin ( Figure 1—figure supplement 3 ) . Genome-wide distribution analysis based on the ENCODE annotation ( Consortium , 2011 ) , revealed that Tat binding is heavily enriched at promoters ( 24% , p-value 3 . 4×10-324 ) and 5’-UTR ( 1 . 2% , p-value 5 . 3 × 10-193 ) relative to their genomic proportions . These genomic domains are underrepresented in the genome ( about 1% ) making Tat’s relative abundance at these sites highly significant ( Figure 1B ) . Although Tat also binds introns ( 36 . 4% , p-value 5 . 1 × 10-4 ) and intergenic domains ( 35% , below the expected level of enrichment ) , defined as sites located more than 1 Kb from the nearest annotated gene , these two genomic domains are highly represented in the genome and the relative Tat binding frequency here is not nearly as striking as at promoters and promoter-proximal regions ( Figure 1B and Table 1 ) . 10 . 7554/eLife . 08955 . 014Table 1 . Genome-wide distribution of Tat in CD4+ T cells . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 014Genomic domainPeaksPercentGenome fractionChIP ( p-value ) Total6117100Intergenic21413552 . 7%IntragenicPromoter ( –1 kb–TSS ) 1469241 . 1%3 . 4 × 10-323Exon1282 . 11 . 9%1 . 4 × 10-3Intron222736 . 442 . 4%5 . 1 × 10-45’ UTR731 . 20 . 4%5 . 3 × 10-1933’ UTR791 . 31 . 5%1 . 4 × 10-2ChIP-seq , chromatin immunoprecipitation sequencing; TSS , transcription start site; UTR , untranslated region . To gain insights into the roles of Tat in cellular transcriptional control , we integrated the FLAG ChIP-seq dataset with a whole transcriptome generated by RNA-seq ( Figure 1C ) . RNA-seq revealed 2013 differentially expressed genes ( DEGs ) using a q-value cutoff <0 . 05 , 456 of which also appeared in the set of Tat-bound genes generated by our ChIP-seq analysis . We refer to this dataset as direct targets . Remarkably , inactivating mutations that abolish Tat recruitment to host cell chromatin also impair gene expression changes , indicating that the effect of Tat on target gene transcription is direct and requires Tat binding and activity . Specifically , reduced Tat C22A binding to promoters of Tat target genes correlates with decreased gene expression changes ( Figure 1—figure supplement 3 ) , thus providing direct evidence of Tat function . Besides the direct target genes , we identified another set referred to as indirect targets . These are genes that are differentially expressed in the presence of Tat ( RNA-seq ) , but are not directly bound by Tat ( Figure 1C ) . Tat might regulate these genes through downstream effects or alternative mechanisms ( i . e . signaling pathways ) or they could be targets not identified by ChIP-seq because of the high-confidence threshold used during peak calling . In this work , we only focused on the direct target genes to study the mechanisms of transcriptional control by Tat . Further analysis of the direct Tat targets revealed that 244 genes are up-regulated and 212 genes down-regulated , and we refer to them as Tat stimulated genes ( TSG ) and Tat downregulated genes ( TDG ) , respectively ( Figure 1C ) . Importantly , we validated the expression of several direct target genes using quantitative real-time polymerase chain reaction ( qRT-PCR ) assays , confirming the reliability of RNA-seq ( Figure 1D ) . Notably , protein expression analysis revealed that the changes detected at the RNA level are also reflected at the protein level ( for example CD69 expression at the cell surface in the presence of Tat ) indicating that gene expression changes are functional ( Figure 1—figure supplement 4 ) . We further analyzed the distribution of Tat binding sites within the direct target genes to probe for enrichment in particular genomic domains . We found that Tat is equally recruited to both promoters ( 39% ) and intragenic domains ( 39% ) at TSG , whereas the majority of occupied domains at TDG are at promoters ( 61% ) ( Figure 1—figure supplement 5 ) . Inspection of genome browser tracks showed that , in addition to binding TSG promoters ( CD69 ) , Tat is also recruited to gene body regions with enrichment at introns ( ADCYAP1 ) ( Figure 1E ) . Notably , this mode of Tat binding appears to be functionally relevant because it correlates with RNA abundance changes as revealed by RNA-seq . In addition , Tat binds the target genes using discrete ( CD69 and ADCYAP1 ) or broad ( CD1E ) distribution patterns , probably due to the different modes or mechanisms of recruitment to chromatin ( Figure 1E ) . Importantly , we have validated the FLAG ChIP-seq dataset by performing extensive ChIP-qPCR on several direct targets including two TSG ( CD69 and ADCYAP1 ) and two TDG ( CD1E and RAG1 ) in the Tat and GFP cell lines ( Figure 1—figure supplement 6 ) . Given that our model was built on the ectopic expression of Tat in target cells , we performed an infection experiment to test whether Tat reprograms cellular transcription in a similar way in the context of infection . Importantly , the TSG ( CD69 , FAM46C and PPM1H ) and TDG ( CD1E , EOMES and FBLN2 ) tested are also modulated early during HIV infection of Jurkat T cells , albeit with different kinetics ( Figure 1—figure supplement 7 ) , supporting the view that the cellular reprogramming by Tat is functional and not simply an artifact of RNA-seq or the ectopic expression of Tat outside the context of the virus . Given that we proposed that Tat effects on host cell gene expression might be relevant to normal HIV biology we asked whether the TSG and TDG are also modulated in response to viral infection of primary CD4+ T cells . To this end , we isolated naïve CD4+ T cells from the blood of healthy donors and generated central memory T cells ( TCM ) ( Figure 1—figure supplement 8A ) . After infection of TCM with replication competent X4 trophic virus ( NL-GFP ) or mock infection and sorting infected cells , we isolated RNA and performed qRT-PCR analysis on several TSG and TDG . We observed that HIV infected TCM cells showed the differential gene expression signature ( at least for the 12 direct targets examined ) that we previously observed with the ectopic expression of Tat or HIV infection of Jurkat T cells ( Figure 1—figure supplement 8B ) . This ex vivo experiment clearly demonstrates the robustness of our minimalistic setting to study Tat functions in the host cell . If the genes directly modulated by Tat are involved in biologically relevant processes then we would expect them to share functional annotations . To explore whether the TSG and TDG have any common biological functions we examined their gene ontology ( Figure 1F ) . To provide statistical robustness , we used cluster analysis and a control set of genes depleted in the Tat ChIP-seq experiment . Gene categories significantly enriched in the set of TSG include positive regulation of immune system process , cell activation and regulation of lymphocyte differentiation , while TDG include negative regulation of cell aging , regulation of myeloid cell differentiation and processes of DNA/RNA biogenesis ( Figure 1F ) . Consistently , network analysis indicates that TSG are significantly enriched in T-cell receptor ( TCR ) pathway , cell cycle and focal adhesion , while TDG enrich processes relevant for DNA/RNA processes , ribosome and proteasome control , among others ( Figure 1—figure supplement 9 ) . With respect to T-cell activation , CD69 exhibits a rather central role , because its upregulation promotes T-cell stimulation and differentiation ( TCR pathway cluster ) ( Sancho et al . , 2005 ) . Another stimulated process involves components of the cell cycle ( CDK6 ) together with cyclinD3 ( CCND3 ) and cyclin-dependent kinase inhibitor 1B ( CDKN1B ) ( cell cycle cluster ) that appear to be controlled by phosphorylation via the lymphocyte-specific protein tyrosine kinase ( LCK ) from the TCR complex , as one of the central node in the network . Another controller node assembles the ataxia-telangiectasia-mutated ( ATM ) serine/threonine kinase , which is best known for its role as an activator of the DNA damage response ( HIV infection cluster ) . The activity of HIV integrase stimulates an ATM-dependent DNA damage response , and ATM deficiency sensitizes cells to retrovirus-induced cell death . In addition , ATM inhibition is capable of suppressing the replication of both wild-type and drug-resistant HIV ( Lau et al . , 2005 ) , thus demonstrating the importance of this TSG in controlling viral processes . With respect to down-regulated processes , ribosomal proteins centered around RPS9 ( ribosome cluster ) , together with translation initiation factors ( EIF3b ) and the nucleolar and coiled-body phosphoprotein NOLC1 , as well as components of the spliceosome such as SF3B5 , SNRPB , SNRNP200 , LSM4 , and PCBP2 ( spliceosome cluster ) , suggest negative regulation of these processes ( Figure 1—figure supplement 9 ) . Because we proposed that the predicted GO biological processes of the direct Tat target genes are essential for the viral life cycle ( to promote a permissive state for viral replication ) , we further analyzed whether they are retained in the context of infection . To test this , we assembled a collection of 62 publicly available datasets including 48 gene-sets from 13 publications containing information on DEGs identified during HIV infection together with 14 datasets from the Molecular Signatures Database ( MSigDB ) of the Broad Institute ( Subramanian et al . , 2005 ) ( Supplementary file 1A ) . After executing gene-set enrichment analysis with Bonferroni correction for multiple testing , we identified five datasets that significantly enriched the TSG ( with FDR ≤ 0 . 05 ) ( Supplementary file 1B ) . In addition , TDG also enrich two gene sets from the HIV relevant MSigDB sets ( Figure 1—figure supplement 10A and Supplementary file 1B ) . Together , the analysis provides evidence that the direct Tat target genes ( and thus the predicted GO biological processes ) are retained in the context of viral infection , supporting the model that Tat-mediated host cell reprogramming occurs during infection . It is noteworthy this proposal is also consistent with our infection data on Jurkat and primary TCM cells ( Figure 1—figure supplements 7 , 8 ) . Our network analysis indicated that TSG and TDG were enriched in specific biological processes that might promote viral functions ( including replication ) ( Figure 1—figure supplements 9 , 10 ) . To answer how those functional annotations could promote viral infection , we employed a variety of methods to identify the functions of the direct Tat target genes , and relate them to the biology of HIV . Functional annotation by GO classes provides an overview of biological processes and functions enriched by the gene-sets ( Figure 1—figure supplement 10B ) . Furthermore , we have obtained additional information from the ‘canonical pathway’ collection of MSigDB . By annotating clusters of the response network with GO and MSigDB pathways we related biological processes with network content . For example , the TCR pathway cluster annotated with MSigDB includes ITK , LCK , LCP2 , PRKCQ and VAV3 , among other targets . This cluster not only reveals that those targets interact with each other , but also with the Tyrosine kinase LCK , which is known to phosphorylate both LCP2 and PRKCQ , implying physical and functional interactions . The data clearly indicates that TSG are enriched in datasets from many pathways that correlate with stimulation of viral replication , such as the ‘TCR pathway/stimulation’ ( CD3 , CD247 , INPP5D , LCP2 , and PTEN ) and ‘downstream TCR signaling/pathway’ ( PIK3CA and PRKCQ ) , ‘T-cell co-stimulation’ ( CTLA4 and CD28 ) , ‘cell motility signaling/pathway’ ( RAC1 ) , ‘generation of second messengers’ ( EVL , ITK , and LCK ) , and ‘phosphatidylinositol signaling’ ( INPP4A , INPP5D , PIK3CA , PLCB1 , PRKCB , and PTEN ) , among others ( Figure 1—figure supplement 10B and Supplementary file 1B ) . The prime example of pathways enriched in TSG is T-cell signaling/activation and co-stimulatory signals ( CD28 ) that provide additional control mechanisms to prevent inappropriate and hazardous T-cell activation . The data is consistent with the fact that , in early stages of infection , the viral encoded proteins ( particularly Tat ) mimic T-cell signaling pathways , resulting in sustained viral replication within infected T cells . This T-cell activation provides new targets for HIV replication , creating a favorable environment for further virus-mediated damage to the immune system and chronic consumption of the pools of naïve and resting memory cells ( Hazenberg et al . , 2000; Hellerstein et al . , 2003 ) . Our data suggest that HIV uses Tat to directly induce several pathways ( including T-cell signaling ) for productive infection of immune cells . Both the biological data and our prediction suggest that TSG play important roles in these processes . On the other hand , TDG are primarily enriched in datasets related to ‘metabolism of RNA’ ( HSPA8 , LSM4 , and PRMT5 ) , in particular ‘ribosome’ ( RPL and RPS variants ) , ‘proteasome’ ( PSMB3 , PSMB4 , PSMB5 , PSMC2 , PSMD1 , and PSMD8 ) , and ‘regulation of apoptosis’ ( DAPK1 , together with above proteasomal genes ) , among others identified by the HIV life cycle/host interaction gene sets from Reactome ( Croft et al . , 2014 ) ( Figure 1—figure supplement 10B and Supplementary file 1C , D ) . However , further studies will be required to precisely define the role of individual TSG and TDG on viral replication . Taken together , we have established the first comprehensive framework of target genes in the human genome directly bound and regulated by Tat . Below we study the molecular mechanisms of transcriptional control by examining genome-wide changes of Pol II and chromatin signatures . With these functional insights we focused on direct target genes to delineate the mechanisms by which Tat regulates cellular transcription . Profiling of histone modifications has revealed fundamental concepts in the regulation of transcription ( Barski et al . , 2007; Zhou et al . , 2011 ) . To investigate whether Tat acts at the initiation or elongation steps , we performed ChIP-seq of several histone modifications ( in the absence and presence of Tat ) that demarcate distinct genomic domains such as promoters , coding units and enhancers and provide ( in general ) information regarding their transcriptional status ( Table 2 ) . 10 . 7554/eLife . 08955 . 015Table 2 . Genome-wide distribution and function of histone modifications . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 015Histone modificationLocationFunctionH3K4me3PromotersTranscription activationH3K79me3Gene bodiesTranscription activationH3K36me3Gene bodiesTranscription activationH3K4me1EnhancersDoes not demarcate active status , but locationH3K27AcEnhancersTranscription activationH3K9me3UbiquitousTranscription repression Active promoters usually have adjacent nucleosomes bearing an H3K4me3 transcription initiation mark ( Bernstein et al . , 2002; Guenther et al . , 2007; Mikkelsen et al . , 2007; Santos-Rosa et al . , 2002 ) , and genes marked by H3K4me3 display significant amounts of transcriptionally competent Pol II at promoters ( Min et al . , 2011 ) . Thus , we first determined the levels of H3K4me3 in both cell lines to test whether Tat modulates transcription initiation ( Figure 2A ) . After fixing TSS selection for a subset of TSG such as ADCYAP1 and ARHGEF7 , where transcription starts at a short , internal isoform ( Figure 2—figure supplement 1A , B ) , we observed that the majority of TSG ( excluding ATP9A , CD244 , ADCYAP1 , CD226 , SERINC2 ) are already marked with variable ( low-to-high ) H3K4me3 levels and , as expected , its distribution mirrors the location of the promoter-adjacent nucleosomes ( Figure 2B ) . We thus defined two groups of genes based on H3K4me3 fold change levels in the presence and absence of Tat ( Tat/GFP ) . While the first group of genes ( n = 17 ) , referred to as class I TSG , has undetectable or low H3K4me3 levels in the absence of Tat ( Figure 2A and Figure 2—figure supplement 1C , D ) , the second group of genes , referred to as class II TSG , shows medium-to-high H3K4me3 levels surrounding the TSS ( Figure 2A , B ) . 10 . 7554/eLife . 08955 . 016Figure 2 . Global analysis of chromatin signatures reveals that Tat activates the transcription initiation and elongation steps . ( A ) Dot plots of H3K4me3 log2 fold change in the region encompassing ±3 Kb from the TSS of all TSG . Genes are divided into two clusters: class I and II based on increased ( >1 . 5-fold change ) or no change/decreased H3K4me3 levels in the presence of Tat . Selected TSG examples are indicated in red . ( B ) Metagene plots centered on TSS showing H3K4me3 occupancy profiles at both class I ( n = 17 ) and class II ( n = 43 ) TSG in the presence of Tat or GFP . ( C ) Dot plots of H3K79me3 log2 fold change in the region from TSS to TTS ( see Materials and methods ) . ( D ) Metagene analysis showing average H3K79me3 ChIP-seq signals at both class I and II TSG in the presence of Tat or GFP . Units are mean tags per million ChIP-seq reads per bin across the transcribed region of each gene with 2 kb upstream and downstream flanking regions . ( E ) Dot plots of H3K36me3 log2 fold change in the region from TSS to TTS . ( F ) Metagene plots showing average H3K36me3 ChIP-seq signals at both class I and cII TSG in the presence of Tat or GFP . ( G ) Genome browser views showing ChIP-seq signal at a class I TSG ( NETO1 ) in the GFP and Tat cell lines . ( H ) Genome browser views showing ChIP-seq signal at a class II TSG ( VAV3 ) in the GFP and Tat cell lines . This figure is associated with Figure 2—figure supplements 1–5 . ChIP-seq , chromatin immunoprecipitation sequencing; GFP , green fluorescent protein; TSG , Tat stimulated genes; TSS , transcription start site; TTS , transcription termination site . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 01610 . 7554/eLife . 08955 . 017Figure 2—figure supplement 1 . Tat specifies TSS selection and synthesis of alternate isoforms . ( A ) Genome browser views of FLAG , H3K4me3 and Pol II ChIP-seq tracks in the GFP and Tat cell lines along with the Refseq track for the ADCYAP1 locus . The position of the canonical and Tat-induced ‘novel’ TSS is indicated with arrows . ( B ) Genome browser views of FLAG , H3K4me3 and Pol II ChIP-seq tracks in the GFP and Tat cell lines along with the Refseq track for the ARHGEF7 locus . The position of the canonical and Tat-induced ‘novel’ TSS is indicated with arrows . ( C ) Normalized H3K4me3 tag density ( -/+ 3 Kb respective to the TSS ) for class I TSG in the Tat ( black ) and GFP ( green ) cell lines . ( D ) Genome browser views of H3K4me3 ChIP-seq tracks in the GFP and Tat cell lines for four class I TSG from panel ( C ) . ChIP-seq , chromatin immunoprecipitation sequencing; GFP , green fluorescent protein; Pol , polymerase; TSG , Tat stimulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 01710 . 7554/eLife . 08955 . 018Figure 2—figure supplement 2 . Correlation between gene expression levels and H3K4me3 density surrounding the TSS at class I and II TSG . ( A ) Correlation plot between normalized H3K4me3 tag density ( -/+ 3 kb respective to the TSS ) and total gene expression levels ( based on RNA-seq FPKM ) at class I TSG in the presence and absence of Tat ( Tat/GFP ) . ( B ) Correlation plot between normalized H3K4me3 tag density ( ± 3 kb respective to the TSS ) and total gene expression levels ( based on RNA-seq ) at class II TSG in the presence and absence of Tat ( Tat/GFP ) . GFP , green fluorescent protein; RNA-seq , RNA sequencing; TSG , Tat stimulated genes; TSS , transcription start site . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 01810 . 7554/eLife . 08955 . 019Figure 2—figure supplement 3 . Evidence that Tat increases Pol II and P-TEFb recruitment , and chromatin marks coinciding with transcription initiation and elongation at class I TSG . ( A ) Jurkat-GFP and -Tat cell lines were used in ChIP assays to analyze the occupancy of GFP and Tat ( FLAG ) , H3K4me3 , H3K79me3 , H3K36me3 , Pol II ( total ) , Pol II ( Ser2P-CTD form ) and P-TEFb ( CDK9 ) at the CD69 locus with the three indicated amplicons . The numbers indicate the position of the amplicons respective to the TSS . ( B ) Jurkat-GFP and -Tat cell lines were used in ChIP assays to analyze the occupancy of GFP and Tat ( FLAG ) , H3K4me3 , H3K79me3 , H3K36me3 , Pol II ( total ) , Pol II ( Ser2P-CTD form ) and P-TEFb ( CDK9 ) at the ADCYAP1 locus with the three indicated amplicons . The numbers indicate the position of the amplicons respective to the TSS . For both panels , IP DNA ( % Input ) values represent the average of three independent experiments ( mean ± SEM; n = 3 ) . ChIP , chromatin immunoprecipitation; GFP , green fluorescent protein; Pol , polymerase; P-TEFb , positive transcription elongation factor b; SEM , standard error of the mean; TSS , transcription start site . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 01910 . 7554/eLife . 08955 . 020Figure 2—figure supplement 4 . Transcription initiation correlates with increased elongation chromatin markers . ( A ) Heatmap representation of ChIP-seq binding for H3K4me3 ( blue ) , H3K79me3 ( red ) , and H3K36me3 ( brown ) at class I and class II TSG , rank ordered from lowest to most H3K4me3 density increase from the GFP to the Tat cell line . The asterisk denotes the position of the TTS . ( B ) Genome browser views of H3K4me3 , H379me3 and H3K36me3 ChIP-seq tracks in the GFP and Tat cell lines along with the Refseq track for the ADCYAP1 locus ( class I TSG ) . The position of the TSS is indicated with an arrow . ( C ) Genome browser views of H3K4me3 , H379me3 and H3K36me3 ChIP-seq tracks in the GFP and Tat cell lines along with the Refseq track for the ATP9A locus ( class I TSG ) . ( D ) Genome browsers of H3K4me3 , H379me3 and H3K36me3 ChIP-seq tracks in the GFP and Tat cell lines along with the Refseq track for the CD82 locus ( class II TSG ) . ( E ) Genome browser views of H3K4me3 , H379me3 and H3K36me3 ChIP-seq tracks in the GFP and Tat cell lines along with the Refseq track for the FAM46C locus ( class II TSG ) . ChIP-seq , chromatin immunoprecipitation sequencing; GFP , green fluorescent protein; TSG , Tat stimulated genes; TSS , transcription start site; TTS , transcription termination site . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 02010 . 7554/eLife . 08955 . 021Figure 2—figure supplement 5 . Tat recruits chromatin-modifying enzymes and elongation factors at selected target genes to promote transcription elongation . ( A ) Strep-tagged GFP and Tat were affinity purified from nuclear preparations of the Jurkat cell lines ( 1 × 109 total cells ) and interacting partners were analyzed by western blot with the indicated antibodies . CDK9 was used as a positive protein interacting control . ( B ) ChIP assay to analyze the distribution of histone marks ( H3K79me3 and H3K36me3 ) and chromatin-modifying enzymes ( Dot1L and SetD2 ) at the CD69 locus of the GFP ( green ) and Tat ( black ) cell lines . The position of the amplicons used in ChIP-qPCR is shown with the schematic of the locus . Values represent the average of three independent experiments ( mean ± SEM; n = 3 ) . ( C ) Genome browsers of FLAG , Pol II , H3K4me3 , H379me3 and H3K36me3 ChIP-seq tracks in the GFP and Tat cell lines along with the Refseq track for the CD69 locus . The position of the TSS is indicated with an arrow . ( D ) Knockdown of Dot1L , SetD2 and CDK9 impairs Tat-mediated CD69 transcription activation . qRT-PCR of CD69 in the Jurkat-GFP and -Tat cell lines expressing non-target ( NT ) , Dot1L , SetD2 and CDK9 shRNAs ( mean ± SEM; n = 3 ) . ( E ) Knockdown of Dot1L , SetD2 and CDK9 does not alter RNA steady state levels of RPL19 . qRT-PCR of RPL19 in the Jurkat-GFP and -Tat cell lines expressing non-target ( NT ) , Dot1L , SetD2 and CDK9 shRNAs ( mean ± SEM; n = 3 ) . ( F , G , H ) qRT-PCR validation of Dot1L , SetD2 and CDK9 knockdown in the Jurkat-GFP and -Tat cell lines using gene-specific primers and normalized to RPL19 ( mean ± SEM; n = 3 ) . ChIP-seq , chromatin immunoprecipitation sequencing; GFP , green fluorescent protein; NT , non-target; qRT-PCR , quantitative real-time polymerase chain reaction; SEM , standard error of the mean; shRNA , small hairpin RNA; TSG , Tat stimulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 021 In the presence of Tat , class I TSG experiences a large increase in H3K4me3 density at promoter-proximal regions ( 2-to 10-fold change ) ( Figure 2A , B , and Figure 2—figure supplement 1C , D ) . Interestingly , Tat can selectively promote de novo transcription initiation at a small subset of genes ( such as ADCYAP1 , ATP9A and CD244 ) , as revealed by the large fold changes in H3K4me3 and selection of a non-canonical or novel TSS ( Figure 2—figure supplement 1A , C , D ) . For example , Tat induces H3K4me3 at an internal site ( intron 4 ) of ADCYAP1 but not at the canonical TSS , which coincides with the production of a novel , short isoform of a yet unknown function ( Figure 2—figure supplement 1A ) . In contrast to class I TSG , class II shows no change or , unexpectedly , a slight reduction in H3K4me3 levels in the presence of Tat ( ∼1 . 5–2–fold decrease ) ( Figure 2A , B ) , suggesting a post-initiation role for Tat in activating these genes ( see below ) . The H3K4me3 density surrounding the TSS of class I TSG was typically lower ( at least 5-fold less ) than the signal in actively transcribed genes . As expected , these genes are transcriptionally inactive or show low RNA levels ( <10 Fragments Per Kilobase of transcript per Million mapped reads [FPKM] ) such as signaling peptide hormones ( ADCYAP1 ) , transcription factors involved in immune system maturation ( ZNF521 , RORβ , ETV6 ) and genes essential for T-cell maturation and responses ( CD69 , CD244 ) . Interestingly , we observed a positive correlation ( R2 = 0 . 764 ) between the increase in transcript levels and H3K4me3 density nearby the promoters of these genes , and that highly stimulated genes ( ADCYAP1 and CD69 ) experience larger increase in H3K4me3 density compared with other genes ( Figure 2—figure supplement 2A ) . On the other hand , class II TSG shows a much lower correlation ( R2 = 0 . 177 ) between the increase in transcript levels and H3K4me3 density surrounding promoters ( Figure 2—figure supplement 2B ) , consistent with the idea that they are regulated at a post-initiation step ( see below ) . If increased transcription initiation at these genes correlates with a productive increase in RNA levels , then we would expect to find nucleosome modifications associated with transcription elongation throughout the coding units ( Li et al . , 2007 ) . Previous studies have elucidated that in actively transcribed genes , H3K79me2/3-modified nucleosomes are present at their highest levels shortly downstream of the TSS and that H3K36me3 modifications occupy the entire gene body , with increasing density towards the 3’-end of the gene ( Guenther et al . , 2007; Kolasinska-Zwierz et al . , 2009; Li et al . , 2007; Seila et al . , 2008; Zhou et al . , 2011 ) . To examine transcription states in detail , we carried out ChIP-seq experiments of H3K79me3 and H3K36me3 using validated antibodies ( Egelhofer et al . , 2011 ) . To assess Tat’s role in coupling transcription initiation with elongation at class I TSG , we calculated the fold change in H3K79me3 and H3K36me3 tag density in the presence and absence of Tat ( Tat/GFP ) as well as changes in their gene average distribution in both class I and II TSG ( Figure 2C–F ) . As expected , we observed that class I TSG ( activated at the initiation step ) also shows evidence of transcription elongation , based on the increase in H3K79me3/H3K36me3 , total Pol II and elongating Pol II levels , as well as recruitment of the P-TEFb kinase at two TSG ( CD69 and ADCYAP1 ) ( Figure 2—figure supplement 3 ) . While in the absence of Tat , class I TSG are devoid of H3K79me3 throughout the gene , Tat promotes a robust increase in H3K79me3 ( ∼6 . 2-fold over GFP ) just downstream of the TSS with a progressive decline towards the transcription termination site ( TTS ) ( Figure 2C–F , and Figure 2—figure supplement 4 ) . Similarly , levels of H3K36me3 at class I TSG are low in the gene body and increase towards the 3’-end of the gene in the presence of Tat ( ∼5 . 3-fold over GFP ) ( Figure 2F ) . Remarkably , the average distribution patterns of both H3K79me3 and H3K36me3 in these genes are consistent with previous genome-wide distribution analysis ( Kouzarides , 2007; Li et al . , 2007 ) . Because H3K4me3 levels do not increase in the class II TSG , we reasoned that these genes are regulated by Tat at a post-initiation step . If this were the case , then we would expect to find an increase in nucleosome modifications associated with transcription elongation marks in the presence of Tat , despite the lack of H3K4me3 increase . To test this possibility , we examined H3K36me3 and H3K79me3 distribution and density in these genes . We ignored genes that lack H3K79me3 irrespective of Tat presence as well as intronless genes , which complicated the density and distribution calculations , and thus ended with a more cohesive and consistently behaved group of class II TSG ( n = 43 ) . As expected , most class II TSG showed increased H3K79me3 ( 100% ) and H3K36me3 ( ∼82% ) levels within their gene bodies ( Figure 2C–F ) consistent with a role of Tat in promoting transcription elongation . Together , this analysis suggests that the majority of target genes that do not experience transcription initiation changes in response to Tat do show evidence of increased levels of promoter escape and transition to active elongation based on the global analysis of chromatin signatures ( Figure 2—figure supplement 4 ) and Pol II ( see below ) . Although the increase in H3K79me3 is significant , H3K36me3 changes are smaller , probably because class II TSG is active and their transcribing units already demarcated by H3K36me3 ( Figure 2C–F ) . Again , the average gene distribution of H3K79me3 and H3K36me3 correlates with previous studies suggesting that H3K79me3 is most enriched shortly downstream of the TSS and H3K36me3 increases towards the middle and end of the gene ( Figure 2D–F ) ( Kouzarides , 2007; Li et al . , 2007 ) . Inspection of individual gene tracks from our genome-wide study reveals examples of these two regulatory mechanisms ( Figure 2G , H ) . At the neuropilin and tolloid-like 1 ( NETO1 ) locus ( class I TSG ) , which encodes a transmembrane receptor that plays a critical role in spatial learning and memory , Tat binding at the promoter-proximal region correlates with a sharp increase in H3K4me3 ( ∼8-fold ) with concomitant increases in H3K79me3 and H3K36me3 downstream of the TSS ( Figure 2G ) . Conversely , at the VAV3 guanine nucleotide exchange factor locus ( class II TSG ) , the H3K4me3 signature surrounding the TSS does not appear to change , while the markers of active transcription in coding units ( H3K79me3 and H3K36me3 ) showed large increases in the presence of Tat , indicating a role in transcription elongation ( Figure 2H ) . Heatmap representation of the density of all three chromatin signatures at promoter and gene body of class I and class II TSG demonstrate that this is common for all target genes , albeit with differences in fold change ( Figure 2—figure supplement 4A–E ) , which is in agreement with the broad distribution of gene expression changes ( Figure 1D ) . Given that we defined two classes of genes using the minimalistic Tat ectopic expression system , we asked whether these genes are also modified in a similar manner ( at the initiation or elongation steps ) during HIV infection . To test this , we analyzed the levels of initiating ( promoter-proximal ) and elongating ( promoter-distal ) transcripts in primary TCM cells infected with replication-competent HIV versus mock infection . Interestingly , we observed that two class I TSG ( CD69 and PPM1H ) and class II TSG ( VAV3 and ANXA1 ) showed evidence of increased initiation or elongation , respectively ( Figure 1—figure supplement 8C , D ) . This implies that the model of cellular reprogramming by the ectopic expression of Tat alone is mirrored ( at least for the target genes examined ) during HIV infection of primary T cells . Several enzymes are known to regulate histone modifications associated with distinct epigenetic states . If Tat promotes cellular reprogramming by modifying the epigenetic landscape then we would expect Tat to interact with chromatin-modifying enzymes associated with the respective histone modifications . While H3K79me3 is generated by the Dot1L complex ( Nguyen and Zhang , 2011 ) , the H3K36me3 mark is imposed by the SetD2 methyltransferase recruited by elongating Pol II and enriched within the body of transcriptionally active genes ( Guenther et al . , 2007; Krogan et al . , 2003a; Krogan et al . , 2003b; Nguyen and Zhang , 2011; Pokholok et al . , 2005 ) . To test if Tat does indeed interact with these enzymes , we first affinity purified ( AP ) Strep-tagged Tat ( or GFP , used as negative control ) from nuclear fractions of the Jurkat T-cell lines and observed that Tat , but not GFP , binds both Dot1L and SetD2 as well as the P-TEFb kinase ( CDK9 ) used as positive control in this assay ( Figure 2—figure supplement 5A ) . To further test whether changes in the epigenetic landscape directly correlate with the recruitment of these chromatin modifiers to specific genes , we performed ChIP followed by quantitative PCR ( ChIP-qPCR ) and found that the Tat-mediated recruitment of Dot1L and SetD2 to the CD69 locus ( class I TSG ) correlates well with the increase in the histone modifications associated with transcription elongation as well as Pol II ( Figure 2—figure supplement 5B , C ) . To provide a functional link between the Tat-mediated recruitment of these chromatin modifiers and gene expression changes , we used short hairpin RNAs ( shRNAs ) to target Dot1L and SetD2 by RNA interference ( Figure 2—figure supplement 5D–H ) . Interestingly , we noted that after knockdown of these chromatin-modifying enzymes as well as of CDK9 , the increase in CD69 RNA levels ( but not RPL19 ) in response to Tat is virtually abolished , implying that Tat-mediated recruitment of Dot1L , SetD2 and P-TEFb to TSG is a requisite for their increased transcription elongation levels , which is in agreement with the ChIP data ( Figure 2—figure supplement 3 ) . The discovery of these key enzymes as Tat targets and the evidence that they play important roles in Tat-mediated cellular gene expression alterations suggest that this regulatory mechanism is part of the reprogramming of target immune cells by Tat . Collectively , we have described a set of human genes stimulated by Tat at two different steps in the transcription cycle ( initiation and elongation ) using selective chromatin-modifying enzymes and the transcription elongation machinery . Pol II regulates the control of transcription initiation and elongation in the context of chromatin ( Fuda et al . , 2009 ) . In fact , the deposition of histone modifications associated with initiation and elongation at promoters and gene bodies has been functionally linked to levels of recruited and transcriptionally engaged Pol II , respectively ( Adelman and Lis , 2012 ) . Therefore , if Tat promotes transcription initiation and elongation as proposed ( Figure 2 ) , we would expect Pol II levels to fluctuate in response to Tat in ways that reflect the specific mode of activation . Pol II is recruited to gene promoters to initiate transcription , but also tends to occupy promoter-proximal regions of genes that show evidence of initiation but no or inefficient elongation , which is referred to as promoter-proximal pausing ( Adelman and Lis , 2012; Fuda et al . , 2009; Rahl et al . , 2010; Wade and Struhl , 2008 ) . To further investigate whether Tat promotes Pol II recruitment ( indicative of transcription initiation ) or increases Pol II levels within gene bodies ( indicative of transcription elongation ) we used ChIP-seq to examine Pol II distribution in the absence and presence of Tat ( Figure 3 ) . 10 . 7554/eLife . 08955 . 022Figure 3 . Tat promotes Pol II recruitment and pause release at two distinct TSG classes . ( A ) Tat binding promotes Pol II recruitment at class I but not class II TSG . Box plots of normalized Pol II tag density at the Tat peak of class I ( n = 17 ) or II ( n = 43 ) TSG in the GFP and Tat cell lines . ( B ) Tat binding at class I TSG induces Pol II occupancy at promoters . Box plots of normalized Pol II tag density at the promoter of class I or II TSG in the GFP and Tat cell lines . ( C ) Pol II normalized tag density relative to the Tat peak at class I TSG . ( D ) Pol II normalized tag density relative to the Tat peak at class II TSG . ( E ) Pol II distribution at class I TSG ( Metagene plots ) in the Tat and GFP cell lines . ( F ) Genome browser views of ChIP-seq data in the Tat and GFP cell lines at a class I TSG ( NETO1 ) . The arrow indicates the position of the FLAG peak called in the Tat cell lines by MACS . ( G ) Pol II distribution at class II TSG ( Metagene plots ) in the Tat and GFP cell lines . ( H ) Genome browser views of ChIP-seq data in the Tat and GFP cell lines at a class II TSG ( VAV3 ) . This figure is associated with Figure 3—figure supplements 1 , 2 . ChIP-seq , chromatin immunoprecipitation sequencing; GFP , green fluorescent protein; MACS , model-based analysis of ChIP-seq; Pol , polymerase; TSG , Tat stimulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 02210 . 7554/eLife . 08955 . 023Figure 3—figure supplement 1 . Tat recruitment induces Pol II and chromatin signatures controlling transcription initiation or elongation at different gene classes . ( A ) Genome browser views of FLAG , Pol II , H3K4me3 , H379me3 and H3K36me3 ChIP-seq tracks in the GFP and Tat cell lines along with the Refseq track for the NETO1 locus ( class I TSG ) . The position of the TSS is indicated with an arrow . ( B ) Genome browsers of FLAG , Pol II , H3K4me3 , H379me3 and H3K36me3 ChIP-seq tracks in the GFP and Tat cell lines along with the Refseq track for the VAV3 locus ( class II TSG ) . The position of the TSS is indicated with an arrow . GFP , green fluorescent protein; Pol , polymerase; TSG , Tat stimulated genes; TSS , transcription start site . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 02310 . 7554/eLife . 08955 . 024Figure 3—figure supplement 2 . Tat controls P-TEFb and Pol II recruitment at class II TSG and class II TDG . ( A–B ) Tat promotes P-TEFb recruitment and Pol II elongation at class II TSG . ( C–D ) Tat precludes P-TEFb recruitment to block Pol II elongation at class II TDG . Jurkat-GFP and -Tat cell lines were used in ChIP assays to analyze the occupancy of GFP and Tat ( FLAG ) , P-TEFb and Pol II at the indicated target genes using the indicated amplicons . Values represent the average of three independent experiments ( mean ± SEM; n = 3 ) . ChIP , chromatin immunoprecipitation; GFP , green fluorescent protein; Pol , polymerase; P-TEFb , positive transcription elongation factor b; SEM , standard error of the mean; TDG , Tat down-regulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 024 We first analyzed levels of Pol II at both Tat binding sites ( Tat peak ) and promoters of target genes in cases where Tat only binds to promoter-distal sites . Interestingly , we found that Tat stimulates Pol II recruitment at both the Tat binding site and promoter-proximal region of class I TSG ( Figure 3A–C ) , in perfect agreement with the Tat-mediated increase in H3K4me3 at those gene promoters ( Figure 2A , B ) . It is noteworthy that this observation is consistent with the model that Tat mediates transcription initiation at class I TSG . Conversely , as expected , we did not see higher levels of Pol II at the promoters of class II TSG in response to Tat ( Figure 3A , B , D ) . This lack of Pol II recruitment further supports our proposed model based on the global analysis of chromatin signatures that class II TSG are not regulated at the initiation step ( Figure 2 ) . Interestingly , a metagene analysis of class I TSG in the absence of Tat showed very low Pol II density levels in both the promoter and transcribing unit . Conversely , Tat induces an increase in Pol II density throughout the gene body with a noticeable peak at the promoter-proximal region , consistent with Pol II recruitment to promoters and transition into elongation ( Figure 3E ) and with the increase in H3K4me3 at the promoter-proximal regions associated with those genes ( Figure 2A , and Figure 2—figure supplement 3 ) . Genome browser views of individual class I TSG such as NETO1 exemplify the metagene analysis depicting that Pol II is strongly recruited to the intragenic Tat binding site and promoter , marked with H3K4me3 in the presence of Tat ( Figure 3F ) , and that transcription initiation correlates with increase in chromatin signatures ( H3K79me3 and H3K36me3 ) associated with active transcription in gene bodies , as well as transcribing Pol II ( Figure 3—figure supplement 1A ) . We have previously defined class II TSG as being primarily regulated at the elongation step because H3K4me3 and Pol II density at the promoter-proximal region do not increase in the presence of Tat ( Figure 2 and 3A–D ) . Examination of two class II TSG ( VAV3 and CD82 ) using Chip-qPCR clearly demonstrates that Tat induces Pol II elongation by recruiting the P-TEFb kinase at those target genes ( Figure 3—figure supplement 2A , B ) . Consistently , a metagene analysis shows that Tat largely increases Pol II density throughout the gene body and 3’-end of class II TSG , but not at the promoter , in agreement with a role of Tat in promoting the transition to elongation ( Figure 3G ) . Therefore , in the absence of Tat stimulation , the majority of Pol II at class II TSG accumulates in the promoter-proximal region with a peak just downstream of the TSS while Tat strongly induces increased Pol II density in the gene body but not in the promoter-proximal region ( Figure 3G ) . Genome browser views of individual class II TSG such as VAV3 are consistent with the elongation function ( Figure 3H ) , and are also in perfect agreement with the analysis of chromatin signatures related to elongation ( H3K79me3 and H3K36me3 ) ( Figure 2 and Figure 3—figure supplement 1B ) . Collectively , the data indicate that Tat controls both Pol II recruitment and pause release to promote initiation and elongation in different gene classes . Tat is recruited to two different genomic domains ( promoters and intragenic ) irrespective of the transcription step regulated ( initiation or elongation ) ( Figure 4A ) , implying that the site of Tat recruitment to its target genes does not dictate the mechanism of transcription activation . To further elucidate how Tat promotes transcription by binding to promoters or intragenic sites , we examined how chromatin signatures associated with these genomic domains change in response to Tat . Promoters bound by Tat are marked with the expected signature: high H3K4me3 and low H3K27Ac ( Figure 4B ) ( Creyghton et al . , 2010; Heintzman and Ren , 2009 ) . Conversely , intragenic sites bound by Tat are marked with the enhancer signature: high H3K27Ac and low H3K4me3 ( Figure 4B ) , as well as high H3K4me1 ( data not shown ) ( Creyghton et al . , 2010; Heintzman and Ren , 2009; Zhou et al . , 2011 ) . This suggests that the intragenic Tat-binding sites appear to be intragenic enhancers , which have been previously proposed to function as alternative elements required for gene activation ( Kowalczyk et al . , 2012 ) . 10 . 7554/eLife . 08955 . 025Figure 4 . Tat induces transcription initiation from distal sites by inducing gene looping . ( A ) Heatmap representation of Tat distribution at TSG promoter or intragenic sites . ( B ) Heatmap representation of H3K4me3 and H3K27Ac tag density centered on the Tat peak at both TSG promoter and intragenic sites in the GFP and Tat cell lines . Promoter sites are marked with high H3K4me3 and low H3K27Ac levels , while intragenic sites are marked by low H3K4me3 and high H3K27Ac levels . ( C ) Heatmap representation of H3K27Ac at class I and class II TSG centered on the Tat peak in the GFP and Tat cell lines . Genes are ranked based on H3K27Ac density . ( D ) Tat binding sharply increases H3K27Ac levels at class I TSG . Box plots showing normalized H3K27Ac density at class I and II TSG in the GFP and Tat cell lines . ( E ) Metagene analysis of H3K27Ac levels surrounding Tat peaks at class I TSG in the GFP and Tat cell lines . ( F ) Metagene analysis of H3K27Ac levels surrounding Tat peaks at class II TSG in the GFP and Tat cell lines . ( G ) ChIP of H3K27Ac and p300 recruitment at an intragenic Tat site at the PPM1H gene . ( H ) ChIP of H3K27Ac and p300 recruitment at an intragenic Tat site at the CD82 gene . ( I ) Top , genome browser views of ChIP-seq at the PPM1H locus in the GFP and Tat cell lines . The arrows indicate the position of two intragenic Tat binding sites . Bottom , 3C assay showing the relative crosslinking efficiency at the PPM1H locus in the GFP and Tat cell lines . The position of the primers used in qPCR assays , restriction sites and fragment generated after digestion are indicated . This figure is associated with Figure 4—figure supplements 1–4 . 3C , chromosome conformation capture; ChIP-seq , chromatin immunoprecipitation sequencing; GFP , green fluorescent protein; RNA-seq , RNA sequencing; TSG , Tat stimulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 02510 . 7554/eLife . 08955 . 026Figure 4—figure supplement 1 . Stimulation of gene expression by Tat correlates with increased H3K27Ac density at the Tat peak . ( A ) Correlation plot between normalized H3K27Ac ( -/+ 1 kb respective to the Tat peak ) and total gene expression levels ( based on RNA-seq FPKM ) at class I TSG in the presence and absence of Tat ( Tat/GFP ) . ( B ) Correlation plot between normalized H3K27Ac ( -/+ 1 kb respective to the Tat peak ) and total gene expression levels ( based on RNA-seq FPKM ) at class II TSG in the presence and absence of Tat ( Tat/GFP ) . FPKM , Fragments Per Kilobase of transcript per Million mapped reads; GFP , green fluorescent protein; RNA-seq , RNA sequencing; TSG , Tat stimulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 02610 . 7554/eLife . 08955 . 027Figure 4—figure supplement 2 . Wild-type Tat but not the C22A non-functional mutant induces gene looping between the promoter and intragenic sites . ( A ) Genome browser views of ChIP-seq at the PPM1H locus in the GFP and Tat cell lines . The arrows indicate the position of two intragenic Tat binding sites . ( B ) 3C assay showing the relative crosslinking efficiency between a promoter anchor primer ( top ) or an upstream control primer ( bottom ) and several distal sites at the PPM1H locus in the GFP and Tat cell lines . The position of the primers used in qPCR assays , restriction sites and fragment generated after digestion are indicated ( mean ± SEM; n = 3 ) . 3C , chromosome conformation capture; ChIP-seq , chromatin immunoprecipitation sequencing; GFP , green fluorescent protein; qPCR , quantitative polymerase chain reaction; SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 02710 . 7554/eLife . 08955 . 028Figure 4—figure supplement 3 . Tat-mediated gene looping does not require transcription activity at enhancers . ( A ) Genome browser views of ChIP-seq at the PPM1H locus in the GFP and Tat cell lines . The arrows indicate the position of two intragenic Tat binding sites . ( B ) 3C assay showing the relative crosslinking efficiency between a promoter anchor primer and intragenic sites at the PPM1H locus in the GFP and Tat cell lines treated with DMSO ( ± flavopiridol [FP] ) . The position of the primers used in qPCR assays , restriction sites and fragment generated after digestion are indicated . ( C ) The GFP and Tat cell lines were treated with DMSO ( ± flavopiridol [FP] ) , and RNA isolated to measure levels of PPM1H mRNAs . ( D ) The GFP and Tat cell lines were treated with DMSO ( -FP ) or FP ( +FP ) , and RNA isolated to measure levels of the intragenic eRNA ( mean ± SEM; n = 3 ) . 3C , chromosome conformation capture; ChIP-seq , chromatin immunoprecipitation sequencing; DMSO , dimethyl sulfoxide; eRNA , enhancer-derived RNA; GFP , green fluorescent protein; qPCR , quantitative polymerase chain reaction; SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 02810 . 7554/eLife . 08955 . 029Figure 4—figure supplement 4 . Class II TSG contain stable gene loops between promoter-enhancer that remain unaltered in response of Tat . ( A ) Genome browser views of ChIP-seq at the CD82 locus in the GFP and Tat cell lines . The arrows indicate the position of two intragenic Tat binding sites . ( B ) 3C assay showing the relative crosslinking efficiency at the CD82 locus in the GFP and Tat cell lines using a promoter anchor primer ( top ) or an upstream control primer ( bottom ) . The position of the primers used in qPCR assays , restriction sites and fragment generated after digestion are indicated ( mean ± SEM; n = 3 ) . 3C , chromosome conformation capture; ChIP-seq , chromatin immunoprecipitation sequencing; GFP , green fluorescent protein; qPCR , quantitative polymerase chain reaction; SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 029 To better define the roles of Tat in activating transcription from these promoter-distal , intragenic sites , we sorted class I and II TSG based on the H3K27Ac density surrounding the Tat peak . Notably , we observed that in the absence of Tat , the intragenic sites at class I TSG have low or undetectable levels of H3K27Ac ( Figure 4C ) , consistent with the idea that these genes are inactive or only minimally transcribed in the basal state without Tat ( Figure 2 ) . However , Tat increases H3K27Ac density near its intragenic binding sites by ∼2–10-fold depending on the gene ( Figure 4C , D ) . Conversely , in the absence of Tat , H3K27Ac levels at the intragenic sites of class II TSG are high and Tat increases their density , albeit with a lower fold-change than in class I TSG ( Figure 4C , D ) . This is consistent with the model that these genes are active in the basal state and Tat activates a post-initiation step , namely transcription elongation . Metagene analysis of class I TSG indicates that the H3K27Ac mark at intragenic binding sites is virtually absent at the immediate binding site itself but high in the nucleosomes directly surrounding the Tat peak , progressively declining further up and downstream from the binding site ( Figure 4E ) . At these sites , Tat increased H3K27Ac levels an average of ∼3-fold . A similar H3K27Ac distribution pattern is observed in class II TSG , even though the magnitude of H3K27Ac increase is smaller ( ∼1 . 5-fold ) because these genes are already active in the absence of Tat ( Figure 4F ) . Given that the histone acetyl transferase p300/CBP is a well-known Tat interactor and that it is recruited to enhancers to facilitate transcription activation through chromatin acetylation ( including H3K27Ac ) ( Hottiger and Nabel , 1998; Jager et al . , 2012; Kim et al . , 2010b ) , we asked whether Tat recruits p300 to these intragenic sites to trigger H3K27Ac . To test this possibility , we performed ChIP assays on one class I TSG ( PPM1H ) and class II TSG ( CD82 ) . Notably , we observed that in the presence of Tat , the increase in H3K27Ac levels at class I intragenic sites in response to Tat mirrors the recruitment of p300 ( Figure 4G ) . However , levels of both H3K27Ac and p300 detected at the intragenic sites of class II TSG are already high in the basal state and are slightly induced ( ∼1 . 2–1 . 4-fold ) in response to Tat ( Figure 4H ) . Remarkably , we observed a sharp correlation ( R2 = 0 . 853 ) between increased H3K27Ac density at the intragenic site and gene expression levels at class I TSG ( Figure 4—figure supplement 1A ) , supporting a model in which Tat is recruited to these sites to induce de novo transcription activation through recruitment of p300 at internal sites . However , the correlation between H3K27Ac density at the intragenic site and gene expression levels at class II TSG is quite low ( R2 = 0 . 0728 ) because these genes are already marked by H3K27Ac in the absence of Tat ( Figure 4—figure supplement 1B ) . These results support the model that Tat binds at intragenic sites of non-productive genes ( class I TSG ) to induce their transcription initiation or is recruited to productive genes ( actively transcribed ) such as class II TSG to augment transcription elongation levels . Because these Tat binding sites are located distally from the promoter in the majority of class I TSG ( 15 out of 17 ) , we hypothesized a model in which Tat controls gene looping to induce spatial proximity between the intragenic site ( putative enhancer ) and the promoter . To further test this possibility in detail we selected one class I TSG ( PPM1H ) , which is about 300 kb in length , thus facilitating the analysis of long-range chromatin interactions ( Figure 4I ) . We performed chromosome conformation capture ( 3C ) to assess the association of two intragenic Tat target sites and the promoter . We also performed a similar analysis between the gene promoter and distal intergenic sites located at a similar distance from the promoter but not bound by Tat . We used anchoring points near the gene promoter to measure the extent of chromatin looping between the promoter and the intragenic Tat-bound sites or the distal control domain . Interestingly , this analysis revealed that prior to Tat stimulation there is no obvious interaction between the gene promoter and the two intragenic Tat-bound sites . However , Tat promotes a strong association between the two sites , albeit with different crosslinking efficiencies ( Figure 4I ) , and this gene looping needs a functional Tat , because the C22A non-functional Tat mutant does not promote this effect ( Figure 4—figure supplement 2 ) . Moreover , the effect of Tat on chromatin looping is specific as there was no detectable association when a control anchor primer was placed about 20 kb upstream from the PPM1H promoter ( Figure 4—figure supplement 2 ) . Gene looping could be a direct consequence of Tat activity at promoters and enhancers , or Tat can simply help activate enhancers , causing them to increase looping/proximity to their target promoters ( Bulger and Groudine , 2011 ) . To distinguish between these two possibilities we used flavopiridol ( FP ) , a CDK9 inhibitor that , in addition to blocking transcription elongation , also inhibits the production of enhancer-derived RNAs ( eRNAs ) , without affecting other molecular indicators of enhancer activity ( such as Pol II binding , H3K4me1 levels , and gene looping ) ( Hah et al . , 2013 ) . We used FP in the Jurkat-GFP and -Tat cell lines and performed 3C experiments to uncouple the assembly of enhancer complexes and gene looping per se from eRNA production . The data suggests that Tat promotes gene looping even in the absence of eRNA synthesis ( Figure 4—figure supplement 3A–D ) , implying that eRNA synthesis occurs after the assembly of Tat and the transcription machinery at the enhancer , and enhancer-promoter communication . We have suggested that class II TSG have paused Pol II , and it is known that promoter-enhancer loops are associated with paused Pol II ( Ghavi-Helm et al . , 2014 ) . We thus wished to test whether gene looping also takes place at class II TSG , and if Tat plays any role . We first analyzed the percentage of class II TSG-containing intragenic Tat-bound sites ( n = 34 out of 43 , 79% ) and then examined whether gene looping is critical for their expression and whether Tat promotes the looping ( Figure 4—figure supplement 4 ) . For this we selected one class II TSG ( CD82 ) , which shows a strong Pol II peak promoter-proximally , has one Tat-bound site intragenically , and shows evidence of looping between the promoter and the intragenic Tat-bound site . We found no evidence that Tat modulates gene looping at the CD82 loci thereby indicating that Tat's effects at class II TSG is at a step post-formation of long-range chromatin interactions ( Figure 4—figure supplement 4 ) , in agreement with a role of Tat in promoting elongation . In conclusion , class I TSG is transcriptionally inactive ( or transcribed at a low rate ) and Tat binds to intragenic sites to induce chromatin looping and transcription activation . On the other hand , class II TSG is actively transcribed and Tat promotes a post-initiation step ( namely Pol II elongation ) by binding to both promoters and/or intragenic sites without affecting gene looping . While at class I TSG Tat promotes the initiation step , at class II TSG , Tat appears to induce Pol II transition into the elongation phase ( Figures 2 , 3 ) . Previously , an algorithm that computes Pol II levels in promoter-proximal regions and gene body ( termed Traveling Ratio [TR] ) has been devised to examine the transition from initiation into elongation ( Figure 5A ) ( Rahl et al . , 2010; Reppas et al . , 2006; Wade and Struhl , 2008; Zeitlinger et al . , 2007 ) . We thought to apply such an algorithm to examine Tat functions in initiation and elongation by comparing Pol II levels found in promoter-proximal regions ( -50 to +1000 respective to the TSS ) versus levels within the gene body . Unexpectedly , we found in our ChIP-seq datasets that in both the presence and absence of Tat , Pol II levels tend to peak not only near the promoter of many genes , but also at certain intronic domains marked by H3K27Ac and H3K4me1 , sometimes at densities similar to those found at the promoter-proximal region ( Figure 5B ) . These sites do not contain any known annotation in the University of California Santa Cruz ( UCSC ) genome browser and does not appear to contain a non-canonical TSS because the H3K4me3 levels are low-to-undetectable . Thus , to examine if these Pol II forms were technical artifacts , we performed ChIP-seq with other Pol II Abs ( total Pol II , CTD , Ser5P-CTD and Ser2P-CTD ) , both in the absence and presence of Tat , and found that , irrespective of the Ab used , this stably paused Pol II form was detected in the gene body , albeit with variable density levels ( Figure 5—figure supplement 1 ) . Similarly , genome browser searches revealed that intragenically poised Pol II is also detected in primary CD4+ T cells , thus ruling out the possibility that it is an artifact of Pol II distribution present in immortalized T cell lines growing in tissue culture . Moreover , examination of our ChIP-seq chromatin signature tracks indicated that the sites of intragenic Pol II pausing correspond to potential enhancers because they are marked with the corresponding enhancer signature ( high H3K4me1/H3K27Ac and low H3K4me3 ) ( Figure 5B ) ( Heintzman and Ren , 2009 ) . 10 . 7554/eLife . 08955 . 030Figure 5 . A modified traveling ratio algorithm reveals Tat roles in transcription elongation . ( A ) TR algorithm used to calculate levels of promoter-proximal paused and elongation Pol II as previously described ( Rahl et al . , 2010; Zeitlinger et al . , 2007 ) . ( B ) Genome browser views of ChIP-seq data at the CD82 locus in the GFP and Tat cell lines . ( C ) mTR algorithm that enables the identification of intragenically paused Pol II at sites of high H3K27Ac levels . ( D ) mTR decreases at TSG in the Tat cell line . Box plots showing mTR calculations at TSG as indicated in panel ( C ) . ( E ) Dual-axis plot of the ratio of paused Pol II ( pPol II ) and traveling Pol II ( tPol II ) at class I ( n = 17 ) and II ( n = 43 ) TSG in the presence and absence of Tat ( Tat/GFP ) . ( F ) Model of Tat-mediated Pol II recruitment and gene looping at class I TSG . Tat binds at promoter and/or intragenic regions ( in some cases inducing gene looping ) , and recruits Pol II to promote transcription initiation from both promoter-proximal and promoter-distal sites . ( G ) Model of Tat-mediated Pol II elongation at class II TSG . Class II TSG are already activated in the absence of Tat , and Tat binds to these genes to primarily stimulate transcription elongation . These genes contain promoter-bound Pol II . This figure is associated with Figure 5—figure supplement 1 . GFP , green fluorescent protein; Pol , polymerase; mTR , modified TR; TR , traveling ratio; TSG , Tat stimulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 03010 . 7554/eLife . 08955 . 031Figure 5—figure supplement 1 . Intragenically paused Pol II is detected in CD4+ T cell lines and primary cells at sites with high H3K27Ac and low H3K4me3 . Genome browser views of H3K4me3 , H3K27Ac and different Pol II forms ( total CTD , Ser5P-CTD and Ser2P-CTD ) in the Jurkat-GFP cell line and primary CD4+ T cells along with the Refseq track for the ABCG1 locus ( class II TSG ) . The position of the different Pol II forms is indicated . GFP , green fluorescent protein; TSG , Tat stimulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 031 Given that we found intragenic enhancers containing a form of Pol II that appears to be paused , and that enhancer–promoter interactions are frequently associated with paused Pol II ( Ghavi-Helm et al . , 2014 ) , we speculated that the original TR algorithm might not accurately describe Tat’s role in transcription elongation in our dataset . The problem arises because the TR algorithm treats all Pol II in the gene body as actively elongating . This assumption is invalid in cases where Pol II appears paused intragenically or is detected at an intragenic enhancer because of chromatin looping . In order to circumvent this problem we devised a custom-made algorithm referred to as ‘modified TR’ ( mTR ) to identify intragenic enhancers marked by H3K4me1/H3K27Ac and paused Pol II . mTR accurately categorizes these sites to more precisely calculate the promoter-proximal and gene body Pol II counts ( Figure 5C ) . We defined paused Pol II ( pPol II ) as Pol II located in both the promoter-proximal region ( -50/+1000 from the TSS ) and at intragenic sites marked with H3K27Ac ( as determined by MACS2 , with a peak cutoff of p-value <0 . 05 ) and having a read density/nt greater than five times that of the average density/nt within the gene body . Of note , we used a window of -50/+1000 from the TSS ( rather than -50/+300 ) because several genes contained accumulation of Pol II beyond the +100 , and up to the +1000 position . Transcribing Pol II ( tPol II ) is defined as Pol II in the remainder of the gene that is not associated with the enhancer signature . Then , the mTR is the ratio of average pPol II ( promoter + gene body ) density to average tPol II ( gene body ) density ( Figure 5C ) . As expected , class II TSG showed decreased mTR in the presence of Tat compared with the GFP cell line ( Figure 5D ) . Despite this compelling observation , the mTR could be altered in response to Tat either by fluctuations in the levels of pPol II , tPol II , or both . Therefore , to more clearly explore Tat’s role in controlling elongation , we computed densities for both Pol II classes ( pPol II and tPol II ) and compared them separately in the GFP and Tat cell lines . For each gene , we calculated the ratio of average pPol II and tPol II densities in the two cell lines ( Tat/GFP ) and plotted them on the X- and Y-axis , respectively ( Figure 5E ) . This plot demonstrates clearly the Pol II behavior in response to Tat , both in terms of changes to initiation and elongation . While class I TSG showed profound increase in pPol II and tPol II as a consequence of transcription initiation/elongation activation , class II TSG showed , albeit with gene-specific differences , higher evidence of elongation ( increased tPol II form in the presence of Tat ) ( Figure 5E ) . Together , the data suggest that Tat has evolved to fine-tune both transcription initiation and elongation steps at functionally different gene classes . At class I TSG , Tat directly mediates Pol II recruitment to gene promoters or binds to intragenic sites to induce chromatin looping ( promoter-enhancer communication ) to trigger transcription initiation ( Figure 5F ) . At class II TSG , Tat binds to already engaged Pol II to promote its escape into the productive elongation phase ( Figure 5G ) . Global analysis of chromatin signatures in stimulated genes revealed that Tat induces both transcription initiation and elongation ( Figure 2 ) . Thus , we computationally examined chromatin signatures throughout promoter-proximal and -distal regions of TDG to determine at which step in the transcription cycle Tat blocks gene activation . Despite the identification of genes having marked changes in both chromatin initiation ( H3K4me3 ) and elongation signatures ( H3K79me3 and H3K36me3 ) such as CD1E and FBLN2 , or only chromatin elongation signatures such as EOMES and HSPA8 , this distinction did not become as clear for genes having low gene expression changes ( <3-fold ) . Therefore , we used a combined chromatin/Pol II signature to identify initiation- ( class I ) and elongation-regulated ( class II ) TDG ( Figure 6 ) . We defined class I TDG as genes having chromatin and Pol II signatures consistent with inhibited transcription initiation , with at least 60% decrease in H3K4me3 surrounding the TSS and 30% decrease in promoter-proximal Pol II with Tat ( fold H3K4me3 Tat/GFP<0 . 4 and fold Pol IITat/GFP<0 . 7 ) . Conversely , class II TDG are inhibited at later stages of transcription because H3K4me3 and Pol II promoter-proximal levels remain virtually unchanged in the presence of Tat ( fold H3K4me3 Tat/GFP>0 . 7 and fold Pol IITat/GFP>0 . 75 ) . Therefore , we defined class II TDG as genes having less than 30% decrease H3K4me3 surrounding the TSS and less than 25% decrease in promoter-proximal Pol II level in response to Tat . Using this signature , we identified 11 class I TDG and 14 class II TDG that strictly pass the above-mentioned criteria . Remarkably , these genes also appear to be controlled at the initiation and elongation level in primary TCM cells infected with HIV ( Figure 1—figure supplement 8E , F ) . Although further research is needed to pinpoint the details , the rest of the genes could be a mixed class I-II population in which Tat simultaneously interfere with both the initiation and elongation steps ( see Discussion ) . 10 . 7554/eLife . 08955 . 032Figure 6 . Global analysis of chromatin signatures reveals the basis of transcriptional repression by Tat . ( A ) Dot plots of H3K4me3 log2 fold change in the region encompassing -3/+3 Kb from the TSS of all TDG . TDG are divided into two classes: class I ( n = 11 ) and II ( n = 14 ) based on decreased or unchanged levels of H3K4me3 in the presence of Tat , respectively . Selected TDG are indicated in red . ( B ) Metagene plots centered on TSS showing H3K4me3 occupancy profiles at both class I and class II TDG in the presence of Tat or GFP . ( C ) Dot plots of H3K79me3 log2 fold change from TSS to TTS . ( D ) Metagene plots showing average H3K79me3 ChIP-seq signals at both class I and class II TDG in the presence of Tat or GFP ( see Materials and methods ) . ( E ) Dot plots of H3K36me3 log2 fold change in the region from TSS to TTS . ( F ) Metagene plots showing average H3K36me3 ChIP-seq signals at both class I and class II TDG in the presence of Tat or GFP . ( G ) Genome browser views showing ChIP-seq signal at a class I TDG ( CD1E ) in the GFP and Tat cell lines . ( H ) Genome browser views showing ChIP-seq signal at a class II TDG ( EOMES ) in the GFP and Tat cell lines . This figure is associated with Figure 6—figure supplement 1 . ChIP-seq , chromatin immunoprecipitation sequencing; GFP , green fluorescent protein; TDG , Tat downregulated genes; TSS , transcription start site; TTS , transcription termination site . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 03210 . 7554/eLife . 08955 . 033Figure 6—figure supplement 1 . Different modes of Tat repression at class I and II TDG . ( A ) Tat blocks transcription initiation of CD1E by interfering with assembly of the pre-initiation complex at the promoter . ( B ) Tat prevents elongation of EOMES by precluding P-TEFb loading at the promoter . ChIP-qPCR experiments with the indicated antibodies , in the GFP and Tat cell lines . The position of the amplicons used in ChIP-qPCR and their distance to the TSS ( arrow ) is indicated . Values represent the average of three independent experiments . The SEM is less than 10% and not shown for simplicity . ChIP-qPCR , chromatin immunoprecipitation-quantitative polymerase chain reaction; GFP , green fluorescent protein; SEM , standard error of the mean; TDG , Tat downregulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 033 Class I TDG experience a sharp decrease in H3K4me3 ( ∼4 . 2-fold change ) and in Pol II ( see below ) in the presence of Tat ( Figure 6A , B ) , consistent with a role of Tat in blocking transcription initiation . In support of this view , we observed a statistically significant correlation between the decrease in H3K4me3 levels surrounding the TSS and RNA levels at class I TDG but not class II ( data not shown ) . To further examine how Tat blocks the initiation step , we performed detailed ChIP-qPCR assays to monitor the occupancy of subunits of the transcription pre-initiation complex ( PIC ) . Interestingly , we observed that Tat recruitment to the CD1E gene promoter blocks the step of PIC assembly as denoted by the sharp decrease in the occupancy levels of TBP , Mediator ( MED1 ) , Pol II and the P-TEFb kinase ( CDK9 ) ( Figure 6—figure supplement 1A ) , in agreement with the virtual removal of H3K4me3 ( Figure 6A , G ) . On the contrary , class II TDG showed no significant change in H3K4me3 density surrounding the TSS ( Figure 6B , H ) , which is in agreement with the lack of Tat effects at the transcription initiation level . Consistently , at the class II TDG EOMES , Tat does not interfere with PIC assembly ( as revealed by identical levels of TBP and MED1 at the promoter ) . However , Tat blocks Pol II transition into productive elongation due to dismissal of P-TEFb from both the promoter and transcription unit ( Figure 6—figure supplement 1B ) , which affects levels of total Pol II and the elongating form ( Ser2P ) in the gene body but not at the promoter-proximal region at two class II TDG ( EOMES and HSPA8 ) ( Figure 3—figure supplement 2C , D ) . If Tat blocks transcription initiation , then we would predict that the chromatin elongation signatures would also decrease , as there is no recruited Pol II available for elongation . In agreement , class I TDG showed a virtual elimination of H3K79me3 and H3K36me3 levels in the transcribed unit , reflecting potent inhibition of these genes ( Figure 6C–F ) . The metagene analysis clearly indicated that all class I TDG showed reduced levels of both chromatin elongation signatures ( Figure 6D , F ) . However , the difference in magnitude at different genes was very broad , most likely due to the fact that these genes are transcribed at different levels and their abundance ( based on RNA-seq ) is quite disparate ( Figure 6C , E ) . Genome browser inspection of individual genes such as CD1E , which is expressed at high levels , provide evidence that Tat is a very potent transcription initiation repressor , as revealed by the large drop in all chromatin signatures profiled ( Figure 6G ) and inhibition of PIC assembly at the promoter , as revealed by the dismissal of TBP , MED1 , CDK9 and Pol II at the CD1E loci ( Figure 6—figure supplement 1A ) . In contrast to class I TDG , class II showed no significant change in H3K4me3 levels surrounding the TSS ( Figure 6A , B ) . However , a subset of class II genes such as EOMES , HSPA8 and CDK6 , showed reduced chromatin elongation signatures ( H3K79me3 and H3K36me3 ) throughout the transcribing unit ( Figure 6C , E , H ) . Genome browser inspection of individual class II TDG such as EOMES demonstrate that Tat blocks the transition to elongation because Pol II density largely decreases at the transcribing unit without significant alterations in the promoter-proximal region , consistent with decreased chromatin elongation signatures but no H3K4me3 effects ( Figure 6H ) . However , surprisingly , a few genes showed no change in H3K79me3 modification , or even a small increase , right after the TSS ( Figure 6C ) , possibly related to the fact that Dot1L-mediated establishment of H3K79me3 also has been linked with transcriptional repression ( Nguyen and Zhang , 2011; Onder et al . , 2012 ) . However , this alternative function of Dot1L in transcription repression will require further investigation . Taken together , it is evident that Tat functions as a transcriptional repressor blocking Pol II recruitment and pause release as well as promoting the removal of chromatin modifications associated with initiation and elongation at different gene classes . The fact that Tat modifies the epigenetic landscape to repress cellular transcription prompted us to examine Tat effects on Pol II distribution changes ( Figure 7 ) . If Tat represses transcription initiation by blocking PIC assembly such as in the CD1E gene then we would expect Pol II levels to decrease at gene promoters and other functionally relevant sites in the presence of Tat . We first analyzed levels of Pol II at both Tat binding sites ( Tat peak ) and promoters of associated genes for both class I and II TDG . Interestingly , we found that Tat blocks Pol II recruitment at both the Tat binding site and promoter-proximal region of class I TDG ( Figure 7A–C ) , in perfect agreement with the Tat-mediated decrease in H3K4me3 at those promoters ( Figure 6 ) . Remarkably , these observations are in agreement with the model that Tat blocks transcription initiation at class I TDG . Conversely , as expected based on the global analysis of chromatin signatures , Tat does not appear to largely block Pol II recruitment at the Tat peak and promoter of class II TDG ( Figure 7A , B , D ) , which we proposed to be regulated at the elongation level based on the global analysis of chromatin signatures and Pol II levels ( Figure 6 ) . In fact , this is in agreement with the lack of H3K4me3 density changes at promoters of class II TDG in the presence of Tat . 10 . 7554/eLife . 08955 . 034Figure 7 . Tat blocks Pol II recruitment and pause release to repress gene transcription at different gene classes . ( A ) Tat binding blocks Pol II recruitment primarily at class I TDG . Box plots of normalized Pol II tag density at the Tat peak of class I or II TDG in the GFP and Tat cell lines . ( B ) Tat binding at class I TDG blocks Pol II density at promoters . Box plots of normalized Pol II tag density at the promoter of class I ( n = 11 ) or class II ( n = 14 ) TDG in the GFP and Tat cell lines . ( C ) Normalized Pol II tag density relative to the Tat peak at class I TDG . ( D ) Normalized Pol II tag density relative to the Tat peak at class II TDG . ( E ) Pol II distribution at class I TDG ( Metagene plots ) in the Tat and GFP cell lines . ( F ) Genome browser views of ChIP-seq data at a class I TDG ( CD1E ) in the Tat and GFP cell lines . ( G ) Pol II distribution at class II TDG ( Metagene plots ) in the Tat and GFP . ( H ) Genome browser views of ChIP-seq data at a class II TDG ( EOMES ) in the Tat and GFP cell lines . This figure is associated with Figure 7—figure supplement 1 . ChIP-seq , chromatin immunoprecipitation sequencing; GFP , green fluorescent protein; Pol , polymerase; TDG , Tat downregulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 03410 . 7554/eLife . 08955 . 035Figure 7—figure supplement 1 . mTR reveals that Tat blocks both Pol II recruitment and pause release at TDG . Calculations of the ratio of paused Pol II ( pPol II ) and traveling Pol II ( tPol II ) at class I and class II TDG in the presence and absence of Tat ( Tat/GFP ) . GFP , green fluorescent protein; mTR , modified traveling ratio; Pol , polymerase; TDG , Tat downregulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 035 Metagene analysis of class I TDG showed high Pol II density levels in both the gene and transcribing unit in the absence of Tat . However , Tat reduces Pol II density at the promoter , consistent with a drop in transcribing Pol II at the gene body ( Figure 7E ) . Genome browser views of individual class I TDG such as CD1E clearly exemplify the metagene analysis depicting that Pol II recruitment at the promoter is strongly blocked , leading to the virtual disappearance of Pol II signal throughout the gene body . The transcription initiation signature H3K4me3 is also eliminated ( Figure 7F ) , and the blockage of transcription initiation correlates with decrease in chromatin signatures associated with transcription elongation ( H3K79me3 and H3K36me3 ) throughout the gene body ( Figure 6 ) . We have defined class II TDG as being primarily controlled at the elongation step because H3K4me3 and Pol II density at the promoter-proximal region does not appear to be modified by Tat ( Figures 6 and 7 ) . Consistently , a metagene analysis shows that Tat decreases Pol II density throughout the gene body and 3’-end , but not at the promoter-proximal region , in agreement with a role of Tat in blocking the transition to elongation ( Figure 7G ) . Genome browser views of individual class II TDG such as EOMES are consistent with the block of Pol II pause release ( Figure 7H ) , and with the global analysis of chromatin signatures related to elongation ( Figure 6 ) . Reduced elongation is due to the block of P-TEFb recruitment to promoters in class II TDG ( EOMES and HSPA8 ) ( Figure 3—figure supplement 2 ) . Moreover , simultaneous analysis of the poised and transcribing Pol II forms ( pPol II and tPol II ) using our mTR algorithm provides further evidence that , in the majority of class II TDG , Tat dampens the level of tPol II without largely modifying or even increasing pPol II ( Figure 7—figure supplement 1 ) , consistent with a role in primarily blocking the transition into elongation . In conclusion , the data indicate that Tat precisely modulates Pol II recruitment and transition into the gene body to block transcription initiation or elongation , respectively , at different gene classes . Given that Tat activates transcription from the HIV promoter by associating with the TAR structure that is formed at the nascent chain of viral pre-mRNAs ( Frankel and Young , 1998 ) ( Figure 8A ) , we investigated the possibility that Tat is recruited to the host chromatin through interaction with TAR-like structures . To test whether the Tat sites at the direct target genes ( both TSG and TDG ) contain TAR-like structures , we searched for TAR-like motifs using a custom algorithm and the input sequence X ( 2 ) GATX ( 1 , 2 ) GAX ( 4 , 40 ) TCTCX ( 2 ) as query ( Figure 8B ) , where X denotes any nucleotide , and the numbers in brackets represent the minimal and maximal allowed positions with the corresponding secondary structure pattern XX ( ( XX ( ( X ( 4–40 ) -X ) ) ) ) XX including the di-/tri-nucleotide bulge within the stem and loop ( both critical determinants for TAR binding ) ( Frankel and Young , 1998 ) . Locations of TAR-like motifs near target genes were cataloged and compared to distribution of Tat peaks in our ChIP-seq dataset . While ∼20% of the TSG and TDG contain TAR-like motifs within a very large window ( <10 Kb ) from the Tat site , an insignificant number ( ∼1% ) contain a TAR-like motif in close proximity ( <0 . 1 Kb ) to the Tat site identified by ChIP-seq ( Figure 8C , D ) . The data suggest that there is no significant presence of TAR-like motifs at or near Tat peaks , and only minimal examples of a TAR-like motif within 10 Kb of a Tat peak . Thus , it seems unlikely that the more prevalent mechanism of Tat recruitment to its target genes is through interaction with HIV TAR mimics . 10 . 7554/eLife . 08955 . 036Figure 8 . Enrichment of master transcriptional regulators , but not TAR-like , motifs on Tat sites at the direct target genes . ( A ) TAR sequence and scheme of the secondary structure . ( B ) Schematic of the sequence query used to search for TAR-like motifs within the direct Tat target genes . ( C ) TAR-like motifs are not found at or significantly near Tat peaks in TSG . Fraction of TSG containing TAR-like motifs at <10 kb , <1 kb or <0 . 1 kb from the Tat peak . ( D ) TAR-like motifs are not found at or significantly near Tat peaks in TDG . Fraction of TDG containing TAR-like motifs at <10 kb , <1 kb or <0 . 1 kb from the Tat peak . ( E ) MEME analysis of 200-bp windows surrounding Tat peaks in TSG reveals high-confidence motifs related to transcription factors ETS1 and RUNX1 . ( F ) MEME analysis of 200-bp windows surrounding Tat peaks in TDG reveal different motifs for ETS1 and RUNX1 . ( G ) Comparison of ETS1 ChIP-seq peak locations in Jurkat , as determined by Hollenhorst et al . ( Hollenhorst et al . , 2009 ) , to Tat peak locations reveal that 73% of TSG contain an ETS1 peak within 100-bp of a Tat peak . ( H ) Comparison of ETS1 ChIP-seq peak locations to Tat peak locations reveal that 80% of TDG contain an ETS1 peak within 100-bp of a Tat peak . This figure is associated with Figure 8—figure supplement 1 . ChIP-seq , chromatin immunoprecipitation sequencing; TAR , trans-activating RNA; TDG , Tat downregulated genes; TSG , Tat stimulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 03610 . 7554/eLife . 08955 . 037Figure 8—figure supplement 1 . The interaction between Tat and host cell chromatin appears to be primarily dictated by protein–protein interactions . ( A ) Fractionation scheme of Jurkat-GFP and Jurkat-Tat cells by increased salt extraction in the absence ( – ) or presence ( + ) of RNase A . Ch denotes the chromatin fraction . ( B ) Western blots of the samples prepared as in panel ( A ) with the indicated antibodies . The efficiency of RNase treatment was verified by electrophoresis of the purified RNA in an agarose gel stained with ethidium bromide . ( C ) ChIP assays to analyze the occupancy of GFP and Tat ( FLAG ) at the CD69 promoter ( -63 amplicon ) in the absence ( – ) and presence ( + ) of RNase . ( D ) ChIP assays to analyze the occupancy of GFP and Tat ( FLAG ) at the FAM46C promoter ( -182 amplicon ) in the absence ( – ) and presence ( + ) of RNase . ( E ) ChIP assays to analyze the occupancy of GFP and Tat ( FLAG ) at the CD1E promoter ( +4 amplicon ) in the absence ( – ) and presence ( + ) of RNase . ( F ) ChIP assays to analyze the occupancy of GFP and Tat ( FLAG ) at the EOMES promoter ( -57 amplicon ) in the absence ( – ) and presence ( + ) of RNase ( mean ± SEM; n = 3 ) . ChIP , chromatin immunoprecipitation; GFP , green fluorescent protein; SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 037 Given that Tat’s interaction with its target genes appears not to rely on TAR-like motifs , we reasoned that Tat might be recruited through interaction with master transcriptional regulators . To identify candidates , we submitted 200-bp windows surrounding Tat peaks from TSG and TDG to the DREME ( Discriminative Regular Expression Motif ) analysis tool ( Bailey , 2011 ) , which returned motifs matching T cell master regulators ( ETS1 , RUNX1 ) and GATA3 transcription factors for TSG , and ETS1 and RUNX1 for TDG , with statistically significant higher p-values in TSG ( 2 . 5 × 10-–18 and 9 . 6 × 10–14 , respectively ) ( Figure 8E , F ) , without enrichment of other factors expressed in T cells such as members of the signal transducer and activator of Transcription ( STAT ) family . Remarkably , enrichment of both ETS1 and RUNX1 at TSG correlates with the observed functional annotation and their known role in modulating T cell activation ( Figure 1 ) ( Hollenhorst et al . , 2009; Hollenhorst et al . , 2011 ) . Given that ETS1 motifs are present at both TSG and TDG , we examined whether the presence of these motifs inform about the mode of Tat effect on cellular genes . To test this , we looked to see whether ETS1 motifs are more prevalent in class I or II TSG or TDG , and observed that the number of target genes containing ETS1 binding sites within 100-bp from the center of the Tat peak is: 11/17 ( 64 . 7% ) for class I TSG , 36/43 ( 83 . 72% ) for class II TSG , 8/14 ( 57 . 1% ) for class I TDG , and 6/11 ( 54 . 5% ) for class II TDG . This analysis indicates that the presence of an ETS1 motif at the Tat target genes is not by itself sufficient to determine the mechanism of action ( stimulation or downregulation ) or whether Tat modulates the initiation or elongation step of transcription . ETS1 may help recruit Tat to chromatin but it needs another determinant for specificity or mode of action , most likely related to the transcriptional activity status or yet unidentified co-factors ( protein and/or long-non coding RNA ) , which might function in a gene-specific manner . Given that the highest-confidence motif returned for both TSG and TDG is ETS1 , we examined whether this was indicative of co-occupancy of ETS1 with Tat in these genes . We first retrieved ETS1 ChIP-seq data in Jurkat T cells ( Hollenhorst et al . , 2009 ) and compared locations of ETS1 peaks to the locations of Tat peaks in our dataset ( Figure 8G , H ) . Contrary to our results when comparing TAR-like motifs sites to Tat ChIP-seq peaks , we discovered that 73% of TSG and 80% of TDG contain an ETS1 ChIP-seq peak within 100-bp of a Tat ChIP-seq peak . Together , the data suggest a model where Tat is recruited to host cell chromatin through interaction with T-cell identity factors such as ETS1 and not through TAR-like motifs , thus revealing unique and unexpected recruitment mechanisms . ETS1 ChIP-seq revealed 19 , 049 sites in the genome ( Hollenhorst et al . , 2009 ) . Interestingly , nearly 52% of the Tat peaks detected by ChIP-seq ( 3203/6117 ) harbor an ETS1 binding event within 100-bp , significantly more than expected from random occurrence ( p-value 1×10-8536 , Hypergeometric test ) . Although a large number of both TSG and TDG contain ETS1 motifs , the enriched motifs are different between both classes ( 5’-AGGAAG/AT/C-3’ and 5’-CNGGAA-3’ , respectively ) ( Figure 8E , F ) , even though both contain the signature motif for ETS1 binding ( 5’-GGAA-3’ ) . Thus , it would be interesting to determine whether these different motifs dictate a diverse binding mode and could inform about the mechanism of activation and repression through the same transcription factor , and whether the DNA binding site directs recruitment of specific cofactors differentially targeted by Tat at TSG and TDG . Given that we found no significant evidence of Tat interaction with TAR mimics in the human genome as well as evidence of Tat interaction with ETS1 , we reasoned that one clear mechanism by which Tat is recruited to chromatin is through direct protein-protein interactions . To biochemically test that Tat is recruited to chromatin in a RNA-independent manner , we performed a cellular fractionation by sequential salt extraction in the absence and presence of RNase ( Figure 8—figure supplement 1A ) and observed that: ( i ) Tat is present in the nucleoplasm ( 100 mM salt fraction ) as well as bound to chromatin ( eluted at 600 mM salt ) from Jurkat nuclear extracts , while GFP is mainly detected in the 100 mM salt fraction; and ( ii ) the interaction between Tat and chromatin is primarily dictated by protein-protein interaction , although a minor fraction appears to be RNA-dependent ( Figure 8—figure supplement 1B ) . To test this possibility further , we performed ChIP-qPCR assays at select TSG and TDG by preparing samples in the presence and absence of RNase and observed that the treatment does not affect Tat binding to four different target gene promoters ( Figure 8—figure supplement 1C ) . Although we cannot strictly rule out that Tat can combinatorially interact with target proteins and RNA species present at different genomic domains , our data indicate that Tat interaction with chromatin is primarily dictated by protein-protein interactions . Given that we found a significant enrichment of ETS1 motifs in Tat sites ( Figure 8 ) we reasoned that ETS1 mediates Tat recruitment to chromatin . To test whether the two proteins interact we performed Strep affinity purifications ( AP ) using nuclear fractions prepared from the Jurkat T-cell lines followed by western blot analysis . The data indicate that Tat , but not the C22A non-functional mutant or GFP , interacts with endogenous ETS1 , as well as with the P-TEFb kinase ( CDK9 ) used as positive control , thereby showing that the protein-protein interaction is specific ( Figure 9A and Figure 9—figure supplement 1A ) . The facts that Tat and ETS1 interact , ETS1 motifs are enriched at the Tat target genes , and ETS1 binds these genes ( as revealed by ETS1 ChIP-seq in Jurkat T cells [Hollenhorst et al . , 2009] ) prompted us to test whether both proteins co-occupy target genes . To test this possibility we performed ChIP assays on the Jurkat-GFP and -Tat cell lines using primer-pairs that amplify both promoter-proximal and promoter-distal regions of different Tat target genes . We found that Tat and ETS1 co-occupy the CD69 promoter but not gene body ( Figure 9B ) , consistent with the motif enrichment analysis and a previous ETS1 ChIP-seq dataset ( Hollenhorst et al . , 2009 ) . Furthermore , Tat binding at the CD69 promoter does not appear to alter the occupancy levels of ETS1 , since the density of ETS1 in the absence and presence of Tat is similar , if not identical ( Figure 9B ) . In addition , Tat binds some target genes lacking ETS1 motifs such as CD1E ( Figure 9C ) . These results provide evidence that Tat also is recruited in an ETS1-independent manner thereby indicating that additional recruitment mechanisms exist . 10 . 7554/eLife . 08955 . 038Figure 9 . Tat is recruited to its target genes through interaction with the master transcriptional regulator ETS1 . ( A ) Western blots showing interactions between Tat and ETS1 . CDK9 was used as a positive control in the interaction . Strep-tagged Tat and GFP were AP from the Jurkat cell lines using Strep beads , and analyzed by western blot using the indicated antibodies . ( B ) Tat is recruited to target genes marked by ETS1 . ChIP assays showing that Tat and ETS1 co-occupy the promoter-proximal but not promoter-distal region of CD69 , in agreement with the location of Tat and ETS1 peaks found by ChIP-seq and the presence of ETS1 motifs ( as revealed by enrichment analysis ) . ( C ) Tat is also recruited to target genes using ETS1-independent mechanisms . ChIP assays showing that Tat but not ETS1 occupies the promoter-proximal region of the CD1E target gene , in agreement with the Tat and ETS1 ChIP-seq dataset and motif prediction analysis . ( D ) Generation of Jurkat-GFP and -Tat cell lines expressing non-target ( NT ) or ETS1 shRNAs . ( – ) denotes the parental untransduced Jurkat cell lines . Protein lysates were analyzed by western blot using the indicated antibodies to verify for the RNAi efficiency . ( E ) ETS1 knockdown impairs Tat recruitment to the CD69 promoter . ChIP-qPCR assays showing the density of ETS1 or FLAG at the CD69 promoter in the Jurkat-GFP or -Tat cell lines transduced with the NT or ETS1 shRNAs . ( F ) ETS1 knockdown does not abolish Tat recruitment to the CD1E promoter . ChIP-qPCR assays showing the density of ETS1 or FLAG at the CD1E promoter in the Jurkat-GFP or -Tat cell lines transduced with the NT or ETS1 shRNAs . This figure is associated with Figure 9—figure supplement 1 . AP , affinity purified; ChIP-seq , chromatin immunoprecipitation sequencing; GFP , green fluorescent protein; NT , non-target; qPCR , quantitative polymerase chain reaction; shRNA , small hairpin RNA . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 03810 . 7554/eLife . 08955 . 039Figure 9—figure supplement 1 . The C22A non-functional Tat mutant fails to bind ETS1 and evidence that ETS1 is critical for transcription activation of Tat target genes . ( A ) Western blots showing interactions between Tat ( but not GFP or the C22A non-functional Tat mutant ) and ETS1 . CDK9 was used as a positive control in the interaction . Strep-tagged GFP , Tat , and C22A were AP from the Jurkat-GFP and -Tat cell lines using Strep beads , and analyzed by western blot using the indicated antibodies . ( B ) qRT-PCR analysis of CD69 expression in the indicated cell lines . ( C ) qRT-PCR analysis of ADCYAP1 expression in the indicated cell lines . ( D ) qRT-PCR analysis of VAV3 expression in the indicated cell lines . ( E ) qRT-PCR analysis of FAM46C expression in the indicated cell lines . ( F ) qRT-PCR analysis of RPL19 expression in the indicated cell lines . ( G ) qRT-PCR analysis of 7SK expression in the indicated cell lines . Expression of the indicated genes ( panels B–G ) was normalized to ACTB . Values represent the average of three independent experiments ( mean ± SEM; n = 3 ) . AP , affinity purified; GFP , green fluorescent protein; qRT-PCR , quantitative real-time polymerase chain reaction; SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 08955 . 039 Given that Tat and ETS1 interact and co-occupy target gene promoters , we asked whether ETS1 ( already bound to DNA elements ) recruits Tat to chromatin to reprogram gene transcription . To test this , we generated Jurkat-GFP and -Tat cell lines expressing ETS1-specific shRNAs to efficiently knockdown ETS1 or a non-target ( NT ) shRNA as negative control ( Figure 9D ) . Remarkably , we observed that ETS1 knockdown consistently diminishes ( ∼6-fold ) ETS1 density at the CD69 promoter in both GFP and Tat cell lines with a concomitant loss of Tat signal ( ∼6-fold ) at the promoter ( Figure 9E ) . Importantly , ETS1 knockdown does not alter the recruitment of Tat to the CD1E promoter , which is regulated in an ETS1-independent manner ( Figure 9F ) . Given that Tat binds ETS1 and both co-occupy promoter-proximal regions of selected target genes , we examined whether ETS1 is critical for transcriptional activation of CD69 and three other TSG and observed that ETS1 knockdown interferes with the Tat-mediated increase in target gene stimulation for both class I and II TSG ( Figure 9—figure supplement 1B–E ) without affecting RNA steady-state levels of non-target genes such as RPL19 and 7SK ( Figure 9—figure supplement 1F , and G ) . The evidence that ETS1 is critical for Tat's transcriptional activation of these four selected TSG correlates with the Tat-mediated increase in the levels of Pol II and the chromatin marks coinciding with transcription initiation and elongation at both class I and II TSG ( Figures 2–5 ) . Together , we provide compelling evidence that Tat is recruited to chromatin through interaction with the T-cell identity factor ETS1 to reprogram cellular transcription . Further studies are needed to determine whether ETS1 functions as a scaffold to promote Tat recruitment or whether Tat induces protein conformational changes to activate or repress ETS1 regulatory transcriptional programs ( see Discussion ) . Several studies have reported that Tat binds the human genome to modulate cellular gene expression to alter the biology of immune cells ( dendritic and CD4+ T ) and generate a permissive environment for viral replication and/or spread ( Huang et al . , 1998; Izmailova et al . , 2003; Kim et al . , 2010a; 2013; Kukkonen et al . , 2014; Li et al . , 1997; Marban et al . , 2011 ) . However , a comprehensive description of the direct target genes and the nature of the regulatory mechanisms have yet to be discovered . In this article , we report a technically improved Tat ChIP-seq assay that dramatically increases sensitivity compared with previous methodologies . Furthermore , by simultaneously assaying transcriptome changes and Tat’s genome-wide distribution we combinatorially identified direct Tat targets in the human genome with high confidence . In contrast to previous studies , we were able to elucidate a large proportion of Tat binding sites in the human genome that correlate with marked gene expression changes ( activation or repression ) at both the RNA and protein levels . Importantly , the Tat target genes shared functional annotations , and are regulated ( for the most part ) by a common set of master transcriptional regulators . Transcription factors coordinate the activation and maintenance of transcriptional programs by regulating one or multiple steps in the transcriptional cycle . While some sequence-specific DNA binding transcription factors recruit Pol II and the basal transcription apparatus to promote transcription initiation ( Hahn , 2004 ) , others function at the elongation step by allowing Pol II transition from a promoter paused state to the productive elongation phase ( Adelman and Lis , 2012; Rahl et al . , 2010; Zhou et al . , 2012 ) . By performing a global analysis of chromatin signatures that are generally associated with different genomic domains ( promoter , coding units and enhancers ) together with genome-wide Pol II distribution data in the absence and presence of Tat , we described , for the first time , the precise nature of the mechanisms of transcriptional activation and repression by Tat . Strikingly , we found that Tat directly controls both the initiation and elongation steps to transcriptionally reprogram the cell . Tat promotes Pol II recruitment at promoters and transcribing units to stimulate the transcription initiation and elongation steps , respectively . Conversely , Tat blocks Pol II recruitment to promoters or transcribing units to prevent the initiation and elongation steps , respectively , thereby leading to gene repression . Remarkably , the global analysis of chromatin signatures is consistent ( for the most part ) with the proposed mechanisms based on Pol II occupancy changes . For example , class I TSG ( regulated at the initiation step ) showed a large increase in both Pol II and H3K4me3 at promoter-proximal regions , consistent with their functional link in activating gene transcription ( Shilatifard , 2012 ) . Although we have not searched for the H3K4 methylase , it would be interesting to define whether Tat hijacks one or more of Set/MLL methylases and whether DoT1L and the SEC are co-recruited to link transcription initiation with elongation ( Luo et al . , 2012; Nguyen and Zhang , 2011; Shilatifard , 2012 ) . In addition , the mechanism by which Tat reduces H3K4me3 density at some class II TSG remains yet unknown , but it may be possible that Tat recruitment to the promoter region interferes with the H3K4me3 signature , as has been seen with the NS1 protein during influenza virus infection ( Marazzi et al . , 2012 ) . Although we have described that Tat associates with and recruits chromatin-modifying enzymes to target genes , one important aspect of the mechanism that we have not clarified yet is that the histone modifications may represent ( in some cases ) indirect effects of changes in transcription and not a direct consequence of Tat function on the genome . Despite the canonical view of transcriptional regulation through transcription factor-promoter DNA interaction , it has recently become evident that internal promoters , intragenic enhancers and other genomic elements contribute to specify transcriptional programs through the formation of three-dimensional structures ( Bulger and Groudine , 2011; Creyghton et al . , 2010; Ghavi-Helm et al . , 2014; Heintzman et al . , 2007; Jin et al . , 2013 ) . Consistent with these discoveries , we found that Tat exploits the human genome by binding not only at promoters but also at intergenic and intragenic sites to modulate long-range chromatin interactions and transcription activation from these promoter-distal sites . We observed that at a large number of class I TSG ( regulated at the initiation step ) Tat binds at intragenic sites to modulate enhancer activity . In the absence of Tat , the nucleosomes surrounding these sites contain low or undetectable histone modifications related to enhancer activity ( H3K4me1 and H3K27Ac ) . Tat binding increases their density in a manner that is proportional to transcriptional levels . Notably , this increase also correlates with the recruitment of chromatin-modifying enzymes ( p300/CBP ) at the Tat site , chromatin looping between the internal site and the promoter , and Pol II recruitment to trigger de novo transcription initiation . The role of intragenic Pol II pausing requires further investigation but it may be possible that this form of Pol II is required for active transcription elongation by promoting template DNA circularization or acting after gene looping . This Pol II form might not have been observed in embryonic stem cells because most , if not all , genes are incompetent for elongation and Pol II is primarily paused in the promoter-proximal region . Recent evidence suggested that internal sites marked with H3K27Ac appeared to be a sort of intragenic enhancers ( Kowalczyk et al . , 2012 ) , implying that Tat directly dictates gene activation by binding at these sites and is likely to be a major determinant of the overall architecture and the composition of histone modifications of these internal sites . The gene looping hypothesis provides a mechanistic explanation for the molecular effects observed at the promoter ( namely Pol II recruitment and increase in H3K4me3 density ) in response to Tat binding at promoter-distal sites , even without detectable Tat at the promoter-proximal region . Although we have shown that the C22A non-functional mutant ( which does not dimerize ) does not promote gene looping , it would be interesting to further test whether the Tat-mediated long-range chromatin interactions are controlled by protein homo-dimerization , a feature of Tat that was originally described by Frankel and Pabo ( Frankel et al . , 1988 ) , but whose function has since then remained largely elusive . Furthermore , the role of intragenic enhancers in activating gene expression is an attractive possibility , and might help recruit chromatin-modifying enzymes for compartmentalization purposes and/or ‘on site’ activation . Although in this report we did not thoroughly examine Tat binding at intergenic regions and their functional role in transcriptional control , it is plausible that Tat binds these genomic domains ( enhancers ) to control long-range interactions and the assembly of transcription complexes to modulate transcription activation/repression . Given that transcription from enhancers is a widespread regulatory mechanism to modulate the activity of nearby genes , further research is needed to understand the role of Tat in controlling transcription through enhancers and the potential role of the newly identified class of enhancer-derived non-coding RNAs ( eRNAs ) in transcription activation or repression ( Kim et al . , 2010b; Kim and Shiekhattar , 2015 ) . Although we have detected Tat-induced RNAs from enhancers that are co-regulated with nearby genes , further investigation is also needed to define the molecular mechanisms by which these Tat-induced non-coding RNAs function during the reprogramming process . Although the majority of transcription factors described to date contact DNA directly , Tat is unique because it binds the nascent RNA structure ( TAR ) formed at the HIV promoter . Surprisingly , in contrast to activation of the HIV genome , we found insignificant enrichment of TAR-like motifs at Tat binding sites in the human genome . Although we cannot completely exclude the possibility that nascent RNA chains ( folding or not into TAR-like structures ) help Tat recruitment to host cell chromatin , we unexpectedly found that Tat occupies sites bound by master transcriptional regulators ( ETS1 ) with high frequency . Notably , we provided biochemical and genetic evidence that ETS1 recruits Tat to chromatin to modulate ( activate or repress ) gene transcription . ETS1 is a member of a large family of transcription factors ( ETS ) that play important roles in T-cell stimulation and differentiation ( Hollenhorst et al . , 2011 ) . ETS1 requires combinatorial interactions with other factors ( such as RUNX1 ) to activate transcription . We propose that , similarly to RUNX1 , Tat uses ETS1 as a scaffold to promote its recruitment to target genes and modulate gene transcription . Given that ETS1 is found at both activated and repressed genes , it is evident that ETS1 does not dictate per se the mode of regulation ( activation or repression ) . For the mechanism of gene activation , a likely scenario would be that Tat is recruited to ETS1 to relieve its auto-inhibition and recruit Pol II , elongation factors and chromatin-modifying enzymes . For the mechanism of gene repression , Tat might compete off factors pre-associated with ETS1 ( such as RUNX1 ) and block PIC assembly ( in the case of transcription initiation blockage ) or prevent the action of elongation factors such as P-TEFb ( in the case of transcription elongation repression ) . Although ETS1 appears to play a central role in the recruitment of Tat to several direct target genes ( irrespective of the transcriptional outcome ) , it is completely possible that several other co-factors including long non-coding RNAs and/or proteins function combinatorially to specify target loci identification and Tat function . Although we have provided important insights into the recruitment mechanisms , further investigation is needed to clarify the molecular basis of the Tat-ETS1 protein-protein interaction in the regulation of gene activation and repression . One particular observation derived from the motif analysis is that the ETS1 motif identified at both TSG and TDG contain a common sequence ( 5’-GGAA-3’ ) but differ in the -1/-2 positions . Of note , transcription factor binding sites slightly differing in sequence have been shown to modulate various steps of the transcription cycle including transcription factor binding affinity and conformational changes upon their recruitment to target sequences , as well as co-factor recruitment ( Meijsing et al . , 2009 ) . Therefore , the difference in the ETS1 motifs at TSG and TDG might contain information utilized by Tat ( and/or Tat co-factors ) to activate or repress transcription . Undoubtedly , further research is required to elucidate the precise molecular basis . Despite the widespread role of ETS1 in recruiting Tat to chromatin , it is most likely that other recruitment strategies exist because not all Tat target genes identified are regulated by ETS1 . Defining all the details will require higher-resolution approaches such as ChIP-exo and ChIP-nexus to improve the definition of Tat target sites and mechanisms of recruitment to chromatin ( He et al . , 2015; Rhee and Pugh , 2011 ) . In addition , the analysis of transcriptome changes using RNA-seq ( which measures steady-state RNA levels and not transcription per se ) has some caveats , for example difficulties in detecting low abundance or highly unstable RNAs . Potentially , this may help explain why we also found a large fraction of binding events that do not have a correlate with gene expression changes . Other tools such as GRO-seq ( Core et al . , 2008 ) , which measure levels of nascent RNAs , will also be needed to further improve the definition of direct and indirect target genes . Moreover , a more recent and powerful technique named NET-seq yields transcriptional activity at nucleotide resolution and thus outperforms GRO-seq ( Mayer et al . , 2015 ) . Certainly , these tools will provide much higher-resolution to define target sites , modes of Tat recruitment and improved functional insights . Nonetheless , our work provides the first comprehensive view of how Tat modulates the biology of immune target cells by mediating key transcriptional changes . Interestingly , Tat can potentially perform different functions depending on the target cell type . Thus , it would be informative to determine how host adaptability is harnessed by a viral protein to selectively reprogram transcription in a cell lineage-specific manner . In conclusion , despite the complexity of transcriptional regulatory mechanisms in the cell , Tat precisely controls Pol II recruitment and pause release to fine-tune the initiation and elongation steps , respectively . It is possible that the diversity of mechanisms employed by Tat to reprogram the host cell arise from the intrinsic complexity of transcriptional regulatory strategies of human genes . Finally , our data provide yet another example on how a virus with a limited coding capacity optimized its genome by evolving a small but ‘multi-tasking’ protein to simultaneously control viral and cellular transcription using distinct regulatory strategies . All scripting was performed using python 2 . 7 . 6 . All ChIP-seq binding events were loaded into a custom MySQL database to allow for efficient comparisons of multiple factors’ binding loci .	The human immunodeficiency virus ( HIV ) reproduces and spreads throughout the body by hijacking human immune cells and causing them to copy the virus’s genetic information . As the virus multiplies , it also causes the death of the immune system cells that help the human body recognize and eliminate viruses . This allows the virus to multiply unchecked . Studies of the genetic material of HIV – which is in the form of single-stranded RNA molecules and contains only a handful of genes – have begun to reveal how the virus can wreak such havoc to the human immune system . A small protein encoded by the virus , called Tat , boosts the expression of HIV genes in infected immune cells by binding to a structure that forms on newly synthesized viral RNAs . Recent evidence suggests that HIV also changes the expression of human genes to make immune cells more hospitable to the virus . However , it was not known exactly which specific genes are targeted , or how the virus alters their expression . Now , Reeder , Kwak et al . reveal how the Tat protein alters the expression of more than 400 human genes . Rather than bind to the same structure seen in newly forming HIV RNAs , Tat turns on or off the expression of its human target genes by interacting with proteins that regulate human gene expression . In doing so , Tat is able to precisely control the activity of an enzyme called RNA Polymerase II that is necessary for the early steps of gene expression . Tat’s multitasking ability – boosting HIV gene expression at the same time as reprogramming human gene expression – helps explain how a virus with so little genetic material of its own can perform such a wide range of activities in infected cells . The work of Reeder , Kwak et al . suggests that Tat reshapes the human genome to position target genes in ways that allow them to be efficiently turned on or off . Future studies will further reveal how Tat accomplishes this genome remodeling during different stages of infection . In addition , further research is also necessary to look closely into the sets of genes targeted by Tat to find patterns of genes that work together to alter cell behavior , and investigate how these new behaviors allow HIV to thrive .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2015	HIV Tat controls RNA Polymerase II and the epigenetic landscape to transcriptionally reprogram target immune cells
The ascidian embryo is an ideal system to investigate how cell position is determined during embryogenesis . Using 3D timelapse imaging and computational methods we analyzed the planar cell divisions in ascidian early embryos and found that spindles in every cell tend to align at metaphase in the long length of the apical surface except in cells undergoing unequal cleavage . Furthermore , the invariant and conserved cleavage pattern of ascidian embryos was found to consist in alternate planar cell divisions between ectoderm and endomesoderm . In order to test the importance of alternate cell divisions we manipulated zygotic transcription induced by β-catenin or downregulated wee1 activity , both of which abolish this cell cycle asynchrony . Crucially , abolishing cell cycle asynchrony consistently disrupted the spindle orienting mechanism underpinning the invariant cleavage pattern . Our results demonstrate how an evolutionary conserved cell cycle asynchrony maintains the invariant cleavage pattern driving morphogenesis of the ascidian blastula . In chordate embryos the functional pattern of cells is generated before gastrulation such that a fate map for all chordate embryos at the blastula stage predicts that cells in different positions will give rise to new cell types and layers that are important for morphogenesis ( Kourakis and Smith , 2005 ) . Invertebrate chordate embryos of the ascidian display a similar fate map to other chordates even though their blastulae are composed of only 64 cells rather than several thousand cells typical of other chordates ( Kourakis and Smith , 2005; Lemaire et al . , 2008 ) . Due to its conserved fate map yet small cell number , the ascidian embryo is an ideal system to elucidate mechanisms underpinning cell positioning during morphogenesis of a chordate blastula . Because ascidian embryos display an invariant cleavage pattern with no cell migration or cell death up to the time of gastrulation , cell division plane orientation is important for pattern formation ( McDougall et al . , 2015 ) . In addition , although the invariant-cleavage pattern displayed by asicidian embryos is specific to ascidians , the lophotrochozoa also display stereotyped spiral cleavage patterns that may employ similar rules as the ones we address here in the ascidian ( Rabinowitz and Lambert , 2010; Brun-Usan et al . , 2017 ) . All blastomeres are fate restricted in the 64 cell ascidian blastula and most cell divisions at the 32 cell stage are fate segregating asymmetric divisions ( Lemaire et al . , 2008; Lemaire , 2009 ) . Such fate-segregating asymmetric cell divisions rely on precisely regulated cell divisions partitioning maternal determinants ( muscle lineage ) or allowing local cell-cell contacts for polarised inductive cell fate specification e . g . neural lineage induction ( Kumano and Nishida , 2007; Lemaire , 2009 ) . The invariant cleavage pattern of ascidian embryos is a relatively simple morphogenetic process operating at the level of the whole embryo that is amenable to genetic analysis in order to identify the gene regulatory networks ( GRNs ) controlling cell division orientation . The overall invariant temporal and spatial pattern of cell divisions in ascidians is even conserved between distantly-related species . Ascidians are split into three orders that diverged more than 350 millions years ago ( aplousobranch , phlebobranch and stolidobranch ) , and it has been estimated that non-coding DNA sequences from two distinct ascidian species can be as different from each other as fish are from mammals ( Stolfi et al . , 2014 ) . Distantly-related species of ascidian also show the same relative cell cycle asynchrony since the 24 cell stage embryo is common to both phlebobranch ( Ciona/Phallusia ) and stolidobranch ( Halocynthia/Styela ) ascidians . In Phallusia this cell cycle asynchrony is induced by a GRN dependent upon nuclear accumulation of β-catenin in six vegetal cells of the 16 cell stage embryo ( Dumollard et al . , 2013 ) . How such stereotyped cell cycle asynchrony has been conserved in distantly-related ascidians is presently unknown , but it is interesting to note that β-catenin becomes nuclear in vegetal blastomeres in both Ciona and Halocynthia embryos at the 16 cell stage ( Kawai et al . , 2007; Hudson et al . , 2013 ) . Mitotic spindles align relative to a number of cues that display a competitive hierarchal relationship with one another . For example , an underlying mechanism known as the long-axis rule based upon microtubule behavior and motors ( reviewed in Minc and Piel , 2012 ) causes animal cell to divide orthogonal to their long axis as was noted more than a century ago by Hertwig ( Hertwig , 1893 ) . This geometric long-axis rule can be altered by cortical polarity cues such as lateral junctions ( Nakajima et al . , 2013; Ragkousi and Gibson , 2014 ) or the apical cortex in asymmetrically dividing Drosophila neuroblasts ( Siller and Doe , 2009 ) . During planar cell divisions in epithelia and endothelia , a lateral belt of LGN/NuMA coupled with the exclusion of LGN/NuMA from the apical cortex causes planar spindle orientation ( Zheng et al . , 2010; Morin and Bellaïche , 2011 ) . After acquiring a planar orientation the spindle rotates in the apical plane to find its final position at metaphase . Spindle orientation in the apical plane will set cell position in the epithelium and is regulated by apical cell shape ( Ragkousi and Gibson , 2014 ) . Because of mitotic cell rounding in cultured cells and some epithelia , apical cell shape at metaphase may become completely round ( Lancaster and Baum , 2014 ) . In these cells , the spindle aligns with the long axis of the cell during interphase which is memorized during mitotic cell rounding via retraction fibers in cultured cells ( Théry and Bornens , 2008 ) or LGN/NuMA-rich tricellular junctions in Drosophila epithelia ( Bosveld et al . , 2016 ) . Alternatively , mitotic cell rounding is less pronounced in the squamous epithelia such as the enveloping cell layer ( EVL ) of Zebrafish gastrulae which maintain a long axis at metaphase to orient the mitotic spindle ( Campinho et al . , 2013 ) . Mitotic cell rounding does not seem to occur in the Xenopus blastula ( Strauss et al . , 2006 ) and remains poorly documented in blastulae of other species ( Xiong et al . , 2014 ) . A computational approach revealed very recently that the first 4 cell divisions in ascidian embryos may follow a geometric rule in a similar manner to early Xenopus , Zebrafish or sea urchin embryos ( Pierre et al . , 2016 ) . Major cell shape changes have been noted during the 32–44 cell stage in ascidian ( Ciona ) embryos ( Tassy et al . , 2006 ) suggesting that mitotic cell rounding may occur in cells of the ascidian blastula . However , the impact of cell cycle asynchrony or mitotic cell rounding on mitotic spindle orientation in cells of the ascidian blastula have not been studied so far . Ascidian embryos display an invariant cleavage pattern up to the 64 cell stage such that both the orientation of cell division and the relative timing of cell division in the different lineages are predictable . In order to determine if mitotic spindles aligned with the cells longest length in the apical plane we extracted the apical plane of every blastomere and assessed with a 2D computational model whether geometry of the apical surface can regulate spindle orientation in the apical plane to set cell positioning . Finally , since both phlebobranch and stolidobranch ascidians display a 24 cell stage indicating that cell cycle asynchrony begins at the 16 cell stage , we assessed what impact reducing this cell cycle asynchrony has on spindle orientation and the invariant cleavage pattern . We perturbed the GRN driving this cell cycle asynchrony to create quasi-synchronous cell cycles and determined the impact on spindle orientation in the apical plane . In this study we provide evidence that spindles align parallel to the apical surface and along the longest length of the apical surface of the blastomeres . We also show that the invariant cleavage pattern is disrupted when the asynchronous cell cycles are made more synchronous . Ascidian embryos undergo a very stereotyped development consisting of unequal cleavages and symmetric cell divisions ( Lemaire , 2009; McDougall et al . , 2011 ) . At the 64 cell stage all cells of the ascidian blastula face the outside of the embryo ( Figure 1A ) suggesting that all cell divisions are parallel to the apical surface and that embryonic cleavage proceeds through planar cell divisions to generate the ascidian blastula . In unstretched epithelia ( and in early embryos ) daughter cells divide orthogonally to the mitotic orientation of their mother cell reflected in one cell generating a square of 4 cells upon two planar cell divisions ( Wyatt et al . , 2015 ) . Cell division is said to be oriented ( called oriented cell division: OCD ) when daughter cells divide in the same orientation as their mother ( Strome , 1993; Wyatt et al . , 2015 ) . By analysing the pattern of 4 grand-daughter cells generated by two successive cell divisions from the 16 cell stage in Phallusia embryos ( see McDougall et al . , 2015for details ) we found that some blastomeres do not divide orthogonally to their mother . Figure 1A shows a virtual Phallusia mammillata embryo ( Tassy et al . , 2006 ) with color-coded lineages . When following two successive cell divisions from the 16 cell stage it can be observed at the 64 cell stage that some groups of 4 grand-daughters form a square pattern ( lineages b5 . 3 , b5 . 4 and A5 . 2 shown in blue and pink , Figure 1A ) suggesting that two cell divisions orthogonal to each other occurred . In contrast , the grand-daughters of B5 . 1 , B5 . 2 and a5 . 3 ( brown ) form a T pattern suggesting that the spindle of one of the two daughter cells is in the same orientation as the spindle of its mother ( indicating OCD in this cell ) . Finally the grand-daughters of A5 . 1 and a5 . 4 ( orange ) form a line indicating that two OCDs occurred in these lineages ( Figure 1A and B , see also McDougall et al . , 2015 ) . Using this strategy we could identify three cells undergoing OCD ( asterisks in Figure 1A ) at the 16–24 cell stage ( a5 . 3; a5 . 4; B5 . 2 ) and seven cells at the 32–44 cell stage ( A6 . 1; A6 . 2; a6 . 6; a6 . 7; a6 . 8; B6 . 2 and B6 . 3 ) . Strikingly the square , T and linear patterns observed at the 64 cell stage are perfectly conserved in Styela partita ( Figure 1B ) and Ciona intestinalis ( McDougall et al . , 2015 ) , showing that the pattern of planar cell divisions in early ascidian embryos may be perfectly conserved . 10 . 7554/eLife . 19290 . 003Figure 1 . Predicted oriented cell divisions ( OCD ) in ascidian embryos . ( A ) Images taken from virtual Phallusia mammillata embryos ( obtained from http://www . aniseed . cnrs . fr/aniseed/download/download_3dve ) showing the different embryonic stages . Top row: Animal hemisphere , bottom row: Vegetal hemisphere . The right side of embryos is color coded for germ layers at the 16 cell stage: Ectoderm is in green , endomesoderm in red and germ lineage in yellow . The left side of embryo is color coded according to type of lineages . Lineages displaying square arrangements of 4 cells at 64 cell stage are shown in blue ( b5 . 3 , A5 . 2 ) and pink ( b5 . 4 ) . Lineages displaying T arrangements are depicted in light ( B5 . 1 ) and dark ( a5 . 3 , B5 . 2 ) brown . Lineages displaying linear arrangements of cell are depicted in orange ( a5 . 4 , A5 . 1 ) . ( B ) Schematic drawing showing 64 cell stage embryos of Phallusia mammillata ( left , outlines from an embryo stained with Cell Mask Orange ) and of Styela partita ( Conklin , 1905 ) . The names of each blastomere are depicted to show conservation of cell positions between the two distant ascidian species . Same color coding as in A . ( C ) Spindle rotation in the ectoderm ( Animal hemisphere ) at the 44 cell stage . Time lapse epifluorescence imaging of a P . mammillata embryo expressing MAP7::GFP to monitor mitotic spindles and H2B::mRFP1 to monitor DNA ( superimposed on the BF image ) . In red are the cell’s outline drawn using the BF image during the running of the computational model . Blue circles joined by a green bar represent mitotic spindles predicted by the computational model . Scale bar = 20 µm . Bar graph showing quantification of the angle difference between observed and predicted spindles ( orienting deviation ) . Black asterisks denote statistical difference with the value for a6 . 6 at prophase/prometaphase ( student test; p<0 . 05; *p<0 . 0001 ) . n represents the number of blastomeres analysed with the computational model . DOI: http://dx . doi . org/10 . 7554/eLife . 19290 . 003 Time lapse imaging of mitotic spindles in live ascidian embryos revealed that spindle rotation accompanies unequal cleavages in the germ lineage ( B5 . 2; B63; Prodon et al . , 2010 ) but also in several other lineages where we predicted OCDs ( for a5 . 3; b5 . 3; a6 . 6; a6 . 7; a6 . 8 see Video 1 and Figure 1C and for A6 . 1; A6 . 2 see Negishi and Yasuo , 2015 ) . In the experiments depicted in Figures 1C and 2D epifluorescence imaging is used and only the plane of imaging is analysed . The apical surface of a6 . 6 , a6 . 8 and b6 . 8 cells can be imaged by our 2D imaging protocol as both spindle poles remain in the plane of imaging during the whole of mitosis . We consistently observed spindle rotation in the apical plane of a6 . 6 and a6 . 8 which are predicted to undergo OCD ( Figure 1C , Video 1 ) . Spindle rotation in A6 . 1 , A . 62 ( Negishi and Yasuo , 2015 ) and a6 . 7 ( data not shown ) which also display OCD is also limited to the apical plane of the cell . In contrast , blastomeres not displaying OCD such as b6 . 8 blastomeres show no major spindle rotation ( Figure 1C , Video 1 ) . 10 . 7554/eLife . 19290 . 004Figure 2 . Changing cell shape in the embryo by compressing embryos or removing cell adhesion . ( A ) Compressing embryos: Left: CellMask images of 4 cell-stage embryos in control ( 4 cell control ) or compressed ( 4 cell compress ) conditions . A3 , B3 are names of blastomeres , arrowheads indicate the position of the CAB ( marking the posterior pole of the embryo ) . In the images of 4 cell compressed embryos , predicted spindles are shown ( blue circles joined by a green line ) . Scale bar = 30 µm . Middle: sagittal views of control and compressed embryos after 3D rendering on Imaris . Plasma membrane is in green and spindles in red , arrowheads indicate the position of the CAB . Table shows orienting deviation measured in A3 and B3 blastomeres in compressed embryos ( n = 8 cells taken from five embryos ) . Right: Top: CellMask images of control embryo at the 8 cell stage with a Vegetal layer of 4 cells and an Animal layer of 4 cells ( same embryo shown ) . CellMask images of 2 different compressed embryos at the 8 cell stage showing one layer of 8 cells . Arrowheads show the position of the CAB . Scale bar = 30 µm . ( B ) Culture in Ca2+ free sea water to remove cell adhesion: CellMask image of 4 cell stage embryo cultured in Ca2+ free sea water from the one cell stage . Like in a control embryo , the 4 cells are arranged in one plane . FSW: 8 cell stage embryo cultured in filtered sea water ( FSW ) exhibiting 2 layers of 4 animal and four vegetal cells ( n = 8/8 embryos ) . Ca-Free SW: 8 cell stage embryos cultured in Ca2+-free sea water from the 2 cell stage exhibiting variable morphologies comprising either wild type morphology ( 4 animal and four vegetal cells , n = 6 out of 17 embryos ) or affected morphologies with 6–5 animal and 2–3 vegetal cells ( n = 8 out of 17 embryos ) or even one layer of 8 cells ( 3 out of 17 embryos ) . Scale bars = 30 µm . ( C ) DA-aPKC: Image of a 4 cell stage embryo showing embryo morphology ( BF image ) and spindle positions ( imaged with Venus::Tpx2 ) as well as predicted spindle positions superimposed ( blue circles joined by a green line: for those cells where both spindle poles were in the imaging plane . Scale bar = 30 µm . 8 cell stage embryos expressing DA-aPKC::Venus which exhibit variable morphologies comprising either wild type morphology ( 4 animal and four vegetal cells , n = 2 out of 14 embryos ) or affected morphologies with 6–5 animal and 2–3 vegetal cells ( n = 10 out of 14 embryos ) or 7 and 1 cells ( 2 out of 14 embryos ) . Scale bar = 30 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 19290 . 00410 . 7554/eLife . 19290 . 005Video 1 . Spindle rotation in the ectoderm in cells undergoing oriented cell divisions . Movie showing live imaging of a Phallusia mammillata embryo injected with RNAs coding for MAP7::GFP ( in green ) and H2B::mRfp1 ( in red ) . ( z-stacks taken 2 min apart ) . View of the ectoderm showing mitoses of 24–32 cell stage and 44–64 cell stage . Spindle rotation is clearly visible in six blastomeres at mitosis 44–64 cell stage ( a6 . 6; a6 . 7; a6 . 8 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19290 . 005 To test whether spindle rotation is influenced by the geometry of the apical surface of the cell , we used a 2D computational model that predicts the preferred spindle orientation based on cell shape ( Minc et al . , 2011; Minc and Piel , 2012; Campinho et al . , 2013; Bosveld et al . , 2016 ) . Using the outline of cells ( in red ) , the computational model outputs a predicted spindle position and orientation ( blue circles joined by a green line ) which do not change between prophase and anaphase ( Figure 1C ) . By comparing the angle between the observed and predicted spindle orientations ( orienting deviation ) it can be observed that at prophase/prometaphase ( i . e . , before spindle rotation ) , apical cell geometry poorly predicts spindle orientation in a6 . 6 and a6 . 8 ( orienting deviation: 61 . 0° in a6 . 6 and 47 . 0° in a6 . 8 ) whereas apical cell geometry predicts reliably spindle position in b6 . 8 ( orienting deviation: 14 . 7° ) . In contrast , at metaphase , all observed spindles show an orienting deviation of less than 20° indicating that apical cell shape predicts reliably spindle orientation in the apical plane ( Figure 1C ) . These observations suggest that mitotic spindles in a6 . 6 and a6 . 8 are not aligned with the long length of the cell in the apical plane at prophase and they rotate during mitosis to align with the long length of the cell’s apical surface at metaphase . In contrast , in b6 . 8 ( not displaying OCD ) the spindle is aligned in the long length of the cell’s apical surface already in prophase . In order to confirm that cell shape can influence spindle positioning in early ascidian embryos we changed blastomeres shapes by compressing embryos or by inhibiting cell adhesion ( Figure 2 ) . Mitosis from 4 to 8 cells transforms a single layered 4 cell stage embryo into an 8 cell embryo made of 2 layers of cells comprising 4 animal and four vegetal cells as all mitotic spindles align along the Animal-Vegetal axis of the embryo ( Figure 2 , Negishi et al . , 2007; Pierre et al . , 2016 ) . Compressing the 4 cell stage embryo along the Animal-Vegetal axis creates a flattened embryo ( Figure 2A , ‘4-cell compressed’ ) which , upon cell division , gives rise to a single layer of 8 cells . Computational analysis shows a good prediction of spindle positioning by cell shape in these embryos ( orienting deviation: 22 . 7 ± 5 . 6° for A3 blastomere and 19 . 5 ± 3 . 4° for B3 ) ( Figure 2A ) . Therefore mitotic spindle orientation in these compressed embryos tends to follow the newly created long axis of the cell . We then removed cell adhesion between blastomeres by culturing embryos in Ca2+-free sea water from the one-cell stage ( Figure 2B ) or by inhibiting basolateral membrane formation using a dominant active form of aPKC ( Sabherwal et al . , 2009 , Figure 2C ) . Both treatments reduced drastically cell adhesion from the 2 cell stage resulting in rounder blastomeres . Spindle orientation did not seem affected during the mitosis from 2 to 4 cell stage ( data not shown ) and 4 cell stage embryos with reduced cell adhesion show no blastomere positional changes as they are still made by a single layer of 4 cells ( Figure 2B and C ) . However , the invariant cleavage pattern is affected in these embryos during the 4 to 8 cell mitosis which normally creates 2 layers of 4 animal and four vegetal ( Figure 2B and C ) . For example , downregulating cell adhesion causes embryonic morphologies at the 8 cell stage ranging from wild type patterns ( 2 layers of 4 cells ) to one layer of 8 cells with all intermediate patterns ( Figure 2B and C ) . Computational analysis of the cell divisions occurring in the plane of imaging revealed a good prediction of spindle positioning by the computational mode ( orienting deviation: 9 , 64 ± 4 . 16° in DA-aPKC expressing embryos , Figure 2C ) suggesting that cell shape still regulates spindle orientation in these cells . Finally cell divisions at 16 and 32 cell stages were random giving rise to 64 cell blastula of highly variable morphologies ( data not shown ) preventing further manual analysis . These observations show that cell shape can orient spindle positioning in early ascidian embryos and indicates that cell adhesion and regulated apicobasal polarity are absolutely vital to maintain the cell shapes supporting the invariant cleavage pattern of early ascidian embryos . Having revealed that mitotic spindles aligned with the long length of the apical surface during metaphase causing some mitotic spindles to rotate through approximately 90° while others remained relatively fixed in position , we wished to determine if apical cell shape can predict spindle orientation in the apical plane in all blastomeres up to the 64 cell stage . Defining the apical plane of an irregular polyhedral shape created by living cells is challenging since 2D imaging provides inherently false information as mitotic spindles in three dimensional embryos rarely align within the imaging plane . We therefore performed confocal 3D + time lapse imaging of living ascidian embryos stained with Cell Mask Orange ( Invitrogen ) to monitor cell membranes and spindle poles at metaphase in all blastomeres ( see McDougall et al . , 2015 for a detailed protocol ) . The apical plane of each blastomere ( defined as the 2D plane containing the poles of the mitotic spindle which can separate most of the apical surface from the basolateral surface ) was systematically extracted from 3D-rendered blastomeres using Imaris software following a specific protocol ( see Materials and methods and McDougall et al . , 2015 ) for detailed protocols ) and the position of the mitotic spindle in the extracted 2D plane was compared with the position of a mitotic spindle predicted by the computational model ( Minc et al . , 2011 ) . 3D rendering and in silico isolation of each blastomere at interphase , prophase and metaphase revealed drastic cell shape changes during the cell cycle ( Figure 3 ) . Sphericity of each blastomere was determined using Imaris statistics ( see Materials and methods ) to estimate the complexity of polyhedral shape of in silico isolated blastomeres . Sphericity of blastomeres at metaphase was found to be significantly different from a round standard from the 4 cell stage and was particularly pronounced at the 16–24 and 32–44 cell stages during which spindle rotations occur ( Figure 3A ) . At the 32 cell stage , in silico isolated blastomeres were found to be columnar during interphase and to flatten along the apical direction at metaphase ( Figure 3B ) without significantly changing cell volume ( data not shown ) . This was reflected in an increase in the average sphericity of blastomeres at metaphase indicating that , at this stage , blastomeres partially round up at mitosis ( Figure 3B ) . We then segmented manually apical ( red ) from basolateral ( green ) membranes on each 3D-rendered blastomere and calculated the apical surface ratio ( apical surface divided by total surface ) and found that a large increase in apical surface area accompanies mitotic cell rounding ( from 0 , 15–0 , 23 at interphase to 0 , 42–0 , 56 at metaphase , Figure 3B ) similarly to what is observed in differentiated epithelia ( Ragkousi and Gibson , 2014 ) . These observations document partial mitotic cell rounding at the sixth cell cycle which results in blastomeres increasing their apical surface by apical expansion at prophase and metaphase . Therefore the apical surface ratio is a better quantitative measure of cell shape changes than sphericity as some cell shape changes observed from a columnar cell ( with a small apical surface ) to a ‘brick’ shaped cell ( with a larger apical surface ) are not associated with changes in sphericity . 10 . 7554/eLife . 19290 . 006Figure 3 . Changes in 3D cell shape during development and during the cell cycle . ( A ) Quantification of polyhedral shape complexity of in silico isolated blastomeres at metaphase ( using sphericity measurements of Imaris software ) . While blastomeres at the 2 cell stage have a similar sphericity than a spherical standard ( standard: 0 . 975+/0 . 001 ( n = 4 ) ; 2 cell: 0 . 950 ± 0 . 010 ( n = 4 ) ) , from the 4-cell-stage on sphericity significantly decreases compared to standard ( p<0 . 05 , black asterisk ) . Average values are: 4 cell: 0 . 897 ± 0 . 013 ( n = 8 ) ; 8 cell: 0 . 889 ± 0 . 009 ( n = 8 ) ; 16 cell ( vegetal ) : 0 . 846 ± 0 . 006 ( n = 8 ) , 24 cell ( animal ) : 0 . 779 ± 0 . 007 ( n = 8 ) , 32 cell ( vegetal ) : 0 . 851 ± 0 . 012 ( n = 8 ) ; 44 cell ( animal ) : 0 . 825 ± 0 . 012 ( n = 8 ) . Note that animal blastomeres ( 24 cell stage ) have a significantly more complex polyhedral shape than their vegetal counterparts ( 16 cell ) ( p<0 . 05 , red asterisk ) . An example of an in silico isolated blastomere is depicted above each bar of the graph . ( B ) Quantification of cell shape changes during the cell cycle at the 32–44 cell-stage . Left: 3D views of manually segmented blastomeres at interphase , prophase and metaphase ( 32 cell stage ) showing cell shape changes between interphase , prophase and metaphase ( inset: apical surface ratio of cell shown ) . Green is basolateral and red is apical . Scale bar as indicated . Top right: quantification of cell sphericity at interphase , prophase and metaphase . 6 blastomeres of the animal ( red ) and vegetal ( blue ) hemisphere of the 32 cell stage were averaged . The sphericity was significantly higher at metaphase than at interphase ( black asterisk , p<0 , 05 ) . Bottom right: quantification of the apical surface ratio at the same time points . The same blastomeres as in the sphericity graph were used to average apical surface ratio in the animal ( red ) and vegetal ( blue ) hemisphere . The apical surface ratio at prophase and at metaphase were significantly higher than at interphase ( black asterisks , p<0 , 05 ) . The apical surface ratio was higher in vegetal blastomeres than in animal ones ( § sign , p<0 , 05 ) . ( C ) Pipeline for predicting spindle position using 2D computational model ( Minc et al . , 2011 ) . See McDougall et al . ( 2015 ) for the full protocol of apical plane extraction . Top row shows examples of 3D rendered , in silico isolated blastomeres . Bottom row shows the extracted apical plane of the corresponding blastomeres with spindle predictions ( blue circles joined by a green line ) . Scale bars = 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 19290 . 006 We then extracted the apical plane of each blastomere from the 2 cell stage to the 44 cell stage . The cell outline drawn from the 2D extracted apical plane of each blastomere was then computed to predict spindle orientation ( Figure 3C , Figure 4 ) . Figure 4A shows the observed and predicted spindle positions in the extracted apical plane of each blastomere from the fourth mitosis ( 8–16 cell ) to the sixth mitosis ( 32-44-64 cell ) and reveals a good prediction of spindle orientation by apical cell shape in almost all blastomeres . Quantification of the deviation between the predicted center of spindles and the observed center of spindles ( centering deviation ) from the 2 cell stage to the 44 cell stage confirms that all spindles are centered within 20% except for the B4 . 1 and B6 . 3 blastomeres of the germ cell lineage which undergo CAB-dependant unequal cleavages ( Figure 4B ) . Strikingly we found that orienting deviation is under 30° in all cells except in B5 . 2/B6 . 3 and A6 . 3 ( Figure 4C ) . By comparing the orienting deviation observed in 149 cell divisions ( excluding the germ cell precursors that divide unequally ) to randomly generated angles between 0 and 90° we found that the distribution of observed orienting deviations is non uniform and significantly different from a random distribution ( Figure 4D ) . In our data set , 88% of cells show an orienting deviation of 30° or less , 78% of cells have an orienting deviation under 20° and 52% of cells show an orienting deviation of less than 10° ( Figure 4D ) . Therefore our model robustly predicts that spindle aligns with the long length of the apical surface in most cells with a precision of 30° and 20° . In contrast spindle orientation could not be predicted reliably in B5 . 2 , B6 . 3 and A6 . 3 blastomeres ( i . e . orienting deviation is above 30° ) suggesting that in these three blastomeres apical cell shape does not regulate spindle positioning in the apical plane . This was expected in B5 . 2 and B6 . 3 which display spindle rotation during unequal cleavage ( Prodon et al . , 2010 ) , but not in A6 . 3 which do not show spindle rotation nor unequal cleavage . However A6 . 3 cells undergo an asymmetric division segregating endoderm and mesoderm fates driven by asymmetric Ephrin and MAPK signalling ( Shi and Levine , 2008 ) suggesting that , like in the germ lineage , cues other than apical cell shape might regulate spindle orientation in A6 . 3 blastomeres . 10 . 7554/eLife . 19290 . 007Figure 4 . Computational model predicts spindle position and orientation in the apical plane of each blastomere . ( A ) 3D views of Phallusia embryos from 8 cell stage to 44 cell stage and extracted apical plane of each blastomere . Observed spindle poles are depicted by white circles/balls . Predicted spindles are depicted with blue circles joined by a green line . The red outline of each cell is the shape used by the computational model to predict spindle position . a=anterior , p=posterior , scale bars are all 20 µm . Lineages are color coded like in Figure 1 . ( B ) Mean centering deviation in each blastomere . n = 6 cells analysed for each blastomere except A6 . 4 ( n = 4 ) ; B6 . 1 ( n = 5 ) ; a6 . 7 ( n = 4 ) , b6 . 5 ( n = 4 ) . Red asterisk denote cells undergoing unequal cleavage . ( C ) Mean orienting deviation in each blastomere . n = 6 cells analysed for each blastomere except A6 . 4 ( n = 4 ) ; B6 . 1 ( n = 5 ) ; a6 . 7 ( n = 4 ) , b6 . 5 ( n = 4 ) . Red asterisks denote cells undergoing unequal cleavage . Blue asterisks denote blastomeres undergoing OCD . Triple black asterisks denote that orienting deviation in the grouped B5 . 2 , A6 . 3 , and B6 . 3 cells are statistically greater than those in other lineages ( Wilcoxon rank sum test with continuity correction , p-value=4 . 78910−7 ) . ( D ) Quantification of orienting deviation: cumulative percentage graph of measured data ( blue dots , n = 149 cell divisions , black dots denote the six A6 . 3 cells analysed ) and randomly generated data ( orange dots , n = 149 ) . The measured data are not uniform and significantly different from the random data ( One-sample Kolmogorov-Smirnov test , p<2 . 2*10−6 ) . The numbers indicated under each graph is the proportion of cells with orienting deviations under the considered threshold ( 10° , 20 , and 30° ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19290 . 007 In the early ascidian embryo , unequal cleavages in the germ lineage are regulated by a maternal factor located in the posterior pole of the embryo called the Centrosome Attracting Body ( CAB ) . The CAB is a cortical complex composed of cell polarity proteins ( Patalano et al . , 2006 ) which is segregated in the germ lineage and is responsible for spindle orientation in the B5 . 2/B6 . 3 mitoses in a cell autonomous manner ( Nishikata et al . , 1999; Prodon et al . , 2010 ) against apical cell shape ( Figure 4C ) . Conversely , zygotic transcription is required for germ layer patterning starting from the 16 cell stage in ascidians ( Lemaire , 2009; Hudson et al . , 2013 ) and also for generating cell cycle asynchrony from the 16 cell stage ( Dumollard et al . , 2013 ) . Removing the precursor of the CAB by dissecting the contraction pole ( CP ) at the zygote stage prevents unequal cleavage in the germ lineage as well as β-catenin stabilization in the endomesoderm giving rise to a synchronous hollow blastula that cannot gastrulate ( Nishida , 1996; Dumollard et al . , 2013 ) . For example , removing the CP prevents unequal cleavage in CAB-containing blastomeres at the 8 cell stage ( B4 . 1 pair ) resulting in more centered spindles ( Figure 5B ) and the absence of small cells in the vegetal posterior pole of the embryo of CP-ablated embryos ( Figure 5A ) . The general morphology of radialized embryos indicates that all cell shapes in the whole embryo are affected at the 16–32 cell stages ( Figure 5A ) . Computational analysis of the cell divisions occurring in the imaging plane shows orienting deviations below 30° at 8 , 16 and 32 cell stages ( Figure 5B ) suggesting that cell shape still regulates spindle orientation in CAB-depleted embryos . Therefore , the activity of the maternal CAB impacts not only unequal cleavage in the germ lineage but also cell shape and hence cell divisions in the rest of the embryo . 10 . 7554/eLife . 19290 . 008Figure 5 . Removing the maternal CAB prevents unequal cleavage and radializes embryos . ( A ) Images showing embryonic morphology and cell shapes in Control ( top row ) and CP-ablated ( bottom row ) embryos . Top row: Images showing mitotic spindles ( MAP7::GFP ) and cell membranes ( PH::GFP , CM-orange: Cell Mask Orange; Ech::Ve: Echinoid::Venus ) . Scale bar = 15 µm . Animal hemisphere ( An ) is depicted in green while the vegetal hemisphere ( Veg ) is depicted in red . Purple asterisks indicate unequal cell division in the germ line . Bottom row: images showing mitotic spindles and cell membranes ( MAP7::GFP and PH::GFP ) in radialised embryos in which the contraction pole ( CP ) was removed ( CP-ablation ) . These embryos ( n = 5 ) are completely radialised and do not bear small cells in the vegetal posterior pole of the embryo . Scale bar = 15 µm . ( B ) Left: Quantification of centering deviation showing that it is less than 10% in a4 . 2 , b4 . 2 and A4 . 1 blastomeres whereas it is over 10% in B4 . 1 of control embryos . In CP-ablated embryos centering deviation is not affected in A4 . 1 but is decreased in B4 . 1 . Triple asterisk indicates a significant difference with A4 . 1 ( student , p=0 . 00004 ) . Right: Quantification of orienting deviation showing that it is below 30° in both control and CP-ablated embryos at the 8 cell stage ( a4 . 2 , b4 . 2 , A4 . 1 , B4 . 1 ) and at the 16 and 32 cell stages in CP-ablated embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 19290 . 008 We then assessed the specific impact of zygotic transcription on the ascidian invariant cleavage pattern by either blocking all zygotic transcription using Pem1 expression ( see Kumano et al . , 2011; Shirae-Kurabayashi et al . , 2011 and Dumollard et al . ( 2013 ) for details on the effects of PEM1 ectopic expression ) or by blocking β-catenin transactivation using DN-Tcf ( which does not affect non transcriptional functions of β-catenin , see Dumollard et al . , 2013 for details ) . Embryo morphology at 16 cell stage and CAB-dependant unequal cleavages are maintained in embryos expressing Pem1 ectopically as shown previously ( Negishi et al . , 2007; Kumano et al . , 2011; Dumollard et al . , 2013 ) . However , all mitotic spindle rotations previously observed at the 16 and 32 cell stages are strongly impaired in Pem1 and DN-Tcf expressing embryos resulting in cell divisions with different orientation from the invariant cleavage pattern ( termed ‘misoriented cell divisions’ ) ( Figure 6 ) . Most strikingly linear and T patterns of two sister cell spindles at the 32-44-64-cell mitosis were often replaced by two parallel spindles ( giving rise to square pattern of cells ) in these embryos ( outlined in red in Figure 6A ) . Counting the occurrence of misoriented cell divisions in embryos expressing Pem1 or DN-Tcf ( Figure 6B ) shows that the blastomeres undergoing spindle rotation displayed more occurrence of misoriented cell divisions ( a5 . 3 , b5 . 3 , A6 . 1 , A6 . 2 , a6 . 6 , a6 . 7 , a6 . 8 marked with an asterisk in Figure 6B ) . b6 . 5 and b6 . 6 blastomeres also showed a strong incidence of misoriented cell divisions but as a consequence of their mother ( b5 . 3 ) being misoriented . The shape of the apical surface appears altered both in Pem1 and DN-Tcf expressing embryos and computational analysis of cell divisions in a6 . 8 ( normally displaying spindle rotation ) shows that misoriented cell divisions are still predicted reliably by apical cell shape ( Figure 6C ) . Therefore the processes supporting spindle orientation in the long length of the cell in the apical plane are maintained in these embryos and misoriented cell divisions are probably due to altered cell shape in manipulated embryos . 10 . 7554/eLife . 19290 . 009Figure 6 . Impact of zygotic transcription on the invariant cleavage pattern . ( A ) Images showing metaphase spindle ( green ) and nuclei ( red ) in control , Pem1::Ve and DN-Tcf expressing embryos . Unaffected cell divisions are surrounded by a colored line ( in control and manipulated embryos ) . Misoriented cell divisions are surrounded by a red line . Scale bars = 20 µm . dt32c indicates the difference in timing of mitotic entry between animal and vegetal hemispheres at the 32–44 cell stage . ( B ) Graph plotting the incidence of misoriented cell divisions in control ( black bars ) , Pem1::Ve ( yellow bars ) and DN-Tcf ( red bars ) embryos . Blue asterisks denote blastomeres undergoing OCD . ( C ) Images showing embryonic morphology in the ectoderm ( animal ) at the 32–44 cell stage . Blastomeres with asterisks are a6 . 8 for which cell outline ( in red ) and spindle prediction ( blue circles joined by a green bar ) are depicted . Orienting deviation in a6 . 8 blastomeres displaying OCD in control embryos and misoriented cell divisions in a6 . 8 blastomeres in Pem1 and DN-Tcf embryos are indicated under the images . n indicates the number of a6 . 8 blastomeres analysed . Scale bars are 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 19290 . 009 Since both PEM1 overexpression and DN-Tcf reduce cell cycle asynchrony ( Dumollard et al . , 2013 ) we next sought to directly alter cell cycle duration to assess the impact cell cycle asynchrony has on the invariant cleavage pattern . In Xenopus , Drosophila or Zebrafish embryos early embryonic cell cycles are regulated by a balance of the two cell cycle regulators Wee1 and Cdc25 which inhibit or activate cdk1 respectively to set interphase length ( reviewed in Farrell and O'Farrell , 2014 ) . Wee1 maternal RNAs are present in the unfertilized oocyte in Phallusia mammillata ( Figure 7A , see also RNA-seq data in Aniseed database: http://www . aniseed . cnrs . fr/ with Gene Id: phmamm . CG . MTP2014 . S423 . g08568 ) where it is enriched in the vegetal cortex . Then maternal Wee1 transcripts can be observed in the CAB similarly to PEM-type genes ( Prodon et al . , 2007 ) . Zygotic Wee1 transcripts first appear at late 32 cell stage in the endoderm lineage ( 64–76-stage ) just before these cells invaginate . After the onset of gastrulation ( 112 cell stage ) Wee1 transcripts are expressed in mesodermal cells just before they invaginate ( Figure 7A ) . We modified Wee1 activity in the whole embryo ( see Materials and methods , Murakami et al . , 2004; Farrell and O'Farrell , 2014 ) and found , as anticipated , that Wee1 inhibition speeded up the slow cell cycle in the ectoderm at the 16 cell stage without affecting the cell cycle duration of the vegetal endomesoderm thereby eliminating the transient 24 cell stage ( Figure 7B ) . At the 32 cell stage , cell cycle asynchrony between ectoderm ( animal ) and endomesoderm ( vegetal ) was 6–8 min in Wee1-manipulated embryos compared to 15 min in control embryos ( Figure 8A ) . Analysis of cell division orientation at the 16–24 and 32–44 cell stages in Wee1-manipulated embryos revealed deviations from the invariant cleavage pattern as well as variability between individuals ( Figure 8A and B ) . In synchronized embryos , misoriented cell divisions are particularly obvious at the 32–64 cell stage where the linear arrangements of A6 . 1/A6 . 2 and a6 . 8/a6 . 7 daughter cells observed in control embryos were changed to square arrangements of cells ( circled in red in Figure 8A ) . Counting the occurrence of misoriented cell divisions revealed that the blastomeres b5 . 3 , A6 . 1 , A6 . 2 , a6 . 7 and a6 . 8 were the most affected ( Figure 8B ) . The orientation of cell division in b6 . 5 and b6 . 6 was also affected ( Figure 8B ) . Therefore the blastomeres most affected by synchronization were the ones displaying oriented cell division ( b5 . 3 , A6 . 1 , A6 . 2 , a6 . 7 and a6 . 8 ) or daughters of cells displaying OCD ( ie b6 . 5 , b6 . 6 ) . Apical surface area at metaphase was altered in Wee1-manipulated embryos and misoriented cell divisions in a6 . 8 blastomeres still aligned with the long length of the apical surface ( orienting deviation: 10 . 17 ± 4 . 15° , Figure 8B ) . Therefore , inhibition of cell cycle asynchrony did not seem to disrupt the mechanisms supporting spindle orientation in the apical plane and misoriented cell divisions may rather be due to altered apical surface area . 10 . 7554/eLife . 19290 . 010Figure 7 . Inhibiting cell cycle asynchrony in the ascidian blastula . ( A ) Images showing in situ hybridizations of Pm-Wee1 from the unfertilized egg ( Unf ) to the late 112 stage . Wee1 is expressed in the unfertilized oocytes where it is slightly enriched in the vegetal cortex of the egg . Then a maternal signal can be observed in the early stages with a specific enrichment in the CAB region hosting the germ plasm . Zygotic Wee1 is expressed in the endoderm precursors from the late 32 cell stage to the 76 cell stage and in the muscle precursor from the 112 cell stage ( i . e . , just before they gastrulate ) . In some images Hoechst staining of nuclei is shown in blue . Scale bar is 20 µm . ( B ) Quantification of cell cycle length at the 16–32 cell stage ( MBT , cell cycle 5 ) in manipulated embryos: control embryos ( black bars , n = 10 ) , embryos expressing Wee1KD::Ve ( stripped bars , n = 9 ) , embryos injected with wee1 MO ( diamond bars , n = 9 ) . wee1KD and wee1 MO both speed up the ectoderm cells ( EctoD ) without affecting the endomesoderm ( Endo-MesoD ) or the germ line ( Germ L ) . % increase in cell cycle time relative to cell cycle timing at the 8 cell stage . A 20% increase means that the cell cycle timing has increased by 20% compared to the previous cell cycle at 8 cell stage ( see Dumollard et al . , 2013 for details ) . Triple black asterisks indicate p=0 . 0003 for wee1KD and p=0 . 000009 for wee1MO . DOI: http://dx . doi . org/10 . 7554/eLife . 19290 . 01010 . 7554/eLife . 19290 . 011Figure 8 . Impact of cell cycle asynchrony on the invariant cleavage pattern . ( A ) Images showing metaphase spindles in control and wee1 MO injected embryos ( top two rows ) or showing nuclei in wee1KD::Ve and control embryos ( bottom two rows ) . Unaffected cell divisions are surrounded by a colored line in control embryos . Misoriented cell divisions occurring in manipulated embryos are surrounded by a red line . Scale bar = 20 µm . dt32c indicates the difference in timing of mitotic entry between animal and vegetal hemispheres at the 32–44 cell stage . ( B ) Graph plotting the incidence of misoriented cell divisions in control ( black bars ) , wee1KD ( pink bars ) and wee1 MO ( purple bars ) injected embryos . Blue asterisks denote blastomeres undergoing spindle rotation , green asterisks indicate daughters of b5 . 3 ( undergoing spindle rotation ) . Inset: Images showing embryonic morphology in the ectoderm ( animal ) at the 32–44 cell stage . Blastomeres with asterisks are a6 . 8 for which cell outline ( in red ) and spindle prediction ( blue circles joined by a green bar ) are depicted . Orienting deviation in a6 . 8 blastomeres displaying OCD in control embryos ( n = 6 cells ) and misoriented cell divisions in wee1-KD ( n = 8 cells ) embryos is indicated in the images . Scale bars are 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 19290 . 011 The ascidian embryo provided a convenient experimental system in which to test the contribution of apical cell shape to spindle orientation by a computational mathematical model ( Minc et al . , 2011; Minc and Piel , 2012 ) . We demonstrate here for the first time that a partial mitotic cell rounding accompanies cell division in the ascidian blastula whereas it was not observed in Xenopus ( Strauss et al . , 2006 ) and is still not clearly documented in Zebrafish ( Xiong et al . , 2014 ) . Interestingly , in these two species , cell deformations have been observed at MBT but only in isolated blastomeres ( called ‘cell motility’ in Newport and Kirschner , 1982 and Kane and Kimmel , 1993 ) . However , even though the isolated blastomeres with the longest cell cycle showed more pronounced cell deformations , cell cycle dependence of such cell deformations remains unclear . Mirror-like cell shape changes in 32–44 cell stage ascidian embryos were noted previously ( Tassy et al . , 2006 ) , prompting us to test the idea that cell cycle asynchrony may be involved in the mirror-like cell shape regulation . We demonstrate here that these cell shape changes consist in cell rounding and apical expansion during mitosis and they correspond to the mitotic cell rounding observed in epithelial cells ( Ragkousi and Gibson , 2014 ) . In contrast to cells in differentiated Drosophila epithelia which completely round up at mitosis ( Bosveld et al . , 2016 ) but similarly to the epibolizing enveloping cell layer ( EVL ) of Zebrafish embryos ( Campinho et al . , 2013 ) , mitotic cell rounding in the ascidian blastula is incomplete . In consequence , blastomeres at the 16–24 and 32–44 cell stage do not become completely round and the shape of the apical surface remains anisotropic during metaphase . After confirming in ascidian embryos that , like in C elegans , sea urchin , Zebrafish or Xenopus embryos ( Wildwater et al . , 2011; Minc et al . , 2011; Strauss et al . , 2006; Pierre et al . , 2016 ) , cell shape impacts the orientation of cell division , we show that a computational model based on the shape of the apical surface predicts spindle orientation in the apical surface in all cells of the embryo except in the germ lineage ( B5 . 2 , B6 . 3 ) and A6 . 3 . 88% of these cells can be predicted by the model with a 30° precision and 78% of these cells are still predicted by the computational model with 20° precision ( Figure 4D ) . These observations strongly indicate that mitotic spindles align with the long axis of the cell in the apical plane to implement the invariant cleavage pattern of planar cell division thus creating the distinctive topographical organization of cells in the ascidian blastula . In the ascidian blastula such a general geometric rule is overridden only in the germ lineage where a maternally derived cortical polarity complex ( the CAB ) attracts one spindle pole duing unequal cell division ( Prodon et al . , 2010 ) . It would be interesting to extend this mathematical analysis to other developmental stages and to distantly-related ascidian species , since we may have unveiled a conserved apical cell shape dependent mechanism underpining the invariant cleavage pattern displayed by all ascidians . By following the position of clones of cells from the 16 cell stage to the 64 cell stage we inferred the occurence of OCDs ( when daughter cell spindle is not orthogonal to the spindle orientation of its mother ) in specific cells . By imaging spindles during mitosis in these cells we discovered that they all perform OCD through spindle rotation rather than migration/rotation of the nucleo-centrosomal complex . A6 . 1 is the only exception to this observation where it is the nucleo-centrosomal complex rather than the spindle that rotates , but OCD in this lineage still depends on apical cell shape ( see Negishi and Yasuo , 2015 for details ) . Thus we argue that , together with apico-basal polarity which likely enforces the planar orientation of cell division , apical cell shape is a major driver of cell positioning in the ascidian blastula . This hypothesis implies that apical cell shape at metaphase must be very stereotyped and conserved in ascidian early embryos . Although perhaps surprising , this finding should be considered with the fact that cell position in ascidian embryos at the 64 cell stage is crucial since it is at this time that the fate map is established ( Lemaire , 2009 ) . More specifically , precise cell position in the anterior part of the embryo determines the number of animal marginal cells ( a6 . 5 , b6 . 5 ) that will receive a local cell induction from contacting vegetal cells to create a neural ectoderm of 6 cells at the 64 cell stage ( Lemaire , 2009 ) . Mesodermal cells of the vegetal hemisphere located in the marginal region segregate , during mitosis 32–44 cell , neural and notochord fates in the anterior part of the embryo ( A6 . 1 , A6 . 2 ) and muscle and mesenchyme ( B6 . 2 ) fates in the posterior part of the embryo ( Kumano and Nishida , 2007 ) . Therefore , the invariant cleavage pattern and oriented cell divisions might be part of the mechanism enforcing fate segregation in the ascidian blastula . A central finding of this study is the causal relationship between cell cycle asynchrony and the orientation of cell division . We suggest that cell cycle asynchrony impacts the spatial pattern of planar cell divisions by regulating the shape of the cell’s apical surface at metaphase . First , we found that every mitotic spindle tends to align with the long length of the cell’s apical plane at metaphase . Second , abolishing the asynchrony that causes the appearance of the 24 cell stage altered the invariant cleavage pattern . Third , misoriented cell divisions in synchronized embryos are still reliably predicted by apical cell shape . Critically , either causing vegetal cell cycles to become slower at the 16 cell stage by inhibiting zygotic transcription , or making animal cell cycles faster by inhibiting Wee1 kinase activity both had the same overall effect: all cells divided synchronously and the invariant cleavage pattern was disrupted . We conclude that overall cell cycle duration is not important , and rather it is the asynchrony between the animal and vegetal halves of the embryo that is crucial . Since ascidian embryos live in a marine environment that does not have a constant temperature , absolute cell cycle duration is likely less important than the cell cycle asynchrony which is maintained over a range of tempartures in different species of ascidian . Therefore the zygotic GRN driven by nuclear β-catenin that patterns germ layers is also responsible for causing cell cycle asynchrony , which in turn enforces the invariant cleavage pattern through cell shape dependent mitotic spindle orientation in the apical plane at the 16-32-44-cell stages . We find here that entry into mitosis is accompanied with partial cell rounding via apical expansion ( and hence a reduction in cell adhesion ) in the ascidian blastula . It was found in Ciona embryos that ‘between the early 32-cell stage and the mid 44-cell stage , the elongation factors of opposing animal and vegetal cells change in precisely opposite manner with time’ ( Tassy et al . , 2006 ) . We hypothesize that such mirror behaviors between animal and vegetal blastomeres is brought about by asynchronous mitotic cell rounding between animal and vegetal blastomeres . Precise shape of the apical surface during mitosis may be a function of not only adhesion with neighbouring cells but also of the remaining adhesion between interphasic and mitotic cells in opposite hemispheres since during cell division in ascidians the adhesion between blastomeres remains . We found that apical cell shape was altered in animal cells when we slowed down the cell cycle of vegetal cells with DN-TCF or PEM1 ( Figure 6 ) . Likewise , we report that apical cell shape was altered in vegetal blastomeres when we speeded up the cell cycle of animal cells by inhibiting Wee1 ( Figure 8 ) . We therefore conclude that the shape of the apical surface of animal cells is affected by the cell cycle state of the vegetal cells , and likewise that the shape of the apical surface of vegetal cells is affected by the cell cycle state of the animal cells . It is interesting to note that the overall cell cycle asynchrony between the animal and vegetal cells is about 15 min . which is about the duration of M phase . We wonder whether this may be one of the selective pressures leading to the retention of the cell cycle asynchrony between distantly-related ascidians . Indeed , both phlebobranchs and stolidobranchs ascidians display nuclear β-catenin in vegetal cells at the 16 cell stage ( Ciona: Hudson et al . ( 2013 ) , Halocynthia: Kawai et al . , 2007 ) , and it is a GRN controlled by β-catenin that causes cell cycle asynchrony starting at the 16 cell stage ( Dumollard et al . , 2013 ) . It will therefore be interesting to elucidate the entire GRN that controls cell cycle duration in the ascidian at the 16–24 cell stage , and to determine how conserved that GRN is between distantly-related ascidian species . Importantly the removal of the axial determinant ( the pre-CAB or centrosome-attracting body ) that generates unequal cell division in the germ lineage ( Nishikata et al . , 1999; Patalano et al . , 2006 ) not only prevents unequal cleavage in the germ lineage but also affects cell division orientation in the whole embryo as CAB-ablated ascidian embryos are completely radialized ( Nishida , 1996 , this study ) . Unequal cleavage of the two vegetal posterior blastomeres at the 16 cell stage thus affects the shape of every cell in the early embryo . Such effect of unequal cleavage on cell division of distant cells is supported by cell adhesion-dependant mechanical coupling between blastomeres which was found to be necessary to maintain the invariant cleavage pattern . It is noteworthy that regulated apicobasal polarity is crucial to maintain cell adhesion in the ascidian embryo to propagate individual cell deformations to the rest of the embryo and implement the invariant cleavage pattern . Given the important role played by the shape of the apical surface in spindle orientation in ascidian early embryos it is evident that physical cellular properties that minimize energy during cell packing likely play an important role in cell division plane specification . Further studies will be required to understand how the apical surface of every blastomere is interdependent on neighboring cells due to packing constraints . Such interdependence between cell division plane orientation and apical cell shape is involved in a number of morphological processes . In vertebrate embryos oriented tissue strain generated by the gastrulating mesoderm determines the global axis of planar polarity in the Xenopus ectoderm ( Chien et al . , 2015 ) , while cell division is oriented by tissue tension in the Zebrafish ectoderm to improve epithelial spreading over the yolk layer during epiboly ( Campinho et al . , 2013; Xiong et al . , 2014 ) . In ascidian blastulae , we hypothesize that the shape of cells in mitosis is a function of the tension generated between the dividing cells and their neighboring interphasic cells coupled with the tension between the cells that are dividing . This may exert stereotyped deformations of adhering mitotic cells to generate the invariant cleavage pattern . However , further studies are needed to assess whether global tissue tension deforms blastomeres of the ascidian embryo or whether , on the contrary , autonomous cell cycle-driven cell shape changes are transmitted in the embryo via cell adhesion . Eggs from the ascidians Phallusia mammillata were harvested from animals obtained in Sète and kept in the laboratory in a tank of natural sea water at 16°C . Egg preparation and microinjection have been described previously ( see detailed protocols in McDougall et al . , 2014 , 2015 ) . All imaging experiments were performed at 19°C . Microtubules and mitotic spindles were imaged using our characterised constructs of either MAP7::GFP or Ensconsin::3XGFP ( McDougall et al . , 2015 ) . DNA and nuclei were imaged with H2B::mRfp or the nuclear proteins ( wee1KD::Ve , Ve::cdc45 , Dumollard et al . , 2013 ) . Plasma membrane was stained using PH::GFP or PH::dTomato or Cell Mask Orange ( Invitrogen , see protocols in McDougall et al . , 2015 ) . Control embryos in Figure 2A are embryos stained with cell mask and cultured in filtered sea water ( FSW ) . Control embryos in Figures 2B , C , 5 , 6 , 7 and 8 are embryos injected with cRNAs coding for MAP7::GFP or Ens::3XGFP or PH::GFP or Ve::cdc45 ( in green ) or H2B::mRfp ( in red ) or a combination of two of these markers . We have found that all these staining procedures have no impact on the invariant cleavage pattern ( McDougall et al . , 2015 ) . To inhibit embryonic patterning Pm-Pem1 ( Shirae-Kurabayashi et al . , 2011; Kumano et al . , 2011 ) and DN-Tcf ( kindly provided by Yasuo Hitoyoshi ( UMR7009 , LBDV ) ) were used exactly as in Dumollard et al . ( 2013 ) . Ci-wee1 ( gene Id: KH . S256 . 1 ) was amplified from a Ciona intestinalis Gateway-compatible cDNA library using PCR ( Roure et al . , 2007 ) . To speed up cell cycle the activity of the Wee1 kinase was inhibited using a kinase dead form of Wee1 ( Wee1KD ) which was generated by introducing a stop codon inside the catalytic domain of the protein ( resulting in a deletion of aa 532 to 633 ) . Such a construct was shown to have a dominant negative effect on endogenous wee1 ( Murakami et al . , 2004 ) . Alternatively , a morpholino target to Pm-wee1 ( CAGGACCATATAAACTCCTACTGCT ) was injected to decrease wee1 activity in the whole embryo . All constructs were made using pSPE3 ( Roure et al . , 2007 ) and the Gateway cloning system ( Invitrogen ) unless otherwise stated ( see McDougall et al . , 2014 , 2015 for detailed protocols ) . Synthetic RNAs were injected in unfertilized eggs or in one blastomere of a 2 cell stage embryo . To remove cell adhesion , Phallusia zygotes stained with Cell Mask-Orange or expressing MAP7::GFP and PH::GFP ( to image cell membranes and spindle poles ) were cultured in Ca2+-free sea water ( supplemented with 1 mM EDTA , as described in Sardet et al . , 2011 ) and imaged up to the 64 cell stage . Dominant active aPKC: Pm-aPKC ( GenBank: AY987397 . 1 , Patalano et al . , 2006 ) was cut at K146 to remove the N-terminal regulatory domain of aPKC and tagged with Venus using our Gateway cloning system ( McDougall et al . , 2015; Roure et al . , 2007 ) . Removal of the N-terminal regulatory domain of aPKC results in a constitutively active form ( DA-aPKC ) . This construct can expand the apical domain of superficial cells at the expense of the basolateral domain in Xenopus embryos ( Sabherwal et al . , 2009 ) and could significantly reduce cell adhesion in ascidian embryos . Ve::Tpx2 was expressed together with DA-aPKC in order to monitor spindles during mitosis ( McDougall et al . , 2015 ) . 4 cell stage embryos stained with Cell Mask-Orange were compressed between slide and coverslip . Only embryos showing a compressed Animal-Vegetal axis of 45 µm ( confirmed by confocal imaging ) or less were used for analysis . Time-lapse imaging of Venus , GFP , mRfp1 , Cherry and Tomato constructs was performed on a Zeiss Axiovert200 and a Zeiss Axiovert100 inverted microscopes set up for epifluorescence imaging . Sequential brightfield and fluorescence images were captured using a cooled CCD camera ( Micromax , Sony Interline chip , Princeton Instruments , Trenton NJ ) and data was collected using MetaMorph software ( Molecular Devices , Sunnyvale CA ) essentially as described in McDougall et al . ( 2014 ) , ( 2015 ) . Time series were reconstructed and analysed by MetaMorph and Image J ( NIH , USA ) software packages . 4D confocal imaging was performed on Leica CSLM SP2 through a long distance 40X ( NA = 0 . 8 ) objective to obtain 3D embryos over time ( 30–35 z-planes imaged every minute ) that were manually segmented and 3D rendered using Imaris 3 . 7 . 2D imaging was performed for Figures 1C , 2 , 5 , 6 and 8 and only the cells whose spindle remains in the imaging plane during the whole of mitosis are analysed . 3D confocal imaging and 3D rendering were performed only for Figures 3 and 4 . The complete protocol for 3D rendering of confocal stacks using Imaris ( x64 , version 7 . 7 . 2 , Bitplane ) is published in details in McDougall et al . ( 2015 ) and can be downloaded from http://www . biodev . obs-vlfr . fr/~dumollard/protocols/Segmentation-manuelle-Imaris-En . pdf . All cell contours were drawn manually on 2D-slices before 3D rendering and analysis by Imaris . Sphericity of the 3D shapes was calculated by Imaris software ( ‘statistics’ function ) for each blastomere and compared to the sphericity of a spherical standard ( an in vivo isolated blastomere at metaphase from a 2 cell stage which showed a sphericity of 0 . 975 ± 0 . 001 , n = 4 cells ) . Sphericity was found to be significantly lower than spherical standard at the 4 cell stage ( 0 . 897 ± 0 . 013 , n = 8 cells ) . To calculate the apical surface ratio ( apical surface related to total surface ) , the contact-free surface ( apical in red ) and the surface of contacts with other cells ( basolateral in green ) were manually segmented by cutting and duplicating the 3D surfaces of each blastomere ( the protocol used for cutting 3D objects using Imaris can be downloaded from http://www . biodev . obs-vlfr . fr/~dumollard/protocols/protocole-surface-apicale-En . pdf ) . In Figure 3B , the apical surface ratio and sphericity of the eight quasi-synchronous animal cells was compared to the six quasi-synchronous vegetal cells at early interphase , prophase and metaphase . B6 . 3 and B6 . 4 were not analysed in this experiments because they were delayed with the rest of the embryo ( by ~10 min ) . The protocol for apical plane extraction in each blastomere is published in McDougall et al . ( 2015 ) . Briefly , the apical plane ( i . e . the plane parallel to the apical surface ) considered as the 2D plane comprising the two spindle poles which can separate most of the apical and basolateral membranes was cut ( using clipping plane function of Imaris ) from 3D rendered blastomeres . Prediction of spindle position in 2D extracted planes was performed with MatLab using scripts which may be dowloaded at: http://www . minclab . fr/research/ ( Minc et al . , 2011 ) . This model postulates that MTs radiating from spindle poles reach the cell cortex and pull with forces that scale to MT length . The model assays all possible orientations in the 2D shape and computes the evolution of the torque with spindle orientation , which informs on the mechanical equilibrium corresponding to stable spindle orientation . To implement the model with the current study , the outline of the cells and the experimental positions of the spindle poles were drawn manually as inputs . The script returned the difference between the centers of observed and predicted spindles ( computed as a ‘centering deviation’ in % ) and the difference in angle orientation between predicted and observed spindles ( computed as an ‘orienting deviation’ in degrees ( ° ) ) . A protocol for how this script was used in this study may be downloaded from http://www . biodev . obs-vlfr . fr/~dumollard/protocols/Minc-prediction-En . pdf . For in situ hybridization of mRNAs , embryos were fixed in 100 mM MOPS pH7 . 6/0 . 5 M NaCl/4% formaldehyde ON at 4°C and then washed in PBS , dehydrated in ethanol and stored at −20°C . Fixed embryos were then processed as described in Sardet et al . ( 2011 ) . Bar graphs in all figures except in Figures 6B and 8B show mean with error bars indicating s . e . m . The number of cell analysed for each graph bar is indicated in the figure legend . Statistical difference was evaluated by an unpaired two-tailed Student t-test ( with Excell ) and the Wilcoxon rank sum test , or the one sample Kolmogorov-Smirnov test with R . software package , and P values are depicted in the figure legends . Bar graphs in Figures 6B and 8B show percentage of misoriented cell divisions ( i . e . cell divisions showing a different orientation to the invariant cleavage pattern ) in each blastomere calculated as the ratio of misoriented cell division divided by the total number of cell divisions analysed ( indicated in the graph legend ) .	The position of cells within an embryo early in development determines what type of cells they will become in the fully formed embryo . The embryos of ascidians , commonly known as sea squirts , are ideal for studying what influences cell positioning . These embryos consist of a small number of cells that divide according to an “invariant cleavage pattern” , which means that the positioning and timing of the cell divisions is identical between different individuals of the same species . The pattern of cell division is also largely the same across different ascidian species , which raises questions about how it is controlled . When a cell divides , a structure called the spindle forms inside it to distribute copies of the cell’s genetic material between the new cells . The orientation of the spindle determines the direction in which the cell will divide . Now , by combining 3D imaging of living ascidian embryos with computational modeling , Dumollard et al . show that the spindles in every equally dividing cell tend to all align in the long length of the cell’s “apical” surface . Such alignment allows the cells to be on the outside of the embryo and implements the ascidian invariant cleavage pattern . The cells in the embryo do not all divide at the same time . Indeed , the shape of the cells ( and especially their apical surface ) depends on two layers of cells in the embryo not dividing at the same time; instead , periods of cell division alternate between the layers . A network of genes in the embryo regulates the timing of these cell divisions and makes it possible for the cells to divide according to an invariant cleavage pattern . However , this network of genes is not the only control mechanism that shapes the early embryo . A structure found in egg cells ( and hence produced by the embryo’s mother ) causes cells at the rear of the embryo to divide unequally , and this influences the shape of all the cells in the embryo . Thus it appears that maternal mechanisms work alongside the embryo’s gene network to shape the early embryo . The next step will be to determine how physical forces – for example , from the cells pressing against each other – influence the position of the embryo’s cells . How do gene networks relay the biomechanical properties of the embryo to help it take shape ?	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2017	The invariant cleavage pattern displayed by ascidian embryos depends on spindle positioning along the cell's longest axis in the apical plane and relies on asynchronous cell divisions
Stem cells fuel the development and maintenance of tissues . Many studies have addressed how local signals from neighboring niche cells regulate stem cell identity and their proliferative potential . However , the regulation of stem cells by tissue-extrinsic signals in response to environmental cues remains poorly understood . Here we report that efferent octopaminergic neurons projecting to the ovary are essential for germline stem cell ( GSC ) increase in response to mating in female Drosophila . The neuronal activity of the octopaminergic neurons is required for mating-induced GSC increase as they relay the mating signal from sex peptide receptor-positive cholinergic neurons . Octopamine and its receptor Oamb are also required for mating-induced GSC increase via intracellular Ca2+ signaling . Moreover , we identified Matrix metalloproteinase-2 as a downstream component of the octopamine-Ca2+ signaling to induce GSC increase . Our study provides a mechanism describing how neuronal system couples stem cell behavior to environmental cues through stem cell niche signaling . Animal tissues are built from cells originally derived from stem cells ( Spradling et al . , 2001 ) . During normal development and physiology , this robust stem cell system is precisely regulated ( Drummond-Barbosa , 2008 ) . Conversely , the dysregulation of these cells can result in abnormal tissue integrity and lead to deleterious diseases . Previous studies have revealed that many types of stem cells reside in a specialized microenvironment , or niche , where they are exposed to local signals required for their function and identity ( Morrison and Spradling , 2008; Spradling et al . , 2001 ) . Recently , researchers have demonstrated how stem cell activity is regulated by tissue-extrinsic signals , such as hormones and neurotransmitters . For instance , in mammals , hematopoietic stem cells , mammary stem cells , muscle stem cells , and neural stem cells are influenced by sex hormones such as estrogen ( Asselin-Labat et al . , 2010; Bramble et al . , 2019; Kim et al . , 2016; Nakada et al . , 2014 ) . Retinoic acid and thyroid hormone play essential roles in the differentiation of testicular stem cells and neural stem cells , respectively ( Gothié et al . , 2017; Ikami et al . , 2015 ) . In addition , mesenchymal stem cell proliferation is stimulated by adrenaline ( Wu et al . , 2014 ) . However , the when , how , and why these humoral factors are produced , circulated , and received during stem cell regulation remain to be elucidated . The ovaries of the fruit fly Drosophila melanogaster are an excellent model system on how stem cell lineages are shaped by both local niche signals and tissue-extrinsic signals ( Drummond-Barbosa , 2019 ) . D . melanogaster ovary is composed of 16–20 chains of developing egg chambers called ovarioles . The anterior-most region of which , known as the germarium , contains germline stem cells ( GSCs ) that give rise to the eggs ( Figure 1A and B ) . GSCs are adjacent to the somatic niche cells , which comprises cap cells , escort cells , and terminal filament cells ( Figure 1A ) . After GSC divides , one daughter cell that remains attached to the niche cells retains its GSC identity , whereas the remaining daughter cells are displaced away from the niche cells and differentiate into cystoblast ( CB ) . Each CB then undergoes differentiation into 15 nurse cells and one oocyte in each egg chamber , which is surrounded by somatic follicle cells . GSC niche produces and secretes several local niche signals that regulate the balance between GSC self-renewal and differentiation ( Hayashi et al . , 2020; Kirilly and Xie , 2007; Spradling et al . , 2011 ) . For example , bone morphogenetic protein ( BMP ) ligands Decapentaplegic ( Dpp ) and Glass bottom boat ( Gbb ) are produced from the niche cells and directly activate BMP receptors in GSCs , leading to the repression of the differentiation inducer , bag-of-marbles ( bam ) ( Morrison and Spradling , 2008; Zhang and Cai , 2020 ) . Recent D . melanogaster GSC studies have also contributed to understanding of the systemic regulation of stem cell proliferation and maintenance in response to external environmental cues ( Ables and Drummond-Barbosa , 2017; Drummond-Barbosa , 2019; Lin and Hsu , 2020; Yoshinari et al . , 2019 ) . For example , protein restriction results in a reduction in GSC division , which is mediated by Drosophila insulin-like peptides ( DILPs ) ( LaFever , 2005 ) . In addition , nutrients influence GSC maintenance via the adipocyte metabolic pathway ( Armstrong and Drummond-Barbosa , 2018; Matsuoka et al . , 2017 ) . Besides nutrients , we have recently found out that mating is another external cue that significantly affects D . melanogaster GSC increase . Mated females show a dramatic increase in egg production , as well as GSC , which is induced by a male-derived peptide from the seminal fluid called sex peptide ( SP ) ( Kubli , 2003; Yoshinari et al . , 2019 ) . SP is received by its specific receptor , sex peptide receptor ( SPR ) , in a small subset of SPR-positive sensory neurons ( SPSNs ) , which are located in the uterine lumen and send afferent axons into the tip of the abdominal ganglion ( Häsemeyer et al . , 2009; Yapici et al . , 2008 ) . The SP-SPR signaling in the SPSNs stimulates the biosynthesis of the ovarian insect steroid hormones ( ecdysteroids ) , which play an essential role in mating-induced GSC increase ( Ameku et al . , 2017; Ameku and Niwa , 2016; Uryu et al . , 2015 ) . Because SPSNs do not directly innervate into the ovary , it is hypothesized that a signal and its signaling pathway are involved in bridging the gap between SPSNs and GSCs . However , it is still unclear how mating information is transmitted from SPSNs to GSCs at the molecular and cellular levels . Here , we present a series of new findings that reveal a novel and fundamental neuronal mechanism connecting SPSNs and GSCs to regulate mating-induced GSC increase . We demonstrate that a small subset of neurons directly innervating into the ovary plays an indispensable role in regulating mating-induced GSC increase . These neurons produce the monoamine neurotransmitter , octopamine ( OA ) , the insect equivalent of noradrenaline ( Roeder , 2005 ) . We also show that the neuronal activity of the OA-producing neurons is required for mating-induced GSC increase . Moreover , we find that the OA directory activates GSC increase through its receptor , octopamine receptor in mushroom body ( Oamb ) , followed by Ca2+ signaling in the ovarian escort cells . Furthermore , OA/Oamb signaling requires Matrix metalloproteinase 2 ( Mmp2 ) to activate GSC increase in the ovarian escort cells . Finally , we show that SPSNs relay the mating signal to the ovary-projecting OA neurons via nicotinic acetylcholine receptor signaling . Taken together , we propose a novel efferent neuronal pathway that transmits mating stimulus to the GSC to control stem cell number . Our study provides a mechanism describing how neuronal system couples stem cell behavior to environmental cues , such as mating , through stem cell niche signaling . As a candidate signal that bridges between SPSNs and GSC increase , we focused on the biogenic amine , OA , because a part of the octopaminergic neurons innervate to the ovary and the oviduct ( Heifetz et al . , 2014; Rezával et al . , 2014 ) . Moreover , it has been reported that OA and Oamb signaling regulate ovulation process and ovarian-muscle contraction ( Deady and Sun , 2015; Monastirioti , 2003; Rezával et al . , 2014 ) . We first conducted transgenic RNAi screen against 4 OA receptor genes with c587-GAL4 , which is active in the ovarian-somatic cells , including the escort cells of the germarium ( Manseau et al . , 1997 ) . In control females , the mated ones exhibited an increase in GSC number as we have reported previously ( Ameku and Niwa , 2016; Figure 1C ) . In contrast , c587-GAL4–mediated Oamb knock-down ( c587 >OambRNAi ) showed significantly impaired mating-induced GSC increase ( Figure 1C ) . This phenotype was observed with two independent UAS-Oamb-RNAi strains ( OambRNAi1 and OambRNAi2 ) ( Figure 1C ) , each of which targeted a different region in the Oamb mRNA . The specificity of Oamb was also confirmed by the fact that the c587-GAL4–driven transgenic RNAi of other octopamine receptor genes ( Octβ1R , Octβ2R and Octβ3R ) ( Ohhara et al . , 2012 ) had no significant effect on the GSC number between virgin and mated females ( Figure 1C ) . Therefore , Oamb has a pivotal role in mating-induced GSC increase . We next examined in which ovarian-somatic cells Oamb regulates mating-induced GSC increase with several GAL4 lines that are active in the specific ovarian-somatic cells . Previous studies have demonstrated that Oamb in mature follicle cells and in the oviduct has a significant role in ovulation ( Deady and Sun , 2015; Lee et al . , 2009; Lee et al . , 2003 ) . Therefore , it is possible that Oamb may indirectly induce GSC increase via Oamb-mediated ovulation processes . However , mating-induced GSC increase was not impaired by Oamb RNAi in the stage-14 follicle cells by R44E10-GAL4 ( Deady and Sun , 2015 ) ( R44E10 >OambRNAi ) , in the stage 9–10 follicle cells by c355-GAL4 , c306-GAL4 , and slbo-GAL4 ( Barth et al . , 2012 ) , or in the common oviduct by RS-GAL4 ( Lee et al . , 2003 ) ( RS >OambRNAi ) ( Figure 1—figure supplement 1A , B and E ) . These data suggest that mating-induced GSC increase is independent from the ovulation process . Consistent with the observation using c587-GAL4 , Oamb RNAi by Traffic jam ( Tj ) -GAL4 ( Olivieri et al . , 2010 ) ( Tj >OambRNAi ) , R13C06-GAL4 , and 109–30-GAL4 ( Sahai-Hernandez and Nystul , 2013 ) , which are active in the pan-ovarian-somatic cells , the escort cells , and the germarium follicle cells , respectively , also resulted in the failure of mating-induced GSC increase ( Figure 1—figure supplement 1C and E ) . On the other hand , Oamb RNAi in the cap cells ( bab >OambRNAi ) and germ cells ( nos >OambRNAi ) had no effect on GSC increase ( Figure 1—figure supplement 1D ) . These results suggest that Oamb in the escort cells or the follicle cells of the germarium plays an essential role in mating-induced GSC increase . It must be noted that c587-GAL4 and Tj-GAL4 are expressed not only in the ovarian-somatic cells but also in the nervous system ( Ameku et al . , 2018 ) . Moreover , Oamb is expressed in the nervous system ( Han et al . , 1998 ) . However , Oamb RNAi in the nervous system ( nSyb >OambRNAi ) did not affect the mating-induced GSC increase ( Figure 1—figure supplement 1D ) , suggesting that the impairment of GSC increase of c587 >OambRNAi or tj >OambRNAi is not due to gene knock-down in neuronal cells but rather in the ovarian-somatic cells . These data also support our idea that Oamb in the ovarian-somatic cells regulate mating-induced GSC increase . To confirm the role of Oamb in mating-induced GSC increase , we generated a Oamb complete loss-of-function genetic allele by Clustered Regularly Interspaced Short Palindromic Repeats ( CRISPR ) /CRISPR-associated protein 9 ( Cas9 ) technology ( Kondo and Ueda , 2013; Figure 1—figure supplement 1F ) . Similar to Oamb RNAi females , Oamb homozygous mutant females ( c587>+; OambΔ/OambΔ ) did not exhibit mating-induced GSC increase ( Figure 1D ) . In addition , the GSC increase of OambΔ/OambΔ was restored by overexpression of Oamb in the escort cells ( c587 >OambAS; OambΔ/OambΔ ) . These findings are all consistent with the idea that Oamb in escort cells modulates GSC increase after mating . We also examined Oamb expression in the ovarian-somatic cells by two Oamb-knock-in GAL4 lines ( Deng et al . , 2019; Kondo et al . , 2020 ) . However , we could not detect any reliable signals in the germarium or the mature follicle cells via these GAL4 lines with UAS-GFP and UAS-Stinger lines ( Figure 1—figure supplement 2A , B , C and D ) . We speculate that this may be due to lower amounts of Oamb transcript in the germarium . Because mating-induced GSC increase is accompanied by GSC division ( Ameku and Niwa , 2016 ) , We next examined whether Oamb in the escort cells is involved in GSC division after mating . We determined the number of GSCs during the M phase by staining using anti-phospho-Histone H3 ( pH3 ) in control and c587 >OambRNAi females . In control female flies , mating increased the frequency of GSCs in the M phase ( Figure 1E ) , whereas in c587 >OambRNAi flies , this was not observed . We also monitored the fraction of apoptotic cells in the germarium by staining with anti-cleaved death caspase-1 ( Dcp-1 ) , a marker for apoptotic cells ( Song et al . , 1997 ) . The number of apoptotic cells in the germarium did not change in c587 >OambRNAi female flies compared with controls ( Figure 1F ) , suggesting that Oamb activates GSC increase by pushing the cell cycle of GSCs and that the lack of mating-induced GSC increase in Oamb RNAi is not due to the enhancement of cell death . Our previous studies revealed that mating-induced GSC increase is mediated by GSC niche signals ( Ameku et al . , 2018; Ameku and Niwa , 2016 ) . In particular , Decapentaplegic ( Dpp ) , the fly counterpart to bone morphogenetic protein ( BMP ) , is the essential niche signal ( Spradling et al . , 2011; Xie and Spradling , 1998 ) . We therefore examined whether Oamb knock-down affects Dpp signaling in GSCs by measuring the level of phosphorylated Mad ( pMad ) , a readout of Dpp signaling activation ( Chen and McKearin , 2003; Raftery and Sutherland , 1999 ) . We confirmed that mating induced the increase in pMad level in GSCs , whereas mating did not increase pMad levels in c587 >OambRNAi animals ( Figure 1G and H ) . We also confirmed that mating increased the signal intensity of Daughters against dpp ( Dad ) -LacZ , a reporter gene that reflect an expression of the BMP target gene Dad , while Oamb knock-down in the escort cells impaired the increase of Dad-LacZ signal after mating ( Figure 1—figure supplement 3A and B ) . These results suggest that mating activates the BMP signal in GSCs through Oamb in the escort cells , thereby resulting in the increase in GSCs . Further , we determined the number of cap cells , which are critical components of the GSC niche ( Xie and Spradling , 1998 ) . c587 >OambRNAi did not change the number of cap cells in virgin nor mated female flies , suggesting that Oamb knock-down does not affect the overall architecture of the niche ( Figure 1I ) . Overall , Oamb in the escort cells plays a pivotal role in mating-induced GSC increase . To examine whether OA is received in the ovary but not in other organs to induce GSC increase , we cultured dissected virgin ovaries ex vivo with or without purified OA in the culture medium . After incubation for 12 hr , the ovaries cultured with OA had more GSCs as compared to those without OA ( Figure 2—figure supplement 1A ) . Whereas the minimal OA concentration to induce ex vivo GSC increase was 1 μM , hereafter we used 100 μM OA because the GSC number plateaued with this concentration ( Figure 2—figure supplement 1A ) . This OA-mediated ex vivo GSC increase was not observed in c587 >OambRNAi virgin ovaries ( Figure 2A ) . We also examined whether OA treatment affects Dpp signaling in GSCs ex vivo by measuring the level of pMad . We confirmed that OA treatment was sufficient to induce the increase in pMad level in GSCs , even in the ex vivo culture system ( Figure 2B ) . On the other hand , OA treatment did not increase pMad levels in c587 >OambRNAi ovaries ( Figure 2B ) . These results suggest that OA activates the BMP signal in GSCs through Oamb in the escort cells . Upon OA binding , Oamb evokes Ca2+ release from the endoplasmic reticulum ( ER ) into the cytosol , leading to a transient increase in intercellular Ca2+ concentration ( [Ca2+]i ) ( Han et al . , 1998 ) . To determine whether OA induces GSC increase via affecting [Ca2+]i in escort cells , we first monitored [Ca2+]i using a genetically encoded calcium sensor , GCaMP6s ( Nakai et al . , 2001; Ohkura et al . , 2012 ) . We dissected the virgin ovaries , in which GCaMP6s transgene was expressed driven by Tj-GAL4 or c587-GAL4 , cultured them ex vivo , and then observed GCaMP6s fluorescence ( Figure 2C ) . We found that 100 μM of OA treatment evoked an increase in GCaMP6s fluorescence in the escort cells and follicle cells , whereas the control medium treatment ( 0 μM of OA ) did not show any increase in fluorescence ( Figure 2D and Figure 2—figure supplement 1B; also see Videos 1 and 2 ) . In addition , the OA-mediated increase in GCaMP6s fluorescence intensity was not observed in the germarium of Oamb RNAi flies ( Figure 2E ) , suggesting that the OA-dependent increase in [Ca2+]i in the germarium is required for Oamb . Notably , the timeline of signal increase was very slow as this fluorescence increases progressively . This response to OA treatment was similar to a report featuring mature follicle cells of stage-14 oocytes ( Deady and Sun , 2015 ) . We also employed another approach utilizing the light-gated cation channel , CsChrimsonn ( Klapoetke et al . , 2014 ) . We prepared c587-GAL4 >CsChrimson flies in combination with nSyb-GAL80 ( Harris et al . , 2015 ) , allowing the expression of CsChrimson gene only in the germarium but not in the nervous system . Because CsChrimson requires all trans-retinal ( ATR ) to form its proper protein conformation ( Wang et al . , 2012 ) , we utilized the flies fed with and without ATR supplement as experimental and control groups , respectively . When we irradiated an orange-light to c587-GAL4 , nSyb-GAL80 >CsChrimson flies to induce Ca2+ flux in the germarium ( Figure 2F ) , GSC increased in the virgin females ( Figure 2G ) . In addition , the ratio of GSC in the M phase increased by CsChrimson activation ( Figure 2H ) . Moreover , the pMad level in GSCs was increased by CsChrimson activation , suggesting that the forced Ca2+ flux is sufficient to induce GSC increase through the upregulation of BMP signaling ( Figure 2I ) . We next confirmed whether the downstream component of Ca2+ signaling is involved in the mating-induced GSC increase . The knock-down of Inositol 3-receptor ( c587 >Insp3RRNAi ) encoding a protein that releases the stored Ca2+ from ER suppressed GSC increase in mated females ( Figure 2J ) . Conversely , the overexpression of Insp3R in the escort cells increased the GSC number in virgin females ( Figure 2J ) . Furthermore , OA-mediated ex vivo GSC increase was not observed in c587 >Insp3RRNAi virgin ovaries ( Figure 2A ) . Overall , we demonstrated that OA signaling regulates the mating-induced GSC increase by controlling Ca2+ signaling in escort cells , which is thereby necessary and sufficient to induce GSC increase . The biosynthesis and signaling of ecdysteroid in the ovary are required for the mating-induced GSC increase ( Ameku et al . , 2017; Ameku and Niwa , 2016 ) . Therefore , we next examined whether ecdysteroid signaling has a pivotal role in the OA-Oamb-Ca2+-dependent GSC increase . As we have previously reported , the RNAi of neverland ( nvd ) , which encodes an ecdysteroidogenic enzyme ( Yoshiyama-Yanagawa et al . , 2011; Yoshiyama et al . , 2006 ) in the escort cells , suppressed the mating-induced GSC increase ( Figure 3A; Ameku et al . , 2017; Ameku and Niwa , 2016 ) . We also found a similar phenotype in the RNAi of ecdysone receptor ( EcR ) in the escort cells ( Figure 3A ) . To assess the requirement of ecdysteroid biosynthesis and signaling in OA-induced GSC increase , we employed an ex vivo experiment . Interestingly , the OA-mediated GSC increase was not observed in nvd RNAi ovaries ( c587 >nvdRNAi ) ( Figure 3B ) . Moreover , this impairment of GSC increase in c587 >nvdRNAi was restored with the administration of 20-hydroxyecdysone ( 20E ) in the culture media . On the other hand , 20E treatment without OA did not induce GSC increase in the control ovaries ( Figure 3B ) . This observation is consistent with our previous study showing that wild-type virgin females fed with 20E did not exhibit any increase in GSCs ( Ameku and Niwa , 2016 ) . Therefore , 20E is required for the OA-mediated increase in GSCs; however , by itself , 20E is not sufficient to induce GSC increase . To assess the role of EcR in downstream OA signaling , we utilized the temperature-sensitive and null alleles of EcR ( EcRA483T and EcRM54fs , respectively ) ( Bender et al . , 1997 ) . At a restrictive temperature of 31°C , the mating-induced GSC increase was suppressed in EcRA483T/EcRM54fs flies , consistent with our previous report ( Ameku and Niwa , 2016; Figure 3C ) . In the ex vivo experiment , the OA-dependent GSC increase was also suppressed in EcRA483T/EcRM54fs ovaries ( Figure 3D ) . These results suggest that OA-Oamb-Ca2+ signaling requires ovarian ecdysteroid signaling . So far , we have identified four components with indispensable roles in mating-induced GSC increase , namely OA , Oamb , Ca2+ , and ecdysteroids . Interestingly , recent studies have reported that they are also essential in the follicle rupture in D . melanogaster ovary ( Deady and Sun , 2015; Knapp and Sun , 2017 ) . In this process , Matrix metalloproteinase 2 ( Mmp2 ) , a membrane-conjugated proteinase , acts downstream of the OA-Oamb-Ca2+ signaling pathway in mature follicle cells ( Deady et al . , 2015 ) . Because recent studies have indicated the expression of Mmp2 in niche cells , including escort cells ( Pearson et al . , 2016; Wang and Page-McCaw , 2014 ) , we examined Mmp2 function in the escort cells using RNAi of Mmp2 with c587-GAL4 . Similar to c587 >OambRNAi , c587 >Mmp2RNAi impaired mating-induced GSC increase ( Figure 4A ) . Notably , Mmp2 knock-down in cap cells by bab-GAL4 also suppressed GSC increase , suggesting that Mmp2 acts in both escort cells and cap cells to induce GSC increase ( Figure 4—figure supplement 1A ) . We also found that Mmp2 RNAi in the ovarian-somatic cells by another GAL4 ( Tj >Mmp2RNAi1 ) resulted in the failure of mating-induced GSC increase , whereas Mmp2 RNAi in the nervous system ( nSyb >Mmp2RNAi1 ) and in the follicle cells of stage 14 oocytes ( R44E10 >Mmp2RNAi1 ) had no effect on GSC increase ( Figure 4—figure supplement 1B ) . These results suggest that Mmp2 in the escort cells and cap cells are necessary to induce post-mating GSC increase . In the GSC niche cells in D . melanogaster , Tissue inhibitors of metalloproteinases ( Timp ) gene encoding an endogenous proteinase inhibitor of Mmp2 is expressed ( Gomis-Rüth et al . , 1997; Page-McCaw et al . , 2003; Pearson et al . , 2016 ) . The knock-down of Timp in escort cells induced GSC increase even in virgin females , whereas its overexpression suppressed mating-induced GSC increase ( Figure 4B ) . Consistent with Mmp2 knock-down , Timp knock-down in cap cells ( bab >TimpRNAiRNAi2 ) increased the GSC number in virgin females , whereas its knock-down in the follicle cells of stage 14 oocyte ( R44E10 >TimpRNAiRNAi2 ) had no effect ( Figure 4—figure supplement 1C ) . These data suggest that Mmp2 activity in the GSC niche cells is necessary for mating-induced GSC increase . We next examined whether Mmp2 is necessary for OA-induced GSC increase . Our ex vivo culture experiment revealed that OA-induced GSC increase was suppressed in c587 >Mmp2 RNAi flies , suggesting that Mmp2 acts downstream of OA signaling ( Figure 4C ) . Moreover , the OA-dependent upregulation of pMad level in GSCs was suppressed in c587 >Mmp2 RNAi flies ( Figure 4D ) . Notably , in virgin females , Mmp2 RNAi affected neither the GSC number nor the cap cell number , indicating that Mmp2 RNAi does not influence the overall niche architecture ( Figure 4—figure supplement 1D ) . Mmp2 in the mature follicle cells cleaves and downregulates collagen VI , also known as Viking ( Vkg ) and is the major component of the basement membrane ( Deady et al . , 2017; Wang et al . , 2008 ) . However , Mmp2 RNAi had no effect on Vkg::GFP level around the cap cells ( Figure 4—figure supplement 1E ) . Therefore , the suppression of mating-induced GSC increase in Mmp2 RNAi does not likely depend on the collagen VI level . To examine the epistasis of OA/Oamb-Ca2+ signaling , ecdysteroid signaling , and Mmp2 , we knocked down Oamb , nvd , or Mmp2 in c587 >Insp3ROE genetic background , where Ca2+ signaling was forcedly upregulated . Whereas the Oamb RNAi did not suppress GSC increase in virgin female ( c587 >Insp3ROE , OambRNAi1 ) , the RNAi of nvd or Mmp2 suppressed GSC increase even when Ca2+ signaling were activated ( c587 >Insp3ROE , nvdRNAi or c587 >Insp3ROE , Mmp2RNAi1 ) ( Figure 4E ) . These results suggest that ecdysteroid signaling and Mmp2 act downstream of Ca2+ signaling in OA-induced GSC increase . Taken together , both the OA-Oamb signaling and the downstream Ca2+ signaling regulate the mating-induced GSC increase via Mmp2 and ecdysteroid signaling ( Figure 4F ) . To examine the in vivo role of OA in mating-induced GSC increase , we silenced the expression of Tyrosine decarboxylase 2 ( Tdc2 ) and Tyramine β hydroxylase ( TβH ) genes , which code for enzymes responsible for OA biosynthesis ( Cole et al . , 2005; Monastirioti et al . , 1996; Figure 5A ) , with Tdc2-GAL4 driver-mediated RNAi ( Tdc2 >Tdc2RNAi1 , Tdc2 >TbHRNAiRNAi1 ) . Similar to the phenotype of Oamb RNAi , Tdc2 or TbH RNAi with Tdc2-GAL4 or nSyb-GAL4 impaired the mating-induced GSC increase ( Figure 5A and Figure 5—figure supplement 1A–B ) . Moreover , the impairment of GSC increase was restored when Tdc2 or TbH RNAi flies were fed with food supplemented with OA ( Figure 5A ) , supporting our hypothesis that OA is responsible for the mating-induced GSC increase in vivo . Because Tdc2-GAL4 is active in the nervous system ( Busch et al . , 2009; Pauls et al . , 2018 ) , we then identified which neurons secrete OA to the escort cells . D . melanogaster has more than 70–100 OAergic neurons dispersed throughout the nervous system ( Monastirioti , 2003; Schwaerzel et al . , 2003; Zhou et al . , 2008 ) . Among them , we were particularly interested in a small subset innervating the reproductive system ( Figure 5B ) as several recent studies have revealed that these neurons regulate mating behavior , egg laying , and ovarian-muscle contraction ( Heifetz et al . , 2014; Lee et al . , 2003; Middleton et al . , 2006; Rezával et al . , 2014; Rubinstein and Wolfner , 2013 ) . The ovary-projecting OAergic neurons are doublesex ( dsx ) + and Tdc2+ double-positive ( Rezával et al . , 2014; Figure 5B ) . Therefore , to manipulate the gene expression of dsx+ Tdc2+ neurons only , we implemented a FLP/FRT intersectional strategy using dsx-FLP ( Rezával et al . , 2014 ) . We could detect GFP expression only in the dsx+ Tdc2+ neurons innervating to the ovary , whose cell bodies are located on a caudal part of the abdominal ganglion ( Rezával et al . , 2014; Figure 5—figure supplement 1C–E ) . We next knocked down Tdc2 in dsx+ Tdc2+ neurons by RNAi ( tub >GAL80>Tdc2RNAi; dsx-FLP ) and found that these RNAi flies failed to increase GSC number after mating . Given the fact that the intersectional strategy using dsx-FLP does not label any neurons in the central nervous system ( Rezával et al . , 2014 ) , these data suggest that only a small subset of dsx+ Tdc2+ neurons controls the mating-induced GSC increase . To assess whether the activity of dsx+ Tdc2+ neurons affects the GSC number , we overexpressed TrpA1 , a temperature-sensitive cation channel gene , in the dsx+ Tdc2+ neurons only . In Tdc2 >stop >TrpA1;dsx-FLP flies , we can tightly control the TrpA1 expression in the dsx+ Tdc2+ neurons only ( Rezával et al . , 2014 ) . Both the control flies and TrpA1-overexpressing flies at permissive temperature ( 17°C ) had the normal GSC number in virgin and mated females . On the other hand , at the restrictive temperature ( 29°C ) , the TrpA1-overexpressing flies , even the virgin ones , had more GSCs ( Figure 5D ) . We also found that Tdc2 >stop >TrpA1; dsx-FLP virgin females at the restrictive temperature had increased GSC frequency in the M phase ( Figure 5E ) . Importantly , the TrpA1-mediated activation of Tdc2 neurons did not induce the GSC increase in loss-of -Oamb-function females ( Tdc2 >TrpA1; OambΔ/OambΔ ) ( Figure 5F ) , suggesting that the TrpA1-mediated GSC increase requires Oamb . Furthermore , we employed the Tetanus toxin light chain ( TNT ) to inhibit neuronal activity ( Sweeney et al . , 1995 ) . When we overexpressed TNT in dsx+ Tdc2+ neurons only , the mating-induced GSC increase was suppressed in mated females as compared with the control , whose inactivated TNTin was overexpressed ( Figure 5G ) . Taken together , these findings suggest that the mating-induced GSC increase is mediated by the neuronal activity of dsx+ Tdc2+ neurons innervating to the ovary . Because the dsx+ Tdc2+ neuronal activity has a significant role in mating-induced GSC increase , we next examined whether these neurons change their activity before and after mating . We monitored the neuronal activity using an end-point Ca2+ reporting system , the transcriptional reporter of intracellular Ca2+ ( TRIC ) ( Gao et al . , 2015 ) . TRIC is designed to increase the GFP expression in proportion to [Ca2+]i . We classified female Tdc2+ neurons in the caudal part of the abdominal ganglion into three clusters based on their location and morphology . We designated the three clusters of these Tdc2+ neurons as the Tdc2+ median , Tdc2+ dorsal , and Tdc2+ caudal clusters ( Figure 5H ) . Among them , the position of first two clusters are not similar to that of dsx+ Tdc2+ neurons , whereas that of the Tdc2+ caudal cluster is similar ( Rezával et al . , 2014 ) . In virgin females , we detected robust TRIC signals in the Tdc2+ median and Tdc2+ dorsal clusters but not in the Tdc2+ caudal cluster ( Figure 5I–L ) . In contrast , 24 hr after mating , we observed a significant increase in the TRIC signal in the Tdc2+ caudal cluster , whereas those in the Tdc2+ median and Tdc2+ dorsal clusters were not changed in virgin and mated females ( Figure 5I–L ) . This result suggests that the Tdc2+ caudal cluster , which are likely dsx+ Tdc2+ neurons , is significantly activated after mating . Our previous study revealed that the mating-induced GSC increase is mediated by the male seminal fluid protein SP ( Ameku and Niwa , 2016 ) . SP is received by SPR in a small number of SPSNs , followed by a neural silencing of SPSNs ( Häsemeyer et al . , 2009; Yapici et al . , 2008 ) . Notably , SPSNs project their arbors into a caudal part of the abdominal ganglion , where the cell bodies of the dsx+ Tdc2+ cluster neurons are located ( Rezával et al . , 2014; Rezával et al . , 2012 ) . Therefore , we examined whether SPSNs physically interact with Tdc2+ neurons in the abdominal ganglion by performing the GFP Reconstitution Across Synaptic Partners ( GRASP ) analysis ( Feinberg et al . , 2008; Gordon and Scott , 2009 ) , in which two complementary fragments of GFP were expressed in SPSNs and Tdc2+ neurons . GRASP signals were detected in the abdominal ganglion ( Figure 6A ) , suggesting that the axon termini of SPSNs and the cell bodies and/or dendrites of Tdc2+ neurons contact each other likely through synaptic connections . Because SPSNs have been implied as cholinergic neurons ( Rezával et al . , 2012 ) , we next examined the expression of Choline acetyltransferase ( ChaT ) -GAL4 in SPSNs . ChaT encodes an acetylcholine biogenic enzyme ( Greenspan , 1980 ) . The SPSNs located on the oviduct , which also co-express pickpocket ( ppk ) and fruitless ( fru ) , are particularly crucial for inducing the major behavioral changes in female flies after mating ( Ameku and Niwa , 2016; Rezával et al . , 2012 ) . By using UAS-mCD8::RFP with ChaT-GAL4 alongside ppk-EGFP , we confirmed that the ppk-EGFP–positive population near the oviduct were co-labeled by RFP ( Figure 6B ) , consistent with the speculation that SPSNs are cholinergic . We next counted the GSC number in ChAT RNAi flies using ppk-GAL4 . ppk >ChAT RNAi virgin flies had more GSCs compared with the control ( Figure 6C ) . In addition , mating did not induce GSC increase in ppk >ChATRNAi flies , suggesting that the acetylcholine released from SPSNs is responsible for suppressing the GSC increase . To further ascertain whether the acetylcholine released from SPSNs is received by dsx+ Tdc2+ neurons to mediate mating-induced GSC increase , we focused on the fast-ionotropic nicotinic acetylcholine receptors ( nAChR ) , which belong to the Cys-loop receptor subfamily of ligand-gated ion channels ( Breer and Sattelle , 1987; Gundelfinger and Hess , 1992; Lee and O’Dowd , 1999 ) . In D . melanogaster , 10 genes coding nAChR subunits have been identified . Among these , we focused on nAChRα1 , nAChRα2 , nAChRα3 , nAChRβ1 and nAChRβ2 because the knock-down of these genes in dsx+ Tdc2+ ( tub >GAL80> , dsx-FLP; Tdc2-GAL4 ) or Tdc2 neurons ( Tdc2-GAL4 ) increased the GSC number in virgin females similar to ppk >ChATRNAi ( Figure 6D and Figure 6—figure supplement 1A ) . We then confirmed the expression of these acetylcholine receptor genes in Tdc2 neurons by generating a knock-in T2A-GAL4 line as previously described ( Kondo et al . , 2020; Ihara et al . , 2020 ) for each 5 nAChR subunits and observed their expression with UAS-mCD8::GFP . All of the five knock-in-GAL4 expressions were detected in anti-Tdc2 positive neurons around the ovary , suggesting that the ovary-projecting dsx+ Tdc2+ neurons expresses these nAChRs ( Figure 6—figure supplement 1B–F ) . To confirm the role of nAChR in mating-induced GSC increase , we generated nAChRα1 complete loss-of-function genetic alleles by CRISPR/Cas9 technology ( Kondo and Ueda , 2013; Figure 6—figure supplement 2A ) . Similar to nAChRα1 RNAi females , the nAChRα1 transheterozygous mutant virgin females ( nAChRα1228/nAChRα326 ) had more GSCs compared with the controls ( Figure 6—figure supplement 2B ) . In addition , the GSC increase of nAChRα1228/nAChRα326 was restored by the overexpression of nAChRa1 in Tdc2+ neurons ( Tdc2 >nAChRα1; nAChRα1228/nAChRα326 ) ( Figure 6—figure supplement 2C ) . These data support our hypothesis that acetylcholine signaling in Tdc2 neurons has a negative role in mating-induced GSC increase . We next assessed relationship between SPSNs , Tdc2+ neurons , and OA-Oamb-Ca2+ signaling in ovarian cells . The silencing of SPSNs neuronal activity ( SPSNs-LexA and LexAop-shits ) increased the GSC number in virgin females ( Figure 6E ) , consistent with our previous study ( Ameku and Niwa , 2016 ) . Upon SPSNs silencing , Tdc2 RNAi by Tdc2-GAL4 reduced the GSC number ( Figure 6E ) , suggesting that Tdc2+ neurons act downstream of SPSNs . In addition , the GSC increase through the silencing of SPSNs ( SPSNs-LexA >LexAop-kir2 . 1 ) was suppressed by Oamb or Insp3R RNAi in the escort cells ( Figure 6F ) , suggesting that OA-Oamb-Ca2+ signaling in ovarian cells acts downstream of SPSNs . Overall , our findings revealed a novel neuronal relay in response to mating that regulates the female GSC increase in the ovary before/after mating ( Figure 7 ) . Our proposed model is that the OA from dsx+ Tdc2+ neurons is directly received by the escort and follicle cells in the germarium . Our model is supported by two of our observations . First , mating-induced GSC increase is impaired by Oamb RNAi using a GAL4 driver that is active specifically in the germarium cells but not mature follicle cells . Second , OA treatment evokes [Ca2+]i elevation in these germarium cells in an Oamb-dependent manner . However , in this study , we did not address whether the escort cells and/or the follicle cells in the germarium express Oamb , as we failed to observe any clear GAL4 expression in two independent Oamb-T2A-GAL4 drivers ( Figure 1—figure supplement 2A , B , C and D ) . We surmise that this may be due to lower amounts of Oamb transcript in the germarium . We have shown that the activation of the ovary-projecting dsx+ Tdc2+ neurons is necessary and sufficient to induce GSC increase . However , from an anatomical point of view , the dsx+ Tdc2+ neurons project to the distal half of the ovary but not to the germarium ( Figure 5—figure supplement 1E ) . Considering our model described above , this disagreement can be attributed to the characteristic volume transmission of monoamine neurotransmitters . In other words , neurotransmitters act at a distance well beyond their release sites from cells or synapses ( Fuxe et al . , 2010 ) . Therefore , the OA secreted from the terminals of dsx+ Tdc2+ neurons could reach the germarium located at the most proximal part of the ovary . Several previous studies have revealed that OA signaling has a pivotal role in reproductive tissues other than germarium , such as mature follicle cells , oviduct , and ovarian muscle , to promote ovulation , oviduct remodeling , and ovarian-muscle contraction , respectively ( Deady and Sun , 2015; Heifetz et al . , 2014; Lee et al . , 2009; Middleton et al . , 2006; Rezával et al . , 2014 ) . Therefore , it is likely that the dsx+ Tdc2+ neurons orchestrate multiple different events during oogenesis in response to mating stimulus . Because a mated female needs to activate oogenesis to continuously produce eggs in concert with sperm availability , it is reasonable that the ovary-projecting neurons switch on the activity of the entire process of reproduction . Based on our present study and several previous studies ( Deady et al . , 2015; Deady and Sun , 2015; Knapp and Sun , 2017 ) , the OA-Oamb-Ca2+-Mmp2 axis is required for GSC increase and follicle rupture , both of which are induced by mating stimuli in D . melanogaster . In both cases , Mmp2 enzymatic activity is likely to be essential , as the overexpression of Timp encoding a protein inhibitor of Mmp2 suppresses GSC increase , as well as follicle rupture . Mmp2 in mature follicle cells cleaves and downregulates Viking/collagen VI ( Deady et al . , 2017; Wang et al . , 2008 ) . In fact , several previous studies have revealed that Viking/collagen VI is required for GSC maintenance in female D . melanogaster ( Van De Bor et al . , 2015; Wang et al . , 2008 ) . However , we observed no significant change in Viking/Collagen VI levels in the germarium between the control and Mmp2 RNAi flies ( Figure 4—figure supplement 1E ) . Therefore , we concluded that Viking/collagen VI is not a substrate of Mmp2 in the regulation of mating-induced GSC increase . Besides Viking/Collagen VI , Dally-like ( Dlp ) is another basement membrane protein associated with extracellular matrix and known as the Mmp2 substrate ( Wang and Page-McCaw , 2014 ) . Interestingly , dlp is expressed in the escort cells ( Wang and Page-McCaw , 2014 ) . Moreover , Dlp controls the distribution of Dpp and Wnts , both of which significantly affect GSC self-renewal and differentiation ( Wang et al . , 2015; Xie and Spradling , 1998 ) . Future research should decipher the exact substrate by which Mmp2 controls Dpp and/or Wnts to modulate GSC behavior in response to mating stimulus . Another remaining question to be addressed is how Mmp2 function is regulated in GSC increase . Ecdysteroid biosynthesis and signaling in the ovary are necessary but not sufficient for the OA-Oamb-Ca2+–mediated GSC increase and follicle rupture ( Ameku and Niwa , 2016; Knapp and Sun , 2017 ) . We found that in the regulation of mating-induced GSC increase , ecdysteroid signaling acts downstream of Ca2+ signaling ( Figure 4F ) . On the other hand , in the follicle rupture process , ecdysteroid signaling either acts downstream , upstream , or both , of Ca2+ signaling . Further , the precise action of ecdysteroid has yet to be elucidated ( Knapp and Sun , 2017 ) . The Mmp2-GFP fusion protein level in the follicle cells is not changed in the loss-of-Ecdysone receptor-function flies , implying that ecdysteroid signaling might regulate Mmp2 enzymatic activity by an unknown mechanism ( Knapp and Sun , 2017 ) . Considering the involvement of both the OA-Oamb-Ca2+-Mmp2 axis and ecdysteroid biosynthesis , it is very likely that the Mmp2 enzymatic activity is also regulated by the same , unknown mechanism not only in the mature follicle cells to control follicle rupture , but also in the germarium to control mating-induced GSC increase . In many animals , reproduction involves significant behavioral and physiological shifts in response to mating . In female D . melanogaster , several post-mating responses are coordinated by SPSNs and their downstream afferent neuronal circuit ( Wang et al . , 2020 ) , including Stato-Acoustic Ganglion neurons , the ventral abdominal lateral Myoinhibitory peptide neurons , and the efferent dsx+ Tdc2+ neurons ( Feng et al . , 2014; Häsemeyer et al . , 2009; Jang et al . , 2017; Rezával et al . , 2014 ) . Our GRASP analysis indicates a direct synaptic connection between cholinergic SPSNs and OAergic neurons . Moreover , we demonstrated that nAChRs in dsx+ Tdc2+ neurons are responsible for the suppression of their neuronal activity in virgin females . However , nAChRs are the cation channels leading to depolarization upon acetylcholine binding , and therefore usually activate neurons ( Corringer et al . , 2000; Lee and O’Dowd , 1999; Perry et al . , 2012 ) . How is the opposite role of nAChRs in dsx+ Tdc2+ neuronal activity achieved ? One possibility is that acetylcholine-nAChR signaling does not evoke a simple depolarization but rather generates a virgin-specific temporal spike pattern in dsx+ Tdc2+ neurons . Interestingly , recent studies demonstrated that the pattern , instead of the frequency , of neuronal firing is significant in adjusting the neuronal activity of clock neurons in D . melanogaster ( Tabuchi et al . , 2018 ) . The firing pattern relies on control of ionic flux by the modulation of Ca2+-activated potassium channel and Na+/K+ ATPase activity . Because whether mating changes the firing pattern of dsx+ Tdc2+ neurons remains to be examined , the neuronal activity in SPSNs and the dsx+ Tdc2+ neuronal circuit between virgin and mated females are future research areas . In the last decades , there is growing evidence that GSCs and their niche are influenced by multiple humoral factors ( Drummond-Barbosa , 2019; Yoshinari et al . , 2019 ) . Based on the data from our current study and previous studies , there are at least four crucial humoral factors for regulating the increase and/or maintenance of D . melanogaster female GSCs , including DILPs ( Hsu et al . , 2008; Hsu and Drummond-Barbosa , 2009; LaFever , 2005 ) , ecdysteroids ( Ables and Drummond-Barbosa , 2010; Ameku et al . , 2017; Ameku and Niwa , 2016; König et al . , 2011 ) , Neuropeptide F ( NPF ) ( Ameku et al . , 2018 ) , and OA ( this study ) . Notably , all of these come from different sources: DILPs are from the insulin-producing cells located in the pars intercerebralis of the central brain; ecdysteroids from the ovary; NPF from the midgut; and OA from the neurons located in the abdominal ganglion . In addition to these identified humoral factors , recent studies also imply that adiponectin and unknown adipocyte-derived factor ( s ) are essential for GSC maintenance ( Armstrong and Drummond-Barbosa , 2018; Laws et al . , 2015; Matsuoka et al . , 2017 ) . These data clearly indicate that D . melanogaster female GSCs are systemically regulated by interorgan communication involving multiple organs . The additional interorgan communication mechanisms that ensure the faithful coupling of the increase and maintenance of GSC to the organism’s external and physiological environments are essential to be investigated in future studies . To modulate the increase and maintenance of GSC , ecdysteroids are received by both GSCs and niche cells ( Ables and Drummond-Barbosa , 2010; König et al . , 2011 ) , whereas DILPs , NPF , and OA are received by niche cells . A major signal transduction mechanism of each of these humoral factors have been well characterized , namely phosphoinositide 3-kinase pathway for DILPs-InR signaling , EcR/Ultraspiracle-mediated pathway for ecdysteroid signaling , cAMP pathway for NPF-NPFR signaling ( Garczynski et al . , 2002 ) , and Ca2+ pathway for OA-Oamb signaling . However , it remains unclear whether and how each of these signaling pathways control the production and secretion of the niche signal , as well as its distribution and transduction . In addition , it is important to understand whether and how the multiple system signals are integrated to control the mating-induced increase and maintenance of GSCs . In recent years , many studies have revealed that not only local niche signals but also systemic and neuronal factors play indispensable roles in regulating GSC behavior ( Ables and Drummond-Barbosa , 2017; Drummond-Barbosa , 2019; Yoshinari et al . , 2019 ) . In D . melanogaster , ecdysteroid signaling is essential for the proliferation and maintenance of GSCs and neural stem cells ( Ables and Drummond-Barbosa , 2010; Homem et al . , 2014; König et al . , 2011 ) . In this study , we have identified the ovary-projecting OAergic neurons as new regulators of stem cell homeostasis . Both steroid hormones and OA-like monoamines , such as noradrenaline , are also involved in stem cell regulation in mammals . For example , the mammalian steroid hormone , estrogen , is important in regulating cell division and/or maintenance of hematopoietic stem cells , mammary stem cell , neural stem cells , and hematopoietic stem cells ( Asselin-Labat et al . , 2010; Bramble et al . , 2019; Kim et al . , 2016; Nakada et al . , 2014 ) . Moreover , noradrenergic neurons , which directly project to the bone marrow , regulate the remodeling of hematopoietic stem cells niche ( Ho et al . , 2019; Méndez-Ferrer et al . , 2010; Méndez-Ferrer et al . , 2008 ) . Therefore , the steroid hormone- and noradrenergic nerve-dependent control of stem cell homeostasis are likely conserved across animal species . In this regard , the D . melanogaster reproductive system will further serve as a powerful model to unravel the conserved systemic and neuronal regulatory mechanisms for stem cell homeostasis in animals . Flies were raised on cornmeal-yeast-agar medium at 25°C . EcRA483T , temperature-sensitive mutants , were cultured at 31°C for 1 d prior to the assays . w1118 was used as the control strain . The genetic mutant stocks used were EcRA483T ( Bloomington Drosophila Stock Center [BDSC] #5799 ) and EcRM554fs ( BDSC #4894 ) . The protein-trap GFP line of Vkg ( Vkg::GFP ) was obtained from Kyoto Stock Center ( DGRC #110692 ) . Dad-LacZ ( Tsuneizumi et al . , 1997 ) ( a gift from Yoshiki Hayashi , University of Tsukuba , Japan ) . The following GAL4 and LexA strains were used: c587-GAL4 ( Manseau et al . , 1997 ) ( gift from Hiroko Sano , Kurume University , Japan ) , R44E10-GAL4 ( Deady and Sun , 2015 ) ( a gift from Jianjun Sun , University of Connecticut , USA ) , RS-GAL4 ( Lee et al . , 2009 ) ( a gift from Kyung-An Han , Pennsylvania State University , USA ) , nSyb-GAL4 ( BDSC #51941 ) , nSyb-GAL80 ( Harris et al . , 2015 ) ( a gift from James W . Truman , Janelia Research Campus , USA ) , tj-GAL4 ( DGRC #104055 ) , R13C06-GAL4 ( BDSC #47860 ) , 109–30 GAL4 ( BDSC #7023 ) , c355-GAL4 ( BDSC #3750 ) , c306-GAL4 ( BDSC #3743 ) , slbo-GAL4 ( BDSC #6458 ) , bab1-GAL4 ( Bolívar et al . , 2006 ) ( a gift from Satoru Kobayashi , University of Tsukuba , Japan ) , nos-GAL4 ( DGRC #107748 ) , tub >FRT >GAL80>FRT ( BDSC #38879 ) , OambKI-RD-GAL4 ( BDSC#84677 ) ( Deng et al . , 2019 ) , Oamb-KI-T2A-GAL4 , nAChRα1-T2A-GAL4 , nAChRα2-T2A-GAL4 , nAChRα3-T2A-GAL4 , nAChRβ1-T2A-GAL4 , nAChRβ2-T2A-GAL4 ( Kondo et al . , 2020; Ihara et al . , 2020 ) , ChaT-GAL4 ( BDSC #6793 ) , ppk-GAL4 ( Grueber et al . , 2007 ) ( a gift from Hiroko Sano , Kurume University , Japan ) , and SPSNs-LexA ( Feng et al . , 2014 ) ( a gift from Young-Joon Kim , Gwangju Institute of Science and Technology , South Korea ) . The following UAS and LexAop strains were used: 20xUAS-6xGFP ( BDSC #52261 ) , UAS-GFP;UAS-mCD8::GFP ( Ito et al . , 1997; Lee and Luo , 1999 ) ( a gift from Kei Ito , University of Cologne , Germany ) , UAS-Stingar ( BDSC #84277 ) , UAS-mCD8::RFP ( BDSC #32219 ) , UAS-CsChrimson ( BDSC #55134 ) , UAS-Insp3R ( BDSC #30742 ) , UAS-OambAS ( Lee et al . , 2009 ) ( a gift from Kyung-An Han , Pennsylvania State University , USA ) , UAS-Timp ( BDSC #58708 ) ( a gift from Andrea Page-McCaw , Vanderbilt University , USA ) , UAS > stop >dTrpA1mcherry , UAS > stop >TNT , UAS > stop >TNTin ( von Philipsborn et al . , 2011; Yu et al . , 2010 ) , dsx-FLP ( Rezával et al . , 2014 ) ( a gift from Daisuke Yamamoto , Advanced ICT Research Institute , National Institute of Information and Communications Technology , Japan ) TRiC; UAS-mCD8::RFP , LexAop2-mCD8::GFP;nSyb-MKII::nlsLexADBDo;UAS-p65AD::CaM ( BDSC:61679 ) , ppk-eGFP ( Grueber et al . , 2003 ) ( a gift from Tadashi Uemura , Kyoto University , Japan ) , and LexAop-Kir2 . 1 ( Feng et al . , 2014 ) ( a gift from Young-Joon Kim , Gwangju Institute of Science and Technology , South Korea ) . The RNAi transgenic lines used were as follows: UAS-LacZRNAi ( a gift from Masayuki Miura , The University of Tokyo , Japan ) , UAS-OambRNAi1 ( BDSC #31171 ) , UAS-OambRNAi2 ( BDSC #31233 ) , UAS-OambRNAi3 ( Vienna Drosophila Resource Center [VDRC] #106511 ) , UAS-Octβ1RRNAi ( VDRC #110537 ) , UAS-Octβ2RRNAi ( VDRC #104524 ) , UAS-Octβ3RRNAi ( VDRC #101189 ) , UAS-Insp3RRNAi ( BDSC #25937 ) , UAS-EcRRNAi ( VDRC #37059 ) , UAS-Mmp2RNAi1 ( BDSC #31371 ) , UAS-Mmp2RNAi2 ( VDRC #330303 ) , UAS-TimpRNAi1 ( BDSC #61294 ) , UAS-TimpRNAi2 ( VDRC #109427 ) , UAS-Tdc2RNAi1 ( VDRC #330541 ) , UAS-Tdc2RNAi2 ( BDSC #25871 ) , UAS-TβhRNAi1 ( VDRC #107070 ) , UAS-TβhRNAi2 ( BDSC #67968 ) , UAS-ChATRNAi1 ( VDRC #330291 ) , UAS-ChATRNAi2 ( BDSC #25856 ) , UAS-nAChRα1RNAi ( VDRC #48159 ) , UAS-nAChRα2RNAi ( VDRC #101760 ) , UAS-nAChRα3RNAi ( VDRC #101806 ) , UAS-nAChRβ1RNAi ( VDRC #106570 ) , UAS-nAChRβ2RNAi ( VDRC #109450 ) , UAS-nvdRNAi1 , and UAS-nvdRNAi2 ( Yoshiyama et al . , 2006 ) . The mutant alleles OambΔ ( Figure 1—figure supplement 1F ) , nAChRα1228 , and nAChRα1326 ( Figure 6—figure supplement 2A ) were created in a white ( w ) background using CRISPR/Cas9 as previously described ( Kondo and Ueda , 2013 ) . The following guide RNA ( gRNA ) sequences were used: Oamb , 5ʹ-GATGAACTCGAGTACGGCCA-3ʹ , and 5ʹ-GCGATCTCTGGTGCCGCATT-3ʹ; nAChRα1228 , 5ʹ-GGACATCATGCGTGTGCCGG-3ʹ; nAChRα1326 , 5ʹ-GGGCAGGTAGAAGACCAGAA-3ʹ . The breakpoint detail of OambΔ is described in Figure 1—figure supplement 1F , whereas those of nAChRα1228 and nAChRα1326 are described in Figure 6—figure supplement 2A . The pcDNA3 . 1 plasmid containing the wild-type D . melanogaster nAChRα1 coding sequences ( nAChRα1-pcDNA3 . 1 ) was synthesized previously described ( Ihara et al . , 2018 ) . Briefly , nAChRα1-pcDNA3 . 1 was digested with EcoRI and NotI , and then the digested nAChRα1 fragment was ligated with a EcoRI-NotI–digested pWALIUM10-moe plasmid ( Perkins et al . , 2015 ) . Transformants were generated using the phiC31 integrase system in the P{CaryP}attP40 strain ( Groth et al . , 2004 ) . The w+ transformants of pWALIUM10-moe were established using standard protocols . Flies were reared at 25°C and aged for 5–6 d . Virgin female flies were mated overnight to w1118 male flies at 25°C ( 10 males and 5–8 females per vial ) . For the thermal activation assays , flies were first reared at 17°C for 6 d and transferred to 29°C . In the case of EcR mutant assays , flies were transferred to 31°C for 24 hr before mating or ex vivo culture . For OA feeding , newly eclosed virgin females were aged for 4 d in vials with standard food containing 7 . 5 mg/mL of OA ( Monastirioti et al . , 1996; Rubinstein and Wolfner , 2013 ) . Tissues were dissected in phosphor buffer serine ( PBS ) and fixed in 4% paraformaldehyde in PBS for 30 to 60 min at room temperature ( RT ) . The fixed samples were washed three times in PBS supplemented with 0 . 2% Triton X-100 , blocked in blocking solution ( PBS with 0 . 3% Triton X-100% and 0 . 2% bovine serum albumin [BSA] ) for 1 hr at RT , and incubated with a primary antibody in the blocking solution at 4°C overnight . The primary antibodies used were chicken anti-GFP ( Abcam #ab13970; 1:4 , 000 ) , rabbit anti-RFP ( Medical and Biological Laboratories PM005; 1:2 , 000 ) , mouse anti-Hts 1B1 ( Developmental Studies Hybridoma Bank [DSHB]; 1:50 ) , rat anti-DE-cadherin DCAD2 ( DSHB; 1:50 ) , rabbit anti-pH3 ( Merck Millipore #06–570; 1:1000 ) , rabbit monoclonal anti-pMad ( Abcam #ab52903; 1:1000 ) , mouse anti-Lamin C LC28 . 26 ( DSHB; 1:10 ) , rabbit cleaved Dcp-1 ( Cell Signaling Technology #9578; 1:100 ) , rat anti-Vasa ( DSHB; 1:50 ) , mouse anti-LacZ ( β-galactosidase ) ( DSHB#40-1a; 1:50 ) , rabbit anti-Tdc2 ( Abcam #ab128225; 1:2000 ) , Alexa Fluor 546 phalloidin ( Thermo Fisher Scientific #A22283; 1:200 ) , and Alexa Fluor 633 phalloidin ( Thermo Fisher Scientific #A22284; 1:200 ) . After washing , fluorophore ( Alexa Fluor 488 , 546 or 633 ) -conjugated secondary antibodies ( Thermo Fisher Scientific ) were used at a 1:200 dilution , and the samples were incubated for 2 hr at RT in the blocking solution . After another washing step , all samples were mounted in FluorSave reagent ( Merck Millipore #345789 ) . GSC numbers were determined based on the morphology and position of their anteriorly anchored spherical spectrosome ( Ables and Drummond-Barbosa , 2010; Ameku et al . , 2018; Ameku and Niwa , 2016 ) . Cap cells were identified by immunostaining with anti-Lamin C antibody as previously described ( Ables and Drummond-Barbosa , 2010 ) . We used 5–6-day-old females . The ovaries were dissected in Schneider’s Drosophila medium ( Thermo Fisher Scientific #21720024 ) and isolated from oviduct using forceps . Approximately 5–6 ovaries were immediately transferred to a dish containing 3 mL of Schneider’s Drosophila medium supplemented with 15% fetal calf serum and 0 . 6% penicillin-streptomycin with/without the addition of OA ( Sigma , final concentration of OA is 0–1000 μM ) and 20E ( Enzo Life Sciences; final concentration of 20 nM ) . The cultures were incubated at RT ( except for EcR mutant flies , Figure 3D ) for 16 hr , and the samples were immunostained to determine the GSC number . We employed the previously described imaging methods to visualize GSC behavior ( Morris and Spradling , 2011; Reilein et al . , 2018 ) . For the live imaging , the ovaries dissected from adult virgin female flies were placed on a glass bottom dish ( IWAKI #4970–041 ) with 3 mL of Schneider’s Drosophila medium and 100 µL of the test reagent ( Schneider’s Drosophila medium containing 300 mM OA ) placed directly at the center of each dish . The images were obtained with a × 40 objective lens ( water-immersion ) using a Zeiss LSM 700 confocal microscope and were recorded every 4 s . The GCaMP6s fluorescence intensity in the escort cell was then calculated for each time point . The ratio of fluorescence ( ΔF ) at each time point was calculated by normalizing the fluorescence with the initial fluorescence ( F0 ) . The initial fluorescence ( F0 ) is the average GCaMP6s fluorescence intensity before adding the test reagent . Red-shifted channelrhodopsin CsChrimson ( Klapoetke et al . , 2014 ) was used to increase the [Ca2+]i in the escort cells by light . UAS-CsChrimson was expressed using c587-GAL4 with nSyb-GAL80 . All crosses and the early development of flies were performed under dark conditions . The experiment was done at 25°C . Adult flies were raised with standard food for 3 d after eclosion and then with standard food with 1 mM all-trans-retinal ( ATR ) for 3 d . Subsequently , they were kept in the presence of orange–red light from LED for 24 hr . LED light was shone from the outside of the plastic chamber covered by aluminum foil to enhance light intensity . All experiments were performed independently at least twice . Fluorescence intensity in confocal sections was measured via ImageJ . For pMad quantification , signal intensity was calculated by measuring the fluorescence intensity in GSCs and CBs , which were co-stained with anti-Vasa antibody to visualize their cell boundaries . Sample sizes were chosen based on the number of independent experiments required for statistical significance and technical feasibility . The experiments were not randomized , and the investigators were not blinded . All statistical analyses were carried out using the ‘R’ software environment . The P value is provided in comparison with the control and indicated as * for p≤0 . 05 , for p≤0 . 01 , * for p≤0 . 001 , and ‘NS’ for non-significant ( p>0 . 05 ) .	Stem cells have the unique ability to mature into the various , specialized groups of cells required for organisms to work properly . Local signals released by the tissues immediately surrounding stem cells usually trigger this specialization process . However , recent studies have revealed that external signals , such as hormones or neurotransmitters ( the chemicals used by nerve cells to communicate ) , can also control the fate of stem cells . This is particularly the case during development , or in response to events such as injury . In the right conditions , germline stem cells can specialize into the egg or sperm required for many animals to reproduce . In fruit flies for example , the semen contains proteins that activate a cascade of molecular events in the female nervous system , ultimately resulting in female germline stem cells multiplying in the ovaries after mating . Yet , exactly how this process takes place was still unclear . To investigate this question , Yoshinari et al . focused on nerve cells in the fruit fly ovary which produce a neurotransmitter called octopamine . The experiments assessed changes in the ovaries of female fruit flies after mating , piecing together the sequence of events that activate germline stem cells . This showed that first , mating triggers the release of octopamine from the nerve cells . In turn , this activates a protein called Oamb , which is studded through the membrane of cells present around germline stem cells . Turning on Oamb prompts a cascade of molecular events which include an enzyme called Matrix metalloproteinase 2 regulating the signal sent from the local environment to germline stem cells . As mammals use a neurotransmitter similar to octopamine , future fruit fly studies could shed light on how neurotransmitters activate stem cells in other animals . Ultimately , unravelling the way external signals trigger the specialization process may offer insight into how diseases arise from uncontrolled stem cell activity .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2020	Neuronal octopamine signaling regulates mating-induced germline stem cell increase in female Drosophila melanogaster
Kinetochore fibers ( K-fibers ) of the mitotic spindle are force-generating units that power chromosome movement during mitosis . K-fibers are composed of many microtubules that are held together throughout their length . Here , we show , using 3D electron microscopy , that K-fiber microtubules ( MTs ) are connected by a network of MT connectors . We term this network ‘the mesh’ . The K-fiber mesh is made of linked multipolar connectors . Each connector has up to four struts , so that a single connector can link up to four MTs . Molecular manipulation of the mesh by overexpression of TACC3 causes disorganization of the K-fiber MTs . Optimal stabilization of K-fibers by the mesh is required for normal progression through mitosis . We propose that the mesh stabilizes K-fibers by pulling MTs together and thereby maintaining the integrity of the fiber . Our work thus identifies the K-fiber meshwork of linked multipolar connectors as a key integrator and determinant of K-fiber structure and function . Accurate mitosis is essential to eukaryotic life . It requires the correct assembly of a bipolar array of microtubules ( MTs ) into a mitotic spindle which , in concert with hundreds of different motors and non-motor proteins , segregates the duplicated sister chromatids to the two daughter cells . Many of the chromosome movements in mitosis are governed by the kinetochore fibers ( K-fibers ) of the spindle apparatus . In human cells , K-fibers are bundles of 20–40 parallel MTs that typically run from the kinetochore to the spindle pole ( McDonald et al . , 1992; Mastronarde et al . , 1993; McEwen et al . , 1997; Booth et al . , 2011; Sikirzhytski et al . , 2014 ) . K-fibers can be thought of as coherent units: their constituent MTs are held together throughout their length , as well as being focused at either end ( Rieder , 1981; Spurck et al . , 1997 ) . The coherence of the K-fiber is thought to be crucial for accurate chromosome segregation in mitosis . However , the ultrastructural and molecular basis of K-fiber coherence is not well understood . Lateral MT connectors are important for the function of MT arrays , such as K-fibers , in cells ( Brangwynne et al . , 2006; Ward et al . , 2014 ) . Classic electron microscopy ( EM ) studies uncovered the presence of electron density between MTs of the K-fibers ( Wilson , 1969; Hepler et al . , 1970; Witt et al . , 1981; Bastmeyer and Fuge , 1986 ) . These inter-MT bridges appear as bipolar struts laterally connecting two MTs in 2D electron micrographs . The bridges are typically ∼5 nm thick and range from 6–20 nm in length . The morphology of bridges is heterogeneous , and they are likely composed of a variety of proteins ( Booth et al . , 2011 ) . The mitotic spindle is an ensemble of hundreds of MT-associated proteins ( Sauer et al . , 2005 ) , some of which are candidates for inter-MT bridges ( Manning and Compton , 2008 ) . The number of inter-MT bridges scales with the number of MTs and also with the number of paired MTs in a bundle ( Bastmeyer and Fuge , 1986 ) , and it was proposed that it is the inter-MT bridges that hold K-fiber MTs together ( Witt et al . , 1981 ) . The connection between the kinetochore and K-fiber MTs has been studied by 3D-EM ( Dong et al . , 2007; McIntosh et al . , 2008 ) . However , a similar 3D investigation of inter-MT connections away from the kinetochore has not been reported . One molecular candidate for inter-MT bridges in K-fibers is TACC3–ch-TOG–clathrin . The assembly of this complex is regulated by phosphorylation of TACC3 at serine 558 by Aurora-A kinase ( Fu et al . , 2010; Hubner et al . , 2010; Lin et al . , 2010; Booth et al . , 2011 ) . This allows two domains from TACC3 and clathrin to come together in space , making a single , composite MT interaction surface ( Booth et al . , 2011; Cheeseman et al . , 2011; Hood et al . , 2013 ) . Previous work showed that this complex is necessary for K-fiber stabilization ( Royle et al . , 2005; Booth et al . , 2011; Cheeseman et al . , 2013 ) and that the complex forms a distinct class of short inter-MT bridges ( Booth et al . , 2011 ) . In this study , we set out to examine the 3D ultrastructure of K-fiber inter-MT bridges . Surprisingly , we found that inter-MT bridges are not simply bipolar connections between two MTs but are a network of interconnected struts that contact multiple MTs . We term this structure ‘the mesh’ . This novel subcellular structure positions MTs within the K-fiber and is required for normal mitosis . Inter-MT bridges were first described in 2D electron micrographs as faintly stained fine threads projecting from the surface of the tubules ( Hepler et al . , 1970 ) . For example , 2D views of inter-MT bridges connecting adjacent MTs within a K-fiber are shown in Figure 1A . The morphology of inter-MT bridges is heterogeneous ( Booth et al . , 2011 ) , as noted in the original report ( Hepler et al . , 1970 ) . In order to examine the morphology of bridges in more detail , we used electron tomography of sections taken from mitotic HeLa cells fixed by high-pressure freezing/freeze substitution ( HPF/FS ) . This gave a 3D view of the K-fiber MTs and the material that connected them ( Figure 1B , Video 1 ) . Careful examination of the tomograms revealed that inter-MT bridges are not just simple bipolar struts connecting two adjacent MTs: they are interconnected and can contact multiple MTs within the K-fiber . This network of inter-MT connectors is best illustrated by rendering the densities seen in the tomogram and visualizing the resultant computer model ( Figure 1B ) . We term the interconnecting material ‘the mesh’ . 10 . 7554/eLife . 07635 . 003Figure 1 . Intermicrotubule connectors in K-fibers are ‘bridges’ in 2D and a ‘mesh’ in 3D . ( A ) 2D views of inter-MT bridges in sections taken orthogonally or longitudinally to the spindle axis . In the annotated version , MTs ( green ) and inter-MT bridges ( yellow ) are shown on a blue background . Zoom of longitudinal section is a 2× expansion of the lower part of the 2D EM view . ( B ) Orthoslice of a tomogram generated from a tilt series of a single section through a K-fiber preserved using HPF/FS ( i ) . Overlaid is a hand-rendered 3D representation of MTs ( green ) ( ii ) and associated mesh ( yellow ) ( iii ) . The model is shown alone ( iv ) . For ( v–viii ) , the tomogram is tilted to show the MT axis . In v and vi , the overlay and model are shown . In vii and viii , the same view is shown but with the mesh detected by the automated segmentation method . Note that sections taken >1 µm away from the kinetochore are shown in this and all subsequent figures . Tomogram thickness , 45 . 6 nm . For scale , MTs are 25 nm in diameter . DOI: http://dx . doi . org/10 . 7554/eLife . 07635 . 00310 . 7554/eLife . 07635 . 004Figure 1—figure supplement 1 . The mesh is visible , but not well preserved , in chemically-fixed K-fibers . Orthoslice of a tomogram generated from a tilt series of a single section through a K-fiber fixed with glutaraldehyde ( i ) . Overlaid is a hand-rendered 3D representation of MTs ( green ) ( ii ) and mesh segmented by automated detection ( yellow ) ( iii ) . The model is shown alone ( iv ) . Tomogram thickness , 51 . 3 nm . For scale , MTs are 25 nm in diameter . DOI: http://dx . doi . org/10 . 7554/eLife . 07635 . 00410 . 7554/eLife . 07635 . 005Figure 1—figure supplement 2 . The mesh is associated with K-fiber MTs . ( A ) 3D rendering of ‘non-mesh’ ( brown ) in a tomogram of a single K-fiber . Single orthoslice ( i ) , with added rendering ( MTs , green; mesh , yellow ) ( ii ) and model alone ( iii ) , is shown . Rendering the mesh involves segmenting density that is attached to the MTs . Here , density to the upper right of the K-fiber was rendered , although it was not touching MTs . The result is a particulate 3D structure that is dissimilar to the K-fiber mesh . ( B ) Translation of MT map from the tomogram in A , to the upper right corner . Mesh detection was performed as described in ‘Materials and methods’ . Single orthoslice view ( i ) , with added rendering ( MTs , green; mesh , yellow ) ( ii ) . Zoomed views of two MTs and associated ‘pseudo-mesh’ are shown in ( iii and iv ) . Note that the mesh passes through the MTs and is more particulate than the K-fiber mesh . Tomogram thickness , 35 . 2 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 07635 . 00510 . 7554/eLife . 07635 . 006Video 1 . Example of the K-fiber mesh from a normal HeLa cell at metaphase . Tomogram of a K-fiber . The mesh ( yellow ) is shown by manual rendering and by automated rendering . MTs ( green ) were rendered by hand . All segmentation was smoothed in Amira . DOI: http://dx . doi . org/10 . 7554/eLife . 07635 . 006 To obtain an unbiased view of the mesh , we developed a semi-automated segmentation method for 3D model building ( Figure 1B , Video 1 ) . This method was used for all subsequent quantification . We found that mesh preservation was superior using HPF/FS compared with chemical fixation ( Figure 1—figure supplement 1 ) . Moreover , the mesh was associated with K-fiber MTs . In contrast , segmentation of non-K-fiber areas resulted in distinct , particulate density and translating MTs to a non-K-fiber area meant that globular density , which passed through the MTs , was detected ( Figure 1—figure supplement 2 ) . Having established a detection method , we next explored the anatomy of this novel subcellular structure . Three types of connectors within the mesh could be distinguished in these 3D models . Bipolar , tripolar , or quadrupolar connectors could be selected using the criterion of uninterrupted density that connects two , three , or four MTs , respectively ( Figure 2A ) . These connectors were highly heterogeneous in size . Where the volumes of individual connectors could be determined easily , we found that bipolar , tripolar , and quadrupolar connectors had mean volumes of 6 , 689 , 17 , 370 , and 32 , 376 nm3 , respectively ( Figure 2B ) . Therefore , there is not a linear relationship between the number of struts in a connector and the connector volume . The volume of mesh within each fiber was , on average , equivalent to the volume of MTs that it encapsulates and larger than the volume of MT walls alone ( Figure 2C ) . This observation establishes that the mesh is a major component of every K-fiber . 10 . 7554/eLife . 07635 . 007Figure 2 . Inter-MT linkages are defined connectors with heterogeneous MT-mesh contacts . ( A ) Examples of bipolar , tripolar , and quadrupolar connectors within the mesh . Single orthoslices from tomograms showing examples of different connectors . MTs are hand-rendered ( green ) , mesh is automatically detected ( yellow , see ‘Materials and methods’ ) , and example connectors are highlighted orange . Tomogram thickness , 51 . 3 nm ( bipolar ) and 35 . 2 nm ( tripolar and quadrupolar ) . ( B ) Box plots to show the volume of bipolar , tripolar , and quadrupolar connectors within the mesh from multiple tomograms . ( C ) Box plot to show the ratio of the volume of mesh relative to MTs ( walls only ) or MTs + lumens ( ‘filled-in’ MTs ) . Box plots show the median , 75th and 25th percentile , and whiskers show 90th and 10th percentile . ( D ) Heterogeneity of contacts between mesh and MTs . Two pairs of MTs are shown with a simple bipolar connector ( above ) or with a more complex connection ( below ) . MTs ( green ) are shown with a single component of the mesh ( orange ) . The contact points between the selected mesh and the MTs are shown in pink . Note the extensive mesh-MT contacts in the lower example . Tomogram thickness , 66 . 4 nm ( simple ) 64 . 8 nm ( network ) . For scale , MTs are 25 nm in diameter . DOI: http://dx . doi . org/10 . 7554/eLife . 07635 . 007 We next examined the contacts made between the mesh and MTs to determine if there were preferred locations on the MT for mesh attachment , for example , on the seams of MTs . We found that mesh-MT contacts were highly heterogeneous ( Figure 2D ) . For example , a crosslink between two MTs could be a simple bipolar strut with a small footprint on both connected MTs , or at the other extreme , the crosslink could be a network of multiple-linked struts , with large mesh contact areas on both MT surfaces . These larger , composite footprints extended for some distance along both linked MTs , but these were not confined to a co-axial line on the MT wall . Moreover , several mesh connectors could contact the same MT at the same axial position ( e . g . , Figure 1B ) , which further argues that the mesh has no preference for the MT seam . Previous work indicated that one component of the mesh is a complex of TACC3–ch-TOG–clathrin whose assembly is regulated by Aurora-A kinase ( Booth et al . , 2011; Hood et al . , 2013 ) . Moreover , the amount of this complex on K-fibers can be increased simply by overexpressing TACC3 ( Booth et al . , 2011 ) . In order to experimentally manipulate the mesh , we therefore made a stable inducible HeLa cell line where GFP-TACC3 could be overexpressed in a controlled manner ( Figure 3—figure supplement 1 ) . The most obvious effect of this manipulation was to alter MT organization . K-fibers in cells expressing GFP-TACC3 had more MTs per fiber , and the cross-sectional area that those MTs occupied was larger , compared to fibers in control uninduced HeLa cells ( Figure 3A ) . The MT density was similar between the two groups suggesting that the fiber area scaled with the number of MTs ( Figure 3A ) . 10 . 7554/eLife . 07635 . 008Figure 3 . Analysis of MT packing within a K-fiber . ( A ) Box plots showing the number of MTs per K-fiber , the cross-sectional K-fiber area , and the density of MTs in the K-fiber . Nfiber = 26 ( control ) , 37 ( TACC3 OE ) . Box plots show the median , 75th and 25th percentile , and whiskers show 90th and 10th percentile . p values from Welch's t-test are shown . ( B ) Spatial maps of MTs ( circles ) in a representative K-fiber from a control ( left ) or TACC3 overexpressing cell ( right ) color-coded to show ( i ) the distance to the nearest neighbor or ( ii ) overlaid on a heatmap to show the number of MTs within 105 nm ( center-to-center distance ) , 80 nm ( edge-to-edge distance ) . ( C ) Histogram to show the frequency of distances to the nearest neighboring MT ( i ) . A representation of the change in average ( median ) MT spacing to the nearest neighboring MT that is caused by TACC3 overexpression ( ii ) . NMTs = 500–1324 . ( D ) Analysis of tubulin intensity in the vicinity of kinetochores by confocal microscopy . Box plot to show the distribution of median values per cell of tubulin intensities in a sphere surrounding the kinetochore ( i ) . Box plots show the median , 75th and 25th percentile , and whiskers show 90th and 10th percentile . p values from one-way Anova with Tukey's post hoc test are shown . Ncell = 20–25 , Nkinetochore = 2823–3533 . Schematic diagram of the changes in K-fibers induced by TACC3 overexpression ( ii ) . K-fibers are thicker because MTs that are not stably attached to the kinetochore are recruited to the K-fiber bundle . DOI: http://dx . doi . org/10 . 7554/eLife . 07635 . 00810 . 7554/eLife . 07635 . 009Figure 3—figure supplement 1 . Inducible expression of GFP-TACC3 in HeLa cells to alter the composition of mesh . ( A ) Western blot to show the extent of overexpression of TACC3 caused by inducing the expression of GFP-TACC3 in a stable TetOn HeLa cell line using 0 . 5 μg/ml doxycycline for 24 hr . The blot was probed for TACC3 and tubulin ( as a loading control ) . ( B ) Widefield fluorescence micrographs of cells inducibly expressing GFP-TACC3 . Note the strong fluorescence on K-fibers of the mitotic spindle . ( C ) Workflow to show the preparation of metaphase cells for analysis by 3D EM . Cells were synchronized and GFP-TACC3 expression was induced before release from RO3306 for 30–40 min . Mitotic cells were shaken off , pelleted , resuspended , and frozen under high pressure . Following freeze substitution and embedding , cells were trimmed and sectioned before imaging by electron microscopy . For details , see ‘Materials and methods’ . DOI: http://dx . doi . org/10 . 7554/eLife . 07635 . 009 To look in more detail at MT packing within a K-fiber , we analyzed the distance from each MT to its nearest neighbor , and the number of neighboring MTs found within 80 nm . Both analyses showed that local MT packing density within the fiber had increased substantially in TACC3 overexpressing cells compared to uninduced controls ( Figure 3B ) , although the fibers themselves were larger overall . A simple manifestation of this tighter local packing was the increased frequency of doublet and triplet MTs within TACC3 overexpressing K-fibers ( Figure 3B ) . The median distance to the nearest neighboring MT had decreased from 56 . 1 to 48 . 1 nm ( Figure 3C ) , a change in edge-to-edge proximity from 31 . 1 to 23 . 1 nm . This means that in TACC3 overexpressing cells , the average nearest neighboring MT is less than the width of one MT away . Because the overexpression of TACC3 alters the MT packing within the K-fiber , these experiments suggested to us that the mesh might influence MT spacing within the K-fiber . We return to the hypothesis that the mesh has an important role in MT spacing below . Are the additional MTs in TACC3 overexpressing K-fibers stably attached to the kinetochore ? To address this question , we used a 3D confocal microscopy assay of tubulin staining in the vicinity of kinetochores ( Cheeseman et al . , 2013 ) . In agreement with the EM analysis , we detected a higher tubulin signal in cells expressing GFP-TACC3 compared to those expressing GFP alone . Following cold treatment , the tubulin intensity in the vicinity of kinetochores was reduced to comparable levels in both conditions , suggesting that the additional MTs in TACC3 overexpressing K-fibers are attached by mesh to the rest of the K-fiber but were not stably attached to the kinetochore ( Figure 3D ) . Overexpression of TACC3 also increased the volume of mesh between K-fiber MTs in a tomogram to 5 . 2 ± 1 . 0 × 106 nm3 ( mean ± s . e . m . ) . This corresponds to 9 . 1 ± 0 . 01% of the fiber volume in the tomogram , whereas control mesh was 5 . 7 ± 0 . 01% . This change is somewhat difficult to interpret because of the significant increase in the number of MTs per fiber and the tighter local packing . More MTs per fiber might push up the volume of mesh , but the closer proximity of MTs limits the space available for mesh to be present . One defining characteristic of the mesh is that it connects multiple MTs within K-fibers . This interconnectivity means that a MT that is contacted by the mesh is connected to one or more MTs and each of these , in turn , may be connected to one or more MTs and so on . We defined these connected MTs as ‘chains’ . In uninduced cells , chain sizes were small , containing at most 6 MTs ( Figure 4A ) . By contrast , cells overexpressing TACC3 had chains containing up to 12 MTs ( Figure 4A ) , suggesting that the MTs were more interconnected as a result of TACC3 overexpression . Although MTs in TACC3 overexpressing K-fibers were more interconnected , the constitution of the connectors within the mesh was not noticeably altered ( Figure 4B ) . In both conditions , the mesh was composed of a predominance of bipolar connectors and similar proportions of tripolar and quadrupolar connectors ( Figure 4B ) . 10 . 7554/eLife . 07635 . 010Figure 4 . Analysis of MTs captured by mesh , their connectivity , and proximity . ( A ) Histogram to show frequency of MT chain sizes detected in single section tomograms . MT chains are collections of MTs that are connected by mesh within a section . Note that single MTs do not feature . ( B ) Bar chart to show the proportion of mesh that is bipolar , tripolar , and tetrapolar , as a percentage of total mesh connections . ( C ) Analysis of MT connectivity and proximity . ( i ) Spatial maps of MTs ( circles ) in a representative K-fiber from a control ( left ) or TACC3 overexpressing cell ( right ) . Chains of MTs are shown in color , gray circles show MTs that are not detectably connected to other MTs by mesh . Orange lines represent the mesh connections schematically ( see Figure 4—figure supplement 1 ) . ( ii ) Spatial maps are shown overlaid on heatmaps for the same fibers to show the number of MTs within 105 nm ( center-to-center distance ) , 80 nm ( edge-to-edge distance ) . Note that single ( unchained ) MTs are attached to the mesh , but that the MT to which they are attached is outside of the section . Tomogram thickness , 43 . 6 nm ( Control ) , 28 . 8 nm ( TACC3 OE ) . ( D ) Histogram to show frequency of MTs with a given number of neighboring MTs within 100 nm ( center-to-center distance ) . MTs connected to others by mesh are shown in color , and single MTs are shown in white . Inset shows the p-value for comparisons calculated using search radii from 20 nm to 120 nm ( center-to-center distance ) . Dotted lines show the same analysis following randomization of the chain membership data . DOI: http://dx . doi . org/10 . 7554/eLife . 07635 . 01010 . 7554/eLife . 07635 . 011Figure 4—figure supplement 1 . Examples of MTs chains and associated mesh . Two example tomograms , which were annotated by 3D rendering and used to make the 2D maps illustrating MT position and chain membership in Figure 4C . MTs are colored as in Figure 4C , mesh associated with chained MTs is orange , other mesh is shown in yellow . Single orthoslice from the tomogram with rendered material is shown above , the MTs and mesh associated with chained MTs is shown below . Note that single ( unchained ) MTs are attached to the mesh , but that the MT/MTs to which they are attached are outside of the section . Tomogram thickness , 43 . 6 nm ( Control ) , 28 . 8 nm ( TACC3 OE ) . For scale , MTs are 25 nm in diameter . DOI: http://dx . doi . org/10 . 7554/eLife . 07635 . 011 We next wondered if the larger chains in TACC3 overexpressing cells were the result of the tighter local MT packing . Accordingly , we constructed 2D MT maps , where each MT's chain membership was displayed , and we compared these to the heat maps of local packing as previously described ( Figure 4C ) . The entire dataset was analyzed computationally in order to test the idea that chain membership depended on MT proximity . For each MT , we calculated the number of neighboring MTs within a given search radius and compared these values for single ( ‘unchained’ ) MTs and those that were part of a chain ( Figure 4D ) . This analysis showed that in controls , the single MTs and chained MTs had similar numbers of neighbors , that is , meshed MTs were not more likely to be in tightly packed regions of the fiber . However , in the TACC3 overexpressing fibers , the MTs that were part of a chain had significantly more neighbors than single MTs ( Figure 4D ) , this pattern was seen for search radii >50 nm and was not seen if the dataset was randomized ( Figure 4D , inset , see ‘Materials and methods’ ) . The closer proximity of chained MTs vs single MTs indicates that the mesh influences MT positioning , effectively pulling them closer together . An alternative possibility is that the mesh is passive and only encapsulates MTs if they are in close proximity . However , this possibility is less likely because MTs in control fibers exhibit no difference in their proximity relative to their chain status . The two possibilities for mesh stabilization of K-fibers are shown in ( Figure 5A ) . To test the possibility that the mesh can influence MT positioning in K-fibers , we turned to an in vitro assay . Fluorescently labeled MTs assembled in vitro and stabilized with taxol were incubated with proteins , and any effect on MT bundling was observed by light or EM . To reconstitute the mesh component containing TACC3 , a protein mixture comprising clathrin , TACC3 , ch-TOG , and GTSE1 was prepared and phosphorylated by Aurora-A kinase ( see ‘Materials and methods’ ) . As a positive control , we used PRC1 , which is known to bundle MTs ( Mollinari et al . , 2002 ) . As a negative control , we used an equivalent amount of GST and MBP-His6 protein as in the test condition , phosphorylated by Aurora-A kinase . Thick bundles of MTs could be seen by fluorescence microscopy for the TACC3-containing complex and for PRC1 but not for the negative control ( Figure 5B ) . These images indicated specific bundling activity for the TACC3 complex . 10 . 7554/eLife . 07635 . 012Figure 5 . Bundling of MTs in vitro by mesh components . ( A ) Cartoon to illustrate two potential models for mesh function . In the passive model ( left ) , MT distances are set by some other factor , and the mesh fills in the gaps to connect MTs . In the influential model ( right ) , the mesh bundles MTs by favoring close spacing , this in turn influences MT spacing . ( B ) Representative fluorescence micrographs of rhodamine-labeled MTs incubated with the indicated proteins . Control ( 210 nM MBP-His6 , 100 nM GST ) , PRC1 ( 2 µM His6-PRC1 ) , and Mesh complex ( 10 nM MBP-ch-TOG-His6 , 100 nM clathrin , 100 nM GST-TACC3-His6 , 100 nM MBP-GTSE1-His6 ) , all phosphorylated by TPX2 ( 1–43 ) /Aurora-A . Scale bar , 10 µm . ( C ) Representative electron micrographs of MTs incubated with the indicated proteins as described in A . Pelleted material was chemically fixed , embedded in resin , sectioned , and imaged . Scale bar , 50 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 07635 . 01210 . 7554/eLife . 07635 . 013Figure 5—figure supplement 1 . MT bundling using purified components . Representative fluorescence micrographs of rhodamine-labeled MTs incubated with the indicated proteins . In all experiments , control was 210 nM MBP-His6 and 100 nM GST , and the total amount of protein was supplemented with MBP-His6 and GST to reach equimolarity . All samples were phosphorylated with Aurora A kinase and GST-TPX2 ( 1–43 ) . ( A ) MT bundling induced by higher concentrations of MBP-ch-TOG-His6 , a range of 1–100 nM is shown . ( B ) Variability in MT bundling induced by higher concentrations of MBP-GTSE1-His6 , a range of 1–100 nM is shown for two experiments . In one experiment , bundling is seen at 50 nM , and in another , no bundling is seen for concentrations up to 100 nM . ( C ) Combination of 10 nM MBP-ch-TOG-His6 with 1 , 5 , or 10 nM MBP-GTSE1 does not induce bundling . ( D ) Significant bundling observed with clathrin ( 100 nM ) , GST-TACC3-His6 ( 100 nM ) , MBP-ch-TOG ( 10 nM ) , and either 1 , 5 , or 10 nM MBP-GTSE1-His6 . C and D were performed in parallel . Scale bar , 10 µm . ( E ) Coomassie blue-stained gels of purified protein components separated by SDS-PAGE . Molecular weights are indicated to the left . DOI: http://dx . doi . org/10 . 7554/eLife . 07635 . 01310 . 7554/eLife . 07635 . 014Figure 5—figure supplement 2 . MT bundling using mesh complex immunoisolated from mitotic HeLa cells . ( A ) Schematic diagram of the procedure to release mitotic spindle proteins from mitotic spindles at metaphase ( described in Booth et al . , 2011 ) . ( B ) Western blot to show specific co-immunoprecipitation of TACC3/clathrin from spindle fractions ( F5-7 ) but not from mitotic cytosol ( F1 ) using an anti-clathrin light chain ( CON . 1 ) . No immunoprecipitation of clathrin or TACC3 was seen with a control antibody ( anti-myc ) . Western blots were probed with anti-TACC3 ( rabbit polyclonal ) or anti-clathrin heavy chain ( CHC , TD . 1 ) . ( C ) Representative fluorescence micrographs of rhodamine-labeled MTs incubated with beads from the immunoprecipitation shown in B . Significant bundling was seen for clathrin IPs from F5-7 only . No bundling activity was seen with control IPs or clathrin IPs from F1 . Beads are out of frame for clarity . Scale bar , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07635 . 014 To look in more detail at the bundled MTs , we analyzed the control , PRC1 , and TACC3 complex conditions by EM . In the PRC1 and TACC3 complex conditions , pairs of MTs were interconnected by short electron dense connectors ( Figure 5C ) . In control conditions , MTs were randomly oriented , but on occasions when they were in close proximity , no density was seen . These MT-bundling experiments were complicated by the purification of several large proteins ( Figure 5—figure supplement 1 ) . As an alternative source , we immunoprecipitated protein complexes , which contained clathrin and TACC3 directly from mitotic spindle of HeLa cells at metaphase ( Figure 5—figure supplement 2A , B ) . We observed bundling of Taxol-stabilized MTs in vitro , using this complex specifically ( Figure 5—figure supplement 2C ) . Together these experiments indicate that the TACC3 complex can crosslink MTs and drive the recruitment of stabilized MTs into bundles . They are not compatible with a passive role for the mesh in responding to MT positioning . In the context of the K-fiber , this suggests a role for the mesh in maintaining fiber integrity by tethering MTs together and influencing MT positioning . To look further for any evidence of a role for the mesh in MT positioning , we examined the trajectories of MTs within each fiber . K-fibers are bundles of parallel MTs , and we sought to characterize how parallel the MTs are and to test if this is altered by TACC3 overexpression . Deviations from parallel would suggest that the mesh tethers MTs closer together and interferes with their trajectory . Qualitatively , we could see that MTs in TACC3 overexpressing K-fibers were more disorganized ( Figure 6A , Video 2 and Video 3 ) . To measure parallelism more rigorously and to allow comparison of multiple K-fibers , we used several computational approaches ( described in ‘Materials and methods’ ) , which allowed us to normalize the 3D trajectory of each fiber and then examine the deviation of MT trajectories from this vector . K-fibers in cells overexpressing TACC3 had a higher proportion of MTs that deviate from parallel . They frequently had significantly larger polar angles compared to MTs in K-fibers from control cells ( Figure 6B , C ) . The deviation of MTs from a completely parallel condition is easier to visualize , by examining the Cartesian intersection coordinates of each MT vector with an x-y plane at a given distance ( z ) from a common origin ( see ‘Materials and methods’ ) . These scatter plots show that MTs in TACC3 overexpressing cells deviate further from parallel than those in control cells ( Figure 6D ) . The fraction of MT vectors intersecting this plane at radial distances of <10 nm from the center is 0 . 39 and 0 . 18 in control and TACC3 overexpressing cells , respectively . In other words , there are half as many parallel MTs after TACC3 overexpression . 10 . 7554/eLife . 07635 . 015Figure 6 . Analysis of MT trajectories within a K-fiber . ( A ) Representative ‘side views’ of rendered MTs in single tomograms of individual K-fibers from control ( left , red , Video 2 ) or TACC3 OE ( right , blue , Video 3 ) cells . Tomogram thickness , 66 . 4 nm ( Control ) trimmed to 34 . 2 nm , 34 . 2 nm ( TACC3 OE ) . For scale , MTs are 25 nm in diameter . ( B ) Aide memoire of the spherical coordinate system . Trajectories of MTs ( green line ) in K-fibers were defined and normalized via Euler's rotation ( see ‘Materials and methods’ ) such that the overall trajectory of all fibers pointed to the zenith . Measurements of polar angle ( θ ) and azimuthal angle ( φ ) were made from the normalized sets and are presented in C–E . ( C ) Violin plots of the polar angles for MTs in K-fibers from control and TACC3 overexpressing cells . Box plots show the median , 75th and 25th percentile , and whiskers show the minimum and maximum . Violins show a kernel density estimate of the data and are trimmed to the minima and maxima . Comparison of the median MT angle from each K-fiber , Wilcoxon Rank Sum Test , p = 0 . 007 , Nfiber = 12–15 . ( D ) Illustration of the deviation of MTs from the fiber axis caused by TACC3 overexpression . Cartesian coordinates of the intersection of individual MT vectors that start from a common origin ( 0 , 0 , 0 ) , with an x-y plane at z = 100 nm . To account for the difference in number of MTs , the insets show a normalized 2D histogram of these coordinates cropped to a 40 × 40 nm square centered at 0 , 0 , 100 . ( E ) Plots of azimuthal angle ( above ) and polar angle ( below ) as a function of distance from the fiber center ( defined as described in the ‘Materials and methods’ ) . Line of best fit is shown with 95% confidence bands ( dashed line ) , r2 = 0 . 07 ( control ) , and 0 . 007 ( TACC3 overexpression ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07635 . 01510 . 7554/eLife . 07635 . 016Video 2 . Example of MT organization in a K-fiber from a control cell . Tomogram of a K-fiber . MTs ( green ) were rendered by hand . All segmentation was smoothed in Amira . DOI: http://dx . doi . org/10 . 7554/eLife . 07635 . 01610 . 7554/eLife . 07635 . 017Video 3 . Example of MT organization in a K-fiber from a GFP-TACC3 expressing cell . Tomogram of a K-fiber . MTs ( green ) were rendered by hand . Pale green is used to highlight two highly deviant MTs . All segmentation was smoothed in Amira . DOI: http://dx . doi . org/10 . 7554/eLife . 07635 . 017 Where are these deviant MTs ? If they resided towards the periphery of the fiber , they may represent ‘extra MTs’ , perhaps non-kinetochore MTs , that became enmeshed in the fiber when TACC3 was overexpressed . Alternatively , if the deviant MTs are throughout the fiber , this could be evidence for a role of the mesh in influencing MT spacing and packing within the fiber . To assess this , we plotted the polar and azimuthal angles as a function of distance from the K-fiber center ( Figure 6E ) . These plots show that deviant MTs in TACC3 overexpressing cells are distributed at all distances from the center of the fiber . In controls , there was a weak tendency for deviant MTs to be at the fiber periphery , but the overall extent of trajectory deviancy was lower than in TACC3 overexpressing cells ( Figure 6C , E ) . The lack of relation between distance from the fiber center and the deviancy from the fiber trajectory of MTs in fibers from TACC3 overexpressing cells indicates that the mesh plays an influential role in organizing MTs within the fiber . Together our results show that the change in composition of the mesh , caused by TACC3 overexpression , results in tighter local packing of MTs , more interconnectivity and disruption of the parallel organization of MTs within a K-fiber . What are the functional consequences of the ultrastructural changes in MT organization within K-fibers ? To address this question , we used live-cell imaging of our stable inducible cells expressing H2B-mCherry , growing asynchronously and compared mitotic progression in cells , where GFP-TACC3 expression was induced vs not induced . Figure 7 shows that TACC3 overexpression increases the time taken for cells to congress all chromosomes to the metaphase plate ( prometaphase-to-metaphase ) . In addition , the time from last chromosome alignment to the onset of anaphase ( metaphase-to-anaphase ) is also longer ( Figure 7 ) . No change was seen in the duration of anaphase ( anaphase-to-telophase ) . Interestingly , these changes in mitotic progression are similar to those seen following depletion of TACC3 ( Lin et al . , 2010; Cheeseman et al . , 2013 ) ( Figure 7—figure supplement 1 ) . We conclude from these experiments that mitosis is sensitive to the levels of TACC3 and that this sensitivity correlates to changes in MT organization within the K-fiber . 10 . 7554/eLife . 07635 . 018Figure 7 . Mitotic consequences of TACC3 overexpression . ( A ) Cartoon representation of the stages of mitosis captured in live-cell imaging experiments . Cells expressing H2B-mCherry were staged as indicated with the transition to prometaphase marked by nuclear envelope breakdown; transition to metaphase marked by alignment of the last chromosome to the metaphase plate; to anaphase marked by first sign of chromosome segregation and transition to telophase marked by decondensation of H2B-mCherry . These events allowed us to assign prometaphase-to-metaphase ( PM-M , dark green ) , metaphase-to-anaphase ( M-A , light green ) , and anaphase-to-telophase ( A-T , light blue ) . ( B ) Mitotic stage duration for individual cells . Time taken to go from prometaphase-to-metaphase ( PM-M ) , metaphase-to-anaphase ( M-A ) , or anaphase-to-telophase ( A-T ) is shown for control ( red , above ) or GFP-TACC3 overexpressing ( blue , below ) cells expressing H2B-mCherry . Each line represents a single cell . The time when the last chromosome aligned was set as 0 ( metaphase ) . Cells that did not achieve metaphase halt at 0 . Note that the time axis is truncated to show a 4 hr window centered on metaphase . Ncell = 163–169 , Nexp = 4 . ( C ) Still images of example cells from live-cell imaging experiments to determine mitotic progression . Detection of H2B-mCherry is shown on an inverted grayscale . Time in min is shown and the colored bars represent the staging , as in A and B . ( D ) Box plots to summarize mitotic progression experiments . Time taken to go from prometaphase-to-metaphase ( PM-M ) , metaphase-to-anaphase ( M-A ) , or anaphase-to-telophase ( A-T ) is shown for control ( red ) or GFP-TACC3 overexpressing ( blue ) cells . Whiskers show 90th and 10th percentiles . DOI: http://dx . doi . org/10 . 7554/eLife . 07635 . 01810 . 7554/eLife . 07635 . 019Figure 7—figure supplement 1 . Mitotic consequences of TACC3 depletion . Box plots to summarize mitotic progression experiments . Time taken to go from prometaphase-to-metaphase ( PM-M ) , metaphase-to-anaphase ( M-A ) , or anaphase-to-telophase ( A-T ) is shown for control ( orange ) or TACC3-depleted ( blue ) HeLa cells using shRNA-mediated knockdown . Whiskers show 90th and 10th percentiles . Ncell = 122–138 , Nexp = 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 07635 . 019 In this paper , we described the mesh: a network of MT connectors in K-fibers . We showed that the mesh is comprised of different connector types and is a major component of the K-fiber . Overexpression of TACC3 alters the mesh and changes the trajectory of MTs within the fiber , indicating that the mesh is a determinant of K-fiber structural integrity . This manipulation causes defects in mitosis and establishes that TACC3 levels at the spindle must be regulated for optimal mitosis . Since the first EM studies of mitotic spindles , it has been clear that K-fiber MTs are held together throughout their length , presumably to function as an integrated unit ( Hepler et al . , 1970; Rieder , 1981; Witt et al . , 1981 ) . The importance of inter-MT connections along the length of the K-fiber was demonstrated by severing K-fibers at their midpoint ( Spurck et al . , 1997 ) . This resulted in two K-fiber stubs attached to the pole and kinetochore . The stubs do not immediately collapse but instead maintain their structural integrity . Our work identifying the mesh suggests that it is the structure that holds MTs together . This makes sense structurally , but there is a possibility that the mesh is passive and simply connects MTs in the fiber , wherever they may be . However , our work points to a role for the mesh in influencing MT position within the K-fiber . The evidence for this is: first , in K-fibers overexpressing TACC3 , interconnected MTs are closer together than those not contacted by mesh . Second , a complex containing TACC3 was able to bundle MTs in vitro . Third , mesh in TACC3 overexpressing fibers pulls MTs out of alignment . In this way , the mesh plays an active role in MT positioning in K-fibers by favoring inter-MT interactions of defined distances . An extension of this is that the mesh can actively pull MTs together , in an energy-consuming process . We have no evidence for this intriguing possibility , which is compatible with our findings . It seems counterintuitive that overexpression of a mesh component results in skewed MTs; shouldn't it make the MTs in the fiber ‘more parallel’ ? However , it is likely that the mesh is made of other proteins , besides the TACC3–ch-TOG–clathrin complex , for example , CHD4 , HURP/DLGAP5 , and HSET/KIFC1 ( Mountain et al . , 1999; Sharp et al . , 1999; Koffa et al . , 2006; Yokoyama et al . , 2013 ) . We favor the idea that because we have increased only one component of the mesh experimentally and not others , we do not see uniform changes in spacing and packing . This seems plausible as previous analysis determined TACC3–ch-TOG–clathrin was the shortest crosslinks in K-fibers ( Booth et al . , 2011 ) , and so an imbalance of connector sizes in the mesh is predicted to skew MT trajectories . The spindle matrix was originally proposed as a microtrabecular lattice that assists the motors and MTs of the mitotic spindle during chromosome movement ( Pickett-Heaps et al . , 1982 ) . The existence of the matrix has been elusive , although several large macromolecular complexes have been proposed as matrix components ( Pickett-Heaps et al . , 1984; Chang et al . , 2004; Ma et al . , 2009; Johansen et al . , 2011; Schweizer et al . , 2014 ) . Although superficially similar , we do not think that the K-fiber mesh is the ultrastructural correlate of the spindle matrix . The mesh that we have observed exists between MTs within K-fibers ( rather than between K-fibers ) and is most likely composed of MT-associated proteins that function on K-fiber MTs . Most models of the spindle matrix envisage an elastic milieu that encompasses the entire spindle , yet is independent of the MTs ( Scholey et al . , 2001; Yao et al . , 2012 ) . Despite this , micromanipulation experiments have left open the possibility that an anisotropic matrix could exist along K-fibers ( Gatlin et al . , 2010 ) , and this would be compatible with our observations . The K-fiber mesh is in agreement with theoretical and experimental work showing that lateral MT connectors are important for MT arrays to sustain physical forces in cells ( Brangwynne et al . , 2006 ) . We used overexpression of TACC3 as our primary tool to manipulate the mesh . Progression through mitosis is slower under these conditions due to problems in chromosome congression and satisfying the spindle assembly checkpoint . Interestingly , a similar phenotype is observed when TACC3 is depleted from the spindle ( Lin et al . , 2010; Booth et al . , 2011; Cheeseman et al . , 2013 ) . This indicates that mitosis is sensitive to TACC3 levels . We propose that this sensitivity reflects the robustness of the K-fibers . Too little TACC3–ch-TOG–clathrin and the mesh cannot strengthen the fibers adequately . Too much , and the MTs within the fibers become less parallel and oversupported . Previous work has suggested that hyperstabilization of K-fibers is counter-productive , leading to errors in mitosis ( Bakhoum et al . , 2009 ) . TACC3 levels are altered in several human cancers , for example , ovarian , non-small cell lung cancer , and bladder ( Schmidt et al . , 2010 ) , and fusions between TACC3 and FGFR3 are reported in glioblastoma and bladder cancer ( Singh et al . , 2012; Williams et al . , 2013 ) . Moreover , the amount of TACC3–ch-TOG–clathrin at the spindle is limited by the availability of phosphorylated TACC3 ( LeRoy et al . , 2007; Hood et al . , 2013 ) , and since Aurora-A kinase is amplified or upregulated in a wide range of cancers ( Nikonova et al . , 2013 ) , it is possible that hyperstabilization of K-fibers occurs in these cells . Under any of these conditions , erroneous mitosis leading to aneuploidy may contribute to cancer initiation or progression . While we favor the idea that normal mitosis requires optimal K-fiber stabilization by the mesh , it must be noted that TACC3 has at least one additional function in mitosis ( Gutiérrez-Caballero et al . , 2015 ) , and future work will seek to delineate the role of TACC3 in the mesh vs at the MT plus-ends . Identifying and locating proteins within the mesh is the next challenge . Potentially , 3D views of K-fibers in situ will allow us to dock molecular structures of mesh components as has been achieved for other cellular systems ( Lucic et al . , 2013 ) . Methods to unambiguously assign specific electron density to a particular protein will enable us to achieve this , and efforts are currently underway in our lab to do this . HeLa cells ( HPA/ECACC #93021013 ) were maintained in Dulbecco's Modified Eagle's Medium ( DMEM ) plus 10% fetal bovine serum ( FBS ) and 100 U/ml penicillin/streptomycin , in a humidified incubator at 37°C and 5% CO2 . For TACC3 overexpression , HeLa Tet-On cells ( Clontech , Takara , Saint-Germain-en-Laye , France ) were used to inducibly express GFP-TACC3 . HeLa Tet-On cells were maintained in full DMEM with 300 μg/ml G418 , and HeLa Tet-On cells with stably integrated pTRE2hyg-GFP-TACC3 ( KDP ) plasmid were maintained with 300 μg/ml G418 and 200 μg/ml Hygromycin B . Expression of GFP-TACC3 was induced with 0 . 5 μg/ml doxycycline , 24 hr before analysis ( EM analysis or mitotic progression experiments ) . For TACC3 knockdown , cells were transfected with plasmids to co-express GFP and shRNA using GeneJuice according to the manufacturer's instructions 2 days before imaging . For immunofluorescence , HeLa cells on coverslips transfected to express GFP or GFP-TACC3 were either kept at 37°C or cold-treated for 5 min then fixed with ice-cold methanol for 10 min . Cells were then blocked ( PBS with 5% BSA and 5% goat serum ) before immunostaining with rabbit anti-alpha-tubulin ( Thermo , PA5-19489 , 1:500 ) and mouse anti-CENPA ( Abcam , ab13939 , 1:500 ) and Alexa568/Alexa633-conjugated secondary antibodies . For EM analysis , cells were synchronized using 2 mM thymidine block for 16 hr , release for 6 hr , followed by 9 μM RO-3306 for 16 hr ( see Figure 3—figure supplement 1 ) . After 30–40 min release , mitotic cells were collected by shake-off and centrifugation at 300×g for 2 min at 37°C and resuspended in DMEM containing 20% FBS . Cell synchronization was necessary to harvest sufficient numbers of cells for our analysis . Similar results were obtained with cells synchronized using nocodazole , and with no synchronization , using a correlative light-EM approach , where a single cell was targeted for freezing . For inducible expression , GFP-TACC3KDP ( Booth et al . , 2011 ) was inserted into pTRE2hyg vector at NheI and NotI sites . Plasmids for expression of GST-TACC3-His6 and GST-TPX2 ( 1–43 ) were available from previous work ( Hood et al . , 2013 ) . To make MBP-ch-TOG-His6 , the coding sequence of full-length human ch-TOG was amplified by PCR to add BamHI and EagI sites before subcloning into pMALPreHis vector . For MBP-GTSE1-His6 , the coding sequence of human G-2 and S-phase expressed 1 was amplified from an IMAGE clone ( 4138532 ) to add EcoRI and BamHI sites . For His6-PRC1 , the coding sequence of human Protein Regulator of Cytokinesis 1 was amplified from an IMAGE clone ( 2958690 ) and inserted into pRSETB at NheI and EcoRI sites . Plasmids to co-express GFP with shRNA targeting GL2 or TACC3 ( pBrain-GFP-shGL2 and pBrain-GFP-shTACC3 ) were available from previous work ( Booth et al . , 2011; Lopez-Murcia et al . , 2014 ) . Clathrin was purified from rat liver as described previously ( Hood et al . , 2013; Lopez-Murcia et al . , 2014 ) . Recombinant human Aurora-A kinase was purchased from EMD Millipore ( Watford , UK ) . Reagents for in vitro MT assembly were from Cytoskeleton Inc ( Bioquote , York , UK ) . All proteins were purified from BL21pLysS bacteria as described previously ( Hood et al . , 2013 ) . For in vitro MT bundling experiments , the following conditions were used: Control ( 210 nM MBP-His6 , 100 nM GST ) , PRC1 ( 2 µM His6-PRC1 ) , and Mesh Complex ( 10 nM MBP-ch-TOG-His6 , 100 nM clathrin , 100 nM GST-TACC3-His6 , 100 nM MBP-GTSE1-His6 ) , and visualized by fluorescence microscopy and EM . Concentrations refer to the final concentration with MTs . Proteins were incubated at 30°C for 90 min to allow for phosphorylation by TPX2 ( 1–43 ) /Aurora-A , 2 µg/ml each . MTs were polymerized from 100 µM tubulin ( 1:10 labeled to unlabeled ) in general tubulin buffer ( 80 mM K-Pipes , 2 mM MgCl2 , 0 . 5 mM EGTA , pH 7 . 0 ) with 1 mM GTP and 6% glycerol for 15 min at 37°C . Diluted 1:20 in general tubulin buffer supplemented with 10 μM taxol ( paclitaxel , Sigma ) and 1 mM GTP , incubated for a further 5 min at 37°C . To 92 . 4 µl of protein mixture , 69 . 9 µl of MTs were added and incubated at RT for 10 min . Then 7 . 25 µl of this was fixed with glutaraldehyde in BRB80 ( 1 . 45 µl; 0 . 1% final ) , then diluted with 10 µl 70% glycerol/BRB80 . 6 µl was spotted onto glass slides , overlaid with a cover slip , and sealed with nail varnish . The remainder was split into two and spun for 90 min RT 21 , 000×g and the pellets fixed in 1% glutaraldehyde for 2 hr RT , followed by post-fixing with 3% glutaraldehyde , 15 min RT . Following three washes in PBS , samples were osmicated in 1% osmium tetroxide , 1 hr RT , washed in PBS ( 3 × 20 min ) , then dH2O ( 2 × 20 min ) , and into 30% ethanol for 20 min . Samples were incubated in 0 . 5% uranyl acetate in 30% ethanol for 40 min . Samples were dehydrated and infused with resin and cured at 60°C for 2 days . Sections were cut and visualized by transmission EM . The MT bundling assay by fluorescence microscopy alone was carried out more than five times . The experiment where the same samples were visualized by fluorescence microscopy and EM was performed twice using different protein preparations . Immunopurification of the mesh complex was as described previously ( Booth et al . , 2011 ) . Beads from this procedure were used directly for MT-bundling experiments as described above . For the mitotic progression assays , 6 . 8 × 104 cells/ml were seeded in 12 well plates and transfected with H2B-mCherry . GFP-TACC3 expression was induced the day after transfection . 24 hr later , the cells were imaged on a Nikon Ti epifluorescence microscope with 40× ELWD objective ( 0 . 6 NA ) and a Coolsnap Myo camera ( Photometrics ( Tucson , AZ ) ) for 14 hr using NIS Elements AR software; H2B-mCherry was imaged once every 3 min , and GFP-TACC3 monitored once every 30 min . Cells were kept at 37°C , in supplemented CO2-independent medium containing doxycycline ( 0 . 5 μg/ml ) for cells with GFP-TACC3 induction . Light intensity and exposure were minimized to avoid light-induced cell damage . Analysis of mitotic staging was by manual inspection of videos , with automated time look-up . The same microscope was used to assay the bundling of fluorescent MTs . For analysis of MTs in the proximity of kinetochores , a similar method to that previously described was used ( Cheeseman et al . , 2013 ) . Briefly , confocal imaging was performed on a spinning disc confocal system ( Ultraview Vox , Perkin Elmer ) using a 100× ∼1 . 4 NA oil immersion objective lens with a Hamamatsu C10600-10B ORCA-R2 camera . Fixed cells expressing GFP or GFP-TACC3 , stained as described above , were excited at 488 nm , 561 nm , 405 nm , and 640 nm , and z-sections were taken every 0 . 25 μm . OME-TIFF image stacks were transferred to IMARIS ( Bitplane ) and subjected to spot detection in the CREST/CENP-A channel with a minimal spot size of 0 . 58 μm to detect kinetochores . Spots that were not properly assigned ( ∼5% ) were corrected manually . The mean pixel densities in this sphere were background subtracted , and the median intensity per cell was found using IgorPro . HPF was performed using either EM PACT2 or EM HPM100 ( Leica Microsystems , Milton Keynes , UK ) . Mitotic cells were transferred to 100 μm-depth membrane carriers or Type A 100 μm-depth carriers for the EM PACT2 or EM HPM100 , respectively . Once frozen , carriers were transferred in liquid nitrogen to the EM AFS2 ( Leica Microsystems ) FS machine , which was precooled to −90°C . Carriers were immersed in a FS medium of 1% osmium tetroxide , 0 . 5% uranyl acetate , and 5% H2O in acetone . Cells were brought to RT over 67 hr: 27 hr at −90°C; ramp to −60°C over 15 hr; 8 hr at −60°C; ramp to −30°C over 15 hr; 1 hr at −30°C; ramp to 4°C over 1 hr; cells were taken to the fume hood to reach RT . Cells were embedded in epoxy resin ( TAAB ) and polymerized at 60°C for 48 hr . Sections ( 70 nm ) were cut using an EM UC6 ( Leica Microsystems ) and post-stained using uranyl acetate and Reynold's lead citrate . Sections were imaged using the FEI Tecnai G2 Spirit BioTWIN microscope at 100 kV . Putative K-fibers were identified as bundles of more than 10 MTs in sections taken orthogonal to the spindle axis . K-fibers were imaged 1 µm away from the kinetochore , so as not to include any kinetochore-MT linkages ( Dong et al . , 2007 ) . Each tilt series was taken using TIA software ( TEM imaging and Analysis , FEI ) from +50° to −50° in 1° steps . The IMOD etomo package was used to generate tomograms from these tilt series ( Mastronarde , 1997 ) . Tomogram thickness ranged from 28 . 8 to 66 . 4 nm . For visualization , 3D rendering of tomograms was done in Amira 5 . 6 . 0 ( Visualization Sciences Group , FEI , Eindhoven , Netherlands ) using a combination of manual and automated detection . Each MT was rendered individually by hand , and then any 3D density connecting the MTs was automatically detected throughout the volume of the tomogram , using the average gray value of the MTs as an initial threshold for segmentation . The space between MTs in a K-fiber is less granular than the surrounding cytoplasm , which improves mesh detection . This unbiased method superseded our initial attempts at hand-rendering the mesh and allowed for unbiased quantification and visualization of the mesh . Note that measurements are taken from the unsmoothed segmentation maps . The segmentation method detects unbroken density connected to MTs . However , after smoothing for visualization , breaks in the mesh appear . In Amira , MTs were smoothed using unconstrained smoothing , and the mesh was rendered with smooth surface ( 50 iterations , 0 . 6 lambda ) . Hand-rendered mesh in Figure 1B was smoothed with constrained smoothing and then smooth surface generated ( 20 iterations , 0 . 6 lambda ) . In longitudinal sections , K-fibers were defined as bundles of MTs contacting both the kinetochore and the pole ( McDonald et al . , 1992; Booth et al . , 2013 ) . In orthogonal sections , K-fiber bundles were defined as collections of 10 or more MTs , using an 80 nm boundary around each MT ( 105 nm from the MT center ) . This distance corresponds to the longest inter-MT bridges visualized previously by 2D EM ( Booth et al . , 2011 ) . For analysis of MT packing , coordinates of each MT at the midpoint of the tomogram were collected , and the area of a convex hull that enclosed the coordinates was found . The distance to the nearest neighboring MT and the number of MTs within 80 nm ( 105 nm from the MT center ) were calculated . Heat maps were generated by Voronoi interpolation of 3D coordinate sets that comprised the x-y coordinates together with the number of MTs within 80 nm in the z dimension . To analyze the mesh connections and MT chaining , individual connections in rendered tomograms were classified and recorded together with the chain sizes for each K-fiber . Coordinate sets were supplemented with chain membership data . This was used for an automated comparison of the number of neighboring MTs within a given search radius , comparing chained vs single MTs . The analysis was performed from 20 to 120 nm , and non-parametric Wilcoxon–Mann–Whitney two-sample rank test used to test the hypothesis that there was no difference in the number of neighboring MTs . Chain membership data were then randomized and the analysis repeated . Note that chain membership is probabilistic and is a function of tomogram thickness and mesh density/MT interconnectivity . ‘Single MTs’ are those that are not detected to be attached to another but are likely to be attached to other MTs in a subsequent slab of K-fiber . For analysis of MT trajectories , 3D coordinates for each MT at the bottom and top of the tomogram were logged in ImageJ and fed into IgorPro . Custom-written procedures ran through the following steps . First , coordinates were compiled into individual MT matrices . Second , the center of the bundle was found using farthest-point clustering , and then the distance of each MT from this point was calculated . Third , the fiber direction was ‘normalized’ , that is , rotated through 3D space so that most MTs pointed towards the zenith . This was necessary because of the variability in the axes of different K-fibers relative to sectioning and the resultant tomograms . To do normalization , all MTs were multiplied by rotation matricesR=Rz ( α ) Ry ( β ) Rx ( γ ) , for ( in radians ) 0 ≤ α ≤ 2π , 0 ≤ β ≤ π/2 , γ = 0 in 180/π increments . The Cartesian distance of the orthogonal projection of each MT on the reference plane after rotation was summed , to find the rotation that produced the minimum value . The spherical coordinates , given byr=x2+y2+z2 , θ=cos−1 ( zx2+y2+z2 ) , φ= tan−1 ( yx ) , of the MTs set at optimal rotation , were used for plotting . Fourth , a 2D plot to visualize the degree of MT deviancy was generated . To do this , the point at which each 3D MT vector starting at the origin intersects an x-y plane that was set arbitrarily at z = 100 nm was calculated and plotted . Note that the even radial dispersal in Figure 6D and the even spread of azimuthal angles in Figure 6E show that capture of tilt series was randomized , and the data set has no bias towards a particular trajectory . Several other strategies were explored to analyze deviations in trajectory vs the fiber axis . These were: examining the variance in trajectory angles , pairwise comparison of all MTs in the bundle , and comparison to a reference MT that represented the fiber axis , using spherical rotation and rotating by an average value . These produced similar results , however , the one described here was the most robust and represents our best method for this kind of spatial statistical analysis . The computer code used in the main analysis pipeline can be found as a Source code 1 . Images were cropped in Photoshop , and figures were assembled in Illustrator CS5 . 1 . IgorPro 6 . 36 ( Wavemetrics , Portland , OR ) was used for all analysis and plotting .	Before a cell divides , its genetic material must be copied and then equally distributed between the newly formed daughter cells . In the cells of plants , animals , and fungi , a structure known as the spindle pulls the two copies of the chromosomes apart . The spindle is made up of a network of long , protein filaments called microtubules , and the bundles of microtubules that attach to the chromosomes are referred to as ‘K-fibers’ . K-fibers are organized in a way that provides strength . These bundles of microtubules are held together throughout their entire length and , in 2011 , it was suggested that a group of proteins including one called TACC3 could cross-link adjacent microtubules within K-fibers . However , it remained unclear how these proteins achieved this . Now , Nixon et al . —including several of the researchers involved in the 2011 work—have used a technique called 3D electron tomography to analyze what holds the K-fibers together in human cells . This analysis revealed struts or connectors that hold together adjacent microtubules within K-fibers . These connectors can vary in size and a single connector can link up to four microtubules . This means that , in a three-dimensional view , the connectors appear as a ‘mesh’ between the microtubules in the bundle . Nixon et al . then increased the levels of the TACC3 protein and found that the K-fibers became disorganized . The spacing of the microtubules with the K-fibers was reduced so that they were more tightly packed than normal . These observations suggest that ‘the mesh’ influences the microtubule spacing within a K-fiber . Nixon et al . analyzed how disorganized K-fibers affected dividing cells and found that it took longer for the chromosomes to move to the newly forming daughter cells . This suggests that cells must maintain optimal levels of TACC3 to ensure that the K-fibers can effectively separate the chromosomes . Further work is needed to identify the other proteins and molecules that make up the mesh .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2015	The mesh is a network of microtubule connectors that stabilizes individual kinetochore fibers of the mitotic spindle
The mechanisms linking systems-level programs of gene expression to discrete cell biological processes in vivo remain poorly understood . In this study , we have defined such a program for multi-ciliated epithelial cells ( MCCs ) , a cell type critical for proper development and homeostasis of the airway , brain and reproductive tracts . Starting from genomic analysis of the cilia-associated transcription factor Rfx2 , we used bioinformatics and in vivo cell biological approaches to gain insights into the molecular basis of cilia assembly and function . Moreover , we discovered a previously un-recognized role for an Rfx factor in cell movement , finding that Rfx2 cell-autonomously controls apical surface expansion in nascent MCCs . Thus , Rfx2 coordinates multiple , distinct gene expression programs in MCCs , regulating genes that control cell movement , ciliogenesis , and cilia function . As such , the work serves as a paradigm for understanding genomic control of cell biological processes that span from early cell morphogenetic events to terminally differentiated cellular functions . A major goal of biology over the last several decades has been to understand the mechanisms that control differential gene expression . While recent advances in genomic technology have dramatically empowered these studies , we still know comparatively little about the mechanisms linking systems-level programs of gene expression to discrete cell biological processes in vivo . This gap in our understanding is important because organ and tissue function are ultimately executed by the specialized behaviors of individual cells ( e . g . , polarized secretion in excretory organs , coordinated contraction in muscle cells ) . Recent studies of cilia and ciliated epithelial cells highlight the current disconnect between genomics and cell biology: it is clear that hundreds of proteins are required for cilia assembly and function ( Gherman et al . , 2006; Inglis et al . , 2006 ) , and moreover , assembly of new cilia/flagella clearly requires new transcription ( Thomas et al . , 2010 ) . Nonetheless , only a handful of transcription factors have been identified that control cilia structure and function , and their associated gene regulatory networks remain largely undefined , especially in vertebrates . In C . elegans , the sole RFX family transcription factor daf-19 is the central regulator of ciliogenesis , and dozens of target genes are known to effect its action ( Efimenko et al . , 2005; Phirke et al . , 2011; Burghoorn et al . , 2012 ) . The one Rfx factor in Drosophila is likewise well characterized ( Laurencon et al . , 2007; Newton et al . , 2012 ) . By contrast , multiple RFX family members are essential for ciliogenesis in vertebrates , but as yet , there has been no comprehensive genome-wide survey of Rfx-dependent gene expression as it relates to ciliogenesis ( Bonnafe et al . , 2004; Ashique et al . , 2009; El Zein et al . , 2009 ) . This gap in our knowledge of the genomics of RFX factors is made the more important because these proteins also possess cilia-independent functions about which very little is yet known , including control neuronal and pancreatic development ( e . g . , Senti and Swoboda , 2008; Ait-Lounis et al . , 2010; Pearl et al . , 2011; Benadiba et al . , 2012 ) . One vertebrate cell type in which Rfx factors are known to play particularly important roles is the multi-ciliated epithelial cell ( MCC ) . These cells project dozens or hundreds of motile cilia from their apical surfaces , and the polarized beating of these cilia generates fluid flow that is essential for development and homeostasis in many organ systems ( Figure 1 ) . Such cells are central to the normal homeostasis of airway , brain and reproductive tracts ( Worthington and Cathcart , 1963; Yeung et al . , 1991; Lyons et al . , 2006; Fahy and Dickey , 2010 ) , and defective functioning of these cells is associated with pathologies ranging from chronic infection to hydrocephalus ( Afzelius , 1976 ) . Despite these cells’ importance and long history of study ( Sharpey , 1830 ) , the transcription factors that control MCC development and function are only now being elucidated ( You et al . , 2004; Stubbs et al . , 2008; Marcet et al . , 2011; Stubbs et al . , 2012; Tan et al . , 2013 ) , and among these factors is Rfx2 ( Chung et al . , 2012 ) . 10 . 7554/eLife . 01439 . 003Figure 1 . Conserved cell behaviors during multi-ciliated cell development in mammalian airways and Xenopus epidermis . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 003 In this study , we sought to combine systems biology approaches with in vivo cell biological studies in order to better define the genomic control of MCC development and function . We used high-throughput sequencing of Rfx2-regulated transcripts , systematic mapping of Rfx2 chromosomal binding sites , and bioinformatic exploration of functional protein interactions to guide our mechanistic cell biology experiments . This approach provided diverse new insights into the roles played by Rfx2 in ciliogenesis and polarized ciliary beating . Surprisingly , we also find that Rfx2 plays a central , but unanticipated , role in controlling cell movement in newborn MCCs , a process about which almost nothing is currently known . Overall , Rfx2 activates a complex program of gene expression that serves to coordinate several distinct features of MCCs , including cell migration , ciliogenesis , and cilia function , thus serving as a paradigm for genomic control of cell biological processes that span from early differentiation events to terminally differentiated cell functions . To explore the genomic control of cell behavior in MCCs , we turned to the amphibian embryo , which has emerged as a powerful and rapidly assayable in vivo model for this cell type . Indeed , several foundational studies of specification , ciliogenesis , and planar polarization in MCCs have been performed in amphibians ( Deblandre et al . , 1999; Stubbs et al . , 2006; Park et al . , 2008; Mitchell et al . , 2009; Marcet et al . , 2011; Sirour et al . , 2011; Werner and Mitchell , 2011; Stubbs et al . , 2012 ) . Importantly , these studies have consistently prefigured results in mammals ( Marcet et al . , 2011; Morimoto et al . , 2010; Stubbs et al . , 2012; Tsao et al . , 2009; Vladar et al . , 2012 ) . We therefore exploited the Xenopus system to perform parallel RNA transcript sequencing ( RNA-seq ) and chromatin immunoprecipitation deep sequencing ( ChIP-seq ) for Rfx2 , which we showed previously to be essential for the normal development of cilia in MCCs ( Chung et al . , 2012 ) ( Figure 2 ) . 10 . 7554/eLife . 01439 . 004Figure 2 . Rfx2 controls diverse ciliogenic machinery . ( A ) Schematic diagram of this study . ( B ) Prominent cilia-related genes identified as Rfx2 targets in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 00410 . 7554/eLife . 01439 . 005Figure 2—source data 1 . 911 genes corresponding to the directly regulated downstream target genes of Rfx2 . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 00510 . 7554/eLife . 01439 . 006Figure 2—source data 2 . Table of enriched GO terms . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 00610 . 7554/eLife . 01439 . 007Figure 2—figure supplement 1 . Controls for the morpholino antisense oligonucleotides used in this study . ( A ) The amount of Rfx2 protein was reduced in Rfx2 morphants . α-tubulin served as a loading control . ( B ) Cilia length was significantly reduced following Ttc29 knockdown . The phenotype can be partially rescued by co-injection with GFP-Ttc29 mRNA . ( C ) The amount of Ribc2 protein was reduced following Ribc2 knockdown . α-tubulin served as a loading control . ( D ) The expression of slit2 was reduced following Slit2 knockdown . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 00710 . 7554/eLife . 01439 . 008Figure 2—figure supplement 2 . Summary of RNA-seq data . ( A ) Correlation between RNA abundances from the replicate wild-type control samples , ( B ) correlation between RNA abundances from the replicate RFX2 knockdown samples , ( C ) MA-plot showing the relationship between fold-change and average abundance of each gene , and ( D ) Volcano plot showing the relationship of fold-change to FDR ( adjusted p-value for differential expression ) . Genes differentially expressed in Rfx2 knockdown samples are indicated as red dots on ( C ) and ( D ) . No systematic biases were evident among the differential expressed genes . Although the numbers of raw reads differed between samples , their normalized read counts correlated well . ( E ) Summary of sequencing data . The row titled ‘Total reads’ provides the numbers of Illumina Hi-Seq sequencing reads after pre-processing to remove low quality . We used the set of JGI 6 . 0 scaffolds longer than 10 , 000 bp for genome mapping , and the longest isoform of each gene model ( ‘Oktoberfest’ version ) , with bowtie1 ( version 0 . 12 . 7 ) allowing two mismatches on the seed ( -v 2 option ) . For mapping ChIP-Seq data to genomic scaffolds , we considered only unique hits ( -m 1 option ) . For mapping RNA-seq reads to transcript models , we allowed for redundant hits ( -an option ) so as to maximize the signals from the RNA-seq datasets for the purposes of calculating differential gene expression , where redundant hits should not significantly affect the analysis , as each gene model was independently tested across conditions . We normalized across libraries by the total number of reads mapped onto a gene model . Subsequent tests of mapping without allowing redundant hits ( -m 1 option ) against the longest gene model confirmed that the differences between these two options was negligible . It should be noted that RNA-seq reads are paired-end 2 × 50 bp and ChIP-seq reads are single-end 1 × 35 bp . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 00810 . 7554/eLife . 01439 . 009Figure 2—figure supplement 3 . Three examples of RNA-seq and ChIP-seq data . ( A ) ift172 ( B ) ribc2 ( C ) ttc29 . In each figure , the top panel plots ChIP-seq read depths for Rfx2-GFP ( red ) and the GFP control ( gray ) samples across the genomic scaffold , while the middle panel plots RNA-seq read depths for the two replicate control experiments ( red and magenta ) and Rfx2 morpholino knockdown experiments ( blue and cyan ) . All mapping results are reported as raw read counts ( not normalized ) . Each bottom panel indicates the corresponding X . laevis gene model , indicating transcription start sites with black arrows . In each case , Rfx2 binds near the first exon of the transcripts , and gene models match well to exons from the RNA-seq data . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 00910 . 7554/eLife . 01439 . 010Figure 2—figure supplement 4 . Validation of Rfx2-dependent genes . ( A ) Overview of the animal cap assay . ( B ) RT-PCR results . Ctl: control animal caps . Rfx2 MO: Rfx2 morpholino-injected animal caps . α-tubulin expression was not changed . ef1α served as a loading control . ( C ) ccdc63 , ccdc104 , dnal1 , ribc2 , ropn1l , tekt3 , and ttc29 were expressed in MCCs and their expressions were Rfx2-dependent . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 010 A detailed description of this approach is provided in the ‘Materials and methods’ section . Briefly , RNA-seq was performed on isolates of Xenopus mucociliary epithelium grown in organotypic culture , comparing control samples to knockdown using validated Rfx2 morpholino antisense oligonucleotides ( Figure 2A; Figure 2—figure supplement 1 ) ( Chung et al . , 2012 ) . We performed two biological replicates , each containing 100–200 individual tissue isolates; analysis of the correlation found excellent reproducibility between the two replicates ( Figure 2—figure supplement 2 ) . As we are most concerned here with results that will be relevant to human health , we confined our analysis of the RNA-seq data to Xenopus genes for which there were unambiguous human orthologues to facilitate cross-organism comparison . From this list of 11 , 644 genes , we identified 2750 genes that were differentially expressed following Rfx2 knockdown ( >twofold; FDR<0 . 05 ) ( Figure 2A , Figure 2—figure supplement 2 ) . We next used ChIP-seq to ask which of these differentially expressed genes were direct targets of control by Rfx2 . We identified 3465 genes that were significantly bound by Rfx2 and , importantly , this set of genes overlapped significantly with the set of genes differentially expressed in Rfx2 morphants ( Figure 2A , Figure 2—figure supplement 3 ) ( p≤10−6 , hypergeometric test ) . The intersection of these two experiments defined 911 genes ( Figure 2—source data 1 ) , corresponding to the directly regulated downstream target genes of Rfx2 . We tested a subset of these directly bound , differentially expressed genes by in situ hybridization and RT-PCR analysis; 100% were confirmed as Rfx2-dependent genes ( Figure 2—figure supplement 4 ) . We then focused our further studies on this set of 911 direct targets . An unbiased analysis of enriched Gene Ontology ( GO ) terms suggested a key role for Rfx2 in the control of ciliary gene expression ( Figure 2—source data 2 ) , consistent with the known roles for Rfx2 in ciliogenesis ( Chung et al . , 2012 ) . Moreover , direct examination of the 911 Rfx2 target genes revealed that components of essentially all known ciliary machinery are under the control of Rfx2 in MCCs ( Figure 2B ) . Principal among these is the intraflagellar transport ( IFT ) system , the core mechanism for moving cargoes into and out of the cilium . IFT involves two separable complexes , IFT-A and IFT-B ( Pedersen and Rosenbaum , 2008 ) , and we identified the IFT–B complex components ift172 and ift88 as direct Rfx2 targets , consistent with data from flies , worms , and mice ( Thomas et al . , 2010 ) . Moreover , we also identified several additional Rfx2-regulated components of both IFT-B and IFT-A complexes , as well as the IFT-A adaptor tulp3 ( Mukhopadhyay et al . , 2010 ) ( Figure 2B; Supplementary file 1A ) . Additionally , we found that Rfx2 directly controls expression of genes encoding axonemal dynein subunits , components of the transition zone , and a component of the BBSome , and many of these genes are mutated in human ciliopathies ( Figure 2B; Supplementary file 1A ) ( Sharma et al . , 2008; Reiter et al . , 2012 ) . Significantly , our analysis identified several ciliary systems that have not previously been associated with RFX factors . For example , Rfx2 controlled expression of tubulins , enzymes involved in tubulin post-translational modification , and microtubule binding proteins , such as map7 and spef1/clamp , which localize to the proximal and distal axonemes , respectively ( Brooks and Wallingford , 2012 ) ( Figure 2B; Supplementary file 1A ) . Rfx2 also controlled the expression of the planar cell polarity ( PCP ) effector gene fritz/wdpcp , which encodes a protein governing assembly of the septin diffusion barrier at the base of cilia ( Kim et al . , 2010 ) . Rfx2 also controls many centriolar genes required for ciliogenesis , including cep164 ( Graser et al . , 2007 ) . Notably , wdpcp and cep164 are both implicated in human ciliopathies ( Kim et al . , 2010; Chaki et al . , 2012 ) . This analysis thus provides a comprehensive view of Rfx-related ciliary gene expression in vertebrates . These data provide a foundation for understanding Rfx-mediated control of known cilia genes . However , a particularly striking feature of the Rfx2 target genes was the large number whose relationship to cilia ( or lack thereof ) is not known . Because transcriptionally co-regulated genes frequently share functions ( Eisen et al . , 1998; Marcotte et al . , 1999 ) , we next asked if our Rfx2 target gene set might provide a jumping-off point for exploring the molecular biology of ciliogenesis , with an aim towards implicating new proteins in this important process . To focus our search , we used the direct Rfx2 target genes as a seed set to interrogate a probabilistic human gene network ( HumanNet ) that captures functionally linked genes based on observations in large-scale functional genomics and proteomics datasets ( Lee et al . , 2011 ) . Using ‘guilt-by-association’ in this network , we identified ttc29 as a candidate for functional interaction with the IFT machinery ( Figure 3A ) . We tested this prediction and found not only that Ttc29-GFP localized to ciliary axonemes , but also that knockdown elicited substantial defects in ciliogenesis ( Figure 3B–D ) . To test the HumanNet prediction more directly , we used high-speed in vivo confocal imaging of IFT particle dynamics in Xenopus MCCs ( Brooks and Wallingford , 2012 ) . Strikingly , partial knockdown of Ttc29 resulted in a significant decrease in the mean rate of anterograde IFT ( Figure 3E–G; Video 1 ) , consistent with the bioinformatic linkage of ttc29 to components of the anterograde IFT-B complex ( ift88 and ttc30a; Figure 3A ) . Retrograde IFT movement was not significantly affected by Ttc29 knockdown ( Figure 3H ) . Thus , by combining our genomic dataset with functional gene networks , we have not only revealed new links between Rfx2 and ciliogenesis , but we have also identified a specific function for an uncharacterized protein in the regulation of anterograde IFT . 10 . 7554/eLife . 01439 . 011Figure 3 . Ttc29 is required for ciliogenesis of MCCs by regulating Intraflagellar Transport . ( A ) Ttc29 is clustered with IFT components in HumanNet . ( B ) Ttc29-GFP is localized in the axoneme . ( C ) A MCC of a stage 27 control embryo injected with membrane-GFP . Acetylated α-tubulin labels cilia and GFP labels the cell boundary . ( D ) A MCC of a stage 27 Ttc29 morpholino-injected embryo . Note that only a few short axonemes are shown following Ttc29 knockdown . ( E ) Still-frame of a control multiciliated cell expressing GFP-IFT20 . The axoneme shown in the time series ( E′ ) is labeled in orange ( Video 1 ) . ( E′ ) A time-series of a single control axoneme from ( E ) shows processive bi-directional traffic ( the distal tip of the axoneme is to the right; pink arrowheads denote an anterograde train over time , blue arrowheads indicate a retrograde train ) . ( F ) A single still frame from a Ttc29 MO treated multi-ciliated cell expressing GFP-IFT20 ( Video 2 ) . ( F′ ) A time-series of a single axoneme from ( F ) . Note that processive bi-directional traffic is qualitatively normal . ( G ) Quantification of anterograde GFP-IFT20 rates shows a significantly slower average anterograde rate upon Ttc29 MO treatment ( Control: n = 97 IFT trains , 40 axonemes , 21 Cells , 6 embryos . Ttc29 MO: n = 100 IFT trains , 53 axonemes , 20 cells , 6 embryos . p < 0 . 0001 ) . ( H ) Quantification of retrograde GFP-IFT20 rates reveals no significant difference between control and Ttc29 MO conditions ( Control: n = 87 IFT trains , 40 axonemes , 21 cells , 6 embryos . Ttc29 MO: n = 94 IFT trains , 53 axonemes , 20 cells , 6 embryos . p = 0 . 0510 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 01110 . 7554/eLife . 01439 . 012Video 1 . Dynamics of GFP-IFT20 in a control multiciliated cell . A control multiciliated cell expressing GFP-IFT20 is shown . Processive bidirectional traffic can be observed . Also see Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 01210 . 7554/eLife . 01439 . 013Video 2 . Dynamics of GFP-IFT20 in a Ttc29-knockdown multiciliated cell . A multiciliated cell of a Ttc29-knockdown embryo is shown . Anterograde GFP-IFT20 traffic showed a significantly slower average rate following Ttc29 knockdown . Also see Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 013 To further explore the mechanisms of Rfx2 function , we returned to the list of GO terms enriched in Rfx2 target genes . Of these , the most strongly enriched biological process GO term was ‘ciliary or flagellar motility’ ( Figure 2—source data 2 ) . A direct exploration of the Rfx2 targets identified a wide range of genes with known roles in cilia beating , including genes implicated in planar polarization of cilia ( frizzled , prickle ) ( Mitchell et al . , 2009; Vladar et al . , 2012 ) , tubulin polyglutamylases governing inner arm dynein activity ( ttll7 , ttll9 , ttll13 ) , dynein regulatory complex components , radial spoke components , and a variety of poorly-understood cilia beating genes , such as ropn1l , mns1 , pacrg , and hydin ( Lechtreck et al . , 2008; Wilson et al . , 2010; Fiedler et al . , 2011; Zhou et al . , 2012 ) ( Figure 2B; Supplementary file 1B ) . Given this large array of cilia-beating genes controlled by Rfx2 , we turned again to HumanNet , this time with the goal of providing new insights into the molecular control of cilia beating . Unbiased clustering predicted that three known ciliary beating genes ( dnai1 , ropn1l , mns1 ) ( Horvath et al . , 2005; Fiedler et al . , 2011; Zhou et al . , 2012 ) should be functionally linked to ribc2 , which encodes an uncharacterized vertebrate protein with similarity to Chlamydomonas protofilament ribbon proteins ( Linck and Norrander , 2003 ) ( Figure 4A ) . Using a GFP-fusion , we found that Ribc2 localized to axonemes , though knockdown did not substantially affect cilia length ( Figure 4B–D ) . Rather , high-speed in vivo imaging of axonemes and analysis of fluid flow revealed a specific defect in ciliary beating following disruption of Ribc2 function ( Figure 4E–H; Videos 3 and 4 ) . Transmission electron microscopy revealed that loss of Ribc2 did not affect the integrity of outer doublet or central pair microtubules , but did disrupt their organization within the axoneme and the apparent number of dynein arms ( Figure 4I , J , Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 01439 . 014Figure 4 . Ribc2 is required for ciliary motility . ( A ) Ribc2 is clustered in HumanNet with known ciliary beating components , such as Dnal1 , Ropn1l , and Mns1 ( B ) Ribc2-GFP is localized along the axoneme . ( C ) An MCC of a stage 27 control embryo injected with membrane-GFP . ( D ) An MCC of a stage 27 embryo injected with Ribc2 morpholino . Ribc2 is not essential for cilia assembly . Tracking of latex beads moving across the epidermis of the control embryo ( E ) and the Ribc2 morphant ( F ) . An arrow represents the moving distance per time frame . The relative average flow rate is shown in ( G ) . While the average flow rate of control is normalized to 1 ± 0 . 075 ( mean ±SEM ) , it is significantly reduced to 0 . 085 ± 0 . 008 in Ribc2 morphants . ( H ) Quantification of ciliary beating using high-speed confocal ( Videos 3 and 4 ) . Beat frequency is 20 . 59 ± 0 . 410 strokes/s in control whereas only 5 . 29 ± 0 . 635 strokes/s following Ribc2 knockdown . Ultrastructure of axoneme from a control embryo ( I ) and a Ribc2 knockdown embryo ( J ) were visualized using TEM . Lack of dynein arms were observed in Ribc2 morphants . ( K ) A MCC of a stage 27 control embryo injected with Tektin2-GFP and membrane-RFP . Enlarged view of an axoneme is shown in ( K′ ) ( K′′ ) . ( L ) A MCC of a stage 27 Ribc2 morphant . Enlarged view is shown in ( L′ ) ( L′′ ) . ( M ) Tektin2-GFP generally decorates 80% ( ±0 . 8 ) of the axoneme as marked by membrane-RFP; this ratio is significantly reduced , to 48% ( ±1 . 6 ) , following Ribc2 knockdown ( N ) . Nme5-GFP generally decorates 86% ( ±0 . 5 ) of the axoneme; this ratio is significantly reduced , to 66% ( ±1 . 2 ) , following Ribc2 knockdown . *p < 0 . 0001 Mann–Whitney test . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 01410 . 7554/eLife . 01439 . 015Figure 4—figure supplement 1 . Ribc2 is required for axonemal organization . Blind analysis of distance from outer doublets to central pair and visible dynein arms in control and Rfibc2 morphants . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 01510 . 7554/eLife . 01439 . 016Figure 4—figure supplement 2 . Nme5 is required for ciliary motility . ( A and B ) A MCC of a stage 27 embryo injected with Nme5-GFP and membrane-RFP . ( C ) A MCC of a stage 27 control embryo injected with membrane-GFP . ( D ) A MCC of a stage 27 embryo injected with Nme5 morpholino . Ribc2 is not essential for cilia assembly . Tracking of latex beads moving across the epidermis of the control embryo ( E ) and the Ribc2 morphant ( F ) . An arrow represents the moving distance per time frame . The relative average flow rate is shown in ( G ) . While the average flow rate of control is normalized to 1 ± 0 . 036 ( mean ±SEM ) , it is significantly reduced to 0 . 24 ± 0 . 011 in Nme5 morphants . Control: n = 42 , 3 embryos . Ribc2MO: n = 88 , 3 embryos . *p < 0 . 0001 Mann–Whitney test . Scale bar: 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 01610 . 7554/eLife . 01439 . 017Figure 4—figure supplement 3 . Ribc2 is not required for the axonemal localization of Pacrg-GFP . ( A ) ( A′ ) A MCC of a stage 27 control embryo injected with Pacrg-GFP and membrane-RFP . ( B ) ( B′ ) A MCC of a stage 27 Ribc2 morphant . ( C ) The length ratio of Pacrg-GFP to membrane-RFP . The ratio is not significantly different between control embryos and Ribc2 morphants . Scale bar: 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 01710 . 7554/eLife . 01439 . 018Video 3 . Cilia beating of a control multiciliated cell . A control multiciliated cell expressing membrane-GFP is shown . Beat frequency is 20 . 59 ± 0 . 410 strokes/s in control multiciliated cells . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 01810 . 7554/eLife . 01439 . 019Video 4 . Cilia beating of a Ribc2-knockdown multiciliated cell . A Ribc2-knockdown multiciliated cell expressing membrane-GFP is shown . Beat frequency is 5 . 29 ± 0 . 635 strokes/s following Ribc2 knockdown . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 019 Little is known about protofilament ribbon proteins in any system ( Linck and Norrander , 2003 ) , but guided by connections in HumanNet ( Figure 4A ) , we found that Ribc2 is essential for the normal axonemal localization of another protofilament ribbon protein , Tekt2 ( Figure 4K–M ) . Moreover , our approach also led us to assign a cilia beating function to the previously uncharacterized kinase Nme5 ( Figure 4—figure supplement 2 ) and to find that proper localization of this kinase within axonemes is Ribc2-dependent ( Figure 4N ) . These data highlight the power of combining genomic data with functional gene network analysis: The approach identified new roles for Ribc2 and Nme5 in cilia beating and placed these proteins in a functional hierarchy with Tekt2 . By contrast , pacrg is another direct target of Rfx2 that is required for cilia motility ( Wilson et al . , 2010 ) but is not linked to ribc2 in HumanNet , and Ribc2 knockdown did not affect the normal axonemal localization of Pacrg protein ( Figure 4—figure supplement 3 ) . Thus , our combination of genomics , bioinformatics , and in vivo cell biology provide an effective paradigm for understanding the links between system-level gene expression and discrete behaviors of individual cells . Our data demonstrate that Rfx2 governs ciliogenesis and cilia beating via the expression of myriad genes , including many novel genes characterized here for the first time . Interestingly however , of the 911 Rfx2 target genes , only ∼20% are present in known cilia proteomes ( Figure 5A ) . This finding suggests that many of the Rfx2 target genes are NOT involved in ciliogenesis or cilia function , a notion that is intriguing in light of recent reports of cilia-independent roles for ‘ciliary’ RFX genes ( e . g . , Senti and Swoboda , 2008; Ait-Lounis et al . , 2010 ) . 10 . 7554/eLife . 01439 . 020Figure 5 . Rfx2 is essential for the insertion of nascent MCCs into the mucociliary epithelium . ( A ) Overlap of Rfx2 target genes and the ‘cilia proteome’ ( as defined in Gherman et al . , 2006; see ‘Materials and methods’ ) . Out of 911 direct target genes of Rfx2 identified in this study , only 20% of them ( 180 genes ) are annotated as known cilia genes . Right panel represents the Gene Ontology terms significantly enriched among direct targets of Rfx2 ( biological process category only; Benjamini corrected p<0 . 05 ) ( B ) A cross-sectional view of a control embryo labeled with ciliated cell marker ( cyan ) . Apical surface is up . MCCs have inserted into the mucociliary epithelium ( arrows ) . ( C ) A cross-sectional view of an Rfx2 morpholino-injected embryo . MCCs fail to insert into the overlying epithelium ( arrows ) . To observe the insertion of MCCs into the overlying epithelium of control embryos ( E ) and Rfx2 morphants ( D ) , a MCC-specific α-tubulin enhancer element driving expression of Utrophin-GFP was used . ( D ) Note the control MCC first exhibited a star-shaped morphology and cell protrusions probed into overlying cell–cell boundaries ( arrows ) . The probing phase then ceased and apical surface expanded ( Videos 5 and 6 ) . ( E ) Protrusions of the MCC were observed , indicating the initial probing was qualitatively normal following Rfx2 knockdown . However , apical surface expansion was strongly inhibited in MCCs ( Video 7 ) . ( F ) Quantification of apical surface area of MCCs of control embryos and Rfx2 morphants . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 020 Analysis of GO terms associated with Rfx2 target genes suggested a potential role for Rfx2 in cell movement and cell morphogenesis ( Figure 5A , Figure 2—source data 2 ) , which is of interest because both Xenopus MCCs and their counterparts in mammalian airways arise from a population of p63-expressing basal precursor cells ( Figure 1 , red cell at right ) ( Evans and Moller , 1991; Drysdale and Elinson , 1992; Lu et al . , 2001; Rock et al . , 2010 ) . Consistent with a role for Rfx2 in the apical movement of MCCs , we found that strong knockdowns consistently resulted in MCCs being positioned well below the apical surface of the epithelium ( Figure 5B , C ) . This effect was not secondary to a ciliogenesis defect , as milder knockdowns with lower doses of MO did not inhibit MCC emergence but did suppress ciliogenesis ( Chung et al . , 2012 ) . These results suggested a previously unrecognized role for an Rfx factor in cell movement . A role for Rfx2 in apical movement is noteworthy , because this process is a conserved feature of newly born MCCs ( Evans et al . , 1989; Drysdale and Elinson , 1992; Lu et al . , 2001; Rock et al . , 2009 ) , and it must involve apical movement , remodeling of junctions , and assembly of an apical cell surface to which basal bodies can dock prior to ciliogenesis . The mechanisms guiding these crucial aspects of MCC biology are largely unknown , so we used transgenic drivers to direct expression of fluorescent reporters specifically in MCCs and documented these cells’ behavior using 4D confocal imaging in vivo . We identified two broad categories of MCC behavior associated with apical insertion . MCCs first engaged in a probing of neighboring cell–cell boundaries; subsequently , this behavior ceased and MCCs smoothly expanded their apical surfaces ( Figure 5D; Videos 5 and 6 ) . Time-lapse imaging in Rfx2 morphants revealed that the initial probing behavior was qualitatively normal , while the latter phase of apical surface expansion failed completely ( Figure 5E; Video 7 ) . 10 . 7554/eLife . 01439 . 021Video 5 . The development of a control multiciliated cell . A control multiciliated cell expressing membrane-GFP is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 02110 . 7554/eLife . 01439 . 022Video 6 . The insertion of a control multiciliated cell into the overlying epithelium . A control multiciliated cell expressing Utrophin-RFP is shown . Note the control multiciliated cell first exhibited a probing phase and then an apical surface expansion phase . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 02210 . 7554/eLife . 01439 . 023Video 7 . The insertion of a Rfx2-knockdown multiciliated cell into the overlying epithelium . A Rfx2-knockdown multiciliated cell expressing Utrophin-GFP is shown . The initial probing was qualitatively normal following Rfx2 knockdown . However , apical surface expansion was strongly inhibited . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 023 Rfx2 is expressed strongly in MCCs ( Chung et al . , 2012 ) but RT-PCR also detected expression of Rfx2 in the neighboring goblet cells ( not shown ) . Thus , Rfx2 might act in MCCs to control their cell movement and insertion into the epithelium or it may control behaviors in neighboring superficial cells that facilitate the process—or both . To distinguish between these possibilities , we generated mosaics in which Rfx2 was disrupted specifically in MCCs or specifically in the overlying goblet cells . Knockdown of Rfx2 in the MCCs led to a robust failure of MCC apical surface expansion while knockdown in overlying superficial cells had no effect ( Figure 6 ) . Thus our data suggest that Rfx2 cell-autonomously controls the process of MCC apical surface expansion . 10 . 7554/eLife . 01439 . 024Figure 6 . Rfx2 acts cell-autonomously to control insertion of nascent MCCs into the overlying epithelium . ( A ) – ( C ) Illustration of the transplantation experiments . The superficial layer from either control ( A ) ( A′ ) or Rfx2 knockdown embryos ( B ) ( B′ ) was transplanted to the control host embryos . ( C ) ( C′ ) the superficial layer from control embryos was transplanted to the Rfx2 knockdown embryos . At stage 26 , MCCs derived from the control host have intercalated into the outer layer transplanted from either control ( A′ ) or Rfx2 knockdown embryos ( B′ ) . ( C′ ) MCCs , in which Rfx2 was knocked down , failed to insert properly into control outer epithelium . ( C′′ ) A z-view of two MCCs in ( C′ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 024 Insertion of MCCs into the mucociliary epithelium represents a novel cell behavior , the molecular mechanisms of which remain almost entirely obscure . Given the specific role for Rfx2 in this process , we leveraged our genomic dataset to gain insights . Consistent with the observed insertion defects , we found that Rfx2 directly regulates the extracellular matrix component dystroglycan ( dag1 ) , which is a known regulator of MCC insertion ( Sirour et al . , 2011 ) . Interestingly , Dag1 is known to functionally interact with the Slit/Robo signaling system to control cellular morphogenesis in diverse settings ( Medioni et al . , 2008; Wright et al . , 2012 ) , so it was notable that Rfx2 directly controlled transcription of both the slit2 ligand and the intracellular effector srgap2 ( Supplementary file 1C ) . Accordingly , we found that Slit2 knockdown disrupted insertion of nascent MCCs into the mucociliary epithelium and analysis of mosaic embryos revealed that Slit2 acts cell-autonomously in MCC insertion , as was the case for Rfx2 ( Figure 7 ) . Together , these data identify rfx2 , dag1 , and slit2 as a preliminary molecular framework for epithelial insertion of nascent MCCs and reveal a tight genomic coordination of cell movement , ciliogenesis , and cilia motility . 10 . 7554/eLife . 01439 . 025Figure 7 . Slit2 is required for MCC insertion into the overlying epithelium . ( A ) A control embryo injected with membrane-RFP and α-tubulin > membrane-GFP to label MCCs . Injected embryos were then fixed and stained with RFP , GFP , and α-acetylated tubulin . ( B ) Slit2 morpholino was injected with membrane-RFP and α-tubulin > membrane-GFP . Note that MCCs fail to insert into the mucociliary epithelium following Slit2 knockdown ( arrows ) . In addition , MCCs ( arrowheads ) with no Slit2 morpholino insert into the superficial layer containing Slit2 morpholino . These data indicate Slit2 controls MCCs in a cell-autonomous fashion . DOI: http://dx . doi . org/10 . 7554/eLife . 01439 . 025 In addition to Rfx2 , recent work has identified keys roles for the transcription factors Mcidas and Foxj1 in the development of MCCs ( You et al . , 2004; Stubbs et al . , 2008 , 2012 ) . A key challenge going forward will be to understand how Rfx2 fits into this larger gene regulatory network , so we compared our set of direct Rfx2 targets with the downstream transcriptome of these other factors ( Figure 8 , left ) . Mcidas has been implicated in the earliest stages of MCC specification ( Stubbs et al . , 2012 ) , and accordingly , both Rfx2 and FoxJ1 are downstream of Mcidas , while Mcidas is not present in the target gene sets of the other two factors . Interestingly , our comparison suggests that many ciliary machines require the combined action of all or a subset of these three factors . For example , ift80 is downstream of all three factors , while Ift46 and ift57 are present only in the FoxJ1 gene set and ift140 , ift172 , ift81 , and ift88 are targets only of Rfx2 . Such combined action by Rfx factors and FoxJ1 is consistent with recent reports in mammalian airway epithelium ( Didon et al . , 2013 ) and in Drosophila ( Newton et al . , 2012 ) . Nonetheless , our data also suggest that Rfx2 also plays essential roles in ciliogenesis that are independent of FoxJ1 or Mcidas , as many IFT genes and several ciliopathy-associated genes ( cep164 , rpgripl1 and cc2d2a ) are under the sole control of Rfx2 ( Figure 8 , left ) . Moreover , we have discovered a role here for Rfx2 in apical surface assembly of nascent MCCs , and Foxj1 has no known role in this process . Accordingly , the MCC insertion genes dag1 and slit2 were present only in the Rfx2 target gene list and were not identified as downstream targets of FoxJ1 or Mcidas . Interestingly however , Rab11 is also required for MCC insertion ( Kim et al . , 2012 ) , but was not present among our Rfx2 targets . Future studies will be needed to fully understand the gene regulatory network of MCCs , including how other essential transcription factors ( e . g . , Myb [Tan et al . , 2013] ) , collaborate with the factors discussed here . Finally , it will now be of great interest to ask how this emerging gene regulatory network governing cilia structure and function in multiciliated cells compares with that governing the structure and function of primary , non-motile cilia in other cell types . Our analysis revealed that Rfx2 controls cilia structure and function by modulating expression of dozens of known genes , and this dataset serves as an important complement to genomic studies of RFX factors in invertebrates ( Efimenko et al . , 2005; Laurencon et al . , 2007; Newton et al . , 2012; Phirke et al . , 2011; Burghoorn et al . , 2012 ) . However , a large proportion of genes , even in model animals , remain uncharacterized , and our Rfx2 target gene list contains dozens of genes with no known function . To overcome this hurdle , we used a functional gene network to identify important functional interactions among novel genes and known genes in the Rfx2 target gene set and we used in vivo imaging to test these interactions ( Figure 8 , right ) . This approach led us to discover an essential role for Ttc29 in the control of anterograde IFT and ciliogenesis and to begin delineating a functional hierarchy for the novel proteins Ribc2 and Nme5 in ciliary beating . Moreover , guided by our exploration of gene networks , we found that the Rfx2 targets Tekt2 and Nme5 localize in novel proximodistally-restricted patterns along the axoneme , a result that is of interest because precise proximodistal positioning of specific dynein arm proteins along motile axonemes is central to normal cilia beating . Indeed , though defective proximodistal pattern in axonemes is associated with human primary ciliary dyskinesia ( PCD ) ( Fliegauf et al . , 2005; Panizzi et al . , 2012 ) , we know essentially nothing of how these patterns are controlled . Thus , by exploiting our genomic data to guide cell biological inquiry , this study has provided a foundation for future exploration of this problem . The role of RFX factors in ciliogenesis has been a key focus of recent research , so it is notable that this function is not thought to be ancestral for RFX proteins ( Piasecki et al . , 2010 ) . Indeed , the yeast RFX orthologue , Sak1 , controls cell cycle exit ( Wu and McLeod , 1995 ) and even Rfx factors commonly associated with ciliogenesis ( e . g . , Daf-19 , Rfx3 ) perform cilia-independent functions ( Senti and Swoboda , 2008; Ait-Lounis et al . , 2010 ) . Accordingly , we found that Rfx2 also controls the assembly of the apical surface of nascent MCCs as they insert into the epithelium . Rfx2 effects this process via the known regulator dag1 and through slit2 , which we show here to be essential for MCC apical surface expansion . These results are of particular interest both because Dag1 and Slit/Robo signaling also collaborate during axon growth cone guidance ( Andrews et al . , 2006; Wright et al . , 2012 ) and because Rfx3 mutant mice display axon guidance defects ( Benadiba et al . , 2012 ) . These data may also shed light on the mechanism of Rfx action in synapse morphogenesis in C . elegans ( Senti and Swoboda , 2008 ) , as the neuron projection morphogenesis GO term was enriched in our Rfx2 target gene set , reflecting that mediators of synapse morphogenesis , such as PTPδ and Netrin3 , were present in the target set ( Supplementary file 1C ) ( Takahashi et al . , 2012 ) . Finally , both the genomic studies and our in vivo imaging of MCC insertion provide an important foundation for future work , because mammalian MCCs—like the Xenopus MCCs studied here—arise from basally-positioned precursor cells ( Evans and Moller , 1991; Rock et al . , 2010 ) . In conclusion , this study has revealed a central role for Rfx2 at the nexus of cell movement , ciliogenesis , and cilia motility ( Figure 8 ) . The work sheds new light on Rfx2 protein functions specifically and also on the transcriptional control of cell behavior and organelle biogenesis generally . Our combined approach of systems biology , computational biology , and in vivo cell biology provides a generalizable paradigm for exploiting genomic data to advance our understanding of cell biological processes . Capped mRNA was synthesized using mMESSAGE mMACHINE ( Ambion , Austin , TX ) . mRNA and anti-sense morpholino were injected into ventral blastomeres at the 4-cell stage to target the epidermis ( Moody , 1987 ) . Embryos were incubated until appropriate stages and were fixed in MEMFA ( Davidson and Wallingford , 2005 ) . The embryos were embedded in 2% agarose for thick ( 250–300 micron ) sections , which were cut with a Vibratome series 1000 ( Davidson and Wallingford , 2005 ) . Morpholino sequence and the working concentration: Rfx2 morpholino: AATTCTGCATACTGGTTTCTCCGTC , 12 ng Ttc29 morpholino: GTGCACTCATTCTCTTCAAGTTTGC , 40 ng Ribc2 morpholino: CGATAGGCAGATCCAGTCGGTACAT , 21 ng Slit2 morpholino: TTCAGGTCTCTGGGAAAACAGGAAC , 10 ng We used draft version ( version 6 , JGIv6 ) of the Xenopus laevis genome for analyzing genomic and ChIP-seq datasets , obtained from the International Xenopus laevis genome project consortium and now available from the XenBase FTP site ( ftp://ftp . xenbase . org ) . We selected scaffolds longer than 10 , 000 bases for further analysis , for a total of 8426 scaffolds used in this analysis . Because gene models for X . laevis are not yet finalized , we employed interim gene models for the analysis described here , using the released transcriptome-derived gene models ( ‘Oktoberfest’ version ) , also provided by the International Xenopus laevis genome project . Sequences and detailed descriptions of the gene model construction pipeline are available at the project website ( http://www . marcottelab . org/index . php/Xenopus_Genome_Project ) . Total RNA was collected from 100 animal caps each at stage 20 from control embryos and from Rfx2 morphants . After poly- ( A ) -capture , we prepared sequencing libraries using the standard manufacturer’s non-strand specific Ilumina RNA-seq protocol , and sequenced paired-end 2 × 50 bp reads by Ilumina HiSeq 2000 . Reads were mapped to the longest transcripts for each of the ‘Oktoberfest’ gene models using the bowtie mapper ( version 0 . 12 . 7; ‘-a -v 2’ options were applied ) ( Langmead et al . , 2009 ) . We then estimated normalized gene expression values for each sample by calculating RPKMs ( Reads Per Kilobase per Million mapped reads ) . It should be noted that one of the two Rfx2 MO replicates ( M2 ) had significantly fewer reads ( 30 M read pairs ) than the other ( 72 M read pairs ) . However , measured expression levels ( RPKMs ) correlated well between the replicates ( Figure 2—figure supplement 2 ) , so we retained both for subsequent analyses . Using the edgeR package ( Robinson et al . , 2010 ) , we identified significant changes in RNA abundances between control and Rfx2 knockdowns , requiring greater than twofold abundance changes and a false discovery rate ( FDR ) less than 5% . Out of 11 , 644 genes tested , we identified 2750 genes significantly differentially expressed following Rfx2 knockdown . ChIP-seq was performed as described previously ( Kim et al . , 2011 ) to identify direct chromosomal binding sites of Rfx2 . Briefly , ChIP-seq assays were performed with stage 20 X . laevis embryos , injected with in vitro transcribed mRNA coding for either GFP-Rfx2 or GFP alone . Samples were crosslinked with 1% formaldehyde for 1 hr and the reaction stopped by adding glycine ( to 125 mM ) for 10 min . The embryos were rinsed with PBS and resuspended in lysis buffer with protease inhibitor cocktail ( Roche ) . Chromatin was sonicated to an average size of 200–600 base pair using a Branson 450 Sonifier , then immunoprecipitated using protein G magnetic beads ( Invitrogen ) coupled to 5 μg α-GFP antibody ( ab290 ) at 4°C overnight . Magnetic beads were washed , the bound chromatin eluted , and crosslinks reversed . ChIP DNA was extracted with phenol-chloroform and purified with a QIAquick PCR Purification Kit ( Qiagen ) . ChIP-seq libraries were prepared according to the standard manufacturer’s Illumina sequencing protocol and sequenced by Illumina HiSeq . Reads were mapped to the Xenopus laevis draft genome ( version 6 ) using bowtie ( ‘-m 1 -n 2’ options were applied ) and peaks identified using MACS ( version 1 . 4 . 2 ) ( Zhang et al . , 2008 ) with default options . Out of 29 , 448 peaks identified , we selected 6646 peaks for further study that exhibited either a FDR <5% or a fold-enrichment > 20 . Genes were associated with significant ChIP-seq peaks based on proximity in the draft genome , requiring genes ( specifically , the longest transcript for each gene model as mapped to the draft genome sequence ) to lie within 10 , 000 bases from an identified peak . Out of the 6646 significant ChIP-seq peaks , 3465 peaks could be assigned to nearby genes and 911 of those putative target genes also showed significantly different gene expression after Rfx2 knockdown . We focused on these differentially expressed , directly bound genes for subsequent analyses ( Figure 2—source data 1 ) . To better understand the molecular mechanisms of genes regulated by RFX2 , we analyzed the clustering of RFX2 target genes using a human functional gene network ( ‘HumanNet’ ) ( Lee et al . , 2011 ) . To increase potential coverage , we considered X . laevis genes that were either marked by RFX2 binding sites in the ChIP-seq data or that were significantly differentially expressed following RFX2 . For the human orthologs of these genes ( as defined by the International Xenopus laevis genome project using phylogenetic analyses of gene models during the course of annotating X . laevis genes ) , we extracted HumanNet gene–gene linkage information and associated confidence scores ( log likelihood scores; LLS ) . The resulting network contained 4609 genes with 52 , 714 functional linkages , and served as the basis for later analyses . Due to an ancestral genome duplication along the X . laevis lineage , many human genes have ( typically ) two X . laevis orthologs , generally referred to as ‘homeologs’ or ‘alloalleles’ . For the purposes of calculating gene networks among RFX2 target genes , we transferred evidence from either of the alloalleles to the single orthologous human gene . For example , in the case of having one human gene orthologous to two frog genes , if one of the frog homologs was identified as a direct target of RFX2 and the other was not , we considered the human gene as a direct target of RFX2 for the purposes of reconstructing the network . To identify functional modules regulated by RFX2 in unbiased way , we clustered this network using the clusterONE algorithm ( Nepusz et al . , 2012 ) available in Cytoscape ( version 2 . 8 . 3 ) ( Shannon et al . , 2003; Cline et al . , 2007 ) , considering linkage confidence scores during the clustering . All network information is available at the following URL: http://www . marcottelab . org/index . php/ChungKwon2013_RFX2 For the comparison of Rfx2 targets to ciliary proteins in Figure 5A , we used a compiled list of ciliary proteins drawn from several studies using a combination of proteomics and comparative genomics ( Gherman et al . , 2006 ) . The protein set was downloaded from this website , http://v3 . ciliaproteome . org/cgi-bin/protein_browser . php , then converted to EnsEMBL gene IDs using BioMart ( version 63 ) . In situ hybridization was performed as described previously ( Sive et al . , 2000 ) . Bright field and low magnification fluorescence images were captured on a fluorescent stereomicroscope , Leica MZ16FA . Embryos were fixed in MEMFA for 1 hr followed by washing in PTW ( PBS+0 . 1% Tween 20 ) for 30 min ( 3 × 10 min ) . Embryos were then blocked in fetal bovine serum ( FBS ) solution ( TBS containing 10% FBS and 5% DMSO ) for 1 hr at room temperature . Embryos were then incubated with the following primary antibodies at 4°C overnight: monoclonal anti-α-tubulin antibody ( 1:500 dilutions , clone DM1A , Sigma ) , mouse anti-acetylated-α-tubulin ( 1:500 , clone 6-11B-1 , Sigma ) , chicken anti-GFP antibody ( 1:500 , ab13970 ) , and rabbit anti-RFP ( 1:500 , ab62341 ) . After primary antibody incubation , all samples were washed with TBST ( TBS+0 . 1% Triton X-100 ) for 5 hr ( 1 × 5 hr ) . Primary antibodies were detected with Alexa Fluor 488 goat anti-mouse IgG ( 1:500 , Molecular Probes ) , Alexa-555 goat anti-rabbit IgG ( 1:500 , Molecular Probes ) , and Alexa-488 goat anti-chicken IgG ( 1:500 , Molecular Probes ) . After secondary antibody incubation , all samples were washed with TBST for 5 hr . Embryos prepared for confocal imaging as described ( Wallingford , 2010 ) . Images were obtained using Zeiss LSM5 Pascal and Zeiss LSM700 confocal microscope . Cilia lengths were measured with Fiji software . Images used throughout this paper have been enhanced using the Unsharp Mask filter in Adobe Photoshop . High-speed confocal imaging was performed by time-lapse collection of single optical section at a frame rate of 370fps using a Zeiss LSM 5LIVE microscope . Images were collected from living embryos expressing membrane-GFP driven by MCC-specific promoter and from embryos injected with Ribc2 morpholino . For filming the MCC intercalation , living embryos of either control or Rfx2 morpholino-injected animals were put on a round cover glass in custom machined dishes ( Kieserman et al . , 2010 ) . Embryos were gently pushed down by a small piece of cover glass . Images were collected every 3–5 min and were then processed into a time-lapse movie using Fiji software . For IFT imaging , embryos expressing GFP-IFT20 alone or with Ttc29 MO were mounted flank down in 0 . 8% LMP agarose . Single confocal slices were collected at ∼2 fps using an LSM 5LIVE confocal microscope , as previously described in Brooks and Wallingford ( 2012 ) . Embryos were injected ventrally at 4-cell stage with mRNA encoding either membrane-GFP or membrane-RFP . Rfx2 morpholino was injected together with membrane-RFP . At stage 10 , a fine hair was used to peel off the outer layer from a region of the ectoderm of a donor embryo . This outer layer peel was then transferred onto a host embryo after removing a similar patch of outer cells . To help the healing process , a small piece of glass coverslip with clay feet was used to press down embryos . Transplantations were performed in Danilchick’s Solution for Amy ( DFA ) + 0 . 1% BSA . After healing , embryos were then transferred back to 1/3 MMR ( Stubbs et al . , 2006 ) .	Cells that have hundreds of tiny hair-like structures called cilia on their surface have important roles in our airways and also in the brain and reproductive system . By beating in a coordinated manner , the cilia cause fluid to flow in a particular direction . The development of these multiciliated cells is a complex process in which genes are expressed as proteins , with this gene expression being regulated by other proteins called transcription factors . In invertebrates the development of the cilia is controlled by transcription factors from the RFX family , which also appear to be important for development of cilia in vertebrates . However , the details of this process—in particular , the identities of the genes that are involved and how their functions are related—are not well understood in vertebrates . Chung et al . have sought to remedy this by analyzing the network of genes whose expression is controlled by the transcription factor Rfx2 in vertebrates . The results showed that the genes controlled by Rfx2 were involved in all aspects of cilia , including several genes that are known to be mutated in diseases caused by abnormal cilia . Chung et al . also identified genes that were not previously thought to be relevant to cilia . As multiciliated cells are developing , but before they can generate cilia , they must first migrate from the bottom of the epithelium , the layer of tissue in which they function , to the top of this layer . Chung et al . found that Rfx2 was also involved in this process . The approach taken by Chung et al . —which involved a combination of RNA sequence analysis , examination of Rfx2 binding sites on chromosomes , computational predictions of protein interactions and in vivo cellular imaging—could be used to perform similar systems-level analyses of other developmental and biological processes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2014	Coordinated genomic control of ciliogenesis and cell movement by RFX2
Can limb regeneration be induced ? Few have pursued this question , and an evolutionarily conserved strategy has yet to emerge . This study reports a strategy for inducing regenerative response in appendages , which works across three species that span the animal phylogeny . In Cnidaria , the frequency of appendage regeneration in the moon jellyfish Aurelia was increased by feeding with the amino acid L-leucine and the growth hormone insulin . In insects , the same strategy induced tibia regeneration in adult Drosophila . Finally , in mammals , L-leucine and sucrose administration induced digit regeneration in adult mice , including dramatically from mid-phalangeal amputation . The conserved effect of L-leucine and insulin/sugar suggests a key role for energetic parameters in regeneration induction . The simplicity by which nutrient supplementation can induce appendage regeneration provides a testable hypothesis across animals . In contrast to humans’ poor ability to regenerate , the animal world is filled with seemingly Homeric tales: a creature that regrows when halved or a whole animal growing from a small body piece . Two views have historically prevailed as to why some animals regenerate better than others ( Goss , 1992 ) . Some biologists , including Charles Darwin and August Weismann , hold that regeneration is an adaptive property of a specific organ ( Polezhaev , 1972 ) . For instance , some lobsters may evolve the ability to regenerate claws because they often lose them in fights and food foraging . Other biologists , including Thomas Morgan , hold that regeneration is not an evolved trait of a particular organ , but inherent in all organisms ( Morgan , 1901 ) . Regeneration evolving for a particular organ versus regeneration being organismally inherent is an important distinction , as the latter suggests that the lack of regeneration is not due to the trait never having evolved , but rather due to inactivation—and may therefore be induced . In support of Morgan’s view , studies in past decades have converged on one striking insight: many animal phyla have at least one or more species that regenerate body parts ( Sánchez Alvarado , 2000; Bely and Nyberg , 2010 ) . Further , even in poorly regenerative lineages , many embryonic and larval stages can regenerate . In regenerating animals , conserved molecular events ( e . g . , Cary et al . , 2019 , Kawakami et al . , 2006 ) and regeneration-responsive enhancers ( Wang et al . , 2020 ) were identified . Although the hypothesis of convergent evolution cannot be fully excluded ( e . g . , Lai and Aboobaker , 2018 ) , these findings begin to build the case that the ability to regenerate may be ancestral ( Sánchez Alvarado , 2000; Bely and Nyberg , 2010 ) . Regeneration being possibly ancestral begs the question: is there a conserved mechanism to activate regenerative state ? This study explored how , and whether , limbs can be made to regenerate in animals that do not normally show limb regeneration . In adult frogs , studies from the early 20th century and few recent ones have induced various degrees of outgrowth in the limb using strategies including repeated trauma , electrical stimulation , local progesterone delivery , progenitor cell implantation , and Wnt activation ( Carlson , 2007; Lin et al . , 2013; Kawakami et al . , 2006; Herrera-Rincon et al . , 2018 ) . Wnt activation restored limb development in chick embryos ( Kawakami et al . , 2006 ) , but there are no reports of postnatal regeneration induction . In salamanders , a wound site that normally just heals can be induced to grow a limb by supplying nerve connection and skin graft from the contralateral limb ( Endo et al . , 2004 ) , or by delivery of Fgf2 , 8 , and Bmp2 to the wound site followed by retinoic acid ( Vieira et al . , 2019 ) . In neonatal and adult mouse digits , a model for exploring limb regeneration in mammals , bone outgrowth , or joint-like structure can be induced via local implantation of Bmp2 ( bone ) or Bmp9 ( joint; Yu et al . , 2019 ) . Thus far , different strategies gain tractions in different species , and a common denominator appears elusive . However , across animal phylogeny , some physiological features show interesting correlation with regenerative ability ( Hariharan et al . , 2016; Vivien et al . , 2016; Sousounis et al . , 2014 ) . First , regeneration especially in vertebrates tends to decrease with age , with juveniles and larvae more likely to regenerate than adults . For instance , the mammalian heart rapidly loses the ability to regenerate after birth and anurans cease to regenerate limbs upon metamorphosis . Second , animals that continue to grow throughout life tend to also regenerate . For instance , most annelids continue adding body segments and regenerate well , a striking exception of which is leeches that make exactly 32 segments and one of the few annelids that do not regenerate body segments ( Rouse , 1998 ) . Consistent with the notion of regeneration as ancestral , indeterminate growth is thought of as the ancestral state ( Hariharan et al . , 2016 ) . Finally , a broad correlate of regenerative ability across animal phylogeny is thermal regulation . Poikilotherms , which include most invertebrates , fish , reptiles , and amphibians , tend to have greater regenerative abilities than homeotherms—birds and mammals are animal lineages with poorest regeneration . These physiological correlates , taken together , are united by the notion of energy expenditure . The transition from juvenile to adult is a period of intense energy usage , continued growth is generally underlined by sustained anabolic processes , and regulating body temperature is energetically expensive compared to allowing for fluctuation . Regeneration itself entails activation of anabolic processes to rebuild lost tissues ( Hirose et al . , 2014; Naviaux et al . , 2009; Malandraki-Miller et al . , 2018; Takayama et al . , 2018 ) . These physiological correlates thus raise the notion of a key role of energetics in the evolution of regeneration in animals . Specifically , we wondered whether energy inputs can promote regenerative state . In this study , we demonstrate that nutrient supplementation can induce regenerative response in appendage and limb across three vastly divergent species . We reasoned that if there was an ancestral mechanism to promote regeneration , it would likely remain intact in early branching lineages with prevalent regeneration across the species . In Cnidaria , the ability to regenerate is established in polyps , for example , hydras and sea anemones . Some cnidarians , notably jellyfish , not only exist as sessile polyps , but also as free-swimming ephyrae and medusae ( Figure 1a ) . In contrast to the polyps’ ability to regenerate , regeneration in ephyrae and medusae appears more restricted in some species ( Abrams et al . , 2015; Sinigaglia et al . , 2020; Schmid and Alder , 1984 ) . We focused on the moon jellyfish Aurelia coerulea ( formerly A . aurita sp . 1 strain ) , specifically on the ephyra , whose eight arms facilitate morphological tracking ( Figure 1b ) . About 3 mm in diameter , Aurelia ephyrae regenerate the tips of arms and the distal sensory organ rhopalium , but upon more dramatic amputations such as removing a whole arm or halving the body , rapidly reorganize existing body parts and regain radial symmetry ( Figure 1c ) . Observed across four scyphozoan species , symmetrization occurs rapidly within 1–3 days and robustly across conditions ( Abrams et al . , 2015 ) . Ephyrae that symmetrized matured into medusae , whereas ephyrae that failed to symmetrize and simply healed the wound grew abnormally . Intriguingly , in the course of our previous study ( Abrams et al . , 2015 ) , we observed in a few symmetrizing ephyrae , a small bud at the amputation site . To follow this hunch , we repeated the experiment in the original habitat of our lab’s polyp population , off the coast of Long Beach , CA ( Materials and methods ) . Two weeks after amputation , most ephyrae indeed symmetrized , but in 2 out of 18 animals a small arm grew ( Figure 1c ) . This observation suggests that , despite symmetrization being the more robust response to injury , an inherent ability to regenerate arm is present and can be naturally manifest . The inherent arm regeneration presents an opportunity: Can arm regeneration be reproduced in the lab , as a way to identify factors that promote regenerative state ? To answer this question , we screened various molecular and physical factors ( Figure 2a , Figure 2—source data 1 ) . Molecularly , we tested modulators of developmental signaling pathways as well as physiological pathways such as metabolism , stress response , immune , and inflammatory response . Physically , we explored environmental parameters , such as temperature , oxygen level , and water current . Amputation was performed across the central body removing three arms ( Figure 2a ) . Parameter changes were implemented or molecular modulators ( e . g . , peptides and small molecules ) were introduced into the water immediately after amputation . Regenerative response was assessed for 1–2 weeks until the onset of bell growth , which hindered the scoring of arm regeneration ( Figure 2—figure supplement 1 ) . After 3 years of screening , only three factors emerged that strongly promoted arm regeneration ( Figure 2b ) . The ephyrae persistently symmetrized in the majority of conditions tested . In the few conditions where regeneration occurred , arm regenerates show multiple tissues regrown in the right locations: circulatory canals , muscle , neurons , and rhopalium ( Figure 2c–e ) . The arm regenerates contracted synchronously with the original arms ( Video 1 ) , demonstrating a functional neuromuscular network . Thus , arm regeneration in Aurelia that was observed in the natural habitat can be recapitulated in the lab by administering specific exogenous factors . The extent of arm regeneration varied , from small to almost fully sized arms ( Figure 2b ) . The variation manifested even within individuals: a single ephyra could grow differently sized arms . Of the three arms removed , if regeneration occurred , generally one arm regenerated ( 67% ) , occasionally two arms ( 32% ) , and rarely three arms ( 1% , of the 4270 total ephyrae quantified in this study ) . Finally , the frequency of regeneration varied across clutches , that is , strobilation cohorts . Some variability may be due to technical factors , for example , varying feed culture conditions; however , variability persisted even with the same feed batch . We verified that the variability was not entirely due to genetic differences , as it manifested across clonal populations ( Figure 2—figure supplement 2 ) . Thus , there appears to be stochasticity in the occurrence of arm regeneration in Aurelia and the extent to which regeneration proceeds . What are the factors that promote arm regeneration ? Notably , modulation of developmental pathways often implicated in regeneration literature ( e . g . , Wnt , Bmp , and Tgfß ) did not produce effect in the screen ( Figure 2—source data 1 ) —although we do not rule out their involvement in other capacity , for example , in downstream patterning . We first identified a necessary condition: water current . The requirement for current for promoting regeneration is interesting because ephyrae can recover from injury by symmetrizing in stagnant water ( Figure 1c ) . Thus , a specific physiological state is required for enabling regenerative response . Behaviorally , the presence of current promotes more swimming , while in stagnant water ephyrae tend to rest at the bottom and pulse stationarily ( Figure 3—figure supplement 1 and Video 2 show the aquarium setup used to implement current ) . In this permissive condition , the first factor that promotes regeneration is the nutrient level: increasing food amount increases the frequency of arm regeneration . To measure the regeneration frequency , we scored any regenerates with lengths greater than 15% of that of an uncut arm ( Figure 3a ) . This threshold was chosen to predominantly exclude non-specific growths or buds that show no morphological structures ( Figure 3b ) while including small arm regenerates that show clear morphological features , that is , lappets , radial canal , and radial muscle sometimes showing growing ends ( Figure 3b ) . Given the clutch-to-clutch variability , control and treatment were always performed side by side using ephyrae from the same clutch . The effect size of a treatment was assessed by computing the change in regeneration frequency relative to the internal control . Statistical significance of a treatment was assessed by evaluating the reproducibility of its effect size across independent experiments ( Materials and methods ) . With these measurement and statistical methodologies , we found that although the baseline regeneration frequency varied across clutches , higher food amounts reproducibly increased regeneration frequency ( Figure 3c ) . The magnitude of the increase varied ( Figure 3g , 95% confidence interval [CI] [4 . 7 , 12 . 1-fold] ) , but the increase was reproducible ( 95% CI excludes 1 ) and statistically significant ( p<10–4 ) . The second factor that promotes regeneration is insulin ( Figure 3d ) . We verified that the insulin receptor is conserved in Aurelia ( Figure 3—figure supplement 2 ) . Administering insulin led to a reproducible ( Figure 3g , 95% CI [1 . 1 , 5 . 0-fold] ) and statistically significant ( p<0 . 05 ) increase in regeneration frequency . The insulin effect was unlikely to be due to non-specific addition of proteins , since bovine serum albumin ( BSA ) at the same molarity showed no statistically significant effect ( Figure 3—figure supplement 4 ) . Finally , the third promoter of regeneration is hypoxia ( Figure 3e ) . We verified that the ancient oxygen sensor HIFα is present in Aurelia ( Figure 3—figure supplement 2 ) . Hypoxia led to a reproducible ( Figure 3g , 95% CI [1 . 4 , 12 . 0-fold] ) and statistically significant ( p<0 . 01 ) increase in regeneration frequency . To reduce oxygen , nitrogen was flown into the seawater , achieving ~50% reduction in dissolved oxygen level ( Materials and methods ) . We verified that the effect was due to reduced oxygen rather than increased nitrogen , since reducing oxygen using argon flow similarly increased regeneration frequency ( Figure 3—figure supplement 4 ) . The factors can act synergistically ( e . g . , insulin and high nutrient level ) , but the effect appears to eventually saturate ( e . g . , hypoxia and high nutrient level ) . In addition to quantifying the number of ephyrae that regenerate , we further quantified the regeneration phenotypes in each ephyra , that is , the number of arms regenerating , the length of arm regenerates , and the formation of rhopalia ( Figure 3—figure supplements 5 and 6 ) . Nutrient level strikingly improved all phenotypic metrics: not only more ephyrae regenerated in higher nutrients , more ephyrae regenerated multiple arms , longer arms , and arms with rhopalia . Insulin and hypoxia , interestingly , show differential phenotypes . Most strikingly , while insulin induced more ephyrae to regenerate multiple arms , hypoxia induced largely single-arm regenerates , for example , hypoxia experiments 3 and 5 in Figure 3—figure supplement 5c . Thus , while all factors increased the probability to regenerate , they had differential effects on the regeneration phenotypes , suggesting a decoupling to a certain extent between the regulation of the decision to regenerate and the regulation of the subsequent morphogenesis . Of the three factors identified in the screen , nutrient input is the broadest , and prompted us to search if a more specific nutritional component could capture the effects of full nutrients in promoting regeneration . Jellyfish are carnivorous and eat protein-rich diets of zooplanktons and other smaller jellyfish ( Graham , 2001 ) . Notably , all three factors induced growth: treated ephyrae are larger than control ephyrae ( Figure 3—figure supplement 7 ) . The growth effect is interesting because of essential amino acids that must be obtained from food , branched amino acids supplementation correlates positively with protein synthesis and growth , and in particular , L-leucine appears to recapitulate most of the anabolic effects of high amino acid diet ( Lynch and Adams , 2014; Stipanuk , 2008 ) . Motivated by the correlation between growth and increased regeneration frequency , we wondered if leucine administration could promote regeneration . Animals typically have a poor ability to metabolize leucine , such that the extracellular concentrations of leucine fluctuate with dietary consumption ( Wolfson et al . , 2016 ) . As a consequence , dietary leucine directly influences cellular metabolism . Feeding amputated ephyrae with leucine indeed led to increased growth ( Figure 3—figure supplement 6 ) . Assessing arm regeneration in the leucine-supplemented ephyrae , we observed a significant increase in the regeneration frequency ( Figure 3f–g , 95% CI [2 . 5 , 6 . 6-fold] , p<10–4 ) . Furthermore , leucine treatment phenocopies the effect of high nutrients , improving all measured phenotypic metrics: increasing multi-arm regeneration , the length of arm regenerate , and the frequency of rhopalia formation ( Figure 3—figure supplements 5 and 6 ) . These experiments demonstrate that abundant nutrients , the growth factor insulin , reduced oxygen level , and the amino acid L-leucine promote appendage regeneration in Aurelia ephyra . The identified factors are fundamental physiological factors across animals . Might the same factors promote appendage regeneration in other animal species ? To pursue this question , we searched for other poorly regenerating systems , which fortunately include most laboratory models . Drosophila , along with beetles and butterflies , belong to the holometabolans—a vast group of insects that undergo complete metamorphosis , and that as whole , do not regenerate limbs or other appendages as adults ( Das , 2015 ) . Larval stages have imaginal disks , undifferentiated precursors of adult appendages such as the legs and antennae , and portions of imaginal disks have been shown to regenerate ( Worley et al . , 2012 ) . Motivated by findings in Aurelia , we asked if leucine and insulin administration can induce regenerative response in the limb of adult Drosophila . We focused on testing leucine and insulin in this study because of considerations of specificity ( i . e . , nutrients are broad and composition of nutritional needs vary across species ) , pragmatism ( i . e . , administering hypoxia requires more complex setups ) , and in the case of Drosophila specifically , Drosophila being resistant to hypoxia ( Haddad et al . , 1997 ) . We amputated Drosophila on the hindlimb , across the fourth segment of the leg , the tibia ( Figure 4a ) . After amputation , flies were housed in vials with standard food ( control ) or standard food supplemented with leucine and insulin , with glutamine to promote leucine uptake ( Nicklin et al . , 2009 ) ( treated ) ( Figure 4c ) . Each fly was examined multiple times , twice in the first week , and then once weekly over the course of 2–4 weeks . No regrown tibia was found in the 860 control flies examined ( Figure 4b–c ) . In the treated flies , only tibia stumps were observed in the first week after amputation . But remarkably , at 7–21 days post-amputation ( dpa ) , a few regrown tibias were observed ( 1 . 0% , N=387; Figure 4d–e ) . The regrown tibias culminate in reformed joints , articulating from which appears to be the beginning of a next . segment . Control tibia stumps showed melanized clots at the tips within 1–3 dpa ( Figure 4b ) , as expected from normal wound healing process ( Rämet et al . , 2002 ) , while some treated tibias showed no clot ( 12 . 1% , N=387 ) and the tips stained positively with DAPI instead ( 14 out of 16 tibias examined; Figure 4f ) . Induction of regenerative response was observed across genetic backgrounds , in Oregon R , as discussed , and Canton S wild-type strains ( Figure 4—figure supplement 1a ) . Reminiscent of Aurelia , not all regenerative response was patterned , few flies showed non-specific outgrowth ( Figure 4—figure supplement 1b ) . The bulk experiment enables us to assess a large number of flies and can capture dramatic regenerative response , that is , a fully regrown leg segment . But it can miss partial segment regrowth and therefore underestimate the extent of regenerative response . To track regenerative response in individual flies , we housed in each vial a small number of flies that were amputated in different limbs ( Figure 5a ) . The amputated limb position , combined with sex , enabled unique identification of each fly within a vial . Each fly was imaged immediately after amputation and 1–3 additional times over the course of 2–4 weeks . The imaging frequency balanced obtaining time-lapse information with minimizing stress from repeated anesthesia . The single-fly tracking showed pervasive regenerative response induced by nutrient supplementation . Time-lapse pictures of select control and treated flies are shown in Figure 5b–d . For each tibia stump , we measured the length over time ( Figure 5e ) . Control tibia stumps showed near-zero percent change in length ( Figure 5d; mean –0 . 3% , 95% CI [–3 . 8 , 3 . 2%] , N=116 ) . By contrast , 49% of treated flies showed growth beyond the 95% CIs of the control distribution ( Figure 5f; N=150 ) . The rate of regrowth varied; some flies showed large growth during the first 1–2 weeks post-amputation , while others regrow more slowly over 2–3 weeks ( Figure 5c ) . Interestingly , four flies ( 2 . 7% ) showed shortened tibia stumps , which may indicate roles of histolysis in regenerative response . Regenerative response was not only observed from amputations across the tibia , but also from more dramatic amputations across the femur ( Figure 5d ) . We verified that the control and treated length change distributions are statistically different ( p<0 . 0001* , nonparametric Kruskal-Wallis test ) . No sex-based differences in response were observed ( Figure 5—figure supplement 1a ) . In some limbs , the new growth already looks almost indistinguishable from the rest of the leg . In some limbs , the new growth showed what may be intermediate morphologies . For instance , some new growths are pigmented differently and/or have no sensory bristles ( Figure 5g , Figure 5—figure supplement 1d , e , ) or showed white tissues protruding from the end ( e . g . , Figure 5—figure supplement 1c ) , which was remodeled over time ( e . g . , Figure 5—figure supplement 1d ) . Finally , reproducing what was observed in the bulk experiments , five flies showed reformed tibia segments: showing a tapering end ( N=150; Figure 5h ) , what appeared to be a joint ( Figure 5i ) , and the beginning of the next segment ( Figure 5j ) . A scanning electron micrograph of a regrown hindlimb tibia ( the top tibia in Figure 4e , taken 1 week later ) morphologically confirms the regenerated joint as a tibial/tarsal joint . The joint-like structure shows the expected bilateral symmetry of a tibial/tarsal joint ( as opposed to , e . g . , the radially symmetrical tarsal/tarsal joints; Mirth and Akam , 2002 ) with rounded projections at the posterior and anterior end ( arrows in Figure 5k ) . These projections , called condyles , function as points of articulation between opposing leg segments . Indeed , articulating from the regrown condyles appears to be further growth of the next tarsal segment . Finally , a unique feature of the tibial/tarsal joint of the hindlimb ( but not of fore or midlimb ) is an additional ventral projection between the side condyles ( Mirth and Akam , 2002 ) , which serves to restrain bending of the leg upward . The ventral projection is indeed present in the regenerated joint ( arrow in Figure 5k ) . Altogether , these data demonstrate for the first time that patterned regenerative response can be induced from adult Drosophila limbs . The ability of leucine and insulin to induce regenerative response in Drosophila limb and Aurelia appendage motivated testing in vertebrates . One sign that limb regeneration may be feasible in humans is that fingertips regenerate ( Illingworth , 1974 ) . The mammalian model for studying limb regeneration is the house mouse , Mus musculus , which like humans , regenerates digit tips . Although proximal regions of digits do not regenerate , increasing evidence suggests that they have inherent regenerative capacity . In adult mice , implanting developmental signals in amputated digits led to specific tissue induction , that is , bone growth with Bmp4 or joint-like structure with Bmp9 ( Yu et al . , 2019 ) . In neonates , reactivation of the embryonic gene lin28 led to distal phalange regrowth ( Shyh-Chang et al . , 2013 ) . Thus , while patterned phalange regeneration can be induced in newborns , induction in adults so far involves a fine-tuned stimulation , for example , to elongate bone and then make joint , Bmp4 was first administered followed by Bmp9 in a timed manner . Motivated by the findings in Aurelia and Drosophila , we tested if leucine and insulin administration could induce a self-organized regeneration in adult mice . We performed amputation on the hindpaw ( Figure 6a ) , on digits 2 and 4 , leaving the middle digit three as an internal control ( Figure 6b ) . To perform non-regenerating amputation , a clear morphological marker is the nail , which is associated with the distal phalange ( P3 ) . Amputation that removes <30% of P3 length , that cuts within the nail , readily regenerates , whereas amputation that removes >60% of P3 length , corresponding to removing almost the entire visible nail , does not regenerate ( Figure 6c; Chamberlain et al . , 2017; Lehoczky et al . , 2011 ) . We therefore performed amputations entirely proximal to the visible nail—giving , within the precision of our amputation , a range of cut across somewhere between the proximal P3 and the distal middle phalange ( P2 ) ( Figure 6d ) —a range that is well below the regenerating tip region . Note additional morphological markers that lie within the non-regenerating region: the os hole ( ‘o’ in Figure 6c ) , where vasculatures and nerves enter P3 , the bone marrow cavity ( ‘m’ in Figure 6c ) , and the sesamoid bone ( ‘s’ in Figure 6c ) adjacent to P2 . The digit portion removed was immediately fixed to determine the precise plane of amputation . The amputated mice were either provided with water as usual ( control ) or water supplemented with leucine and sucrose ( treated ) ( Figure 6e ) . Both groups were monitored for 7–8 weeks . Sucrose was used because insulin is proteolytically digested in the mammalian gut . The sucrose doses used are lower or the administration duration is shorter than those shown to induce insulin resistance ( Cao et al . , 2007; Togo et al . , 2019 ) . We verified that control and treated mice had comparable initial weights ( 35 . 1±0 . 6 vs . 34 . 1±1 . 1 g , p=0 . 402 , Student’s t-test ) , and that as expected from amino acid and sugar supplementation , treated mice gained more weight over the experimental duration ( 4 . 5±1 . 0 vs . 7 . 8±1 . 0 g , p=0 . 028 , Student’s t-test ) . As expected for amputation proximal to the nail , no regeneration was observed in the control mice ( N=34 digits , 17 mice ) . Amputated digits healed and re-epithelialized the wound as expected ( Figure 6f ) . Skeletal staining shows blunt-ended digit stumps ( Figure 6i ) and in many instances , as expected , dramatic histolysis , a phenomenon where bone recedes further from the amputation plane ( Figure 6—figure supplement 1; Chamberlain et al . , 2017 ) . By contrast , 18 . 8% of the treated digits ( N=48 digits , 24 mice ) showed various extents of regenerative response ( Figure 6—figure supplement 1 ) . The increase in regeneration frequency due to the treatment is statistically significant ( 95% CI [8 , 30%] , p=0 . 0019 , , Student’s t-test ) . We observed , as in Aurelia and Drosophila , an unpatterned response ( Figure 6—figure supplement 1 ) , wherein skeletal staining reveals excessive bone mass around the digit stump , similarly to what was observed in some cases with BMP stimulation ( Yu et al . , 2019 ) . However , we also observed patterned responses ( Figure 6—figure supplement 2 ) . The most dramatic regenerative response was observed in two digits ( Figure 6g–h ) . In one digit , an almost complete regrowth of the distal phalange and the nail was observed ( Figure 6g ) . Skeletal staining of the portion removed from this digit ( Figure 6j ) shows that it was amputated at the proximal P3 transecting the os hole . By 7 weeks , skeletal staining of the regrown digit ( Figure 6j ) shows that the P3 bone was almost completely regrown . The regrown P3 shows trabecular appearance that is similar in general structure but not identical to the original P3 . Another dramatic response was observed from another digit , which began reforming the nail by 7 weeks ( Figure 6h ) . Skeletal staining of the portion removed from this digit shows that it was amputated across the P2 bone , removing the entire epiphyseal cap along with the sesamoid bone ( Figure 6k ) . Skeletal staining of the regenerating digit shows that the epiphyseal cap was regrown , along with its associated sesamoid bone . Moreover , articulating from the regenerated P2 appears to be the beginning of the next phalangeal bone ( arrow , Figure 6k ) . To our knowledge , the regenerative response observed in these digits represents the most dramatic extent of self-organized mammalian digit regeneration reported thus far . Distal phalange regeneration in adults has not been reported , while interphalangeal joint formation from a P2 amputation has been achieved only through sequential Bmp administration ( Yu et al . , 2019 ) and there has been no documentation of the regrowth of the sesamoid bone . In this study , amputations were performed on Aurelia appendage , Drosophila limb , and mouse digit . None of these animals are known to regenerate robustly ( Aurelia ) if at all ( Drosophila and mouse ) from these amputations . Upon administration of L-leucine and sugar/insulin , dramatic regenerative response was observed in all systems . The conserved effect of nutrient supplementation across three species that span more than 500 million years of evolutionary divergence suggests energetic parameters as ancestral regulators of regeneration activation in animals . While we did not test the regenerative effect of hypoxia beyond Aurelia , it is notable that in mice hypoxia coaxes cardiomyocytes to re-enter cell cycle ( Kimura et al . , 2015 ) and activate HIF1α promotes healing of ear hole punch injury ( Zhang et al . , 2015 ) . Notably in Aurelia , the amputation bisected through the body , and more than appendage was in fact regenerated , for example , circular muscle in the body is regrown ( Figure 2e ) . Thus , nutrient supplementation may have regenerative effect in body parts beyond appendage . The diverse physiologies of animals across phylogeny may seem difficult to reconcile with a conserved regulation of regeneration activation , especially in the view of regeneration as recapitulation of development . Growing a jellyfish appendage is different from building a fly leg or making a mouse digit . However , there is another way of looking at regeneration as a part of tissue plasticity ( Galliot and Ghila , 2010 ) . In this view of regeneration , upstream from tissue-specific morphogenesis is a conserved regulation of cell growth and proliferation . In support of this idea , early steps in regeneration across species and organs rely one way or another on proliferation by stem cells or differentiated cells re-entering cell cycle ( Cox et al . , 2019 ) . We propose that in animals that poorly regenerate , high nutrient input turns on growth and anabolic states that promote tissue rebuilding upon injury . That regenerative response can be induced seemingly blurs the boundary between regenerating versus non-regenerating animals , because the factors identified in the study are not exotic . Variations in amino acids , carbohydrates , and oxygen levels are conditions that the animals can plausibly encounter in nature . These observations highlight two potential insights into regeneration . First , regeneration is environmentally dependent . An animal would stop at wound healing under low-energy conditions and regenerate in energy-replete conditions . In this view , for the animals examined in this study , the typical laboratory conditions may simply not be conducive to regeneration . Alternatively , the interpretation we favor , what we observed is inherent regeneration , which can be activated with broad environmental factors . We favor this interpretation because the regenerative response was unusually variable . The variability stands in stark contrast to the robust regeneration in , for example , axolotl , planaria , or hydra . Whereas wild-type processes tend to be robust , mutations produce phenotypes that are sensitive to variations in physiological parameters . Thus , just like mutant phenotypes show varying penetrance and expressivity , the variable regenerative response speaks to us as a fundamental consequence of activating a latent biological module . In this interpretation , the ordinariness of the activators suggests ancestral regeneration as part of a response to broad environmental stimuli . It would be interesting next to identify individual differences that contribute to varying propensities to mount regenerative response . In particular , the conserved effects of nutrient supplementation suggest that regeneration might have originally been a part of growth response to abundant environments . No nutrient dependence has been observed in highly regenerating animal models such as planaria , hydra , and axolotl . Environment-dependent plasticity , however , is pervasive in development , physiology , behavior , and phenology ( West-Eberhard , 2003; Moczek et al . , 2011 ) . We therefore conjecture that environment-dependent plasticity may have characterized the ancestral form of regeneration . In this conjecture , present regenerating lineages might have decoupled the linkage with environmental input and genetically assimilated regenerative response—because regeneration is adaptive or coupled to a strongly selected process , for example , reproduction . In parallel , non- or poorly regenerating animals might have also weakened the linkage with environmental input , but to silence the regenerative response . This predicts an ancient form of a robustly regenerative animal ( like planaria , hydra , and axolotl ) that tunes its regeneration frequency to nutrient abundance . Such plasticity has been reported in the basal lineage Ctenophora ( Bading et al . , 2017 ) . In conclusion , this study suggests that an inherent ability for appendage regeneration is retained in non-regenerating animals and can be unlocked with a conserved strategy . The treatments across species were not exactly identical , and correspondingly there might be differences in the precise molecular mechanisms—in spite of which they could be applied across species in a predictive manner . In line with our findings , the role of nutrients in promoting regeneration was reported in yet another species ( in the Xenopus tadpoles , Williams et al . , 2021 ) . While the observed regenerative response is not perfect , this motivates further investigation into potentially more promoting factors or the possibility of combining broad promoting factors with species- or tissue-specific morphogenetic regulators . Reiterating Spallanzani’s hope , Marcus Singer supposed half a century ago that ‘ . . . every organ has the power to regrow lying latent within it , needing only the appropriate “useful dispositions” to bring it out ( Singer , 1958 ) . ’ The surprise , in hindsight , is the simplicity by which the regenerative state can be promoted with ad libitum amino acid and sugar supplementation . This simplicity demonstrates a much broader possibility of organismal regeneration , and can help accelerate progress in regeneration induction across animals . The polyp population in the study arose from parental polyps collected off the coast of Long Beach , CA ( 33°46′04 . 2″N 118°07′44 . 2″W , GPS: 33 . 7678376–118 . 1289559 ) . Ephyrae were amputated in location and immediately after submersed in the ocean . For submerging the amputated ephyrae in the ocean , a two-layered aquarium was custom-built . Ephyrae were placed in plastic canisters with a 7-cm diameter hole cut in the lid and covered with a 250-μm plastic screen . The canisters were then placed in a thick plastic tank fitted with a 500-μm plastic screen on top . This design offers protection to the ephyrae against predators and strong waves , while at the same time allowing exchange of water , zooplanktons , and other particulates . Ephyrae were collected after 2 weeks . All experiments were performed at 68°F . Two to three days old ephyrae were anesthetized in 400 μM menthol and amputated using a razor blade mounted on an x-acto knife handle . Amputated ephyrae were let to recover in 1 L sand settling cones ( Nalgene Imhoff , Figure 3—figure supplement 1 ) . In each experiment , ~90 animals were amputated for each condition ( e . g . , 90 animals for control and 90 animals for treated ) . Because of the varying baseline across strobilation batches , each experiment was repeated across 2–5 strobilation batches ( biological replicates ) . These sample sizes were chosen to obtain a 95% confidence level on the treatment effect ( statistical analysis described below ) . Hundreds of experimental animals were first amputated , mixed together in a beaker , and then randomly allocated to the control or treatment groups . Regeneration was assessed at various times for 1–2 weeks after amputation , before onset of maturation to medusa . All data were included in the analysis . Among the possible amputation schemes , 3-arm amputation was chosen because it could be performed fastest . Removing one arm requires carefully cutting across the base of the arm while avoiding injuring the surrounding body . Removing two arms is less hard but still requires awkward positioning of the knife . Removing four arms again takes more time because it requires cutting through the large protruding manubrium , which also affects the animal’s feeding ability . The fast 3-arm amputation facilitates testing hundreds of ephyrae per experiment . Amputated ephyrae were fed daily with rotifers . The number of rotifers was estimated using a six-well plate fitted with STEMgrid ( the same principle as using a hemocytometer ) . In this study , low food was ~10–20 rotifers/ephyra and high food was ~40 rotifers/ephyra . To replicate the study , these numbers should only be used as initial estimates , as what is ‘low’ or ‘high’ food amount may easily vary across lab cultures ( e . g . , rotifer culture , differences across Aurelia strains , etc . ) . Most if not all rotifers were typically consumed within an hour ( determined by measuring the rotifers in the water ) . Immediately after amputation , ephyrae were placed in ASW supplemented with 500 nM human recombinant insulin ( Sigma-Aldrich I0908 ) . Insulin was refreshed weekly . To determine the concentration used , a range of concentrations , 10 nM to 3 mM , were tested . The concentration 500 nM was chosen as it maximized regeneration frequency while avoiding solubility problems . To control that the effect of insulin was not due to non-specific additions of proteins , BSA at 500 nM was tested . Immediately after amputation , ephyrae were placed in hypoxic ASW . To create a hypoxic environment , nitrogen or argon , instead of ambient air , was pumped into the bubbler cone , beginning from the day before the experiment and maintained throughout the duration of the experiment . The bubbler cone was sealed with parafilm to maintain the lowered oxygen level . The nitrogen/argon flow was adjusted to achieve 50% reduction in the dissolved oxygen level . Dissolved oxygen level was measured using a Clark-type electrode Unisense OX-500 microsensor . The measurement was normalized to oxygen level in control ASW bubbled normally with ambient air . Oxygen measurement was performed prior to the experiment and subsequently every 3 days . Immediately after amputation , ephyrae were placed in ASW supplemented with 100 μM L-leucine ( Sigma-Aldrich L1002 , the cell-permeable methyl ester hydrochloride form ) . L-leucine was refreshed weekly . To determine the concentration used , a range of concentrations from one to hundreds of mM was tested . The concentration of 100 mM was chosen as it maximized the regeneration frequency without non-specific , negative effects . To assess the statistical significance of the treatments , meta-analysis of effect size was performed ( Borenstein et al . , 2009 ) . The effect size metrics used are determined by the form of the data set . For measurements of frequencies ( e . g . , regeneration frequency ) , the data sets are in the form of a 2×2 table of dichotomous variables . For such 2×2 data sets , in situations where the baseline varies ( e . g . , varying baseline regeneration across clutches ) , the commonly used measures of effect size are the RR , RR= # ephyrae that regeneratetotal # ephyraein treated group# ephyrae that regeneratetotal # ephyraein control group=cc+daa+b and the OR , OR= # ephyrae that regenerate# ephyrae that do not regeneratein treated group# ephyrae that regenerate# ephyrae that do not regeneratein control group=cc+daa+b RR compares the probability of an outcome in treated versus control group , whereas OR compares the odds of an outcome in treated versus control group . For measurements of arm length and body size , the data sets are in the form of continuous variables . For such data , the commonly used effect size is the Response Ratio ( R ) , R= mean arm length in treated groupmean arm length in control group R evaluates the proportionate change that results from a treatment , and is the meaningful effect size to use when the outcome of a treatment is measured on a physical scale , for example , length or area ( as opposed to arbitrary scale , e . g . , happiness level ) . Experiments where regeneration in one of the groups occurred in 0 ephyra were necessarily excluded . Having computed the effect size ( RR , OR , or R ) within each experiment , meta-analysis of the effect size across experiments was performed . The metafor package ( Viechtbauer , 2010 ) in R was used , with fixed-effect model ( for nutrients and leucine ) or random-effect restricted maximum likelihood model ( for insulin and hypoxia , which had different control conditions across the experiments ) . Statistical coefficients were based on normal distribution . All steps were performed at room temperature , unless indicated otherwise . Ephyrae were first anesthetized in 400 μM menthol , which minimizes curling during fixing . Next , ephyrae were fixed in 3 . 7% ( v/v ) formaldehyde ( in phosphate-buffered saline [PBS] ) for 15 min , permeabilized in 0 . 5% Triton X-100 ( in PBS ) for 5 min , and blocked in 3% ( w/v ) BSA for 2 min . For neuron staining , ephyrae were incubated in 1:200 mouse anti-tyrosinated alpha tubulin antibody ( Sigma-Aldrich MAB1864-I ) overnight at 4°C , and then in 1:200 goat-anti-mouse Alexa Fluor 488 ( Life Technologies A11029 ) overnight in the dark at 4°C . Primary or secondary antibodies were diluted in 3% BSA . For actin staining , ephyrae were incubated in 1:20 Alexa Fluor 555 Phalloidin ( Life Technologies A12379 ) overnight or for 2 hr in the dark at 4°C . For nuclei staining , ephyrae were incubated in 1:10 Hoechst 33342 ( Sigma-Aldrich B2261 ) for 30 min in the dark . Ephyrae were imaged anesthetized in menthol . Brightfield images , fluorescent images , and movies were taken with the Zeiss AxioZoom . V16 stereo zoom microscope and AxioCam HR 13-megapixel camera . Optical sectioning was performed with ApoTome . 2 . CantonS wild-type strain was a gift from Peter Lee in Kai Zinn’s lab at Caltech . OregonR was a gift from James McGehee in Angela Stathopolous’ lab at Caltech . OregonR and CantonS flies were reared under standard conditions at 23°C , sometime supplemented with baker’s yeast . Amputation was performed on adult flies 2–7 days after eclosion . Flies were anesthetized with CO2 , placed under a dissection microscope , and tibia amputated using a spring scissors ( Fine Science Tools , 91500-09 ) and superfine dissecting forceps ( VWR , 82027-402 ) . See Figure 4 for detailed description of the amputation plane . Recovering Drosophila were allocated randomly to vials with standard lab fly food ( control ) or standard lab fly food mixed with 5 mM L-Leucine ( Sigma-Aldrich L8000 ) , 5 mM L-Glutamine ( Sigma-Aldrich G3126 ) , and 0 . 1 mg/ml insulin ( human recombinant , MP Biomedicals 0219390080 ) . To introduce the nutrient supplements , the fly food was microwaved in short pulses , such that the topmost layer of the food was liquified . The supplements in aqueous stocks were then pipetted into this liquified layer . Food was allowed to re-set at 4°C for at least 20 min . New food was prepared fresh every 2 days , and flies were moved into freshly prepared treated food every 2 days , throughout the course of the 2- to 3-week experiment . Amputation and treatment were performed as in the bulk regeneration experiments described above , with the following modifications . Canton S flies were amputated in different limbs , and were housed in small groups such that in any given vial , each fly was uniquely identifiable by sex and amputated limb . Typically , 1–6 flies are housed in each vial . 3–5 days treatment produced similar response as sustained treatment; therefore , for simplicity , treatment was performed for 3–5 days . Treated flies , but not control flies , were sometime continued on yeast supplementation . Flies were imaged immediately after amputation and 1–3 additional times over the course of 2–4 weeks . As anesthetized flies jitter , images were taken in a video format , and single frames were then selected for analysis in which the leg stump was in focus . Tibia length was quantified in ImageJ as the diameter of the minimum enclosing circle of the leg to achieve rotation-invariant assessments . Identical imaging and analysis procedures were used for treated and control flies . Blind measurements were performed on one pair of control and treated datasets . Statistical comparison of percent change in length in control and treated leg stumps was performed using the non-parametric Kruskal-Wallis test . Samples were assigned integer ranks from smallest to largest by quantity , and then differences between conditions were assessed with respect to the ranks . The p-value tests the null hypothesis that the data are drawn from the same distribution . Fly tibias were dissected and washed in 70% ethanol ( <1 min ) to decrease the hydrophobicity of the cuticle and washed in PBS with 0 . 3% Triton-X for 10 min . The legs were fixed in 4% paraformaldehyde ( in PBS ) overnight at 4°C and washed five times for 20 min each in PBS with 0 . 3% Triton-X . The legs were equilibrated in Vectashield mounting medium with DAPI ( Vector H-1200 ) overnight at 4°C , and imaged using Zeiss AxioZoom . V16 stereo zoom microscope with AxioCam HR 13-megapixel camera . Confocal imaging was performed using X-Light V2 spinning disk mounted on the Olympus IX81 inverted microscope . Flies anesthetized on a CO2 bed were imaged under a Zeiss SteREO Discovery . V8 stereomicroscope equipped with the Zeiss AxioCam 503 color camera . Environmental scanning electron microscopy ( ESEM ) was performed on a FEI Quanta 200F ( FEI , Hillsboro , OR ) . Whole-live flies were mounted onto the SEM stub with copper tape . ESEM images were attained at a pressure of 0 . 1 mbar and 5 kV at a working distance of 9–12 mm , with water as the ionizing gas . All studies comply with relevant ethical regulations for animal testing and research , and received ethical approval by the Institutional Animal Care and Use Committees at the California Institute of Technology . Adult female ( 3–6 months old ) wild-type CD1 mice ( Charles River Laboratories strain 022 ) were used for all regeneration studies . Digit amputation was performed following the established protocol in the field ( Simkin et al . , 2013 ) . Mice were anesthetized with 1–5% isoflurane ( in oxygen ) in an induction chamber , followed by maintenance on a nosecone . The mouse was positioned on its belly with its hind paws outstretched and the ventral side of the paw facing upwards . Sustained-Release Buprenorphine was administered ( Buprenorphine SR LAB ) at 0 . 5 mg/kg subcutaneously as an analgesic . Blood flow to the hindlimb was stemmed by tying a rubber band around the ankle and clamping it with a hemostat . All surgical procedures were carried out under a Zeiss Stemi 305 dissection microscope . An initial incision , parallel to the position of foot , was made through the ventral fat pad using Vannas spring scissors ( World Precision Instruments , 14003 ) . The length of this incision was determined by the amount of ventral skin needed to seal the digit amputation wound completely . The ventral skin freed in the initial incision was peeled back using surgical forceps , and a no . 10 scalpel ( Sklar , 06-3110 ) was used to amputate and bisect the digit completely through the second or third phalange . Digits 2 and 4 on the right hind paw were operated on in this fashion , while digit three remained unamputated as an internal control . The amputation wound was immediately closed with the ventral skin flap and sealed with GLUture ( Zoetis , Kalamazoo , MI ) . Amputated portions were immediately fixed as control for skeletal staining . Dissolved 1 . 5% L-leucine ( USP grade , VWR E811 ) , 1 . 5% L-glutamine ( USP grade , Sigma-Aldrich G8540 ) , and 4–10% sucrose ( AR ACS grade , Avantor 8360 ) in drinking water was administered to mice in the experimental group ad libitum after amputation . Control mice were given untreated drinking water . Drinking water was refreshed weekly for both control and experimental groups , and treated water was made fresh on the day that drinking water was replaced . The amputated digit stumps were photographed weekly for 7–8 weeks , at which time the digits were dissected for skeletal staining . The sample size in the experiment balanced the aim of achieving >90% confidence level with ethical consideration of minimizing the number of animals used . Animals were randomly allocated to the control or treatment group . No restricted randomization was applied . For weight measurement , the unit of analysis is a single animal . For regeneration phenotype , the unit of analysis is a single digit . Student’s t-test was used to evaluate the null hypothesis that there is no difference between the control and treated groups . 95% CIs were computed assuming normal distribution . All data were included in the analysis . Mice were euthanized and digits 2 , 3 , and 4 were removed with a no . 10 scalpel ( Sklar , 06-3110 ) through the first phalange . Excess skin and flesh were removed with spring scissors ( Fine Science Tools , 91500-09 ) and fine dissecting forceps ( Fine Science Tools , 11254-20 ) . All digits analyzed by whole-mount skeletal stains were prepared with a standard alizarin red and alcian blue staining protocol ( McLeod , 1980 ) . Digits were dehydrated in 95% ethanol for 1 day , and incubated in staining solution ( 0 . 005% alizarin red ( Beantown Chemical , BT144735 ) , 0 . 015% alcian blue ( Acros Organics , AC40046-0100 ) , 5 % acetic acid , 60 % ethanol ) for 1 day at 37 °C . Tissue was cleared in 2 % potassium hydroxide at room temperature for 1 day , 1% potassium hydroxide for 1 day , and then taken through an increasing glycerol series ( 25% , 50% , 75% , and 100% ) . The stained samples were imaged on Zeiss AxioZoom . V16 stereo zoom microscope with a Zeiss AxioCam 503 color camera or a Zeiss Stemi 305 dissection microscope with an iPhone six camera .	The ability of animals to replace damaged or lost tissue ( or ‘regenerate’ ) is a sliding scale , with some animals able to regenerate whole limbs , while others can only scar . But why some animals can regenerate while others have more limited capabilities has puzzled the scientific community for many years . The likes of Charles Darwin and August Weismann suggested regeneration only evolves in a particular organ . In contrast , Thomas Morgan suggested that all animals are equipped with the tools to regenerate but differ in whether they are able to activate these processes . If the latter were true , it could be possible to ‘switch on’ regeneration . Animals that keep growing throughout their life and do not regulate their body temperatures are more likely to be able to regenerate . But what do growth and temperature regulation have in common ? Both are highly energy-intensive , with temperature regulation potentially diverting energy from other processes . A question therefore presents itself: could limb regeneration be switched on by supplying animals with more energy , either in the form of nutrients like sugars or amino acids , or by giving them growth hormones such as insulin ? Abrams , Tan , Li et al . tested this hypothesis by amputating the limbs of jellyfish , flies and mice , and then supplementing their diet with sucrose ( a sugar ) , leucine ( an amino acid ) and/or insulin for eight weeks while they healed . Typically , jellyfish rearrange their remaining arms when one is lost , while fruit flies are not known to regenerate limbs . House mice are usually only able to regenerate the very tip of an amputated digit . But in Abrams , Tan , Li et al . ’s experiments , leucine and insulin supplements stimulated limb regeneration in jellyfish and adult fruit flies , and leucine and sucrose supplements allowed mice to regenerate digits from below the second knuckle . Although regeneration was not observed in all animals , these results demonstrate that regeneration can be induced , and that it can be done relatively easily , by feeding animals extra sugar and amino acids . These findings highlight increasing the energy supplies of different animals by manipulating their diets while they are healing from an amputated limb can aid in regeneration . This could in the future pave the way for new therapeutic approaches to tissue and organ regeneration .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "evolutionary", "biology" ]	2021	A conserved strategy for inducing appendage regeneration in moon jellyfish, Drosophila, and mice
DNA polymerase θ ( Polθ ) promotes insertion mutations during alternative end-joining ( alt-EJ ) by an unknown mechanism . Here , we discover that mammalian Polθ transfers nucleotides to the 3’ terminus of DNA during alt-EJ in vitro and in vivo by oscillating between three different modes of terminal transferase activity: non-templated extension , templated extension in cis , and templated extension in trans . This switching mechanism requires manganese as a co-factor for Polθ template-independent activity and allows for random combinations of templated and non-templated nucleotide insertions . We further find that Polθ terminal transferase activity is most efficient on DNA containing 3’ overhangs , is facilitated by an insertion loop and conserved residues that hold the 3’ primer terminus , and is surprisingly more proficient than terminal deoxynucleotidyl transferase . In summary , this report identifies an unprecedented switching mechanism used by Polθ to generate genetic diversity during alt-EJ and characterizes Polθ as among the most proficient terminal transferases known . DNA polymerases ( Pols ) are essential for life since they are necessary for the propagation and maintenance of genetic information . Intriguingly , bacterial and eukaryotic cells encode for multiple different types of Pols , some of which are intrinsically error-prone due to their relatively open active sites which enables them to tolerate particular DNA lesions ( Foti and Walker , 2010; Lange et al . , 2011; Sale et al . , 2012; Waters et al . , 2009 ) . Such enzymes are referred to as translesion polymerases and are mostly among the Y-family of polymerases ( Foti and Walker , 2010; Lange et al . , 2011; Sale et al . , 2012; Waters et al . , 2009 ) . Although these specialized polymerases are necessary for DNA damage tolerance , they are generally error-prone , and therefore must be highly regulated to prevent unnecessary mutations that can lead to genome instability and tumorigenesis ( Lange et al . , 2011; Sale et al . , 2012; Waters et al . , 2009 ) . The unique A-family DNA polymerase θ ( Polθ ) , encoded by the C-terminal portion of POLQ , tolerates bulky lesions like Y-family polymerases and is therefore also referred to as a translesion polymerase ( Hogg et al . , 2011; Seki et al . , 2004 ) . However , in contrast to Y-family polymerases , Polθ is capable of replicating past the most lethal type of lesion , the double-strand break ( DSB ) ( Chan et al . , 2010; Kent et al . , 2015; Koole et al . , 2014; Mateos-Gomez et al . , 2015; Yousefzadeh et al . , 2014 ) . For example , in recent studies we demonstrated the ability of the polymerase domain of POLQ , herein referred to as Polθ , to perform microhomology-mediated end-joining ( MMEJ ) —also referred to as alternative end-joining ( alt-EJ ) —in the absence of any co-factors ( Kent et al . , 2015 ) . MMEJ requires the ability of the polymerase to perform DNA synthesis across a synapse formed between two opposing single-strand DNA ( ssDNA ) overhangs containing sequence microhomology ( Figure 1A ) ( Kent et al . , 2015 ) . ssDNA overhangs are formed by partial resection of DSBs via Mre11-Rad50-Nbs1 ( MRN complex ) and CtIP , and potentially other factors ( Lee-Theilen et al . , 2011; Truong et al . , 2013; Zhang and Jasin , 2011 ) . Specifically , Polθ was shown to generate MMEJ products by promoting DNA synapse formation of 3’ ssDNA overhangs containing a minimal amount ( ≥2 base pairs ( bp ) ) of sequence microhomology , then using the opposing ssDNA overhang as a template in trans to extend the DNA , resulting in stabilization of the end-joining intermediate ( Figure 1A ) ( Kent et al . , 2015 ) . The polymerase then likely extends the second overhang resulting in gap filling ( Figure 1A ) . Ligase III ( Lig3 ) is required to seal the DNA junction formed during alt-EJ/MMEJ ( Audebert et al . , 2004; Simsek et al . , 2011 ) , presumably after other enzymes such as endonucleases further process the end-joining intermediate ( Figure 1A ) . This end-joining activity appears to be dependent on a unique insertion motif , called insertion loop 2 , that also enables Polθ to bypass of other types of DNA lesions ( Hogg et al . , 2011; Kent et al . , 2015 ) . Thus , although Polθ is an A-family polymerase , which normally exhibit high-fidelity DNA synthesis and lack translesion synthesis activity , its unique sequence composition confers end-joining , translesion synthesis , and low-fidelity DNA synthesis activities ( Arana et al . , 2008; Hogg et al . , 2011; Kent et al . , 2015; Seki et al . , 2003; 2004 ) . 10 . 7554/eLife . 13740 . 003Figure 1 . Polθ exhibits robust template-independent terminal transferase activity in the presence of manganese . ( A-C ) Models of Polθ dependent DNA end-joining . ( A ) Polθ uses existing sequence microhomology to facilitate DNA end-joining . ( B ) Polθ is proposed to extend ssDNA by a template-independent mechanism , then use the newly generated sequence to facilitate DNA end-joining . ( C ) Polθ is proposed to extend ssDNA by using the opposing overhang as a template in trans , then after DNA synapse dissociation Polθ uses the newly generated sequence to facilitate DNA end-joining . ( D ) A denaturing gel showing Polθ extension of poly-dC ssDNA in the presence of indicated dNTPs and 10 mM Mg2+ . ( E , F ) Denaturing gels showing Polθ extension of poly-dC ssDNA in the presence of dTTP and indicated divalent cation concentrations ( E ) and time intervals and temperatures ( F ) . ( G ) Denaturing gels showing Polθ extension of indicated ssDNA in the presence of all four dNTPs and 10 mM Mg2+ or 5 mM Mn2+ . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 00310 . 7554/eLife . 13740 . 004Figure 1—figure supplement 1 . Polθ template-independent activity is stimulated by physiological concentrations of Mn2+ and Mg2+ . ( A ) Denaturing gels showing Polθ extension of poly-dT in the presence of dCTP with indicated concentrations of Mn2+ and Mg2+ . ( B ) Plots of percent ssDNA extension observed in panel A . Percent extension was calculated by dividing the intensity of the sum of the extended products by the sum of the intensity of all DNA in each lane . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 00410 . 7554/eLife . 13740 . 005Figure 1—figure supplement 2 . Optimization of Polθ-Mn2+ template-independent terminal transferase activity . ( A-E ) Denaturing gels showing Polθ-Mn2+ extension of poly-dC ssDNA in the presence of dTTP with indicated [Mn2+] ( A ) , buffers ( B ) , salts ( C ) , detergent and glycerol ( D ) , and Polθ concentration ( E ) . Panels B–D included 5 mM Mn2+ . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 00510 . 7554/eLife . 13740 . 006Figure 1—figure supplement 3 . Sequence analysis of Polθ-Mg2+ template-dependent terminal transferase activity . ( A ) Schematic of method used to sequence Polθ-Mg2+ extension products . ( B ) Sequences of extension products generated by Polθ in the presence of 10 mM Mg2+ , all four dNTPs , and ssDNA RP347 . Initial sequence of RP347 ssDNA is indicated at top . Sequences of extension products are shown in a 5’-3’ direction . Black underline , sequence copied from template . ( C ) Model of how Polθ-Mg2+ repeatedly generates products 1–8 from RP347 ssDNA via snap-back replication . ( D ) Representative sequence traces of products 1–8 demonstrate non-identical sequencing reactions and files . Certain sequences are represented as complements due to their particular orientation during TA-cloning . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 006 Cellular studies show that Polθ is essential for MMEJ/alt-EJ ( Chan et al . , 2010; Kent et al . , 2015; Koole et al . , 2014; Mateos-Gomez et al . , 2015; Yousefzadeh et al . , 2014 ) , which is consistent with biochemical studies ( Kent et al . , 2015 ) . Intriguingly , these cellular studies showed the presence of both templated and non-templated ( random ) nucleotide insertions at alt-EJ repair junctions which were dependent on Polθ ( Chan et al . , 2010; Koole et al . , 2014; Mateos-Gomez et al . , 2015; Yousefzadeh et al . , 2014 ) . The random insertions were suggested to be due to a putative Polθ template-independent activity ( Figure 1B ) ( Mateos-Gomez et al . , 2015 ) . Yet , insofar Polθ template-independent terminal transferase activity has not been demonstrated in vitro . For example , early in vitro studies showed the unusual ability of Polθ to extend ssDNA and partial ssDNA substrates with 3’ overhangs ( pssDNA ) by several nucleotides ( Hogg et al . , 2012 ) . Although it was suggested that this activity might be the result of template-independent terminal transferase activity , the polymerase failed to extend homopolymeric ssDNA templates , which contain one type of base , without the complementary deoxyribonucleoside-triphosphate ( dNTP ) present ( Hogg et al . , 2012 ) . These previous studies therefore demonstrated a lack of template-independent terminal transferase activity by Polθ ( Hogg et al . , 2012 ) . More recent studies confirmed a lack of template-independent activity by Polθ ( Yousefzadeh et al . , 2014 ) . Instead , data was presented that suggests Polθ extends ssDNA by transiently annealing two oligonucleotides together in an anti-parallel manner , resulting in repeated use of the opposing ssDNA as a template in trans ( Yousefzadeh et al . , 2014 ) . Since cellular studies additionally demonstrated the presence of templated nucleotide insertions at alt-EJ junctions , which were also dependent on expression of the polymerase , models of Polθ copying sequences from the opposing overhang were also proposed ( Figure 1C ) ( Chan et al . , 2010; Koole et al . , 2014; Yousefzadeh et al . , 2014 ) . This form of templated extension in trans can conceivably facilitate end-joining by generating short regions of microhomology ( Figure 1C ) . Although previous biochemical studies suggested the possibility that Polθ performs templated extension of ssDNA in trans ( Yousefzadeh et al . , 2014 ) , more recent studies showed that the polymerase extends ssDNA by performing ‘snap-back’ replication on the same template in cis , similar to other end-joining polymerases ( Brissett et al . , 2007; Kent et al . , 2015 ) . Clearly , our understanding of how Polθ extends ssDNA , which is important for alt-EJ and is a unique activity for this polymerase , is very limited . For example , it remains to be determined whether Polθ is capable of performing non-templated extension ( Figure 1B ) , and templated extension in trans in the absence of sufficient microhomology ( Figure 1C ) . Furthermore , whether other factors or co-factors are necessary for activating Polθ terminal transferase activity during alt-EJ , which likely facilitates insertion mutations , remains unknown . Considering that Polθ contributes to the survival of breast and ovarian cancer cells deficient in homologous recombination ( HR ) , is associated with a poor clinical outcome for breast cancer patients , and confers resistance to chemotherapy drugs and ionizing radiation , understanding the enzymatic functions of the polymerase is essential for elucidating its roles in cancer progression and chemotherapy resistance ( Ceccaldi et al . , 2015; Higgins et al . , 2010; Lemee et al . , 2010; Mateos-Gomez et al . , 2015 ) . In this study , we sought to elucidate how Polθ generates insertion mutations during alt-EJ which contribute to genome instability . First , we found that manganese ( Mn2+ ) activates Polθ template-independent terminal transferase activity . Next , we discovered that Polθ generates random combinations of templated and non-templated insertion mutations during alt-EJ by oscillating between three different modes of terminal transferase activity: non-templated extension , templated extension in cis , and templated extension in trans . Lastly , we further characterized Polθ terminal transferase activity and surprisingly found that this activity is more proficient than terminal deoxynucleotidyl transferase ( TdT ) . Together , these data identify an unprecedented switching mechanism employed by Polθ to generate genetic diversity during alt-EJ and characterize Polθ as among the most proficient terminal transferases in nature . A current paradox in our understanding of alt-EJ is that Polθ promotes non-templated ( random ) nucleotide insertions at DNA repair junctions in vivo , but lacks template-independent terminal transferase activity in vitro . For example , similar to previous studies ( Hogg et al . , 2012; Kent et al . , 2015 ) , we found that Polθ fails to extend a homopolymeric ssDNA containing deoxycytidine-monophosphates ( poly-dC ) in the absence of the complementary deoxyguanosine-triphosphate ( dGTP ) under standard buffer conditions with magnesium ( Mg2+Figure 1D ) . This shows that efficient ssDNA extension by Polθ requires the complementary nucleotide , which demonstrates that the template bases facilitate the nucleotidyl transferase reactions by pairing with the incoming nucleotide . Our recent studies suggest that this template-dependent activity is due to ‘snap-back’ replication whereby the polymerase uses the template in cis ( Kent et al . , 2015 ) . A separate biochemical study also indicated that Polθ lacks template-independent activity ( Yousefzadeh et al . , 2014 ) . Thus , it remains unclear how Polθ facilitates random nucleotide insertions during alt-EJ which contribute to genome instability ( Figure 1B ) . Considering that divalent cations other than Mg2+ are present in cells , they may account for the discrepancy between the ability of Polθ to perform template-independent DNA synthesis in vivo but not in vitro . We therefore tested various divalent cations in a reaction including Polθ , poly-dC ssDNA and deoxythymidine-triphosphate ( dTTP ) , in the presence and absence of Mg2+ ( Figure 1E ) . The results showed that Mn2+ , and to a lesser extent Co2+ , activates Polθ extension of poly-dC with dTTP ( Figure 1E ) . For example , in the absence of Mn2+ in Figure 1D , Polθ extended only a small fraction of substrates with dTTP ( lane 4 ) . In contrast , the addition of Mn2+ under the same reaction conditions promoted extension of the same substrate by Polθ even when Mg2+ was abundant ( Figure 1E ) . Since thymidine cannot base pair with cytidine , these data demonstrate that Mn2+ activates Polθ template-independent terminal transferase activity ( i . e . non-templated DNA synthesis ) . Since Polθ DNA synthesis activity is fully supported by Mn2+ ( Figure 1E , lane 25 ) , this indicates that Mn2+ binds to the same positions as Mg2+ within the polymerase active site which is necessary for the nucleotidyl transferase reaction . Consistent with this , recent structural studies show that other metals such as calcium can substitute for Mg2+ in the polymerase active site ( Zahn et al . , 2015 ) . Furthermore , several lines of evidence show that Mn2+ can act as a co-factor for DNA polymerases and RNA polymerases and reduces the fidelity of these enzymes ( Andrade et al . , 2009; Dominguez et al . , 2000; Walmacq et al . , 2009 ) . Hence , the data show that Mn2+ acts as a co-factor for Polθ which promotes template-independent activity and likely reduces the fidelity of the polymerase . Importantly , this template-independent activity was also stimulated 3–8 fold by relatively low concentrations of Mn2+ ( 0 . 2 mM ) and Mg2+ ( 1–2 mM ) which are found in cells ( Figure 1—figure supplement 1 ) ( MacDermott , 1990; Schmitz et al . , 2003; Visser et al . , 2014 ) . Biochemical studies have also shown that Mn2+ is a necessary co-factor for the yeast Mre11-Rad50-Xrs2 ( MRX ) nuclease complex and its mammalian counterpart , MRN , which is essential for generating 3’ overhangs during alt-EJ , presumably by acting with CtIP ( Lee-Theilen et al . , 2011; Trujillo et al . , 1998; Zhang and Jasin , 2011 ) . Thus , these and other lines of evidence strongly indicate a physiological role for Mn2+ as a co-factor for DNA repair enzymes ( Andrade et al . , 2009; Cannavo and Cejka , 2014; Dominguez et al . , 2000; Trujillo et al . , 1998 ) . We identified optimal conditions for Polθ-Mn2+ template-independent terminal transferase activity in Figure 1—figure supplement 2 . Using these optimal conditions at different temperatures , we found that Polθ-Mn2+ exhibits robust template-independent terminal transferase activity ( Figure 1F ) . This suggests Mn2+ promotes the ability of Polθ to generate random nucleotide insertions during alt-EJ in cells . We further found that Mn2+ greatly stimulates Polθ terminal transferase activity on non-homopolymeric ssDNA substrates ( Figure 1G , left and right ) . In contrast , in the presence of Mg2+ Polθ became mostly arrested after transferring ~10–20 nucleotides ( nt ) , but also generated some larger discrete products ( Figure 1G , left and right ) . These data along with those presented in Figure 1D indicate that Mg2+ promotes template-dependent activity which directs the polymerase to repeatedly synthesize a few discrete products as observed for both substrates ( Figure 1G , left and right ) . Consistent with this , we found that Polθ-Mg2+ consistently generated similar DNA sequences from the RP347 ssDNA template , which is likely due to snap-back replication ( Figure 1—figure supplement 3 ) . Mn2+ on the other hand facilitates template-independent activity which enables Polθ to generate random products of different lengths as indicated by a smear ( Figure 1G , left and right ) . To gain more insight into these mechanisms of Polθ terminal transferase activity , we analyzed the sequences of ssDNA extension products generated by Polθ-Mn2+ in the absence of Mg2+ and with a 10-fold excess of Mg2+ which models cellular conditions . As expected , most of the DNA sequence generated by Polθ-Mn2+ in the absence of Mg2+ was random and therefore due to template-independent activity ( Figure 2A ) . This is consistent with the appearance of a smear rather than a few discrete bands as observed with Polθ-Mg2+ ( Figure 1G ) . Intriguingly , some of the sequences contained short regions that were either identical or complementary to the initial ssDNA ( Figure 2A , black underlines ) . Other sequence regions within individual molecules were complementary to one another but not to the original ssDNA template ( Figure 2A , grey and colored lines ) . Next , we analyzed DNA sequences generated by Polθ in the presence of a ten-fold excess of Mg2+ relative to Mn2+ , which more closely resembles physiological conditions ( Figure 2B ) . Again , we observed random sequence , complementary sequences within individual products ( grey and colored lines ) , and short sequence tracts identical or complementary to the initial template ( black underlines ) . Interestingly , Polθ generated more complementary sequences with an excess of Mg2+ ( compare Figure 2A and B ) . Furthermore , the average length of ssDNA extension products was shorter with an excess of Mg2+ ( Figure 2E ) , which is consistent with the results in Figure 1G . 10 . 7554/eLife . 13740 . 007Figure 2 . Polθ oscillates between three different modes of terminal transferase activity . ( A , B ) Sequences of Polθ ssDNA extension products in the presence of indicated divalent cations ( A , 5 mM Mn2+; B , 10 mM Mg2+ , 1 mM Mn2+ ) . Initial ssDNA sequences are indicated at top . Black underlines , sequences copied from either original template or complementary sequences generated from original template; matching colored lines , complementary sequences due to snap-back replication . ( C ) Models of Polθ terminal transferase activities . ( Top ) Polθ preferentially exhibits template-independent activity in the presence of Mg2+ and Mn2+ . Polθ also performs templated ssDNA extension in cis ( bottom left ) and in trans ( bottom right ) , and oscillates between these three mechanisms . ( D ) Models of Polθ terminal transferase activity based on sequences 3 and 8 from panel B . ( E ) Plot showing lengths of ssDNA products generated by Polθ in the presence of indicated divalent cations . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 00710 . 7554/eLife . 13740 . 008Figure 2—figure supplement 1 . Model of Polθ-Mn2+ terminal transferase activity involving template copying in cis and in trans . Model of how Polθ generates sequence tracts identical to the initial template in the presence of Mn2+ . Red , original sequence copied; black , complement of red sequence . The black complementary sequence may also be generated via templated extension in trans . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 00810 . 7554/eLife . 13740 . 009Figure 2—figure supplement 2 . Control experiments for Polθ-Mn2+ template-independent activity . ( A ) Scheme of experimental conditions . ( B ) Model of sequential activity of Polθ-Mn2+ on a primer-template ( top ) . Sequences generated by Polθ-Mn2+ during primer-extension in solid-phase in the presence of 5 mM Mn2+ . Black sequence , template-dependent; red sequence , template-independent; blue sequence , misincorporation; dash , frameshift mutation . Colored lines , complementary sequences generated by snap-back replication . ( C ) Models of Polθ activity on a primer-template in the presence of Mg2+ and Mn2+ ( top ) . Denaturing gels showing Polθ primer-extension products in the presence of 10 mM Mg2+ ( left ) and 5 mM Mn2+ ( right ) . ( D ) Models of Polθ-Mn2+ activity on a primer-template and ssDNA in the presence of dATP ( top ) . Denaturing gels showing template-dependent ( left ) and template-independent ( right ) Polθ-Mn2+ activities on a primer-template ( left ) and primer ( right ) , respectively , in the presence of 5 mM Mn2+ and dATP . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 00910 . 7554/eLife . 13740 . 010Figure 2—figure supplement 3 . Polθ-Mn2+ exhibits de novo DNA and RNA synthesis activities . Denaturing gels showing de novo nucleic-acid synthesis by Polθ in the presence of 5 mM Mn2+ and indicated nucleotides . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 01010 . 7554/eLife . 13740 . 011Figure 2—figure supplement 4 . Polθ-Mn2+ exhibits processive terminal transferase activity . ( A ) Scheme of experiment ( left ) . Denaturing gel showing inhibition of Polθ-Mn2+ terminal transferase activity by a ssDNA trap ( right ) . ( B ) Scheme of experiment ( left ) . Denaturing gel showing a time course of Polθ-Mn2+ terminal transferase activity in the presence and absence of ssDNA trap ( right ) . Assays ( A , B ) included 5 mM Mn2+ . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 01110 . 7554/eLife . 13740 . 012Figure 2—figure supplement 5 . Polθ-Mn2+ oscillates between different terminal transferase activites in the presence of a DNA trap . ( A ) Scheme of experiment performed in solid-phase . ( B ) Bar graph depicting ssDNA product lengths generated by Polθ in the presence ( orange ) and absence ( grey ) of excess ssDNA with 10 mM Mg2+ and 1 mM Mn2+ . ( C , D ) Sequences generated by Polθ incubated with the indicated ssDNA substrate in the presence ( D ) and absence ( C ) of excess ssDNA trap with 10 mM Mg2+ , 1 mM Mn2+ , and all four dNTPs . Black underlines , sequences identical or complementary to initial ssDNA substrate; red underlines , sequences complementary to ssDNA trap; colored lines above text , complementary sequences within individual ssDNA products . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 01210 . 7554/eLife . 13740 . 013Figure 2—figure supplement 6 . Polθ oscillates between templated and non-templated terminal transferase activities in the presence of physiological concentrations of Mg2+ and Mn2+ . Sequences generated by Polθ during ssDNA extension in the presence of 1 mM Mg2+ and 50 µM Mn2+ . Black and grey inderlines , sequence complementary to initial ssDNA substrate; red underline , sequence identical to initial ssDNA substrate; blue lines , complementary sequence generated by snap-back replication; red text without lines , random insertions . Initital ssDNA sequence indicated at top . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 013 Together , these data demonstrate that Polθ exhibits three distinct modes of terminal transferase activity when Mn2+ is present even at 10-fold lower concentrations than Mg2+ ( Figure 2C ) . In the first and predominant mode , Polθ performs template-independent terminal transferase activity ( Figure 2C , top ) . In the second mode , Polθ performs transient template-dependent extension in cis , also called snap-back replication ( Figure 2C , bottom left ) . This mechanism accounts for the appearance of complementary sequences within individual extension products ( Figure 2A , B; grey and colored lines ) . In the third mode , Polθ performs transient template-dependent extension in trans ( Figure 2C , bottom right ) . This accounts for sequence tracts that are identical or complementary to the initial ssDNA substrate ( Figure 2A , B; black underlines ) ; templated extension in cis can also promote sequence complementary to the initial template ( Figure 2C ) . Identical sequence tracts are most likely due to copying in trans of complementary sequence tracts initially formed by templated extension in cis or in trans ( Figure 2—figure supplement 1 ) . Further in vitro and in vivo evidence for these three mechanisms of terminal transferase activity is presented in Figures 3 and 4 , respectively . 10 . 7554/eLife . 13740 . 014Figure 3 . Polθ oscillates between three different modes of terminal transferase activity during alternative end-joining in vitro . ( A ) Scheme for reconstitution of Polθ mediated alt-EJ in vitro ( top ) . Sequences of alt-EJ products generated by Polθ in vitro using 10 mM Mg2+ and 1 mM Mn2+ ( bottom ) . Red text , insertions; black text , original DNA sequence; black and grey underlines , sequences copied from original template; red underlines , complementary sequences due to snap-back replication; red sequence without underlines , random insertions; superscript 1 , suggests sequences were copied from a template portion that was subsequently deleted during alt-EJ; superscript 2 , suggests sequences were copied from the template in more than one way . Original DNA sequences indicated at top . Blue type , mutations . ( B ) Plot of insertion tract lengths generated in panel A . ( C ) Chart depicting percent of individual nucleotide insertion events due to non-templated extension , templated extension in cis and templated extension in trans . t test indicates no significant difference between percent of non-templated and templated in cis insertions . ( D ) Models of Polθ activity based on end-joining products 1 and 2 from panel A . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 01410 . 7554/eLife . 13740 . 015Figure 3—figure supplement 1 . Supporting information for Polθ mediated alt-EJ in vitro . ( A ) Schematic of alt-EJ reaction and subsequent procedures used for amplification and sequencing of end-joining products . Control alt-EJ reactions were performed with 10 mM Mg2+ and 1 mM Mn2+ . ( B ) Non-denaturing gels showing the products of PCR reactions containing either purified DNA from alt-EJ reactions performed in the presence of Polθ and Lig3 ( top left ) , Polθ alone ( top middle ) , and in the absence of Polθ and Lig3 ( top right ) , or no DNA with primers only ( bottom middle ) . Products in the top middle and top right gels are due to primer-dimer events as shown in the primers only control ( bottom middle gel ) . Lanes 1-8 represent PCR reactions performed at the following respective temperatures: 61°C , 60 . 8°C , 60 . 4°C , 59 . 9°C , 59 . 2°C , 58 . 6°C , 58 . 2°C , 58°C . Lanes 9–13 represent PCR reactions performed in the absence of PCR primers RP435 and RP431 and at the following respective temperatures: 61°C , 60 . 4°C , 59 . 9°C , 59 . 2°C , 58 . 2°C . The absence of PCR products in lanes 9–13 show that Taq polymerase cannot amplify original pssDNA templates via end-joining or other mechanisms . ( C ) Plot showing percent of end-joining products observed in cloning vectors following end-joining reactions containing the indicated proteins . Red , end-joining products with insertions; grey , end-joining products without insertions . n = 64 ( +Polθ , +Lig3 ) , n = 72 ( +Polθ , –Lig3 ) , n =12 ( –Polθ , –Lig3 ) . End-joining products in the absence of Polθ and Lig3 are likely due to infrequent byproducts of PCR . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 01510 . 7554/eLife . 13740 . 016Figure 3—figure supplement 2 . Polθ acts processively during alt-EJ in vitro . ( A ) Scheme for reconstitution of Polθ mediated alt-EJ in vitro with ssDNA trap ( top ) . Sequences of alt-EJ products generated by Polθ in vitro using 10 mM Mg2+ and 1 mM Mn2+ ( bottom ) . Red text , insertions; black text , original DNA sequence; grey underlines , sequences copied from original template; red underlines , complementary sequences due to snap-back replication; red sequence without underlines , random insertions; superscript 1 , suggests sequences were copied from a template portion that was subsequently deleted during alt-EJ . Original DNA sequences indicated at top . Blue type , mutations . ( B ) Plot of insertion tract lengths generated in panel A . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 01610 . 7554/eLife . 13740 . 017Figure 3—figure supplement 3 . Polθ generates insertions during alt-EJ in the presence of low concentrations of Mg2+ and Mn2+ . ( A ) Scheme for reconstitution of Polθ mediated alt-EJ in vitro with 1 mM Mg2+ and 50 µM Mn2+ ( top ) . Sequences of Polθ–mediated alt-EJ products with insertions >2 bp ( bottom ) . Red text , insertions; black text , original DNA sequence; black underlines , sequences copied from original template; red and blue lines , complementary sequences due to snap-back replication; red sequence without lines , random insertions . ( B ) Plot of insertion tract lengths illustrated in panel A . ( C ) Plot showing percentage of Polθ–mediated alt-EJ products with and without insertions . n = 32 . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 01710 . 7554/eLife . 13740 . 018Figure 4 . Polθ oscillates between three different modes of terminal transferase activity during alternative end-joining in vivo . ( A ) Scheme for Polθ mediated alt-EJ of site-specific DSBs in mouse embryonic stem cells ( top ) . Sequences of alt-EJ products generated by Polθ in cells ( bottom ) . Red text , insertions; black text , original DNA sequence; black and grey underlines , sequences copied from original template; red and blue underlines , complementary sequences due to snap-back replication; red sequence without underlines , random insertions; Original DNA sequences indicated at top; . . . , large deletions . ( B ) Plot of insertion tract lengths generated in panel A . ( C ) Chart depicting percent of individual nucleotide insertion events due to non-templated extension , templated extension in cis and templated extension in trans . t test indicates no significant difference between percent of non-templated and templated in cis insertions . ( D ) Models of Polθ activity based on end-joining products 1 and 2 from panel A . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 01810 . 7554/eLife . 13740 . 019Figure 4—figure supplement 1 . Large insertions copied from remote donor locations . Scheme for Polθ mediated alt-EJ of site-specific DSBs in mouse embryonic stem cells ( top ) . Insertion sequences of alt-EJ products generated by Polθ in cells ( bottom three panels ) . Probable remote donor sites listed at right based on sequence similarity . The large templated insertions copied from remote donor locations are likely due to strand invasion into duplex DNA followed by D-loop extension and dissociation . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 01910 . 7554/eLife . 13740 . 020Figure 4—figure supplement 2 . Additional sequence analysis of alternative end-joining products generated in vivo . ( A ) Scheme for Polθ mediated alt-EJ of site-specific DSBs in mouse embryonic stem cells ( top ) . Sequences of alt-EJ products generated by Polθ in cells ( bottom ) . Red text , insertions; black text , original DNA sequence; grey underlines , sequences copied from original template; red and blue underlines , complementary sequences due to snap-back replication; red sequence without underlines , random insertions; Original flanking DNA sequences indicated at top in bold; . . . , large deletions . ( B ) Pie chart of insertion tract lengths generated in vivo . n = 118 . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 020 Intriguingly , many of the extension products were generated by more than one mode of terminal transferase activity ( Figure 2B ) , which demonstrates that the polymerase oscillates between these different mechanisms ( Figure 2C ) . We utilized product sequences to specifically trace this enzymatic switching phenomenon at near base resolution ( Figure 2D ) . For example , sequence 8 from Figure 2B demonstrates that Polθ first performs 50 consecutive random nucleotide transfer events , then switches to a transient snap-back replication mode ( templated extension in cis ) . Next , Polθ switches to random mode then after transferring 4 nt switches back to snap-back mode followed by another switch back to random synthesis . Next , Polθ switches to the templated extension in trans mode where it copies 7 nt , then switches back to random mode for an additional 23 nt . Finally , Polθ switches back to snap-back mode , then after transferring 8 nt it ends the reaction by randomly incorporating an additional 5 nt . Sequence 3 from Figure 2B shows similar oscillation between these different mechanisms ( Figure 2D , bottom ) . Here , Polθ performs 55 consecutive random nucleotide transfer events then switches to snap-back mode where it incorporates another 15 nt . Since the melting temperature of this 15 bp duplex is predicted to be 50°C and the reaction was performed at 42°C , Polθ appears to be capable of unwinding duplexes formed during snap-back replication . Polθ then performs three additional switching events , ultimately generating in a 138 nt product composed of a combination of random and templated sequence . Under these conditions , Polθ shows a preference for template-independent terminal transferase activity ( Figure 2C ) , which is more prevalent when Mg2+ is omitted ( compare Figures 2A and B ) . Thus , the ratio of Mn2+ to Mg2+ modulates the balance between these different mechanisms . For example , higher concentrations of Mn2+ promote template-independent transfer events , whereas lower concentrations of Mn2+ reduce random transferase activity while increasing template-dependent activity due to snap-back replication ( compare Figures 2A and B ) . Higher concentrations of Mn2+ also promote longer extension products , which correlates with the polymerase’s preference for template-independent activity under these identical conditions ( Figure 1G; Figure 2E ) . To be certain Polθ-Mn2+ performs template-independent activity rather than highly error-prone template-dependent activity which may be perceived as template-independent , we performed multiple additional controls . First , we analyzed template- dependent and independent activities in the same reaction performed in solid-phase ( Figure 2—figure supplement 2A , B ) . Here , a biotinylated primer-template was immobilized to streptavidin beads , then excess template strand was removed by thorough washing . Primer extension in the presence of Mn2+ was then performed and extension products were sequenced . The results show that the initial template-dependent activity is performed with relatively high fidelity ( Figure 2—figure supplement 2B ) . For example , misincorporation and frameshift error rates of 5 . 6 x 10–2 and 6 . 9 x 10–3 , respectively , were observed on this short template . On the other hand , once Polθ reaches the end of the template mostly random sequence was generated , demonstrating template-independent activity ( Figure 2—figure supplement 2B ) . Consistent with this , we show the ability of Polθ to continue DNA synthesis far beyond the end of the template exclusively in the presence of Mn2+ ( Figure 2—figure supplement 2C ) . We further show that the rate of misincorporation and mismatch extension by Polθ-Mn2+ on a primer-template in the presence of a single nucleotide ( dATP ) is dramatically slower than its activity under identical conditions without the template strand present ( Figure 2—figure supplement 2D ) . Thus , these data demonstrate that Polθ-Mn2+ terminal transferase activity is not the result of misincorporation or mismatch extension . As an additional control for template-independent activity , we tested whether Polθ-Mn2+ performs de novo synthesis in the absence of DNA . Remarkably , Polθ-Mn2+ exhibits de novo DNA and RNA synthesis which unequivocally demonstrates its ability to synthesize nucleic-acids in a template-independent manner ( Figure 2—figure supplement 3 ) . Next , we examined whether Polθ-Mn2+ acts processively during ssDNA extension and whether the polymerase can switch between the three different modes of terminal transferase activity without dissociating from the initial ssDNA template . We tested the processivity of Polθ-Mn2+ on ssDNA by allowing the polymerase to extend the ssDNA for an initial 5 min followed by the addition of a 150-fold excess of unlabeled ssDNA which sequesters the polymerase if it dissociates from the initial radio-labeled ssDNA during the reaction ( Figure 2—figure supplement 4B ) . Remarkably , addition of the ssDNA trap had no effect on Polθ-Mn2+ terminal transferase activity , demonstrating that the polymerase performs ssDNA extension with high processivity . As a control , we show that 150-fold excess of unlabeled ssDNA effectively sequesters the polymerase from solution ( Figure 2—figure supplement 4A ) . Since Polθ-Mn2+ exhibits three different modes of terminal transferase activity under the same conditions ( Figure 2A ) , these results indicate the polymerase switches between these distinct activities without dissociating from the initial ssDNA . To further test the processivity of this switching mechanism , we performed ssDNA extension in the presence and absence of a ssDNA trap in solid-phase which enabled removal of excess unbound polymerase from solution ( Figure 2—figure supplement 5 ) . For example , Polθ was first allowed to bind ssDNA immobilized to streptavidin beads . Then , excess unbound Polθ was removed by thorough washing of the beads . Next , the reaction was initiated by the addition of dNTPs in buffer containing 10 mM Mn2+ and 1 mM Mn2+ . After 15 s , a 150-fold excess of ssDNA trap was added , whereas the negative control reaction contained no trap . Following completion of the reactions , the immobilized ssDNA was isolated and sequenced . Consistent with the results obtained in Figure 2—figure supplement 4 , the ssDNA trap did not suppress Polθ terminal transferase activity . In fact , the data indicate that the addition of excess ssDNA increases the length of ssDNA extension products generated by Polθ in solid-phase ( Figure 2—figure supplement 5 , panels B–D ) . This suggests that use of a template in trans enables Polθ terminal transferase activity rather than suppressing it . Consistent with this , sequence analysis shows that Polθ frequently utilizes the ssDNA trap as a template in trans ( Figure 2—figure supplement 5 , red underlines , panel D ) . The polymerase also performs template-independent and snap-back replication activities when the ssDNA trap is present ( panel D ) . Since Polθ is highly processive during ssDNA extension ( Figure 2—figure supplement 4 ) , these data provide strong support for a model whereby a single polymerase oscillates between the three different modes of teminal transferase activity without dissociating from the initial ssDNA template . Importantly , using intracellular concentrations of Mg2+ ( 1 mM ) and Mn2+ ( 50 µM ) , Polθ remains effective in extending ssDNA and utilizes a combination of templated and non-templated mechanisms during this activity ( Figure 2—figure supplement 6 ) . Next , we examined Polθ terminal transferase activity in the context of alt-EJ . Although cellular studies have shown that Polθ expression is required for the appearance of non-templated and templated insertions at alt-EJ repair junctions , it remains unknown whether additional factors or co-factors facilitate these insertion events . For example , Polθ has been shown to promote what appears to be random nucleotide insertion tracts at alt-EJ repair junctions in mice and flies ( Figure 1B ) ( Chan et al . , 2010; Mateos-Gomez et al . , 2015 ) . Evidence in flies , mice and worms also indicates that Polθ promotes templated nucleotide insertions , which are proposed to be due to a template copy mechanism in trans ( Figure 1C ) ( Chan et al . , 2010; Koole et al . , 2014 ) . To determine whether Polθ is solely responsible for these insertions , and whether the three mechanisms of terminal transferase activity identified herein facilitate these insertions , we reconstituted a minimal alt-EJ system in vitro . Here , two DNA substrates containing a 3’ overhang , herein referred to as partial ssDNA ( pssDNA ) , and a single base pair of microhomology ( G:C ) at their 3’ termini were incubated with Polθ , Lig3 , ATP , and dNTPs in buffer containing a high ratio of Mg2+ to Mn2+ which models cellular conditions ( Figure 3A , top ) . Although Polθ can perform MMEJ without Lig3 by promoting templated extension in trans ( Figure 1A ) ( Kent et al . , 2015 ) , the pssDNA substrates in the current assay lack sufficient microhomology for MMEJ , but contain a 5’ phosphate on their short strands which can support ligation of the opposing 3’ overhang that is extended by the polymerase ( Figure 3A , top ) . Control experiments show that the addition of Polθ and Lig3 is required for efficient alt-EJ , and that insertions depend on Polθ ( Figure 3—figure supplement 1C ) . These results are expected since Lig3 is required for most alt-EJ in cells and therefore likely functions with Polθ which facilitates insertions ( Audebert et al . , 2004; Simsek et al . , 2011 ) . Following termination of the reaction by EDTA , DNA was purified then end-joining products were amplified by PCR and individually sequenced from cloning vectors ( Figure 3—figure supplement 1A , B ) . To gain significant insight into the mechanisms of Polθ terminal transferase activity during alt-EJ , we chose to analyze insertion tracts greater than 2 nt in length which reveal information regarding template dependency . Remarkably , we found that Polθ generated both random and templated nucleotide insertions at repair junctions ( Figure 3A ) , which is similar to the results obtained in Figure 2 . In the case of templated insertions , we observed sequence tracts that appear to be due to both templated extension in cis ( snap-back replication; red underlines ) and in trans ( grey underlines ) . A median insertion length of 7 bp was observed ( Figure 3B ) , and cumulative analysis of individual nucleotide insertion events reveals a roughly equal proportion of insertions due to the three modes of terminal transferase activity identified in Figure 2 , for example non-templated extension , templated extension in cis , and templated extension in trans ( Figure 3C ) . We again modeled Polθ switching activity based on the sequence generated , in this case during alt-EJ ( Figure 3D ) . Consistent with the mechanism identified in Figure 2 , sequence traces strongly suggest spontaneous and rapid switching between the three different terminal transferase activities ( Figure 3D ) . We next examined whether the polymerase acts processively to generate insertions during alt-EJ . To test this , we repeated the alt-EJ reaction in vitro , but added a 150-fold excess of ssDNA trap 15 s after the reaction was initiated . The results show that Polθ generates similar insertion tract lengths in the presence and absence of the ssDNA trap ( compare Figure 3 and Figure 3—figure supplement 2 ) . Thus , these data also indicate that Polθ acts processively during alt-EJ which provides further support for a model whereby a single polymerase oscillates between the different terminal transferase activities prior to dissociating from the initial substrate . Importantly , further alt-EJ experiments show that Polθ generates similar size insertions by a combination of templated and non-templated mechanisms in the presence of 1 mM Mg2+ and 50 µM Mn2+ which model intracellular concentrations ( Figure 3—figure supplement 3 ) . To test whether Polθ uses this switching mechanism to generate insertions during alt-EJ in cells , we analyzed insertion tracts synthesized by Polθ during alt-EJ in vivo ( Figure 4 ) . Here , Polθ dependent alt-EJ in mouse embryonic stem cells promotes translocations between sequence specific DSBs generated in chromosomal DNA by the CRISPR/Cas9 system , as shown in previous studies ( Figure 4A , topMateos-Gomez et al . , 2015 ) . To distinguish between the different Polθ mediated activities during chromosomal translocation , we carefully analyzed junctions of events resulting from the cleavage of chromosomes 6 and 11 , and subsequent formation of Der ( 6 ) and ( 11 ) . Similar to Figure 3 , we chose to analyze junctions containing insertions >2 bp in length . Remarkably , in the cellular alt-EJ system we also observed insertion tracts that appear to be due to all three modes of Polθ terminal transferase activity ( Figure 4A ) . For example , similar to the results obtained in the in vitro alt-EJ system ( Figure 3 ) , cumulative analysis of individual nucleotide insertion events produced in vivo demonstrates that Polθ generates a roughly equal proportion of insertion events due to the three different modes of terminal transferase activity ( Figure 4A–C ) . Templated extension in trans accounts for short sequence duplications ( black and grey underlines ) , whereas templated extension in cis ( snap-back replication ) accounts for the appearance of short complementary sequence tracts ( red and blue underlines ) ( Figure 4A ) . Individual nucleotide insertion events due to non-templated extension appear to be slightly lower in the in vivo system ( 33 . 2% ) compared to the in vitro system ( 39% ) , which is likely due to a lower proportion of Mn2+ to Mg2+ in cells . Consistent with this , events due to templated extension in cis ( snap-back replication ) appear slightly higher in the in vivo system ( 37 . 2% ) compared to the in vitro system ( 28 . 8% ) . We note that DNA deletions were observed in both systems , albeit more frequently in cells which is likely due to nuclease activity . Deletions in the in vitro system likely result from Polθ mediated end-joining at internal sites within the 3’ overhang , as shown previously ( Kent et al . , 2015 ) . This mechanism may also contribute to deletions observed in vivo . Regardless of the specific mechanisms underlying deletion formation in each system , the insertion tracts observed in vitro and in vivo appear similar in nature in regards to template dependency ( compare Figures 3C and 4C ) . Furthermore , the median insertion tract length ( 7 bp ) generated by Polθ in vitro and in vivo was identical ( compare Figures 3B and 4B ) . Thus , these data demonstrate that the reconstituted alt-EJ system closely resembles the mechanism of alt-EJ in cells . We note that some large ( >30 bp ) insertions copied from remote chromosome sites and the CRISPR/Cas9 vector were also observed in the in vivo system ( Figure 4—figure supplement 1 ) . However , these insertions are likely due to a different mechanism such as strand invasion into duplex DNA . Additional analysis of end-joining products generated in vivo demonstrates that Polθ preferentially produces insertions >2 bp in length , and occasionally generates relatively long insertions ( i . e . >25 bp ) ( Figure 4—figure supplement 2 ) . Importantly , sequences of end-joining products generated in vivo support the same mechanism of Polθ switching observed in vitro ( Figure 4D ) . Altogether , the results presented in Figures 3 and 4 along with previous studies showing the requirement for Polθ in forming insertions indicate that Polθ is the main enzyme involved in generating insertions during alt-EJ . These results also indicate that Polθ oscillates between three different modes of terminal transferase activity to generate insertion mutations , and that Mn2+ likely acts as a co-factor for Polθ in vivo . We next further characterized Polθ-Mn2+ terminal transferase activity on a variety of DNA substrates . For example , we further tested Polθ-Mn2+ on homopolymeric ssDNA composed of either deoxythymidine-monophosphates ( poly-dT ) or deoxycytidine-monophosphates ( poly-dC ) , and ssDNA containing variable sequences . The polymerase preferentially extended all of the substrates by more than 100 nt in the presence of deoxyadenosine-triphosphate ( dATP ) , regardless of the sequence context ( Figure 5A , B ) . Polymerases are known to preferentially incorporate deoxyadenosine-monophosphate ( dAMP ) when template base coding is not available , which is referred to as the A-rule . For example , polymerases preferentially incorporate a single dAMP opposite an abasic site or at the end of a template . Thus , the observed preferential incorporation of dAMP by Polθ-Mn2+ is consistent with the A-rule and template-independent activity . Polθ also extended ssDNA in the presence of dTTP , dCTP , and dGTP , however , the lengths of these products were shorter than with dATP ( Figure 5A , B ) . For example , in the case of non-homopolymeric ssDNA , Polθ-Mn2+ transferred ~30–70 nt in the presence of dTTP , dCTP , or dGTP ( Figure 5B ) , which demonstrates that Polθ-Mn2+ terminal transferase activity is relatively efficient even in the absence of the preferred dATP . Notably , the non-homologous end-joining ( NHEJ ) X-family polymerase , Polμ , exhibited minimal terminal transferase activity compared to Polθ under identical conditions ( Figure 5—figure supplement 1 ) . Previous studies similarly demonstrated limited terminal transferase activity by Polμ which is most closely related to TdT ( Andrade et al . , 2009 ) . Thus , to date the data presented insofar indicate that , aside from TdT , Polθ possesses the most robust terminal transferase activity for the polymerase enzyme class . 10 . 7554/eLife . 13740 . 021Figure 5 . Polθ exhibits preferential terminal transferase activity on pssDNA . ( A ) Denaturing gels showing Polθ extension of poly-dC ( left ) and poly-dT ( right ) ssDNA with 5 mM Mn2+ and the indicated dNTPs . ( B ) Denaturing gel showing Polθ extension of the indicated ssDNA with 5 mM Mn2+ and indicated dNTPs . ( C ) Denaturing gel showing Polθ extension of the indicated dsDNA with 5 mM Mn2+ and indicated dNTPs . ( D ) Denaturing gel showing Polθ extension of a primer-template with 5 mM Mn2+ and all four dNTPs . Model of Polθ-Mn2+ activity on a primer-template ( right ) . ( E ) Denaturing gel showing Polθ extension of the indicated pssDNA with 5 mM Mn2+ and indicated dNTPs . ( F ) Denaturing gels showing Polθ extension of ssDNA modeled after telomere sequence with 5 mM Mn2+ and the indicated dNTPs . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 02110 . 7554/eLife . 13740 . 022Figure 5—figure supplement 1 . Comparison of Polθ and Polμ terminal transferase activities with Mn2+ . ( A , B ) Denaturing gels showing Polθ and Polμ extension of poly-dC ( A ) and RP347 ssDNA ( B ) in the presence of Mn2+ and the indicated nucleotides . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 022 We next examined the ability of Polθ-Mn2+ to extend blunt-ended double-strand DNA ( dsDNA ) . The results show that Polθ efficiently extends duplex DNA , however , this is limited to only 1–2 nucleotides which may be due to a lower affinity of the polymerase for blunt-ended DNA ( Figure 5C ) . Interestingly , Polθ efficiently extended a primer-template far beyond the downstream end of the template ( Figure 5D , left ) . Thus , the polymerase performs efficient long-range extension of dsDNA when given a running start ( Figure 5D , right schematic ) . Considering that Polθ is thought to act on DSBs partially resected by MRN and CtIP during MMEJ/alt-EJ ( Kent et al . , 2015 ) , we examined its terminal transferase activity on pssDNA . Remarkably , Polθ-Mn2+ exhibited the most efficient terminal transferase activity on pssDNA ( Figure 5E ) . For example , the polymerase extended the pssDNA substrates to longer lengths with dTTP and dCTP , whereas dGTP was still limiting ( compare Figure 5E with Figure 5B ) . Consistent with its role in promoting alt-EJ of telomeres in cells deficient in telomere protection and NHEJ factors ( Mateos-Gomez et al . , 2015 ) , we found that Polθ exhibits efficient terminal transferase activity on ssDNA modeled after telomeres which are known to contain stable G-quadruplex ( G4 ) secondary structures ( Figure 5F ) . Here again , extension in the presence of dGTP was suppressed . Considering that consecutive dGMP incorporation events limit Polθ terminal transferase activity , we presume the multiple guanosines present in telomere repeats cause a similar inhibitory effect . All other nucleotides were efficiently transferred to the telomeric ssDNA substrate ( Figure 5F ) . Taken together , the results in Figure 5 show that Polθ exhibits the most robust terminal transferase activity on pssDNA which is consistent with its role in MMEJ/alt-EJ , and that the polymerase is also efficient in extending various ssDNA substrates and dsDNA when given a running start . Next , we sought to identify structural motifs that promote Polθ terminal transferase activity . Polθ is a unique A-family polymerase since it contains three insertion loops , and previous studies have shown that loop 2 is necessary for Polθ extension of ssDNA ( Hogg et al . , 2012; Kent et al . , 2015 ) . The position of this motif is conserved in Polθ and is located immediately downstream from a conserved positively charged residue , arginine ( R ) or lysine ( K ) , at position 2254 ( Figure 6A ) . Recent structural studies of Polθ in complex with a primer-template and incoming nucleotide show that loop 2 lies relatively close to the 3’ terminus of the primer , but is likely flexible in this conformation due to a lack of resolution ( Figure 6B ) ( Zahn et al . , 2015 ) . Considering that Polθ ssDNA extension with Mg2+ is likely related to its activity with Mn2+ , we predicted that loop 2 would also confer template-independent terminal transferase activity . Indeed , a loop 2 deletion mutant of Polθ ( PolθL2 ) failed to extend ssDNA under optimal template-independent terminal transferase conditions with Mn2+ ( Figure 6C ) . Similar to previous results , PolθL2 fully extended a primer-template ( Figure 6D ) . Here , PolθWT extension continued beyond the template due to the polymerase’s robust terminal transferase activity with Mn2+ ( Figure 6D ) . 10 . 7554/eLife . 13740 . 023Figure 6 . Conserved residues contribute to Polθ processivity and template-independent terminal transferase activity . ( A ) Sequence alignment of Polθ and related A-family Pols . Conserved positively charged residues ( 2202 , 2254 ) and loop 2 in Polθ are highlighted in yellow and grey , respectively . Black boxes indicate conserved motifs . * = identical residues , : = residues sharing very similar properties , . = residues sharing some properties . Red , small and hydrophobic; Blue , acidic; Magenta , basic; Green , hydroxyl , sulfhydryl , amine . ( B ) Structure of Polθ with ssDNA primer ( PDB code 4X0P ) ( Zahn et al . , 2015 ) . Residues R2202 and R2254 are indicated in blue . Dotted blue lines indicate ionic interactions . Loop 2 is indicated in dark red . Thumb and palm subdomains are indicated . ( C ) Denaturing gel showing PolθWT and PolθL2 extension of ssDNA with 5 mM Mn2+ and all four dNTPs . ( D ) Denaturing gel showing PolθWT and PolθL2 extension of a primer-template with 5 mM Mn2+ and all four dNTPs . Model of PolθWT-Mn2+ and PolθL2-Mn2+ activities on a primer-template ( right ) . ( E ) Denaturing gel showing a time course of PolθWT and PolθRR extension of a primer-template in the presence of 10 mM Mg2+ and all four dNTPs . ( F ) Denaturing gel showing PolθWT ( left ) and PolθRR ( right ) extension of poly-dC ssDNA with 5 mM Mn2+ and the indicated dNTPs . ( G ) Schematic of assay ( left ) . Denaturing gel showing PolθWT and PolθRR extension of an excess of radiolabeled primer-template with all four dNTPs and 10 mM Mg2+ either in the presence or absence of 150-fold excess unlabeled DNA trap . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 023 Structural studies showed that two conserved positively charged residues , R2202 and R2254 , bind to the phosphate backbone of the 3’ portion of the primer ( Figure 6A , B ) ( Zahn et al . , 2015 ) . Since these positively charged residues are conserved in Polθ but not other A-family members ( Figure 6A ) , we envisaged that they might contribute to Polθ terminal transferase activity . We first tested primer-extension of a double mutant version of Polθ in which R2202 and R2254 were changed to alanine ( A ) and valine ( V ) , respectively ( PolθRR ) . Recent studies showed that single R2202A and R2254V Polθ mutants were slightly defective in translesion synthesis ( Zahn et al . , 2015 ) . PolθRR extended the primer in a similar manner to PolθWT ( Figure 6E ) . Yet , PolθRR showed a severe defect in template-independent terminal transferase activity compared to PolθWT under identical conditions with Mn2+ ( Figure 6F ) . Since PolθWT performs terminal transferase activity with high processivity , we wondered whether PolθRR exhibits reduced processivity . Indeed , PolθRR showed a significant deficiency in primer extension compared to PolθWT when a large excess of DNA was present , confirming a reduction in processivity ( Figure 6G ) . These data also suggest that PolθWT exhibits lower processivity during primer-template extension compared to ssDNA extension ( compare Figure 6G and Figure 2—figure supplement 4 ) . Since PolθRR is defective in processivity and template-independent terminal transferase activity , this suggests that the polymerase must be processive on ssDNA to effectively perform template-independent terminal transferase activity . Together , these data identify conserved residues that contribute to Polθ terminal transferase activity by conferring processivity onto the enzyme through binding the 3’ primer terminus . Importantly , terminal transferase activity is widely used to modify ssDNA ends for various types of applications including biotechnology , biomedical research , and synthetic biology . Currently , the only enzyme developed and marketed for these applications is terminal deoxynucleotidyl transferase ( TdT ) whose cellular function is to promote antibody diversity by transferring non-templated nucleotides to V , D and J exon regions during antibody gene maturation ( Motea and Berdis , 2010 ) . We compared the activities of Polθ and TdT in Figure 7A . Remarkably , Polθ exhibited a similar ability to extend ssDNA as TdT assayed under optimal conditions recommended by the supplier ( Figure 7A ) . The results also show that in this reaction Polθ and TdT preferentially utilize dATP and dTTP , respectively , which suggests different mechanisms of action ( Figure 7A ) . 10 . 7554/eLife . 13740 . 024Figure 7 . Comparison of Polθ and TdT terminal transferase activities . ( A ) Denaturing gel showing Polθ-Mn2+ ( lanes 1–5 ) and TdT ( lanes 6–10 ) extension of ssDNA with the indicated dNTPs . ( B ) Denaturing gel showing Polθ-Mn2+ ( lanes 1–6 ) and TdT ( lanes 7–12 ) extension of ssDNA with the indicated ribonucleotides ( rNTPs ) . ( C ) Denaturing gel showing Polθ-Mn2+ ( lanes 1–11 ) and TdT ( lanes 12–22 ) extension of ssDNA with the indicated nucleotide analogs illustrated in panel ( d ) . Boxed lanes indicate nucleotides analogs that are exclusively transferred by Polθ-Mn2+ . ( D ) Nucleotide analogs: 1 , cy3-dUTP; 2 , Digoxigenin-11-dUTP; 3 , Biotin-16AA-dUTP; 4 , Texas Red-5-dCTP; 5 , N6 - ( 6-Azido ) hexyl-ATP; 6 , Cyanine 3-AA-UTP; 7 , 4-Thio-UTP; 8 , Biotin-16AA-CTP; 9 , Ganciclovir Triphosphate; 10 , 5-Hydroxymethyl-2’-deoxyuridine-5’-Triphosphate . Underlined nucleotide analogs ( 4 , 5 , 9 ) are exclusively transferred by Polθ-Mn2+ . ( E ) Denaturing gel showing Polθ-Mn2+ extension of RNA with all four dNTPs in the presence ( lane 3 ) and absence ( lane 2 ) of unlabeled ssDNA ( left panel ) . Denaturing gel showing Polθ-Mn2+ extension of RNA with the indicated nucleotide analogs ( right panel ) . Polθ-Mn2+ extension assays ( A-C , E ) included 5 mM Mn2+ . DOI: http://dx . doi . org/10 . 7554/eLife . 13740 . 024 Many biotechnology and biomedical research applications require ssDNA substrates modified with fluorophores or other chemical groups , such as those that enable DNA attachment to solid surfaces . We therefore examined the ability of Polθ to transfer deoxyribonucleotides and ribonucleotides conjugated with different functional groups to the 3’ terminus of ssDNA . Again , using the supplier’s recommended assay conditions for TdT , and identical concentrations of Polθ under its optimal conditions , we unexpectedly found that Polθ-Mn2+ is more effective in transferring ribonucleotides to ssDNA compared to TdT ( Figure 7B ) . Although previous studies have shown that Polθ strongly discriminates against ribonucleotides ( Hogg et al . , 2012 ) , this fidelity mechanism is largely compromised under our conditions used for terminal transferase activity . Again , using the respective optimal conditions for Polθ and TdT at identical concentrations , we also found that Polθ-Mn2+ is more proficient in transferring most modified deoxy-ribonucleotides and ribonucleotides to ssDNA than TdT ( Figure 7C , D ) . For example , Polθ more efficiently transferred eight out of ten modified nucleotides tested . In some cases , Polθ produced longer extension products than TdT ( Figure 7C ) . In other cases , Polθ transferred nucleotides that TdT was unable to incorporate ( Figure 7C , black boxes ) . For instance , Polθ efficiently transferred a nucleotide containing a linker attached to an azide group which is widely used for “click chemistry” applications ( Figure 7C , lane 6 ) . In contrast , TdT failed to transfer this nucleotide altogether ( Figure 7C , lane 17 ) . Moreover , TdT failed to transfer nucleotides containing a modified sugar and a linker attached to Texas Red , whereas these substrates were efficiently incorporated by Polθ ( Figure 7C , nucleotide analogs 4 and 9 ) . These results show that Polθ efficiently transfers ribonucleotides and deoxyribonucleotides containing modifications on their base moieties , such as fluorophores and functional groups including biotin and digoxigenin , as well as nucleotides containing sugar modifications ( i . e . ganciclovir mono-phosphate ) . Considering that Polθ also exhibits translesion synthesis activity , these results may be attributed to its natural ability to accommodate non-canonical nucleotides in its active site ( Hogg et al . , 2011; Yoon et al . , 2014 ) . Lastly , we investigated whether Polθ exhibits terminal transferase activity on RNA . Surprisingly , Polθ transferred both canonical and modified nucleotides to RNA , albeit less efficiently than to DNA ( Figure 7E ) . Together , the results presented in Figure 7 characterize Polθ as among the most proficient terminal transferases identified and demonstrate that Polθ is more effective than TdT in modifying nucleic-acid substrates for biomedical research and biotechnology applications . Recent studies have discovered that mammalian Polθ is essential for MMEJ/alt-NHEJ , which promotes chromosome rearrangements and resistance to DNA damaging agents , including those used for chemotherapy ( Kent et al . , 2015; Mateos-Gomez et al . , 2015; Yousefzadeh et al . , 2014 ) . Polθ was previously shown to be essential for alt-EJ in flies and worms ( Chan et al . , 2010; Koole et al . , 2014 ) , demonstrating a conserved role for this polymerase in higher eukaryotes . These cellular studies have shown that two types of insertions , non-templated and templated , are generated at alt-EJ repair junctions which are dependent on Polθ expression ( Chan et al . , 2010; Koole et al . , 2014; Mateos-Gomez et al . , 2015; Yousefzadeh et al . , 2014 ) . In the case of non-templated insertions , it has been proposed that Pol promotes random transfer of nucleotides via a putative template-independent terminal transferase activity ( Mateos-Gomez et al . , 2015 ) . Yet , biochemical studies have shown that Polθ lacks template-independent terminal transferase activity , creating a paradox between cellular and in vitro data ( Kent et al . , 2015; Yousefzadeh et al . , 2014 ) . In the case of templated insertions , a copy in trans model has been proposed which also has not been proven in vitro ( Chan et al . , 2010; Koole et al . , 2014; Yousefzadeh et al . , 2014 ) . In this report , we elucidate how Polθ generates both templated and non-templated nucleotide insertion mutations during alt-EJ , and characterize the polymerase as a highly robust terminal transferase for biotechnology and biomedical research applications . We first discover that Polθ exhibits robust template-independent terminal transferase activity in the presence of Mn2+ . Considering that structural studies show that differential binding of divalent cations within the active site of Polθ slightly alters its local conformation ( Zahn et al . , 2015 ) , Mn2+ binding likely facilitates an active site conformation more favorable for non-templated DNA synthesis . Since Polθ dependent non-templated nucleotide insertions are commonly associated with alt-EJ in cells , our findings suggest that Mn2+ acts as a co-factor of Polθ in vivo . For example , although the concentration of Mn2+ is relatively low in cells ( ~0 . 2 mM ) and is considerably less than Mg2+ ( ~1 . 0 mM ) , we show that these concentrations of Mn2+ and Mg2+ stimulate Polθ template-independent terminal transferase activity by 3–8 fold . Thus , cellular concentrations of Mn2+ are likely to activate Polθ template-independent activity . Intriguingly , Mn2+ has been shown to act as a necessary co-factor for the MRX nuclease complex and its mammalian counterpart , MRN , which is also essential for alt-EJ due to its role in generating 3’ ssDNA overhangs onto which Polθ acts ( Cannavo and Cejka , 2014; Trujillo et al . , 1998 ) . Thus , various enzymes involved in DNA repair are likely to utilize Mn2+ as a co-factor in addition to Mg2+ . To our surprise , the Polθ-Mn2+ complex exhibited a higher efficiency of transferring ribonucleotides and most modified nucleotide analogs to the 3’ terminus of ssDNA than TdT at identical concentrations . For example , in the presence of ribonucleotides , Polθ-Mn2+ generated substantially longer extension products , which demonstrates a lower discrimination against ribonucleotides . Polθ-Mn2+ also produced longer extension products than TdT in the presence of most nucleotide analogs , including those that contain large functional groups . Moreover , Polθ-Mn2+ efficiently transfered certain nucleotide analags that TdT failed to utilize as substrates . For instance , we found that Polθ-Mn2+ exclusively transfered a nucleotide conjugated with Texas Red and a nucleotide containing an azide group which is widely used for 'click' chemistry applications . We additionally found that Polθ-Mn2+ is capable of transferring canonical and modified nucleotides to RNA , albeit with lower efficiency than DNA . Based on these unexpected findings , we anticipate that Polθ will be more useful for modifying nucleic acid substrates for biotechnology , biomedical research and synthetic biology applications . Moreover , since Polθ does not require toxic reaction components like TdT , such as Co2+ salts or salts of cacodylic acid , Polθ terminal transferase assays are a safer option for research and biotechnology applications . Our report raises the question why evolution selected for two robust terminal transferases: Polθ and TdT . It is well known that the primary function of TdT is to generate insertion mutations during NHEJ of V , D and J antibody gene regions , which promotes antibody diversity that is necessary for a strong immune system ( Motea and Berdis , 2010 ) . Since a diverse immunological defense is important for survival , a clear selective pressure for TdT existed . In the case of Polθ , it appears that the polymerase has also been selected to generate insertion mutations during end-joining , however , the evolutionary pressure for this particular mechanism is not as clear . For example , although Polθ is essential for alt-EJ , this pathway appears to occur infrequently compared to primary DSB repair processes , such as HR ( Mateos-Gomez et al . , 2015; Truong et al . , 2013 ) . Consistent with this , Polθ is not important for normal cell survival or development . Recent studies of C . elegans , however , surprisingly show that Polθ mediated alt-EJ is a primary form of repair in germ cells ( van Schendel et al . , 2015 ) . Furthermore , it was shown that Polθ mediated alt-EJ promotes a deletion and insertion ( indel ) signature in propogated laboratory strains that is similar to indels found in natural isolates ( van Schendel et al . , 2015 ) . These studies therefore suggest that Polθ is important for generating genetic diversity . Interestingly , human Polθ is highly expressed in testis , suggesting the polymerase might also play a role in facilitating genetic diversity in mammals ( Seki et al . , 2003 ) . Considering that alt-EJ also promotes replication repair as a backup to HR , Polθ likely benefits cell survival at the expense of indels when lethal DSBs fail to be repaired by the primary HR pathway ( Truong et al . , 2013 ) . For example , Polθ mediated alt-EJ in C . elegans was shown to facilitate replication repair at stable G4 structures which may pose problems for the HR machinery and therefore potentially require an alternative and more accommodating error-pone form of repair ( Koole et al . , 2014 ) . Polθ has also been shown to suppress large genetic deletions in C . elegans , which demonstrates an obvious benefit for the polymerase ( Koole et al . , 2014 ) . Yet , whether these various functions of Polθ are conserved in mammals awaits further research . Our studies reveal that Polθ generates nucleotide insertions by oscillating between multiple mechanisms , which portrays a promiscuous enzyme that readily extends ssDNA by almost any means in order to catalyze end-joining products that frequently contain insertion mutations . For example , we observed that Polθ generates nucleotide insertions during alt-EJ in vitro by spontaneously switching between three distinct modes of terminal transferase activity: non-templated extension , templated extension in cis , and templated extension in trans . Importantly , we show that the characteristics of these insertions are nearly identical to those generated by Polθ mediated alt-EJ in cells , which indicates that Polθ also switches between these three mechanisms of terminal transferase activity in vivo . To our knowledge , the ability of a polymerase to spontaneously switch between three distinct modes of DNA synthesis has not been demonstrated . Thus , our data reveal an unprecedented set of mechanisms by which a single polymerase can synthesize DNA , presumably for generating genetic diversity and as a last resort for repairing lethal DSBs at the expense of mutations . 500 nM Polθ was incubated with 50 nM of the indicated 5’ 32P-labeled DNA for 120 min at 42°C ( or other indicated time intervals and temp ) in the presence of 0 . 5 mM of indicated dNTPs in a 10 µl volume of buffer A ( 20 mM TrisHCl pH 8 . 2 , 10% glycerol , 0 . 01% NP-40 , 0 . 1 mg/ml BSA ) with indicated divalent cations; optimal Polθ terminal transferase activity was performed with 5 mM MnCl2 . Reactions were terminated by the addition of 20 mM EDTA and 45% formamide and DNA was resolved by electrophoresis in urea polyacrylamide gels then visualized by autoradiography . Polm terminal transferase reactions were performed using the same conditions as Polθ . 50 nM Polθ was used in experiments employing ssDNA traps . 150-fold excess of unlabled ssDNA trap was added to reactions at indicated time points where indicated . Polθ terminal transferase activity in solid-phase . 50 nM RP347B was immobilized to magnetic streptavidin beads ( Dynabeads M-270 , Invitrogen ) in buffer A supplemented with 100 mM NaCl . Excess unbound DNA was then removed by washing beads 3x with buffer A with 100 mM NaCl . Next , the bead-DNA mixture was washed and resuspended in buffer A containing 10 mM MgCl and 1 mM MnCl . 500 nM Polθ was then added for 10 min to allow for ssDNA binding . Excess unbound Polθ was then removed by washing the beads 4x with 200 µl buffer A supplemented with 10 mM MgCl and 1 mM MnCl . Beads were resuspended in buffer A supplemented with 10 mM MgCl and 1 mM MnCl , then 0 . 5 mM dNTPs were added at 42°C . After 15 s , either dH20 or 7 . 5 µM RP427 was added and the reaction was terminated after 120 min by addition of EDTA . The beads were thoroughly washed to remove excess ssDNA trap . The beads were then resuspended in dH20 followed by boiling for 1–2 min . The supernatant was collected , then another cycle of boiling and supernatant collection was performed . The DNA from the supernatant was purified using Zymo DNA Clean and Concentrator-5 kit . Purified DNA was then ligated to RP430P overnight at room temp using T4 RNA ligase ( New Englan Biolabs ) . RNA ligase was denatured at 65°C , then the DNA was purified using Zymo DNA Clean and Concentrator-5 kit . The ligated DNA was then amplified via PCR using GoTaq Green ( Promega ) and primers RP347 and RP431 . PCR products were purified using QIAquick PCR purification kit ( Qiagen ) . Pure PCR products were then cloned into E . coli plasmid vectors using TOPO TA cloning ( Invitrogen ) . Individual plasmids containing PCR products were amplified in E . coli , isolated , then sequenced . Equimolar concentrations ( 100 nM ) of pssDNA substrates RP429/RP430-P and RP434-P/RP408 were mixed with 50 nM Polθ and 88 . 5 nM Lig3 in buffer A supplemented with 1 mM MnCl2 , 10 mM MgCl2 and 1mM ATP . Next , 10 μM dNTPs were added for 120 min at 37°C in a total volume of 100 μl . Reactions were terminated by incubation at 80°C for 20 min . ( Negative control reactions included: omission of Lig3 , and; omission of Polθ and Lig3 ) . DNA was purified using QIAquick Nucleotide Removal kit ( QIAGEN ) then amplified using PCR Master Mix ( Promega ) and end-joining specific primers RP431 and RP435 . PCR products were purified using GeneJET PCR Purification Kit ( ThermoScientific ) then cloned into the pCR2 . 1-TOPOvector ( Invitrogen ) . DNA was transformed into E . coli DH5α cells , and individual plasmids from single colonies were purified and sequenced . Polθ mediated alt-EJ in Figure 3—figure supplement 3 was performed as described above , however , 1 mM MgCl2 , 50 μM MnCl2 and 100 μM dNTPs were used . Where indicated , 150-fold excess ( 15 µM ) of ssDNA trap ( RP347 ) was added to the reaction at the indicated time point . Polθ mediated alt-EJ involving chromosomal translocation was performed as previously described ( Mateos-Gomez et al . , 2015 ) . Briefly , mouse Embryonic Stem ( ES ) cells were transfected with 3 µg of Cas9-gRNA ( Rosa26;H3f3b ) ( Mateos-Gomez et al . , 2015 ) . After transfection , 5 × 104 cells were seeded per well in a 96-well plate , and lysed 3 days later in 40 µl lysis buffer ( 10 mM Tris pH 8 . 0 , 0 . 45% Nonidet P-40 , 0 . 45% Tween 20 ) . The lysate was incubated with 200 µg/ml of Proteinase K for 2 hr at 55°C . Translocation detection was performed using nested PCR . The primers used in the first PCR reaction include Tr6-11-Fwd:5′-GCGGGAGAAATGGATATGAA-3′; Tr6-11-Rev: 5′- TTGACGCCTTCCTTCTTCTG -3′ , and Tr11-6-Fwd: 5′-AACCTTTGAAAAAGCCCACA-3′ and Tr11-6-Rev:5′-GCACGTTTCCGACTTGAGTT-3′ , for Der ( 6 ) and Der ( 11 ) respectively . For the second round of PCR amplification , the following primers were used: Tr6-11NFwd: 5′-GGCGGATCACAAGCAATAAT-3′; Tr6-11NRev: 5′-CTGCCATTCCAGAGATTGGT-3′ and Tr11-6NFwd:5′-AGCCACAGTGCTCACATCAC-3′ and Tr11-6NRev:5′TCCCAAAGTCGCTCTGAGTT-3′ . Amplified products corresponding to translocation events were subject to Sanger sequencing to determine the junction sequences . TdT terminal transferase reactions were performed on indicated 5’ 32P-labeled DNA using conditions recommended by New England Biolabs: 50 mM potassium acetate , 20 mM Tris acetate , 10 mM magnesium acetate , pH 7 . 9 , with 0 . 25 mM cobalt and incubated at 37°C . Incubation times and DNA concentrations were identical as experiments with Polθ . TdT was either used at concentrations recommended by New England Biolabs ( 0 . 2 units/µl ) or equimolar concentrations as Polθ as indicated in text . DNA products were resolved as indicated above . Polθ ( 500 nM ) was incubated with 50 nM RP347 ssDNA along with 0 . 5 mM dNTPs in 100 μl of buffer A supplemented with either 5 mM MnCl2 or 1 mM MnCl2 and 10 mM MgCl2 for 120 min at 42°C . Reactions were terminated by the addition of 25 μl of 5X non-denaturing stop buffer ( 0 . 5 M Tris-HCl , pH 7 . 5 , 10 mg/ml proteinase K , 80 mM EDTA , and 1 . 5% SDS ) . This was followed by phenol-chlorophorm extraction , ethanol precipitation , then ligation to 5’-phosphorylated RP359-P ssDNA using T4 RNA ligase ( NEB ) . DNA products were ethanol precipitated then dissolved in water . Next , PCR amplification of ligation products was performed using primers RP347 and RP359C and Taq Master Mix ( Promega ) . PCR products were purified using GeneJET PCR Purification Kit ( ThermoScientific ) then cloned into the pCR2 . 1-TOPO vector ( Invitrogen ) . DNA was transformed into E . coli DH5α cells , and individual plasmids from single colonies were purified and sequenced . Polθ-Mg2+ primer-extension was performed as described ( Kent et al . , 2015 ) with either 10 mM MgCl or 5 mM MnCl and indicated dNTPs and time intervals . Primer-extension in solid-phase was performed as follows . A 2:1 ratio of template ( RP409 ) to biotinylated primer ( RP25B ) was annealed then immobilized to magnetic streptavidin beads ( Dynabeads M-270 , Invitrogen ) pre-washed with buffer A supplemented with 100 mM NaCl . Excess unbound DNA was then removed by washing beads 3x with 200 µl of buffer A with 100 mM NaCl . Next , the bead-DNA mixture was washed and resuspended in buffer A containing 5 mM MnCl and 0 . 5 mM dNTPs . 500 µM Pol was then added for 120 min at 42°C . The reaction was then terminated by the addition of 20 mM EDTA followed by boiling for 1–2 min . The supernatant was collected , then another cycle of boiling and supernatant collection was performed . The DNA from the supernatant was purified using Zymo DNA Clean and Concentrator-5 kit . Purified DNA was then ligated to RP430P overnight at room temp using T4 RNA ligase ( New Englan Biolabs ) . RNA ligase was denatured at 65°C , then the DNA was purified using Zymo DNA Clean and Concentrator-5 kit . The ligated DNA was then amplified via PCR using GoTaq Green ( Promega ) and primers RP25 and RP431 . PCR products were purified using QIAquick PCR purification kit ( Qiagen ) . Pure PCR products were then cloned into E . coli plasmid vectors using TOPO TA cloning ( Invitrogen ) . Individual plasmids containing PCR products were amplified in E . coli , isolated , then sequenced . Where indicated primer-extension was performed with either a 1:1 ratio of PolθWT or PolθRR to primer-template ( 50 nM ) , or a 1:25 ratio of PolθWT or PolθRR to primer-template ( 50 nM ) . A 150-fold excess of ssDNA trap ( 7 . 5 µM RP316 ) was added 1 min after initiation of primer-extension where indicated . 500 nM Polθ was incubated with the indicated nucleotides at the following concentrations ( 500 nM ATP , UTP , GTP , dATP , dTTP , dGTP; 97 nM dCTP , [α-32P]- 6000 Ci/mmol 20 mCi/ml ( Perkin Elmer ) ) for the indicated time intervals at 42°C in buffer A supplemented with 5 mM MnCl . Nucleic acid products were resolved in denturing polyacrylamide gels and visualized by autoradiography . PolθWT and mutant proteins PolθL2 and PolθRR were purified as described ( Kent et al . , 2015 ) . Site-directed mutagenesis was performed using QuickChange II Site-Directed Mutagenesis Kit ( Agilent Technologies , Santa Clara , CA ) . TdT was purchased from New England Biolabs ( NEB ) . Polμ and Lig3 were purchased from Enzymax . pssDNA , dsDNA and primer-templates were assembled by mixing equimolar concentrations of ssDNA substrates together in deionized water , then heating to 95–100°C followed by slow cooling to room temp . ssDNA was 5’ 32P-labeled using 32P-γ-ATP ( Perkin Elmer ) and T4 polynucleotide kinase ( NEB ) . DNA ( Integrated DNA technologies ( IDT ) ) and RNA ( Dharmacon ) oligonucleotides ( 5’-3’ ) . RP25: CACAGATTCTGGCAGGCTGCAGATCGC RP25B: Biotin-CACAGATTCTGGCAGGCTGCAGATCGC RP347: CACTGTGAGCTTAGGGTTAGAGATAC RP348: CACTGTGAGCTTAGGGTTAGAGCCGG RP63: CGAAATAGACAGATCGCTGAGGATAGGTGCCTCACTG RP63C: CAGTGAGGCACCTATCCTCAGCGATCTGTCTATTTCG RP271: CATCTTTTACTTCCACCAGCGTTTCTGGG RP271C: CCCAGAAACGCTGGTGGAAGTAAAAGATG RP359: GTGGATGAATTACACATGCTGGGAGACTC RP359C: GAGTCTCCCAGCATGTGTAATTCATCCAC RP266: TTTTTTTTTTTTTTTTTTGCGATCTGCAGCCTGCCAGAATCTGTG RP331: ACTGTGAGCTTAGGGTTAGGGTTAGGGTTAGGGTTAG RP340: CACTGTGAGCTTAGGGTTAGAGATCG RNA-2: AUCGAGAGG RP343-P: /5Phos/CTAAGCTCACAGTG RP429: GGAGGTTAGGCACTGTGAGCTTAGGGTTAGAGATAC RP430-P: /5Phos/CTAAGCTCACAGTGCCTAACCTCC RP434-P: /5Phos/GAGCACGTCCAGGCGATCTGCAGCCTG RP408: GAGCACGTCCAGGCGATCTGCAGCCTGCCAGAATCTGTG RP427: CGCCACCTCTGACTTGAGCG RP409: GAGCACGTCCACGCGATCTGCAGCCTGCCAGAATCTGTG RP347B: Biotin-CACTGTGAGCTTAGGGTTAGAGATAC pssDNA substrates: RP347/RP343-P , RP348/RP343-P , RP340/RP343-P , RP429/RP430-P , RP434-P/RP408 . Telomeric ssDNA , RP331 . Primer-templates , RP25/RP266 , RP25/409 , RP25B/409 . 1 , cy3-dUTP ( Santa Cruz Biotech . ) ; 2 , Digoxigenin-11-dUTP ( Sigma ) ; 3 , Biotin-16AA-dUTP ( TriLink Biotech . ) ; 4 , Texas Red-5-dCTP ( PerkinElmer ) ; 5 , N6 - ( 6-Azido ) hexyl-ATP ( Jena Bioscience ) ; 6 , Cyanine 3-AA-UTP ( TriLink Biotech . ) ; 7 , 4-Thio-UTP ( TriLink Biotech . ) ; 8 , Biotin-16-AACTP ( TriLink Biotech . ) ; 9 , Ganciclovir Triphosphate ( TriLink Biotech . ) ; 10 , 5-Hydroxymethyl-2’-deoxyuridine-5’-Triphosphate ( TriLink Biotech . ) .	DNA polymerases are enzymes that replicate DNA by using single-stranded DNA as a template . DNA replication is needed to duplicate an organism’s genome , and repair it if it is damaged . For example , when DNA double-strand breaks occur in the genome , DNA polymerases help repair these potentially lethal DNA breaks . If not repaired accurately , double-strand breaks in DNA can lead to genetic mutations and cancer . Cells have evolved many different pathways to repair damaged DNA . DNA polymerase θ ( called Polθ for short ) is a key player in a repair mechanism called ‘alternative end-joining’ in mammals . In this pathway , certain enzymes trim back DNA strands on both sides of the double-stranded break to expose overhanging single strands of DNA . Polθ binds to the ends of both overhanging strands and helps them pair up with each other . Polθ then extends each single strand using the opposing overhanging strand as a template . After the gaps are filled , the DNA junction is sealed to form double-stranded DNA by other enzymes . Previous research suggested that during alternative end-joining Polθ extends single-stranded DNA by using a guiding template strand or not using a template strand . As a consequence , Polθ frequently inserts extra pieces of DNA at the repair junction thereby introducing mutations in the DNA . It is poorly understood how this unusual DNA synthesis mechanism takes place . Kent et al . have now investigated how Polθ extends single-stranded DNA and introduces extra DNA segments during alternative end-joining . Biochemical experiments showed that Polθ spontaneously switches between three distinct modes in which single-stranded DNA is extended . The first mode does not use a template; the second uses the opposing overhanging strand as a template; and the third uses the same overhanging strand which folds back ( or snaps-back ) on itself to act a template . Kent et al . also found that extra DNA pieces are inserted in all these three different modes of activity , and that this process occurs in mouse cells too . Additionally , single-stranded extension without a template was shown to be stimulated by manganese ions . Thus by spontaneously switching between three different modes of single-stranded DNA extension , Polθ is able to incorporate diverse DNA sequence segments at DNA repair junctions . Further work is now needed to understand whether abnormal activity of Polθ contributes to cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "biochemistry", "and", "chemical", "biology" ]	2016	Polymerase θ is a robust terminal transferase that oscillates between three different mechanisms during end-joining
Zolpidem produces paradoxical recovery of speech , cognitive and motor functions in select subjects with severe brain injury but underlying mechanisms remain unknown . In three diverse patients with known zolpidem responses we identify a distinctive pattern of EEG dynamics that suggests a mechanistic model . In the absence of zolpidem , all subjects show a strong low frequency oscillatory peak ∼6–10 Hz in the EEG power spectrum most prominent over frontocentral regions and with high coherence ( ∼0 . 7–0 . 8 ) within and between hemispheres . Zolpidem administration sharply reduces EEG power and coherence at these low frequencies . The ∼6–10 Hz activity is proposed to arise from intrinsic membrane properties of pyramidal neurons that are passively entrained across the cortex by locally-generated spontaneous activity . Activation by zolpidem is proposed to arise from a combination of initial direct drug effects on cortical , striatal , and thalamic populations and further activation of underactive brain regions induced by restoration of cognitively-mediated behaviors . Induced recovery of spoken language , cognitive and motor functions following administration of zolpidem , a gamma-aminobutyric acid type A ( GABA-A ) subclass alpha 1 receptor positive allosteric modulator ( Hambrecht-Wiedbusch et al . , 2010 ) in severely brain-injured subjects with disorders of consciousness are rare , but well documented ( Clauss et al . , 2000; Brefel-Courbon et al . , 2007; Cohen and Duong , 2008; Shames and Ring , 2008; Whyte , 2009; Hall et al . , 2010 ) . In some instances , behavioral improvement spans the range from only limited signs of conscious awareness to recovery of conversant language , ambulation and coordinated motor activity within an hour of administration of the medication . However , the mechanisms underlying this phenomenon are poorly understood . Because of the rarity of the phenomenon , as well as logistical impediments , only a small number of prior studies have characterized zolpidem-induced physiological changes these patients ( Clauss et al . , 2000; Clauss and Nel , 2006; Brefel-Courbon et al . , 2007; Cohen and Duong , 2008; Shames and Ring , 2008; Whyte , 2009; Hall et al . , 2010; Nyakale et al . , 2010 ) . Here we investigate the detailed neurophysiological changes present in the resting electroencephalogram ( EEG ) in three patient subjects with such paradoxical behavioral facilitation in response to zolpidem and widely disparate underlying etiologies of severe brain injury and disorders of consciousness ( severe hypoxic-ischemic encephalopathy , severe diffuse axonal injury and multifocal ischemic injuries ) . All subjects were more than 12 months post-injury and had histories of zolpidem response and regular usage of the drug for behavioral facilitation . Despite the wide differences in the etiology of brain injury across subjects , quantitative analysis of resting EEG dynamics reveals a striking commonality of spectral features in the absence of zolpidem , and stereotyped changes in these spectral quantities when the drug is administered . Based on the specific spectral changes observed , we propose a unifying biological mechanism accounting for the observed dynamical features of the resting brain state and the changes in brain dynamics linked with zolpidem-induced behavioral facilitation . Further , our interpretative model implies a mechanistic basis for the differences in EEG dynamics associated with the zolpidem response in subjects with global brain injury and those observed in a stroke patient ( Hall et al . , 2010 ) , and offers predictions for distinguishing these two mechanisms . The three patient subjects were drawn from a wider observational study of recovery after severe brain injuries based on their history of behavioral facilitation with zolpidem; all subjects received zolpidem on a regular basis prior to this study ( see ‘Methods’ Clinical Histories ) . As a first step , the zolpidem-induced behavioral responses were confirmed by formal assessments of behavior ON and OFF drug by the research teams ( ‘Methods’ ) . For subjects 1 and 3 , this assessment was made with the Coma Recovery Scale-Revised , a well-established , standardized rating scale for patients with disorders of consciousness ( Giacino et al . , 2004; Seel et al . , 2010 ) ; the CRS-R captures variations in cognitively-mediated behaviors up to a level of reliable and consistent communication through speech or gesture . For subject 2 , baseline behavior was at ceiling for the CRS-R , so behavioral assessment was carried out by structured clinical observations . All three subjects showed reproducible behavioral improvements following zolpidem . Figure 1 shows CRS-R ratings ( total and subscale ) for Subject 1 , following five doses ( red lines ) of the medication ( four doses with concomitant video/EEG recordings studied here are labeled on Figure with red arrows ) . For assessments in the OFF-drug period ( the first assessment , which followed a 62-hr washout , and at every time point that was at least 4 hr after zolpidem administration ) , the total CRS-R score ranged from 10–15 . After each administration of zolpidem , the CRS-R score rose to ceiling ( total score = 23 ) . This change in CRS-R total score reflects improvements in all subscales , including recovery of functional movements , consistent communication , and elements of executive function . Additional changes not captured by this psychometric instrument included recovery of fluent verbal communication , writing and complex organized movements such as assembling block structures to match arbitrary configurations ( see ‘Clinical histories’ for further clinical details and additional neuropsychological assessments ) . As seen in Figure 1 , maximal total CRS-R scores consistently appeared within 1 hr after drug administration . The period during which the maximal CRS-R score was maintained appeared to be longer for the second dose of the day ( doses two and four ) than the corresponding first dose ( doses one and three ) . 10 . 7554/eLife . 01157 . 003Figure 1 . Behavioral changes associated with zolpidem doses in Subject 1 . Subject 1 demonstrated only a limited range of behaviors in the baseline ( OFF drug ) state including automatic motor responses ( e . g . , reaching to a hand extended for a handshake , CRS-R motor subscore 5 ) , oro-motor behaviors ( e . g . , opening mouth when presented with a spoon CRS-R oro-motor subscore 2 or biting a tongue depressor , CRS-R oro-motor subscore 1 ) , localization of sound with head turning ( CRS-R auditory subscale score 2 ) , and reaching to objects ( CRS-R visual subscale score 4 ) . During all baseline assessments the patient demonstrated no evidence of command following or a communication system ( CRS-R communication subscale score 0 ) . Across three assessments of baseline behavior OFF drug after overnight periods and a 62 hr washout period at the onset of the study total CRS-R scores ranged from 10–15 reflecting a lack of goal-directed behaviors , evidence of any communication systems either verbal or gestural , nor consistent response to command following . Across two assessments of baseline behavior at least 4 hr after a prior dose of zolpidem within a day ( reflecting typical duration of action of the medication ) maximal total CRS-R scores of 18 reflected evidence of command following with inaccurate communication and higher level motor function ( CRS-R subscale motor score of 6 ) or consistent auditory command following ( CRS-R auditory subscale score of 4 ) . Compared with these baseline behavioral assessments , consistent achievement of a maximum possible total CRS-R scores of 23 was obtained during all ON drug periods reflecting a state in which the patient consistently demonstrated behavioral levels not captured by this psychometric instrument including recovery of consistent communication , fluent verbal communication , writing and complex organized movements ( see ‘Clinical histories’ for further clinical details and additional neuropsychological assessments ) . As seen in graph , a maximal total CRS-R scores consistently appeared following drug administration within approximately 1 hr and had a variable duration of maintenance with second daily dose of the medication showing extended time periods of maximal total scores . Red arrows indicate zolpidem doses for which accompanying EEG data were available for analyses . Videos 1 and 2 illustrate aspects of the examinations to show correspondence of numerical ratings and behavior . DOI: http://dx . doi . org/10 . 7554/eLife . 01157 . 00310 . 7554/eLife . 01157 . 004Video 1 . Demonstration of CRS-R Motor subscale Subject 1 during off zolpidem state . Prior to zolpidem administration subject is tested with two common objects ( comb and spoon ) and asked to demonstrate their use . While able to hold each object he is unable to demonstrate their use . This level of behavior corresponds to a 5 on the CRS-R Motor subscale . Note resting tremor of right hand . DOI: http://dx . doi . org/10 . 7554/eLife . 01157 . 00410 . 7554/eLife . 01157 . 005Video 2 . Demonstration of CRS-R Motor subscale Subject 1 following zolpidem administration . Short legend: Following zolpidem administration subject is again tested with two common objects ( comb and spoon ) and asked to demonstrate their use . He is now able to demonstrate the use of each object . This level of behavior corresponds to a 6 on the CRS-R Motor subscale . Note that subject’s posture has changed and that the right hand is used . DOI: http://dx . doi . org/10 . 7554/eLife . 01157 . 005 For subject 2 , the behavioral baseline in the OFF drug state included consistent communication and command-following , and the CRS-R was at ceiling . As documented by structured clinical observations following nine doses , this patient’s zolpidem-induced behavioral facilitation included marked improvements in oro-motor control of chewing and swallowing , and increased verbal fluency . Additionally , there was a suppression of a coarse tremor and restoration of goal-directed fine motor-control of the dominant right arm and hand . This subject had received three daily doses of zolpidem for five years at the time of the study; an earlier report of zolpidem induced behavioral facilitation in this subject from the third year of use of the medication ( Clauss and Nel , 2006 ) indicated a shift from a level II Rancho Los Amigos Cognitive Score to a level VI-VII indicating that at baseline purposeful vocalization was not observed in the OFF drug state at that time . For subject 3 , the behavioral baseline was characterized by reproducible responses to command , visual fixation , automatic motor responses , intelligible verbalization and intentional communication . These behavioral responses , however , could only be elicited with vigorous arousal enhancing maneuvers . After zolpidem administration on a single formal assessment , CRS-R score increased from 15 to 16 . This one-point change reflected a change in the CRS-R arousal subscale score from 1 to 2: a shift from a state that required constant stimulation to maintain arousal to a state characterized by spontaneous eye-opening and behavioral responsiveness . Other CRS-R subscale scores did not change but no longer depended on external stimulation to elicit responsiveness . In addition , the subject also showed increased alertness ( as judged by orientation to sensory stimuli ) and agitated movements , with an increase in speech rate and words that were subjectively rated as more intelligible ( see ‘Methods’ Clinical Histories ) . To characterize changes in spontaneous brain dynamics associated with zolpidem administration , we analyzed the EEG around the times of zolpidem administration . Subjects 1 and 2 were studied over several days in a hospital setting; Subject 3’s evaluation was limited to assessment of a single dose of zolpidem while at home ( ‘Methods’ ) . We begin by comparing the EEG in the hour preceding and following zolpidem administration , and then consider its further evolution in time . Figure 2 shows results from the frontal midline electrode pair Fz-Cz for all three subjects . Prior to zolpidem administration , each subject’s data demonstrates a peak of power in the range 6–10 Hz . In all cases , this peak is sharply attenuated 1 hr following drug administration , and there is an increase in power in the range ∼15–30 Hz ( p<0 . 05 by a z-statistic , the Two Group Test ( Bokil et al . , 2007 ) , see ‘Methods’ for approach to significance testing , Figure 3—figure supplement 1; Tables 1 and 2 ) . Figure 2—figure supplement 1 shows that the changes in spectral shape do not represent a ‘performance confound’ . Specifically , we determined EEG power spectra only from periods in which the patient was at rest and not interactive with persons or objects in their environment and compared this with the power spectra obtained during active interactions with people at the bedside . The level of environmental stimulation did not influence the spectrum , and the zolpidem effect of Figure 2 was similarly present for both periods of either low or high levels of environmental stimulation . 10 . 7554/eLife . 01157 . 006Figure 2 . Power spectra estimated from midline EEG channel Fz-Cz recordings . Power spectra from all three subjects ( mean and 95% confidence intervals ) . Red: average spectral power in the hour prior to each zolpidem dose . Blue: 20–60 min after the zolpidem dose . Narrow , low frequency spectral peaks are apparent in the pre-drug state that are attenuated in the hour post-dose . Beta range peaks ( 20–30 Hz ) are apparent in all post-drug spectra . Changes in ∼6–10 Hz peak between conditions are significant by two-group test ( see ‘Methods’; Table 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01157 . 00610 . 7554/eLife . 01157 . 007Figure 2—figure supplement 1 . Analysis of EEG segments obtained during ON and OFF drug periods from Subject 1’s second zolpidem administration ( cf . Figure 1; Table 1 ) subdivided according to periods of low and high levels of environmental stimulation , a distinction based on detailed review of the simultaneously recorded video . ‘Quiet’ segments were chosen from periods in which the subject appeared to be resting quietly with eyes open and there was no direct engagement of the subject ( verbal engagement or physical manipulation ) . ‘Engaged’ segments were chosen from periods in which the patient was interactive with investigators at the bedside . EEG channels from the midline are analyzed , as these typically have less movement and muscle artifacts , and thus allow comparison during active engaged periods . The level of environmental interaction ( quiet vs engaged ) has no effect on either the background or the post-zolpidem spectra . In contrast , marked changes in spectral shape following zolpidem administration ( first hour vs baseline ) are seen for both quiet and engaged periods . DOI: http://dx . doi . org/10 . 7554/eLife . 01157 . 00710 . 7554/eLife . 01157 . 008Table 1 . Relationship of low frequency peak in OFF zolpidem baseline and time interval between dosesDOI: http://dx . doi . org/10 . 7554/eLife . 01157 . 008SUBJECT 1 center frequency for low frequency peak in hour prior to dose across channels and transitionsBaseline 1SignificantBaseline 2SignificantBaseline 3SignificantBaseline 4Significant ( 62 hr off drug ) On vs off ( 17 hr off drug ) On vs off ( 9 hr off drug ) On vs off ( 5 hr off drug ) On vs offChannelPairs9:30AM , 1stsuppression12:00PM , 1stsuppression6:30PM , 2ndsuppression5:00PM , 2ndsuppressionFpzFp17 . 77 . 37 . 7––FpzFp277 . 37 . 3––Fp1F37 . 77 . 37 . 37Fp2F47 . 37 . 37 . 36 . 3F1FC17 . 37 . 77 . 37FC1C37 . 37 . 777FC2C4––7 . 37––F3FC16 . 77 . 36 . 77 . 3F4FC27 . 37 . 777 . 7F2FC27 . 3n . s . 7 . 777n . s . Fp1AF76 . 37 . 77 . 37Fp2AF87 . 37 . 37 . 3––AF7F76 . 77 . 37 . 36 . 3AF8F877 . 37 . 3––F7FC56 . 77 . 37 . 36 . 3F8FC67 . 77 . 37 . 38 . 3n . s . FC5T367 . 37 . 37 . 3FC6T47 . 3n . s . 7 . 3––––T3CP5––7 . 37 . 39 . 7T4CP67 . 77 . 37 . 7n . s . 9 . 3C3CP1––7 . 777C4CP2––7 . 77 . 37FzCz7 . 37 . 777Anterior Channels Average Center Freq7 . 17 . 47 . 27 . 3 Standard Deviation0 . 50 . 20 . 21 . 0T5PO7––7 . 37 . 36 . 7n . s . T6PO87 . 77 . 37 . 711CP1P37 . 77 . 76 . 76 . 7CP6T67 . 78 . 3––8n . s . CP2P47 . 3n . s . 8––9n . s . CP5T5––7 . 77 . 39 . 7P3O17 . 37 . 77 . 39P4O2––8–-––PO7O17 . 37 . 37 . 79PO8O27 . 77 . 77 . 3-–CzPz––7 . 777CPzPOz7 . 3n . s . ––6 . 76 . 7POzOz7 . 77 . 37 . 3––Posterior Channels Average Center Freq7 . 57 . 77 . 28 . 3 Standard Deviation0 . 20 . 30 . 31 . 5Average across all Average Center Freq7 . 37 . 57 . 27 . 7 Standard Deviation0 . 40 . 30 . 31 . 2Table 1 shows average center frequency of low frequency peak at∼6–10 Hz , if present , across all channels for Subjects 1 in relation to time off zolpidem prior to 1 hr baseline EEG measurements and clock time of each dose . In both patients , shorter zolpidem dosing intervals are associated with higher center frequencies across posterior but not anterior EEG channels . Similarly , an increased standard deviation of the measurement is observed as interval between doses is shorter for the posterior EEG channel measurements . Of note , for both subjects , anterior EEG channel pre-zolpidem dose baselines revealed a consistent∼7 . 4 Hz average center frequency with a small standard deviation . The consistency of these findings despite the wide difference in their underlying etiologies of injury support the proposed common cellular and circuit mechanism; the correspondence of the∼7 . 4 Hz peak and intrinsic oscillation frequency of neocortical Layer V cells ( Silva et al . , 1991 ) also support this model . 10 . 7554/eLife . 01157 . 009Table 2 . Relationship of low frequency peak in OFF zolpidem baseline and time interval between dosesDOI: http://dx . doi . org/10 . 7554/eLife . 01157 . 009SUBJECT 2 center frequency for low frequency peak in hour prior to dose across channels and transitionsBaseline 1SignificantBaseline 2SignificantBaseline 3SignificantBaseline 4SignificantBaseline 5Significant ( 20 hr off drug ) On vs off ( 16 hr off drug ) On vs off ( 6 hr off drug ) On vs off ( 5 hr off drug ) On vs off ( 4 hr off drug ) On vs offChannelPairs4:30PM , 1stsuppression8:30AM , 1stsuppression4:30PM , 2ndsuppression5:00PM , 3rdsuppression12:15PM , 2ndsuppressionFp1F37–n . s . 7 . 37 . 7n . s . 8n . s . Fp2F478 . 57 . 377 . 7n . s . F3C37 . 78 . 57 . 77 . 77 . 7F4C47 . 787 . 77 . 77 . 3F7T37 . 7n . s–7 . 7n . s7 . 7n . s . 6n . s . F8T4 ( No data ) 8n . s . 7 . 37n . s . 7 . 3n . s . Fp1F77n . s–7 . 77 . 7n . s . 7n . s . Fp2F87 . 7–n . s . 76 . 7n . s . 7 . 3n . s . FzCz7 . 78 . 577 . 77n . s . Anterior Channels Average Center Freq7 . 48 . 37 . 47 . 47 . 3 Standard Deviation0 . 40 . 30 . 30 . 40 . 6C3P38 . 38 . 57 . 78 . 38 . 7n . s . C4P4899 . 38 . 37 . 7n . s . P3O17 . 7911 . 38 . 78n . s . P4O28 . 39108 . 78 . 3n . s . T3T578 . 57 . 711 . 7n . s . 11 . 7n . s . T4T6 ( No data ) 99 . 3n . s1010n . s . T5O189108 . 311 . 3n . s . T6O289 . 5109 . 711n . s . CzPz8 . 38 . 598 . 7*7 . 7n . s . Posterior Channels Average Center Freq8 . 08 . 99 . 49 . 29 . 4 Standard Deviation0 . 40 . 31 . 21 . 11 . 6 Average Center Freq7 . 78 . 78 . 48 . 38 . 3 Standard Deviation0 . 50 . 41 . 31 . 21 . 6Table 2 shows average center frequency of low frequency peak at∼6–10 Hz , if present , across all channels for Subjects 2 in relation to time off zolpidem prior to 1 hr baseline EEG measurements and clock time of each dose . In both patients , shorter zolpidem dosing intervals are associated with higher center frequencies across posterior but not anterior EEG channels . Similarly , an increased standard deviation of the measurement is observed as interval between doses is shorten for the posterior EEG channel measurements . Of note , for both subjects , anterior EEG channel pre-zolpidem dose baselines revealed a consistent∼7 . 4 Hz average center frequency with a small standard deviation . The consistency of these findings despite the wide difference in their underlying etiologies of injury support the proposed common cellular and circuit mechanism; the correspondence of the∼7 . 4 Hz peak and intrinsic oscillation frequency of neocortical Layer V cells ( Silva et al . , 1991 ) also support this model . Of note as an outlier is the second baseline for this subject which is the only measurement early in the morning ( 8:30 ) suggesting a possible interaction with diurnal activity of the brain arousal system , however , insufficient data is available to establish this linkage . Figure 3 , Figure 3—figure supplement 1; Tables 1 and 2 show that in each patient , the changes in spectral shape following zolpidem administration are widely distributed across the electrode array . 10 . 7554/eLife . 01157 . 010Figure 3 . Power spectra ( mean and 95% confidence intervals ) estimated from selected EEG channel recordings across the head from all three subjects ( A , B , and C ) . Red: average spectral power in the hour prior to each zolpidem dose . Blue: 20–60 min after the zolpidem dose . Narrow , low frequency spectral peaks are apparent in the pre-drug state that are attenuated in the hour post-dose . Beta range peaks ( ∼20–30 Hz ) are apparent in all post-drug spectra . Changes in ∼6–10 Hz peak between conditions are significant by two-group test ( ‘Methods’; Table 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01157 . 01010 . 7554/eLife . 01157 . 011Figure 3—figure supplement 1 . Significance testing of power across frequencies comparing the OFF and ON zolpidem states for Subject 1 , transition 1 ( baseline vs first hour ) is summarized using two-group test ( see ‘Methods’; Bokil et al . , 2007 for further methods ) . Significant increases in power with zolpidem administration compared to baseline are plotted as red asterisks , significant decreases are plotted as blue asterisks . Comparison of power at different frequencies in the OFF and ON zolpidem states shows that significant suppression of ∼6–10 Hz coherence is evident across a majority of channels with administration of zolpidem . DOI: http://dx . doi . org/10 . 7554/eLife . 01157 . 011 Although some individual channels show additional specific features ( e . g . , a low frequency peak at ∼12 Hz in channel pair C4-CP2 in Subject 1 ) , the low frequency oscillations ∼6–10 Hz are diffusely present across brain regions in the baseline condition in all subjects , and absent following zolpidem . Similarly , widely-distributed increases in ∼15–30 Hz power are observed after zolpidem administration in all three subjects ( Figure 3 ) . Analysis of responses to multiple drug administrations was carried out in Subject 1 ( four administrations ) and Subject 2 ( five administrations ) . Figure 4 shows the spectra at Fz-Cz in order of decreasing washout time of the prior zolpidem dose ( Tables 1 and 2 ) . Consistent with the data shown in Figures 2 and 3 , all transitions show a decrease in power at 6–10 Hz and an increase from 15–30 Hz following drug administration ( p<0 . 05 ) . In both subjects , the low-frequency peak is least prominent when prior washout period is shortest , suggesting that there is a carry-over effect at these dosing intervals . 10 . 7554/eLife . 01157 . 012Figure 4 . Power spectra from midline channels for subjects 1 and 2 across multiple transitions from a washout baseline to ON drug state . Eyes open , awake epochs selected from one hour prior to 1 hr subsequent to each dose ( Red ) are shown and compared to 20–60 min after the zolpidem dose ( Blue ) . Changes in ∼6–10 Hz peak between conditions are significant by two-group test ( Tables 1 and 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01157 . 012 Further analysis of changes in the power spectrum across different washout intervals after each dose revealed three additional findings: ( 1 ) a clear distinction between changes in resting low frequency oscillations present over anterior ( frontocentral-temporal ) and posterior ( centro-parietal-occipitotemporal ) channels , ( 2 ) a common average resting low frequency peak across anterior channels , and ( 3 ) a consistent increase in the average center frequency of the posterior but not the anterior EEG channels with shorter dosing intervals . Comparison of the anterior and posterior channels in both subjects reveal a consistent finding of lower resting average low frequency in anterior channels typically ∼7 . 4 Hz . In Table 1 , the average center frequency for anterior EEG channels in Subject 1 ranges from 7 . 1 Hz to 7 . 4 Hz , while posterior EEG channels show a higher average center frequency ranging from 7 . 5 Hz to 8 . 3 Hz . A sharper difference for anterior and posterior EEG channels is seen in Table 2 for Subject 2; for this subject , anterior EEG channels show an average center frequency of ∼7 . 4 Hz for four of the five measured baselines , with the notable exception of the second baseline measurement . The second baseline measurement in Subject 2 occurred early in morning ( 8:30AM ) and may reflect an independent influence of arousal state , as a similar effect of increased baseline 15–30 Hz power is seen for this pre-dose baseline ( refer to Figure 4 and discussion below ) . Posterior EEG channels in Subject 2 show an average center frequency range from 8 . 0 Hz to 9 . 4 Hz . Tables 1 and 2 show , in order of decreasing dosing ( washout ) interval , the average center frequency of low frequency peaks in the baseline spectral power across EEG channels for Subjects 1 and 2 prior to each administration of zolpidem . For the longest washout intervals compared to the shortest , the baseline spectral peaks appear at lower frequencies across all channels for both Subjects 1 and 2 ( Tables 1 and 2 ) . For the shortest washout intervals ( Tables 1 and 2 ) , however , the posterior channels demonstrate a persistent shift in the average center frequency suggesting a carry-over effect for these regions of the brain . For example , in the baseline prior to dose four for Subject 1 several EEG electrode derivations show an increase in the baseline low frequency peak including an ∼11 Hz peak arising posteriorly in channel pair T6–PO8; for Subject 2 an 11 Hz peak is similarly seen in their 4 and 5 hr washout baselines in the T5–O1 , T6–O2 and T3–T5 channel pairs . There are other differences between the spectral changes during the individual drug administrations , but their interpretation is unclear because of the limited number of drug administrations that we were able to study . In some instances increased power in the ∼15–30 Hz range is also apparent before and after the drug administration . This variation is substantial and suggests a fluctuation in the baseline state possibly linked to differences in arousal state or other factors not controlled for in this observational study . For Subject 2 , a marked change in the power spectrum in the ∼15–30 Hz range is limited primarily to the first dose which followed a 20 hr washout period and was received late in the afternoon ( 4:30 PM ) , suggesting that the regular dosing schedule in this subject of three doses per day may produce carry-over effects that diminish once the subject is off drug for more extended time periods . In addition to diurnal arousal effects , carry-over effects also appear to affect the 15–30 Hz component as a broad ∼15–30 Hz peak also remains present in this range for the shortest washout baselines in Subject 1 ( 4 hr; Figure 4D ) . The main findings of suppression of ∼6–10 Hz activity and increase of ∼15–30 Hz activity with zolpidem administration showed a general robustness across all three subjects and in Subjects 1 and 2 across separate assessments of ON and OFF drug doses ( Tables 1 and 2 ) . Consistent patterns of EEG dynamics across Subjects 1 and 2 are also observed over the 2 . 5–3 hr course of each dose , as shown in Figure 5 . In both subjects , the increase in high frequency power has two phases: maximal increases in power ∼15–30 Hz occur within the first 30–40 min , and this is followed by an attenuation of power and narrowing of frequency range to ∼20–30 Hz . By 2 hr after dosing , the enhanced high-frequency power has noticeably decayed . In contrast , both subjects show suppression of low-frequency power beyond this point ( up to 3 hr in subject 1 , and up to 4 hr in subject 2 ) , suggesting that the prolonged effect on low-frequency activity may underlie the ‘carry-over’ effect . ( Data were not collected for subject 3 beyond the first hour after zolpidem administration and similar analyses are thus not available ) . 10 . 7554/eLife . 01157 . 013Figure 5 . Time-frequency analysis of selected EEG channels from Subject 1 ( A ) and Subject 2 ( B ) . Frontal and posterior midline channels , for both subjects demonstrate a low frequency peak in power ( red arrows ) during the hour prior to zolpidem administration attenuates within the first 10–15 min after the drug is given . This corresponds to the time period when the subject begins to manifest improved behavioral function . Concomitant with the attenuation of the low frequency peak , there appears a broader ∼15–25 Hz peak during the 30 min after the drug is given ( white arrows ) , which narrows and reduces attenuates slowly over the next 2–3 hr . In Subject 1 ( A ) , a ∼10 Hz peak appears approximately 1 hr into the post-dose period in CPz-POz channel ( green arrow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01157 . 013 Of particular interest is a finding observed in two posterior channels from subject 1 in one of the three transitions analyzed ( dose two ) . As seen in the third spectrogram ( bottom panel ) shown in Figure 5A , the posterior channels CPz-POz reveal the appearance of a ∼10 Hz feature in the spectrogram around 1 . 25–1 . 5 hr after administration of the dose ( green arrow ) . This feature of the CPz-POz spectrogram is distinct from the low frequency ∼7 Hz feature ( red arrows ) and high frequency 15–25 Hz feature ( white arrows ) seen in the other EEG channel spectrograms ( Figure 5A ) . While not reproduced in other transitions , this feature suggests a possible restoration of the posterior alpha rhythm arising later in the course of ON drug state . To determine whether zolpidem administration altered the relationship between activity across cortical regions , we examined its influence on EEG coherence . Representative findings are shown in Figure 6A , B , C ( and Figure 6—figure supplement 1 ) for intra-hemispheric coherence , and in Figure 7 for inter-hemispheric coherence ( and Figure 7—figure supplement 1 ) . Consistently across subjects , the OFF drug state was associated with a coherence peak range 0 . 6–0 . 8 at ∼6–10 Hz , the same range as the low-frequency power peak ( p<0 . 05 , two-group test ) . Zolpidem administration reduced the level of coherence for most channels to the range 0 . 3–0 . 5 . In subjects 1 and 2 , this was typically accompanied by an increase in coherence in the range of the high-frequency power peak ( 20–30 Hz ) . These changes are most prominent in the transitions following the longest washout periods for Subjects 1 and 2 ( Doses one to three ) . 10 . 7554/eLife . 01157 . 014Figure 6 . ( A ) . Intra-hemispheric coherences ( subject 1 ) . Pre-drug coherence peaks at 6–10 Hz . This peak is attenuated in the hour after the drug is given . Changes in ∼6–10 Hz peak between conditions are significant by two-group test ( ‘Methods’; Figure 6 Figure Supplement 1 ) . ( B ) ( subject 2 ) . Intra-hemispheric coherences ( subject 2 ) . Pre-drug coherence peaks are evident at ∼ 6–10 Hz . This peak is attenuated in the hour after the drug is given . ( C ) ( subject 3 ) . Intra-hemispheric coherences ( subject 3 ) . Pre-drug coherence peaks at ∼ 6–10 Hz . This peak is attenuated in the hour after the drug is given . DOI: http://dx . doi . org/10 . 7554/eLife . 01157 . 01410 . 7554/eLife . 01157 . 015Figure 6—figure supplement 1 . Significance testing of intra-hemispheric coherences for Subject 1 , transition 2 ( baseline vs first hour ) is summarized using two-group test ( see ‘Methods’; Bokil et al . , 2007 for further methods ) . Significant increases in coherence with zolpidem administration compared to baseline are plotted as red asterisks , significant decreases are plotted as blue asterisks . Comparison of coherence in the OFF and ON zolpidem states shows that significant suppression of ∼6–10 Hz coherence is evident across a majority of intra-hemispheric channel pairs with administration of zolpidem . DOI: http://dx . doi . org/10 . 7554/eLife . 01157 . 01510 . 7554/eLife . 01157 . 016Figure 7 . Inter-hemispheric coherences all three subjects . Pre-drug coherence peaks at 6–10 Hz are seen in frontal and posterior channels comparisons for all subjects . This peak is attenuated in the hour after the drug is given . Changes in ∼6–10 Hz peak between conditions are significant by two-group test ( see ‘Methods’; Figure 7—figure supplement 1 , for Subject 1 coherences for example of coherence changes across all channel pairs in single dose ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01157 . 01610 . 7554/eLife . 01157 . 017Figure 7—figure supplement 1 . Significance testing of inter-hemispheric coherences for Subject 1 , transition 2 ( baseline vs first hour ) is summarized using two-group test ( see ‘Methods’; Bokil et al . , 2007 for further methods ) . Significant increases in coherence with zolpidem administration compared to baseline are plotted as red asterisks , significant decreases are plotted as blue asterisks . Comparison of coherence in the OFF and ON zolpidem states shows that significant suppression of ∼6–10 Hz coherence is evident across a majority of inter-hemispheric channel pairs with administration of zolpidem . DOI: http://dx . doi . org/10 . 7554/eLife . 01157 . 017 As a result of diverse mechanisms of injury , all three severely brain-injured subjects suffered widespread neuronal death and consequent deafferentation of many neurons across the cerebral cortex and subcortical structures . The implications of severe deafferentation of cortical neurons for EEG dynamics can be estimated by comparing two extremes: completely de-afferented ‘slabs’ of cortical tissue , and normal cortical tissue . In the former case ( in vivo studies of completely deafferented ‘slabs’ ) , neocortical neurons typically have an intracellular potential of ∼-70 mV and the dominant frequency of activity measured in far field potentials is ∼1 Hz ( Timofeev et al . , 2000 ) . Intact , fully connected neocortical neurons have an average membrane potential in active wakeful states , that ranges from ∼−65 to −55 mV ( Steriade , 2001 ) and the dominant frequency of far-field potentials is in the 8–12 Hz range . Given these benchmarks it is reasonable to assume that in our patient subjects , all with very significant but incomplete deafferentation , that average membrane potentials are intermediate between these conditions . In this context , the prominent low frequency peaks seen across the power spectra from anterior EEG channels and across different OFF drug baselines of ∼7 . 4 Hz are notable for comparison to findings from in vitro studies of spontaneous oscillations generated by neocortical pyramidal cells . Silva et al . ( Silva et al . , 1991 ) identified that nearly identical frequencies of sustained firing lasting over seconds ( 7 . 4 Hz SD +/−0 . 6 ) are produced by layer V neocortical pyramidal cells in response to very brief ( 4 ms ) current pulses . These sustained oscillations could be initiated under physiologic conditions in vitro that maintain intrinsic membrane potentials near firing threshold at ∼−60–65 mV ( Silva et al . , 1991 ) . As seen in Tables 1 and 2 , an average center frequency of 7 . 4 Hz ( ∼SD +/−0 . 6 ) characterizes all but one of nine separately measured baselines across the anterior EEG channels of Subjects 1 and 2 . Thus , we propose that the markedly deafferented conditions of cortical neurons in the reduced slice preparation may reasonably approximate the markedly deafferented state of the cortex in our three subjects and that self-sustained intrinsic membrane oscillations arising from Layer V pyramidal neurons triggered by random synaptic background activity are the underlying mechanism for the origin of the ∼6–10 Hz activity present in the resting EEG at baseline . In the slice preparation , increases in membrane depolarization with stronger and more sustained injected currents produced shifts of these intrinsically generated oscillatory firing patterns toward higher frequencies ( up to 12 Hz ) when kept below firing threshold ( Silva et al . , 1991 ) indicating that a wider range of frequencies beyond 7 . 4 Hz could in principle still reflect such pathologically driven oscillations ( and in Subject 3 the resting oscillations tend to be at higher frequencies as shown in Figures 1 and 3C ) . Moreover , fully depolarizing these neurons in the slice results in similar shifts to ∼15–30 Hz firing rates ( Silva et al . , 1991 ) . If the intrinsic properties of partially de-afferented cortical tissue accounts for the peak of EEG activity in the 6–10 Hz range , then what could account for its high levels of spatial coherence ? We propose that this coherence arises from a simple straightforward dynamical mechanism: the ‘Huygens’ clock principle’ ( refer to Huygens , 1665; Rosenblum and Pikovsky , 2003 ) . According to this principle , when oscillators with similar frequencies interact weakly , stable synchronization around a consensus frequency is the generic result . Here , we hypothesize that the net effects of residual Layer V output neurons ( efferent to other cortical regions either directly or via subcortical structures ) constitute a weak coupling . The predicted result is widespread coherence of cortical columns at a consensus theta frequency ( ∼7 . 4 Hz ) , as we in fact observe in all three subjects . We speculate that our model for common baseline EEG dynamics may also account for other widespread rhythmic oscillations that are described in the cardiac arrest and other types of severe anoxic brain injury . These oscillations are in the alpha ( Young et al . , 1994 ) and less commonly theta ( 3–7 Hz ) frequency range . Slower frequencies and lack of a change in frequency in response to sensory stimulation are associated with worse outcomes ( Young et al . , 1994 ) , consistent with the notion that this activity results from de-afferentation , and the dominant frequency range indexes its severity . However , we do not propose that all forms of pathological rhythmic activity arise from this mechanism , only to those in which the structural brain injuries are severe and diffuse . Theta ( 3–7 Hz ) activity has also been reported in patients with focal cortical abnormalities or neuropsychiatric disorders without widespread neuronal loss or disconnection . For example , somewhat similar increases in theta power observed in the resting MEG signal recorded from patients with Parkinson’s disease , obsessive compulsive disorder and other syndromes have been proposed to arise in the setting of less severe deafferentation and disfacilitation of thalamic neurons producing a ‘thalamocortical dysrhymia’ syndrome ( Jeanmonod et al . , 1996; Llinás et al . , 1999 ) . These findings are correlated with theta frequency bursting identified in the thalamus ( driven by de-inactivation of low-threshold T-type calcium channels ( Jeanmonod et al . , 1996 ) ) of patients with Parkinson’s disease or central pain ( Jeanmonod et al . , 1996; Llinás et al . , 1999 ) who lack broad global cerebral deafferentation that characterizes our subject population with disorders of consciousness . The distinction between global and focal injuries leading to rhythmic theta activity is important because these scenarios predict opposite effects on locally measured EEG activity in the 15–30 Hz range . Specifically , experimental and clinical studies of thalamocortical dysrrhythmia link abnormal increases in theta power to coincident abnormal increases 15–30 Hz activity produced by lateral inhibition in the cortex ( ‘edge effect’ , Llinás et al . , 1999; Llinás et al . , 2005 ) . Of note , the linkage of elevated 4–10 Hz and 15–30 Hz activity has been observed in a study of a zolpidem responsive stroke patient subject ( Hall et al . , 2010 ) . Hall et al . ( 2010 ) studied MEG signals in a fully conscious , alert , and cognitively normal patient subject with mild aphasia and diminished sensorimotor integration ; these investigators reported that zolpidem administration produced similar changes of power suppression in co-localized areas of abnormally increased 4–10 Hz activity and increased 15–30 Hz activity in the resting MEG signal measured near the boundaries of a large stroke lesion within the damaged hemisphere . This suppression of increased high frequency power with zolpidem is exactly opposite to the changes observed in ∼15–40 Hz power in our subjects and indicates that a different mechanism underlies the changes seen in the present study , compared to that underlying the resting MEG activity in the Hall et al . subject ( Hall et al . , 2010 ) . In the latter subject , global cerebral structures , including both thalami , remained intact and the patient’s state of consciousness was normal ( and there also was no history of a preceding disorder of consciousness ) . Moreover , in the Hall et al . study , no increases were observed in ∼15–40 Hz power outside of MEG channels with abnormal 3–7 Hz power ( where ∼15–40 Hz suppressed with zolpidem administration ) . Since the elevated theta and beta activity seen in the Hall et al . study was focal , it is more likely to be due to bursting of a localized set of deafferented thalamic neurons ( i . e . , thalamocortical dysrhythmia ) , than to widespread cortical de-afferentation . To produce thalamocortical dysrhythmia , low-threshold T-type calcium channels must be de-inactivated ( Jeanmonod et al . , 1996; Llinás et al . , 1999 , 2005 ) . This is far more probable to arise in this setting with a more locally deafferented thalamus in largely structurally intact brain . Taken together with the model proposed above , the distinct pattern of spectral changes observed in the Hall et al . patient leads to a prediction concerning the substrate of the pathological ∼3–7 Hz EEG oscillations in patients with severe brain injuries . In our subjects—with a widely coherent theta rhythm—we expect that thalamic activity is severely suppressed and not generating widespread bursting activity . In contrast , when focal ∼3–7 Hz and ∼15–40 Hz oscillations coexist ( as in the Hall et al . patient [Hall et al . , 2010] ) , we hypothesize that the thalamus is not globally suppressed , and that instead , there is bursting due to focal thalamic deafferentation . The former expectation is supported by metabolic imaging studies in Subject 2 , ( Figure 8 , Table 3 ) , and other studies in awake severely brain-injured subjects that included direct measurement of thalamic activity that demonstrated loss of central thalamus firing activity ( Giacino et al . , 2012 ) . 10 . 7554/eLife . 01157 . 018Figure 8 . FDG-PET measured cerebral metabolism OFF and ON zolpidem for Subject 2 . A marked increase in global cerebral metabolic rate is seen with zolpidem administration , average increase is 1 . 97 times OFF baseline across cerebral structures ( see ‘Methods’ and Table 3 for regional differences in metabolic rates for selected areas ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01157 . 01810 . 7554/eLife . 01157 . 019Table 3 . Percent change in regional standardized uptake values with administration of zolpidemDOI: http://dx . doi . org/10 . 7554/eLife . 01157 . 019RegionLeft ( % ) Right ( % ) Mesial frontal139146Lateral frontal166183Calcarine145153Striatum156156Thalamus148127 In addition to the thalamocortical dysrhythmia model of 6–10 Hz activity , other theoretical models of the expression of these rhythms in the corticothalamic system exist based on network dynamics ( e . g . , Ching et al . , 2011; Drover et al . , 2011; Hughes et al . , 2004; Robinson et al . , 2002 ) . However , these models do not include gross alterations in physiological profiles of neurons in severely-injured brains with average neuronal membrane potentials remaining well below firing thresholds and likely do not capture the relevant physiological conditions present in our subjects . While an intrinsic cortical mechanism appears to be sufficient to account for our observations , such independent or interacting contributions of network mechanisms linked to thalamocortical projections ultimately cannot be conclusively excluded . We next consider possibilities for how zolpidem might reverse the marked down-regulation of anterior forebrain activity . Two studies have demonstrated increases in cerebral perfusion and metabolism in zolpidem responsive subjects with disorders of consciousness ( Brefel-Courbon et al . , 2007; Nyakale et al . , 2010 ) . Brefel-Courbon et al . ( 2007 ) reported broad activation of FDG-PET signal with zolpidem in a patient remaining chronically in minimally conscious state who recovered speech , swallowing and ambulation reliably with zolpidem administration . These changes were marked and bi-hemispheric , demonstrating broad metabolic increases across the frontal lobes , striatum and thalami of both hemispheres ( Subject 2 studied here a ∼50% reduction in global metabolic rates and relatively reduced frontal , striatal and thalamic metabolic expression consistent with this earlier report as shown in Figure 8 ) . Local increases in blood flow within these same anterior forebrain regions have also been reported in the dominant hemisphere of a stroke patient who recovered fluent speech with zolpidem administration ( Cohen et al . , 2004 ) . A‘mesocircuit’ model has been proposed to account for the mechanisms of action of zolpidem in the context of marked down-regulation of the anterior forebrain in the severely injured brain and restoration of activity across the anterior forebrain during recovery ( Brown et al . , 2010; Schiff , 2010 ) . Briefly , this mesocircuit model incorporates the basic observation that all severe brain injuries produce diffuse disfacilitation across corticothalamic systems arising from widespread disconnection or neuronal death ( Adams et al . , 2000; Maxwell et al . , 2006 ) . As a consequence of their broad and widespread point to point connections , neurons in the central thalamus are progressively more deafferented with severity of structural brain injuries ( Maxwell et al . , 2006 ) . Reduction of corticostriatal , thalamocortical and thalamostriatal outflow following multi-focal deafferentation and loss of neurons across the corticothalamic system likely also results in conditions of sufficient loss of afferent input to the medium spiny neurons ( MSNs ) of striatum to prevent these neurons from reaching firing threshold because of their requirement for high levels of synaptic background activity ( Grillner et al . , 2005 ) . Loss of active inhibition from the striatum in turn is expected to result in release of tonic activity from neurons of the globus pallidus interna ( GPi ) that will provide an active inhibition to their synaptic targets including relay neurons of the already strongly disfacilitated central thalamus and possibly also the projection neurons of the pedunculopontine nucleus ( Rye et al . , 1996 ) . Collectively , these effects are expected to produce very broad reductions in global cerebral synaptic activity as reflected in very low cerebral metabolic rates typically measured in severe brain injuries producing disorders of consciousness ( reviewed in Laureys and Schiff , 2012 ) . The main prediction of this model is that across all etiologies of structural injury , metabolic and functional down-regulation of the anterior forebrain should be prominent . The dominance of the ∼7 . 4 Hz activity across fronto-central EEG channels observed here provide support for this prediction . A primary proposed activating effect of zolpidem is suppression of increased firing of the GPi via a direct effect of zolpidem on GABA-A alpha 1 subtype receptors which are present on all neuronal cell types in the human globus pallidus interna ( Waldvogel et al . , 1999 ) . Zolpidem can also be linked directly to the selective binding of GABA-A alpha 1 receptor subtypes in the neocortex , where it may increase thalamocortical and thalamostriatal outflow indirectly as a result of activation of cortical inhibitory interneuronal networks ( McCarthy et al . , 2009 ) . In addition , zolpidem may activate the striatum where GABA-A currents facilitate alpha and beta ( ∼8–30 Hz ) rhythms within the striatum and normal MSN function ( McCarthy et al . , 2011 ) . Finally , a direct stimulatory action of GABAergic agonist at the pyramidal cell axon could produce excitation and persistent gamma-frequency oscillations ( Traub et al . , 2003 ) . Collectively , all of these potential mechanisms of zolpidem’s action would result in the marked increase of thalamocortical and thalamostriatal outflow , restoration of corticothalamic and corticostriatal outflow , as well as suppression of tonic inhibition of the central thalamus and possibly the PPN by the GPi . Zolpidem induced behavioral facilitation and EEG activation in patients has been proposed to directly link to a phenomena generally observed in anesthesia known as ‘paradoxical excitation’ which arises with propofol and other anesthetics ( Brown et al . , 2010 ) . In the context of anesthesia , broad reductions of background synaptic activity occur during initial sedation . During paradoxical excitation normal subjects demonstrate a brief period of agitation and increased power in the 15–30 Hz frequency range after the initial quieting with light sedation . Paradoxical excitation is more common with GABAergic agonists ( Brown et al . , 2010 ) and the phenomenon may reflect a similar process of release of thalamocortical outflow . A consistent but less prominent shift of fronto-central activity toward increased beta range frequencies ( ∼15–30 Hz ) is initiated by zolpidem in the EEG of awake normal subjects receiving the drug but it is not sustained for more than 30 min before returning to baseline ( Patat et al . , 2004 ) . Thus , it can be expected that all patients administered zolpidem will show similar initial dynamical changes in EEG the early period after drug administration ( ∼30 min ) . However , in those brain-injured patients who show behavioral facilitation with zolpidem this initial shift of ‘paradoxical excitation’ is sustained and evolves over time as shown in Figure 5 and is associated with further dynamical changes ( narrowing of increased activity in the 20–35 Hz range ) . The observations of propofol and zolpidem induced increases in 15–30 Hz activity shows that an initial shift of power in this range is expected in all subjects , it remains unclear how this initial change interacts with the overall increase in cortical activation ( as indexed by the loss of the ∼7 . 4 Hz activity ) and remains self-sustaining . Effective behavioral facilitation likely involves a significant recruitment of cerebral activity across many other cortico-cortical and cortical-subcortical pathways that may become self-sustaining and further activating . This interpretation anticipates other contributions from brain arousal systems and ongoing activity within the cerebrum in maintaining the power distribution in the ∼20–30 Hz range across the frontal-temporal and central-parietal regions as observed for both Subjects studied over hours ( refer to Figure 5 ) . Such engagement of the plurality of arousal systems in ongoing wakeful behaviors is a powerful arousal stimulus as demonstrated experimentally even in animals with primary deficits in the orexin wake promoting systems of the hypothalamus ( España et al . , 2008 ) . The evident carry over effects of continued zolpidem use in Subject 2 noted above suggests that restoration of behavior allows additional benefits to be harnessed in drug responders . However , the washout periods here were not varied experimentally , so a more rigorous assessment of this hypothesis awaits future work . The appearance of several posterior EEG channels with increased average center frequency in the 9–10 Hz in both subjects 1 and 2 in association with shorter washout intervals ( Tables 1 and 2 ) is also consistent with a time-evolving recruitment of activity; expression of more normal patterns of organized rhythms may require the sustained cortical activation over time , with shorter dosing intervals allowing the effects of each dose to build up background brain activity . Subjects 1 and 2 were recorded continuously over 24 hr periods ( with occasional interruption for imaging tests ) using video EEG ( vEEG ) at a sampling rate 1024 Hz for subject 1 and 200 Hz for subject 2 using an XLTEK data acquisition system ( Xltek-Tech , Ltd . , Oakville , Ontario , Canada LH65S1 ) . Silver chloride scalp electrodes ( 19 electrodes for subject 2 , 35 electrodes for subject 1 ) were affixed with collodion according to a modified international 10–20 system of electrode placement , with the lead at the FPz location as the ground reference . Impedance checks were carried out periodically to ensure impedance less than 5 kiloOhms in all leads . Subject 3’s EEG recordings were acquired using a 256 active electrode high-density EEG system ( Electrical Geodesics ) sampled at 500 Hz and referenced for acquisition purposes to Cz ( the EEG acquisition system default setting ) . Videotaped recordings were acquired simultaneously and synchronously to the EEG data to confirm the patients’ behavior . For Subjects 1 and 2 video EEG recordings were manually reviewed on an XLTEK NeuroWorks EEG reviewing station and epochs of 3 s duration without movement artifact were identified that fell within 60 min prior to each recorded zolpidem dose or between 20 and 60 min after each dose . Channels with perceptible muscle artifact were excluded from primary analysis . Additional epochs of 3 s duration with no movement artifact and minimal muscle artifact were identified during the 4 hr subsequent to each zolpidem dose . All signals were detrended and line noise removed using standard MATLAB functions . We then estimated power spectra using the multi-taper method ( Thomson , 1982; Thomson and Chave , 1991; Mitra and Pesaren , 1999; Mitra and Bokil , 2007 ) . With five tapers , the effective frequency resolution obtained was 1 . 7 Hz for 3 s epochs and 2 . 5 Hz for 2 s epochs . The resulting power spectra were then averaged for all epochs in a condition , and 95% confidence intervals were computed via taper-based jackknife techniques ( Thomson and Chave , 1991; Mitra and Bokil , 2007 ) . All spectral estimates were calculated using Chronux functions , written in MATLAB ( Goldfine et al . , 2011 ) . Peaks in power were identified by the presence of a maximum in the spectrum within the frequency range of interest ( e . g . 6–12 Hz for low frequency peaks ) . Time-varying power spectra ( spectrograms , Figure 5 ) were derived as follows: for each minute ( 60 s ) of data recorded between one hour prior to and 3–4 hr after zolpidem administration , an epoch of 3 s duration with the least artifact was selected . Power spectra were then estimated as described above for each epoch , and averaging across epochs was not performed . Coherences were estimated as cross-covariance of the individual signals , normalized by the geometric mean of their power spectra . Nine tapers were used , providing an effective frequency resolution of 3 Hz . To determine significant changes between EEG samples obtained from ON and OFF drug conditions we used the Two Group Test ( TGT , 12 ) , as implemented by the Chronux toolbox routine , two_group_test_spectrum ( http://www . chronux . org ) , with a cutoff of p=0 . 05 by the jackknife method . Because spectral estimates within 2 Hz of each other are correlated by the taper functions , a difference identified by the TGT was only considered significant if it was present for all frequencies contiguously over a range greater than 2 Hz . TGT was applied for all comparisons of spectral power and coherence ( Goldfine et al . , 2011 ) shown here ( see Tables 1 and 2 for power significance testing of suppression of low frequency peaks for all channels and doses in Subjects 1 and 2 , Figure 2—figure supplement 1; Figure 6—figure supplement 1 for example of coherence significance testing across all channel pairs ) .	Some individuals who experience severe brain damage are left with disorders of consciousness . While they can appear to be awake , these individuals lack awareness of their surroundings and cannot respond to events going on around them . Few treatments are available , but a minority of patients show striking improvements in speech , alertness and movement in response to the sleeping pill zolpidem . Although the idea of a sleeping pill increasing consciousness is paradoxical , it is possible that in patients with impaired consciousness , zolpidem reduces the activity of an area of the brain that would otherwise inhibit activity in other regions of the brain . However , the precise mechanisms by which zolpidem increases consciousness in these patients , and the reasons why only a minority of individuals respond , are unknown . Now , Williams et al . have used electrodes attached to the scalp to measure changes in brain activity in three patients known to respond to zolpidem . These measurements showed that before the drug was taken , there were two important differences between the brain activity of the patients and that of healthy subjects: first , the patients showed brain waves of a lower frequency than any seen in healthy subjects; second , these brain waves were much more synchronized than brain activity in healthy individuals . After taking zolpidem , this synchronicity was reduced and all of the patients also showed an increase in higher frequency brain waves . Based on the effects of zolpidem on electrical activity throughout the brain , Williams et al . propose a new model to explain the therapeutic action of the drug in some minimally conscious patients . If the correlation between brain waves and zolpidem response holds up in future studies , this relation could be used to predict which patients might benefit from the drug . A better understanding of these processes should also help us to understand , diagnose and develop new treatments for disorders of consciousness .	[ "Abstract", "Introduction", "Results", "Discussion", "Methods" ]	[ "medicine", "neuroscience" ]	2013	Common resting brain dynamics indicate a possible mechanism underlying zolpidem response in severe brain injury
Natural product screening programs have uncovered molecules from diverse natural sources with various biological activities and unique structures . However , much is yet underexplored and additional information is hidden in these exceptional collections . We applied untargeted mass spectrometry approaches to capture the chemical space and dispersal patterns of metabolites from an in-house library of marine cyanobacterial and algal collections . Remarkably , 86% of the metabolomics signals detected were not found in other available datasets of similar nature , supporting the hypothesis that marine cyanobacteria and algae possess distinctive metabolomes . The data were plotted onto a world map representing eight major sampling sites , and revealed potential geographic locations with high chemical diversity . We demonstrate the use of these inventories as a tool to explore the diversity and distribution of natural products . Finally , we utilized this tool to guide the isolation of a new cyclic lipopeptide , yuvalamide A , from a marine cyanobacterium . The marine environment is extraordinarily rich in both biological and chemical diversity . It is estimated that nearly 90% of all organisms living on earth inhabit marine and coastal environments ( Blunt et al . , 2012; Kiuru et al . , 2014 ) . Moreover , marine organisms are recognized as an extremely rich source of novel chemical entities that have potential utility in medicine , personal health care , cosmetics , biotechnology and agriculture ( Gerwick and Moore , 2012; Kiuru et al . , 2014; La Barre , 2014; Rastogi and Sinha , 2009 ) . Marine cyanobacteria ( ‘blue-green algae’ ) are considered the most ancient group of oxygenic photosynthetic organisms , having appeared on earth some three billion years ago . It is thought that their extensive evolutionary history explains , at least in part , the extensive nature of their structurally-unique and biologically-active natural products ( Burja et al . , 2001; Chlipala et al . , 2011; Schopf , 2012; Whitton and Potts , 2012 ) . Indeed , since the 1970s , several hundred unique chemical entities have been discovered from marine cyanobacteria , and these have been found to possess a remarkable spectrum of biological activities including anticancer , anti-inflammatory , antibacterial , anti-parasitic , neuromodulatory , and antiviral , among others ( Blunt et al . , 2015; Nunnery et al . , 2010; Tan , 2007; Villa and Gerwick , 2010 ) . Examples of marine cyanobacterial natural products that are under investigation for anticancer applications include the cyclodepsipeptides apratoxin A and F , the nitrogen-containing lipid curacin A and the lipopeptides dolastatin 10 and carmaphycin B ( Bai et al . , 1990; Blokhin et al . , 1995; Luesch et al . , 2001a , 2001b; Pereira et al . , 2012; Tidgewell et al . , 2010 ) . These natural product templates have inspired the synthesis of lead compounds using medicinal chemistry approaches , and have resulted in one clinically approved drug ( Brentuximab vedotin ) for the treatment of cancer; several others are in various stages of clinical and preclinical evaluation ( Gerwick and Moore , 2012; Luesch et al . , 2001a; 2001b; Tan et al . , 2010; Tidgewell et al . , 2010; Uzair et al . , 2012 ) . Unfortunately , such fruitful natural product materials , rich with so much unexplored potential , oftentimes disappear upon completion of academic careers and closing of laboratories . In this study , we used a mass spectrometry based approach to digitize ( convert to data format that can be stored , shared and analyzed by computational tools ) the chemical inventory of an established marine cyanobacteria and algae collection in order to better evaluate its diversity and probe for novel natural products . Additionally , while a number of biologically potent natural products have been isolated from marine cyanobacteria and algae , the overall molecular profile of these organisms has not been compared to other classes of bacteria or terrestrial/freshwater cyanobacteria ( Chlipala et al . , 2011 ) . To address this , we analyzed the crude extracts of a relatively large number of algae and cyanobacteria , as well as their derived chromatographic fractions , by high resolution liquid chromatography tandem mass spectrometry ( HR-LC-MS/MS ) . We then compared this metabolomics dataset to four existing public datasets [Mass spectrometry Interactive Virtual Environment ( MassIVE ) repository] within the Global Natural Product Social ( GNPS ) platform ( Wang et al . , 2016 ) . The marine samples used in this study were derived from over 300 field collections from locations in the Indian , Indo-Pacific , Central and South Pacific , and Western Atlantic ( Caribbean ) oceans with initial field taxonomic identifications as predominantly cyanobacteria and macro-algae . It should be noted that these are environmental samples , and therefore are inherently communities of organisms and are not single axenic cultures . Additionally , three samples were from non-axenic laboratory cultures of the cyanobacteria Phormidium ( No . 1646 ) and Lyngbya ( No . 1933 , 1963 ) ; the latter two were recently classified as Moorea ( Supplementary file 1 ) ( Engene et al . , 2012 ) . From both a chemical and biosynthetic perspective , these three cultures represent the most intensively studied organisms in this dataset ( Engene et al . , 2012; Kleigrewe et al . , 2015; Mevers et al . , 2014; Pereira et al . , 2010; Williamson et al . , 2002 ) . While these and other of our cultured filamentous cyanobacteria cannot yet be grown axenically , their MS-derived molecular data represent marine cyanobacterial metabolomics markers that aid in data analysis . Furthermore , in exploring the chemical diversity of marine cyanobacterial and algal assemblages , we established a cartographic platform that combines LC-MS data and geographic locations that facilitates the discovery of new chemical scaffolds as well as identifies geographical areas of high chemical diversity ( Boeuf and Kornprobst , 2009 ) . The discovery of such ‘hotspots’ may reveal new patterns of phylogenetic diversity as well as identify geographical areas with enhanced bioprospecting opportunities . Assessment of the chemical diversity of these marine cyanobacterial and algal collections involved four major steps: ( 1 ) Collection - including permits , field collection , transport , extraction , fractionation and metadata recording; ( 2 ) Data acquisition and digitization in public repository - by LC-HRMS/MS to generate molecular fingerprints; ( 3 ) Data analysis and visualization - clustering similarly structured compounds as molecular families within the GNPS platform , identification of known molecules and assessing the richness and diversity between as well as within samples; ( 4 ) Discovery - identification of geographical patterns of distribution , differentiating common from regiospecific natural products , dereplication of new derivatives and the discovery of previously uncharacterized natural products . Few tools allow assessment of the chemical diversity within large and diverse MS datasets such as those in the current study ( Bouslimani et al . , 2014; Charlop-Powers et al . , 2015; Luzzatto-Knaan et al . , 2015; Purves et al . , 2016 ) . To determine whether this collection of marine cyanobacterial and algal communities possessed unique chemical diversity , the LC-MS/MS data were analyzed with publicly available data sets accessed via the GNPS-MassIVE database ( For datasets see methods ) ( Wang et al . , 2016 ) . These datasets include LC-MS/MS data from terrestrial and marine actinobacteria ( 1000 samples ) , lichens ( 132 samples ) , marine sponges and corals ( 260 samples ) and freshwater cyanobacteria ( 535 samples ) . Collection of all datasets was acquired using tandem mass spectrometry to obtain unique fragmentation fingerprints for each detected molecule . To evaluate the chemical diversity of these source materials , we employed diverse MS-based informatic approaches , including multivariate analyses and molecular networking . Multivariate approaches , such as principal component analysis ( PCA ) and partial least squares-discriminant analysis , ( PLS-DA ) , have been widely employed to assess the chemical diversity of metabolomics data ( Worley and Powers , 2013 ) . Here , for comprehensive metabolomics analysis , we applied principal coordinate analysis ( PCoA ) , a statistical method that provides an overview of the dissimilarities between samples based on the metabolome of an individual sample relative to all other samples in the analysis ( He et al . , 2015 ) The relatedness between samples is calculated by the dissimilarity distance matrix and displayed in a three-dimensional plot where each sphere represents a sample with specific MS/MS features . We utilized the Bray-Curtis distance matrix applied to all MS/MS features , as exported from GNPS , using the QIIME platform ( Caporaso et al . , 2010 ) . The separation of the samples in PCoA space highlights the differences between the various origins of the samples ( Figure 1A ) . The lichen samples separate from the marine associated samples , although they overlap the spatial distributions of some cyanobacteria populations , perhaps because most lichens contain cyanobacteria , bacteria and fungi ( Mushegian et al . , 2011 ) . Interestingly , among the five environmental samples , the marine cyanobacteria/algae and freshwater cyanobacteria are more similar in their metabolomics profiles than any other of the source materials , even though the two aquatic environments are quite distinct ( Figure 1A ) . Nevertheless , while presenting the highest similarity compared to the other datasets , the marine cyanobacteria/algae and freshwater cyanobacteria samples show a clear differentiation in their chemical inventories ( Figure 1B ) . To quantify the common and differential chemical features driving these distances , we generated a Venn diagram using the MS/MS features . ( Figure 1C , ) . This analysis revealed that only 13 . 7% of the marine cyanobacteria/algae molecular features overlapped with other datasets , highlighting that 86 . 3% were features unique to these collections . Among the molecular commonalities , a number of the mass spectrometry signals were identified as matches to reference spectra in GNPS . These included a mixture of primary metabolites such as amino acids , fatty acids , structural lipids , pheophytin and pheophorbide A ( chlorophyll derived products ) , as well as common mass spectrometry background signals such as formate clusters , plasticizers and polymers . Remarkably , despite being most similar by PCoA analysis , no reference MS/MS spectra of the known freshwater cyanobacterial natural products matched with our marine cyanobacteria and algal collection dataset . 10 . 7554/eLife . 24214 . 003Figure 1 . MS/MS features as generated in GNPS are shown for marine cyanobacteria and algae ( green ) , marine and terrestrial actinobacteria ( blue ) , lichens ( red ) , freshwater cyanobacteria ( yellow ) and corals ( pink ) . ( A ) PCoA plots of 300 samples randomly selected from each dataset display the distance between samples based on molecular features using Bray-Curtis dissimilarity matrix . ( B ) Marine cyanobacteria and algae and freshwater cyanobacteria . Laboratory cultures of Phormidium 1646 , Lyngbya 1933 are highlighted in red circles . Each sphere represents the full sample metabolome ( C ) Venn diagram display of overlapping MS/MS features . Percentage of overlapping features with the marine cyanobacteria and algae dataset are given in parenthesis in their respective colors ( Supplementary file 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24214 . 003 In exploring the chemical diversity of the marine cyanobacteria and algae communities , we hypothesized that either the organism collected ( cyanobacterial or algal ) or the geographical location was responsible for sample-to-sample metabolomics differences ( Figure 2 ) . Our PCoA analysis did not indicate differentiation based on field annotations as algal versus cyanobacterial ( Figure 2A ) , and the non-significant clustering based on geographical origin , highlights the vast chemical diversity of samples within the same region ( Figure 2B ) . PCA , the most common dimensionality reduction and visualization method used for metabolomics analysis , displayed a parallel outcome with minor trends being observed ( Supplementary file 1 , Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 24214 . 004Figure 2 . MS/MS feature diversity between and within sampling sites . PCoA plots of crude extract molecular features using Bray-Curtis , dissimilarity matrix . Each point represents a single sample and points are colored by metadata . ( A ) PCoA plot shows the distance between samples based on field identified classification . ( B ) PCoAs color coded by geographical origin shows the distance of all locations together ( B: upper panel ) and the individual PCoA plots of the four collection sites with most samples ( B: lower panel ) . Laboratory cultures of Phormidium 1646 , Lyngbya 1933 and Lyngbya 1963 are highlighted in orange circles . DOI: http://dx . doi . org/10 . 7554/eLife . 24214 . 00410 . 7554/eLife . 24214 . 005Figure 2—figure supplement 1 . PCA plot of crude extracts of cyanobacteria and algae collections . PCA was performed using Bruker Profile Analysis software version 2 . 1 . Samples are displayed by collection location as described in the legend . NAC-Curacao , PAB-Panama Bocas del Toro , PAC-Panama Coiba , PAG-Panama Gulf of Chiriqui , PAP-Panama Portobelo , PAL-Palmyra Atoll , PNG-Papua New Guinea . DOI: http://dx . doi . org/10 . 7554/eLife . 24214 . 005 Construction of fraction libraries has become a common strategy to make more efficient the discovery of bioactive molecules in natural products studies so as to reduce sample complexity for biological screening as well as increase the titer of minor constituents ( Bugni et al . , 2008; Harvey et al . , 2015 ) . In the current analysis , the fraction library yielded over five times the number of molecular features compared to the crude extracts , thereby enabling a more thorough assessment of chemical richness and diversity ( Figure 3A ) . The fraction library resulted in 1094 molecular families compared to 132 in the crude extract analysis . This tenfold difference highlights the advantage of using fraction libraries for chemical and biological assay screens . By comparing the number of features as well as the number of underlying spectra for a given molecular family ( Nguyen et al . , 2013 ) , we could identify particular chemical scaffolds that are more diverse in their structural variations and are more abundant in our collections . For example , the barbamide molecular family ( Figure 3B ) is comprised of 40 molecular features represented in 985 individual spectra and found in 75 collections . By contrast , the palmyrolide A molecular family has only two associated features and consists of 32 spectra that are found in five collections ( Figure 3B ) . Using a rarefaction analysis , both the crude extracts and the fraction library were found to level off under the current experimental conditions of analysis , indicating that the sample collections are saturating the chemical diversity of these materials ( Figure 3A ) . 10 . 7554/eLife . 24214 . 006Figure 3 . Chemodiversity and richness of molecular features based on MS/MS data . ( A ) Rarefaction curve of cyanobacteria collection library showing the chemical richness of crude extracts vs . the fraction library . ( B ) Abundance of molecular families: Bars represent the number of features and spectra comprising each molecular family clustered by GNPS molecular networking . ( C ) Bar chart depicting both the total number of extracts from a given location and the percent contribution of unique features to the entire dataset . ( D ) Pie chart representing the percentage of unique molecular features attributed to origin of the sample . HI-Hawaii , NAC-Curacao , PAB-Panama Bocas del Toro , PAC-Panama Coiba , PAG-Panama Gulf of Chiriqui , PAP-Panama Portobelo , PAL-Palmyra Atoll , PNG-Papua New Guinea . DOI: http://dx . doi . org/10 . 7554/eLife . 24214 . 00610 . 7554/eLife . 24214 . 007Figure 3—figure supplement 1 . Molecular network of marine cyanobacterial natural products with annotated molecular families . ( A ) Molecular networking generated using MSV000078568/MSV000078892 datasets by http://gnps . ucsd . edu and visualized in Cytoscape . A-malyngamides [M+H]+ , B-barbamides [M+H]+ , C-curacins [M+H]+ , D-hectochlorins [M+H]+ , E-viequeamides [M+Na]+ , F-carmabins [M+H]+ , G-palmyramides [M+Na]+ , H-palmyramides [M+H]+ , I-dolastatins [M + 2 hr]+ , J- cyanolides [M+H]+ , K-apratoxins [M+H]+ , L-hoiamides [M+H]+ , M-hoiamides [M+Na]+ , N-majusculamides [M+Na]+ , O-hectochlorins [M+Na]+ , P-palmyrolides [M+H]+ . Spectral data dereplicated by fragmentation patterns present in spectral libraries available on GNPS . Nodes are color-coded by the geographical origin of the sample and labeled with parent mass ( m/z ) . Edge thickness represents the cosine similarity score . Insets on right hand side depict the ( B ) barbamide and ( C ) cyanolide molecular families . Shown for each are the standard library spectrum ( upper panel ) and the matching sample spectrum from within the same node ( lower panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24214 . 00710 . 7554/eLife . 24214 . 008Figure 3—figure supplement 2 . Dereplication of the apratoxin molecular family . Identification of apratoxin B based on library standards enabled dereplication of 5 known apratoxins ( A , D , F , G , and H ) . Additional apratoxin derivatives were characterized based on mass differences of functional groups . DOI: http://dx . doi . org/10 . 7554/eLife . 24214 . 00810 . 7554/eLife . 24214 . 009Figure 3—figure supplement 3 . Dereplication and spatial distribution for new derivatives of the barbamide molecular family . Based on barbamide ( 1 ) dereplication , peaks assigned to fragments enable the characterization of N-demethylbarbamide ( 2 ) and a substitution of the Phe and Leu groups as a new barbamide derivative ( 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24214 . 009 To identify locations of high chemical diversity among sampling sites represented in this study ( locations with more unique natural products compared to other regions ) , we compared the MS/MS features found only in a single location to the overall diversity of the entire dataset ( Figure 3C and D ) . To do this , we normalized the contribution of features from a single location to a percentage contribution as some locations were sampled more frequently than others based on metadata information collected across three decades ( Figure 3C ) . For example , approximately 4% of all MS/MS features contributing to the overall diversity are from Panama-Bocas del Toro ( PAB ) whereas about 7–8% are each contributed from Panama-Portobelo ( PAP ) and Palmyra Atoll ( PAL ) , and 20% are from Papua New Guinea ( PNG ) ( Figure 3D ) . To explore the chemical nature of differentially produced metabolites in the marine cyanobacterial and algal sample collections , GNPS based molecular networking was used and the network was visualized in Cytoscape ( For molecular network and Cytoscape file see Materials and methods , Figure 3—figure supplement 1 ) ( Purves et al . , 2016; Shannon et al . , 2003; Watrous et al . , 2012 ) . As MS/MS spectral alignment detects molecular features , a set of structurally related features connected by edges is termed a ‘molecular family’ . Experimental MS/MS spectra were dereplicated ( also described as ‘finding known unknowns’ by the metabolomics community ) against the community contributed mass spectral ‘GNPS library’ and third party spectral libraries including MassBank , HMDB , ReSpect and NIST 2014 housed within GNPS analysis infrastructure ( Horai et al . , 2010; Johnson and Lange , 2015; Sawada et al . , 2012; Wang et al . , 2016; Wishart et al . , 2013 ) . This is currently the largest molecular network generated and only possible through the GNPS interface ( Wang et al . , 2016 ) . From the more than 15 . 6 million spectra collected , 6249 MS/MS spectra were dereplicated to existing publicly accessible reference data , with 91 of the matches to previously characterized cyanobacterial and algal derived natural products from 30 molecular families ( Nguyen et al . , 2013 ) . The spectral library matches from the marine cyanobacteria/algae collection included: 74 matches to the GNPS library , 7 matches to the NIH pharmacologically active compounds library , 3 matches to the NIH natural product library , 1 match to the NIH clinical collection , 1 match to the FDA natural products library and 5 matches to the Faulkner legacy library . This is a match rate of about 0 . 04% which is much lower than database matches of the average untargeted metabolomics experiments ( 1 . 8% , [Wang et al . , 2016] ) and metabolomics analyses of model samples such as E . coli , human cell lines , mice or humans ( da Silva et al . , 2015 ) . The low match rate to existing natural product libraries within GNPS highlights that these marine cyanobacterial/algal communities are significantly underexplored from a chemical perspective . The hundreds of unannotated molecular families from these samples suggest that they contain a significant number of natural products that remain uncharacterized . Molecular networking allows putative structural propagation through a molecular family because most molecules of similar structure fragment similarly . By investigating selected molecular families , we identified previously characterized metabolites and previously undescribed derivatives ( Supplementary file 2 and Figure 3—figure supplements 2 and 3 ) . For example , we identified the spectral cluster representing the apratoxin molecular family by a spectral match to the apratoxin B library standard . Further dereplication of this molecular family using literature information and manual analysis of MS/MS spectra revealed several known apratoxin derivatives ( A , D , F , G , H ) ( Figure 3—figure supplement 2 ) ( Luesch et al . , 2002; Tidgewell et al . , 2010; Watrous et al . , 2012; Yang et al . , 2013 ) . Additional apratoxin derivatives are suggested on the basis of mass differences of common functional groups ( Figure 3—figure supplement 2 ) . Further , GNPS identified a molecular family comprised of the barbamides ( Orjala and Gerwick , 1996 ) . Within this molecular family , we were able to deduce several candidate derivatives based on fragmentation patterns and presence of chlorine isotopes ( Figure 3—figure supplement 1 and Figure 3—figure supplement 3 [1] ) ( Orjala and Gerwick , 1996; Yang et al . , 2013 ) . For example , a 14 Da decrease from the y ion of barbamide most likely results from loss of the amide methyl group , leading to N-desmethyl barbamide ( Figure 3—figure supplement 3 [2] ) . O-desmethyl barbamide was previously published and is identifiable by the loss of 14 Da from the characteristic MS/MS fragment containing the three chlorine atoms ( m/z 242 . 98 ) ( Kim et al . , 2012 ) . The 34 Da reduction from the y ion appears to be associated with the substitution of the Phe group by Leu , a conservative hydrophobic replacement for A-domains that accept Phe in the NRPS-based biosynthesis of barbamide . Such a conservative substitution has previously been observed in other NRPS biosynthetic systems ( Figure 3—figure supplement 3 [3] ) ( Flatt et al . , 2006 ) . This exchange of amino acids gives rise to a new molecular entity; however , it is reminiscent of related compounds , such as dysidin , which was isolated from the cyanobacteria-harboring sponge Dysidea herbacea ( Ilardi and Zakarian , 2011 ) . Because a majority ( 55% , Figure 3D ) of molecules detected in this dataset were found in more than one sampling location , we explored the spatial distribution and abundance of specific natural products using binning networks . For this visualization , the abundance/presence of a metabolite was correlated to the eight collection sites . In the network , each node represents a distribution pattern of molecules , where the node size and color correspond to the relative number of features that share the same distribution ( Figure 4 ) . Of the 256 conceivable regiospecific patterns ( based on the eight collection sites ) some molecules share similar distribution patterns , while some distribution patterns are absent or very minor in their abundance ( light colored nodes ) . To visualize the distribution of select molecules identified via molecular networking , in our eight major sampling sites , MS/MS feature intensity values were projected onto a 2D geographical map based on GPS coordinates using the open-source tool ‘ili ( an open-source tool for 2D and 3D data visualization created by our laboratories , https://github . com/ili-toolbox/ili [Protsyuk et al . , 2017] , available as a Google Chrome app; a copy is archived at https://github . com/elifesciences-publications/ili ) ( see legend of Figure 4 for more information ) . For example , out of the 33 , 481 detected features , dolastatin 10 along with 72 other chemical features share the same widespread distribution pattern ( Curaçao ( NAC ) , PAB , Panama-Coiba ( PAC ) , PAL , PAP , PNG ) , while palmyramide A and 3033 other chemical features are endemic to PAL , at least within the scope of this data set , and are not found in any of the other sampling locations . Additionally , while barbamide is widely distributed , putative analogs are encountered relatively infrequently ( Figure 4B and Figure 3—figure supplement 3 ) . 10 . 7554/eLife . 24214 . 010Figure 4 . Spatial maps showing the geographic distribution of selected cyanobacterial natural products . ( A ) Features were binned by their distribution patterns across the eight main geographical locations , each bin represented by a node with edges linking bins differing by one location . Number of features in each bin is presented according to size and color scale from white ( 0 ) to red ( 10 , 000 ) as indicated by the scale bar on the top . Spatial patterns are represented for selected natural products within these bins . Inserts in each map display zoomed-in sections of Panama , Curaçao and Papua New Guinea . Each sample is designated to a specific coordinate based on GPS coordinates of the collection site ( multiple samples are represented by spots placed around the collection site ) . See the URL provided below . ( B ) Spatial maps display chemogeographical distribution and abundance of barbamide ( 1 ) and two barbamide analogs ( 2 , 3 ) . Relative abundance is presented by Jet color scale from low ( blue ) to high ( red ) . HI-Hawaii , NAC-Curaçao , PAB-Panama Bocas del Toro , PAC-Panama Coiba , PAG-Panama Gulf of Chiriqui , PAP-Panama Portobelo , PAL-Palmyra Atoll , PNG-Papua New Guinea . ( To use the open source tool ‘ili please open the following link in Google Chrome: http://ili-toolbox . github . io/ ? cyano/bg . png;cyano/intensities . csv . Please wait until the data is loaded and visualized , then click on the Mapping submenu , and change the scale to Logarithmic , and Color map to Jet . For flipping through maps corresponding to different molecules , please click on the name of the molecule shown above the colorbar and select another molecule either with a click or with up-down arrow keys . Alternatively , you can refer to the example tab and choose cyanobacteria natural products . If you experience any problems , please contact theodore . alexandrov@embl . de . ) DOI: http://dx . doi . org/10 . 7554/eLife . 24214 . 010 To emphasize the application of MS-based geographic network analyses for the discovery of uncharacterized natural products , we examined several chemical features from locations of high chemical diversity , namely the PAP , PAL and PNG sites . One sample was chosen from PAP and targeted for the isolation of an m/z 536 . 333 molecule from a cluster with no described family members ( For molecular network see Materials and methods , molecular family #483 ) . This specific molecular feature was observed in several cyanobacterial collections of Lyngbya from PAP obtained in different years from 2007 to 2013 ( collections in Dec 2007 , Sept 2010 and Jan 2013 ) . The MS/MS data indicated that this compound was peptidic and contained the amino acids glycine , valine ( or N-methyl glycine ) , leucine or isoleucine , and 2-hydroxyvaleric acid ( Hiv ) . The presence of this latter residue suggested it was the product of a non-ribosomal peptide synthetase ( NRPS ) ( Kehr et al . , 2011 ) . An additional residue was observed corresponding to the fragment mass m/z 149 . 094 , compatible with the formula C10H13O+ . Further analysis of the sequential fragment losses suggested a putative structure of [Gly-Ile/Leu-Hiv-Val-C10H13O]+ . A search for a putative peptide with a molecular mass m/z 536 . 333 in various chemical databases ( Blunt et al . , 2015 , Blunt et al . , 2012; Luzzatto-Knaan et al . , 2015; http://pubs . rsc . org/marinlit/ ) was unsuccessful , suggesting that this natural product was uncharacterized . In order to confirm the mass spectrometry-based partial structural assignment , we isolated and structurally characterized this natural product of mass m/z 536 . 333 ( Figure 5 , Figure 5—figure supplements 1 and 2 , Supplementary file 3 [4] ) . Isolation of compound 4 , given the common name of ‘yuvalamide A’ , was guided by LC-MS analyses . NMR analysis ( DMSO-d6 ) of the pure compound confirmed the presence of Gly , Val , Ile , and hydroxyisovaleric acid , and revealed the presence of 2 , 2-dimethyl-3-hydroxy-7-octynoic ( Dhoya ) as the C10H13O+ fragment ( Figure 5 , Figure 5—figure supplements 1 and 2 , Supplementary file 3 ) . These NMR-based determinations correlate well with the structural predictions obtained by mass spectrometry . 10 . 7554/eLife . 24214 . 011Figure 5 . The molecular family and structure elucidation of a novel natural product yuvalamide A ( 4 ) , isolated from a Panama-Portobelo ( PAP ) cyanobacterial collection . ( A ) Molecular families and the spatial distribution of yuvalamide A [M+H]+ and [M+Na]+ ions are highlighted within yellow circles . MS/MS spectra display the linear structure for the non-ribosomal peptide ( NRP ) fragments Gly-Ile-Hiv-Val . ( B ) for the [M+H]+ and the ( C ) [M+Na]+ ions . ( D ) The full elucidated structure of yuvalamide A ( 4 ) as confirmed by NMR analysis ( Figure 5—figure supplements 1 and 2 , Supplementary file 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24214 . 01110 . 7554/eLife . 24214 . 012Figure 5—figure supplement 1 . Structure identification of yuvalamide A . ( A ) 1H NMR spectrum of yuvalamide A with COSY and key HMBC correlations ( 600 Hz , DMSO-d6 ) . ( B ) HMBC correlation of Dhoya-H3 - Gly-CO . ( C ) HMBC correlation of Dhoya-H6 - C7 and H6-C8 . DOI: http://dx . doi . org/10 . 7554/eLife . 24214 . 01210 . 7554/eLife . 24214 . 013Figure 5—figure supplement 2 . MS/MS fragmentation of yuvalamide A . The theoretical masses of predicted MS/MS fragments of yuvalamide A are shown in the top panel . The observed masses of these MS/MS fragments are labelled in the acquired spectrum . DOI: http://dx . doi . org/10 . 7554/eLife . 24214 . 01310 . 7554/eLife . 24214 . 014Figure 5—figure supplement 3 . Dereplication of yuvalamide molecular family . Additional putative yuvalamide analogs observed in the molecular family . Molecular families of [M+H]+ and [M+Na]+ ions highlighting yuvalamide A ( 4 ) ( Gly-Ile-Hiv-Val-Dhoya ) in yellow and the putative yuvalamide B ( 5 ) characterized as Dhoea analogue ( Gly-Ile-Hiv-Val-Dhoea ) in green circles . Two additional putative analogs characterized 522 . 3 m/z [M+H]+ ( Gly-C10H17NO3-Val-Dhoya ) and 572 . 3 m/z [M+Na]+ ( Gly-Ile-MeHiv-Val-Dhoya ) . One analog with an [M+H]+ m/z 522 . 3156 ( 14 Da smaller than compound 4 ) displayed a similar , but not identical fragmentation pattern to yuvalamide A ( Figure 5—figure supplement 3 ) . The linear order we propose is Gly-[C10H17NO3]-Val-Dhoya , as the −14 Da change is located in the Ile - Hiv residue region when compared to yuvalamide A . This is likely due to the promiscuous behavior of A domains in the NRPS biosynthetic assembly . Such promiscuous behavior where one amino acid is substituted by another , is quite common in non-ribosomal peptides , and results from similar amino acids being recognized by the cognate adenylation domain ( Crawford et al . , 2011 ) . Another yuvalamide analog with a 572 . 3303 m/z ( Figure 5—figure supplement 3 ) that clusters with the sodiated yuvalamide molecular family possess a similar fragmentation pattern to the yuvalamide A but with an increase in 14 Da . This shift is likely due to presence of additional methyl group and the fragmentation suggests this methyl is near the Hiv residue . DOI: http://dx . doi . org/10 . 7554/eLife . 24214 . 014 The other chemical entities in the yuvalamide A molecular family were present at very low abundance , preventing their isolation and structural analysis by NMR . However , based on the structure of yuvalamide A and its fragmentation pattern by LC-ESI-MS/MS , we deduced putative structures of these co-metabolites . An analog of yuvalamide A ( Figure 5—figure supplement 3 compound 5 , yuvalamide B , [M+H]+ m/z 538 . 3487 , [M+Na]+ m/z 560 . 3320 ) was detected by HRMS . Its fragmentation was nearly identical to compound 4 , differing only in the fragment belonging to Dhoya , suggesting this to be the alkene analog 2 , 2-dimethyl-3-hydroxy-7-octenoic acid ( Dhoea ) . Co-occurring analogs possessing the Dhoya or Dhoea unit have been observed previously in marine cyanobacterial natural extracts . The Dhoya subunit is usually the more abundant of the two , as observed in this case for yuvalamide A ( Boudreau et al . , 2012; Han et al . , 2005 , 2011; Luesch et al . , 2001; Sitachitta et al . , 2000; Wan and Erickson , 2001 ) . Moreover , the geographical distribution of yuvalamide B is the same as yuvalamide A ( Figure 5 ) , supporting the proposal that these share a common biosynthetic origin . Two additional yuvalamide analogs were observed and tentative structures were assigned based on MS ( Figure 5—figure supplement 3 ) . The integration of molecular networking with geographical mapping revealed that marine algae and cyanobacteria have widely varying metabolite distributions . As we can see in this dataset most molecules are broadly distributed in disparate physical locations , whereas others are highly specific to particular locations . Whether this specificity to a given location is due to unique species with distinct metabolic capabilities , or to specific environmental factors regulating expression of these compounds , is unknown at present . Moreover , field collections are a complex mixture of microorganisms that hold a unique repertoire of natural products that may not be discoverable under laboratory conditions , even with sequenced strains . Therefore , our approach is uniquely capable of organizing the vast chemical diversity present in these complex environmental samples . Further , this approach could be used to identify locations with a higher level of chemical diversity as well as distribution patterns of natural products , potentially giving insights into their ecological roles . In turn , these perceptions may facilitate the search for novel natural products of utility to biomedicine and biotechnology . A total of 317 diverse marine cyanobacterial and benthic algae collections were hand-collected in Panama , Papua New Guinea , Hawaii , Madagascar , Palmyra Atoll , Curaçao , and a few other tropical marine sites at depths from 0 . 3 to 20 m with the aid of snorkel or scuba gear ( Supplementary file 1 ) . GPS location , preliminary field taxonomic identification of the specimens , and depth of collection were noted for most collections . Samples for subsequent chemical extraction were strained through a mesh bag to remove excess seawater , preserved in equal volumes of seawater and ethanol , transported to the laboratory , and stored at −4°C or −20°C until workup . Live samples of marine cyanobacteria were brought back to the laboratory in vented tissue culture flasks with 0 . 2 µm filtered native seawater and subsequently isolated to mono-cultures . These were cultured in SWBG-11 media with 35 g/L Instant Ocean ( Aquarium Systems Inc . ) . Mono-cultures were grown at 28°C in a 16 hr light/8 hr dark cycle with a light intensity of ~7 µmol photon/s/m2 provided by 40 W cool white fluorescent lights . Specific collection information is described in the Supporting Information . Chemistry samples were extracted repetitively with CH2Cl2:MeOH 2:1 , dried in vacuo , and in most cases , fractionated into nine fractions ( A-I ) by silica gel vacuum liquid chromatography ( VLC ) using a stepwise gradient of hexanes/EtOAc and EtOAc/MeOH . Each fraction and crude extract was re-suspended at 5 mg/mL in pure DMSO and stored in 96-well plates at −20°C until analysis . The extracted metabolites were analyzed with an UltiMate 3000 UHPLC system ( Thermo Scientific ) using a Kinetex 1 . 7 µm C18 reversed phase UHPLC column ( 50 × 2 . 1 mm ) and Maxis Impact Q-TOF mass spectrometer ( Bruker Daltonics ) equipped with an ESI source . The data were acquired in positive ionization mode for parent mass ( MS1 ) and tandem MS for molecular fragmentation ( MS2 ) . Amount of material injected from each sample was 3 . 3 µg . The data were acquired in positive ionization mode for parent mass ( MS1 ) and tandem MS for molecular fragmentation ( MS2 ) . The gradient employed for chromatographic separation was 5% solvent B ( ACN/H2O/formic acid 98%/2%/0 . 1% ) with solvent A ( H2O/ACN/formic acid 98%/2%/0 . 1% ) for 1 . 5 min , a step gradient of 5% to 50% B in 0 . 5 min , held at 50% B for 2 min , a second gradient of 50–100% B in 6 min , held at 100% B for 0 . 5 min , 100–5% B in 0 . 5 min and kept at 5% B for 0 . 5 min at a flow rate of 0 . 5 mL/min throughout the run . The MS analysis was performed on a Maxis QTOF mass spectrometer ( Bruker Daltonics ) , controlled by the Otof Control and Hystar software packages ( Bruker Daltonics ) , and equipped with ESI source . MS spectra were acquired as previously described ( Garg et al . , 2015 ) . A molecular network was created using the online workflow at GNPS ( Wang et al . , 2016 ) . To de-noise , the data was filtered by removing all MS/MS peaks within ±17 Da of the precursor m/z . MS/MS spectra were filtered by choosing only the top six peaks in the ±50 Da window throughout the spectrum . The data were clustered with MS-Cluster with a parent mass tolerance of 1 . 0 Da and a MS/MS fragment ion tolerance of 0 . 5 Da to create consensus spectra . Consensus spectra that contained less than three spectra were discarded . A network was then created where edges were filtered to have a cosine score above 0 . 6 and more than four matched peaks . Further edges between two nodes were kept in the network if and only if each of the nodes appeared in each other's respective top 10 most similar nodes . The spectra in the network were then searched against GNPS spectral libraries . The library spectra were filtered in the same manner as the input data . The networking output was visualized using Cytoscape 2 . 8 . 3 free software ( Shannon et al . , 2003 ) and organized using the FM3 force directed layout ( Hachul and Junger , 2004 ) . Follow the link to the GNPS molecular network of multiple datasets: http://gnps . ucsd . edu/ProteoSAFe/status . jsp ? task=e69c25e3fbcf4c09b6e0c75633565f95 . Links to the cyanobacteria/algae datasets: ftp://massive . ucsd . edu/MSV000078568 , ftp://massive . ucsd . edu/MSV000078892 and the corresponding network on GNPS: http://gnps . ucsd . edu/ProteoSAFe/status . jsp ? task=0837888b3dab43efa5bf9a50254c7c8f . For the Cytoscape file ftp://massive . ucsd . edu/MSV000078568/other/ . The MS/MS data tables generated by GNPS for all analyzed extracts and fractions was converted to a BIOM table ( http://biom-format . org/ ) ( McDonald et al . , 2012 ) . Each of the tables was then used for calculation the sample-sample distance metrics using binary metric ( Binary-Jaccard ) using QIIME ( Caporaso et al . , 2010 ) version 1 . 9 . Principal coordinates were calculated and visualized based on various metadata categories ( region , organism etc . ) using EMPeror ( Vázquez-Baeza et al . , 2013 ) . Rarefaction curves of 20 random iterations were generated by GNPS . Spatial molecular maps were created with background map from a Google Map screenshot with several regions of interest enlarged and plotted as insets to the overview map . By visual examination , we assigned to each sampling spot the ( x , y ) pixel coordinates on the background map based on GPS coordinates taken at the time of collection . For multiple samples from the same location , spots were placed in adjacent locations . For each node of the molecular network generated in GNPS , we considered the area under the curve of all mapped spots , scaled them from 0% to 100% and assigned colors to all intensities by using the ‘jet’ color map , color-coding low intensities with blue , high intensities red , and other intensities with a colored gradient between blue and red . The rendering of the maps was performed by the ‘ili tool for 2D and 3D spatial mapping ( https://github . com/ili-toolbox/ili ) . The color was made gradually disappearing with highest intensity at the center of the sampling location and almost no intensity at the boundary of the spot . For this , a logarithmic combination of the default color and assigned color was used , with the coefficient exponentially depending on the distance from the sampling spot center . A preliminary structure was deduced based on a HRMS ( C28H45N3O7 , 8° unsaturation ) and HRMS/MS ( Bruker Daltonics Maxis Impact qTOF ) fragment analysis . Exact masses observed were [M+H]+ m/z 536 . 3335 and [M+Na]+ m/z 558 . 3155 corresponding to a molecular formula of C28H45N3O7 ( <0 . 9 ppm ) . Neutral losses of glycine , leucine/isoleucine and valine were observed . An additional loss matching the formula of 2-hydroxyisovaleric acid ( Hiv ) was observed suggesting yuvalamide was a non-ribosomal peptide . This MS-based fragment analysis suggested a Gly-Ile/Leu-Hiv-Val sequence with an additional C10H13O fragment . The peptide appeared to be cyclic based on degrees of unsaturation , and by database query , was a new compound . Analysis of 2D NMR data ( gCOSY , TOCSY , HSQC , HMBC ) confirmed three of the five substructures ( Val , Hiv , Gly ) and clarified the structures of the remaining two ( Figure 5—figure supplement 1 , Supplementary file 3 ) . The Ile/Leu substructure was identified as Ile from COSY correlations between the beta protons at 1 . 96 ppm and a methyl group at δ 0 . 86 , the beta protons at δ 1 . 96 and a pair of germinal protons at δ 1 . 49 and 1 . 13 , and between the geminal protons to a second methyl group at δ 0 . 81 . Finally , a 2 , 2-dimethyl-3-hydroxyoctynoic acid ( Dhoya ) residue was similarly identified . The gem-dimethyl arrangement was revealed by key HMBC correlations from CH3-9/10Dhoya to the carbonyl carbon C-1Dhoya , quaternary carbon C-2Dhoya , and oxygenated methine CH-3Dhoya . This latter resonance was connected by COSY to a linear sequence of three methylene groups , the terminal methylene of which was long range coupled to alkynyl carbons ( δC 71 . 7 and 84 . 5 ) , one of which ( δC 84 . 5 ) showed a HMBC correlation to a terminal alkynyl proton ( δH 2 . 78 ) . The assembly of these substructures into an overall cyclic depsipeptide structure was enabled by a key HMBC correlation between Dhoya H3 ( δH 4 . 71 ) and the Gly carbonyl ( δC 168 . 6 ) ( Figure 5—figure supplement 1/2 , Supplementary file 3 ) . Additional HMBC correlations linking all the substructures verified the MS/MS-deduced sequence described above ( Figure 5—figure supplements 1 and 2 , Supplementray file 3 ) .	Cyanobacteria and algae are found in all oceans around the globe . Like plants , they can use sunlight as a source of energy in a process called photosynthesis . As a result , these organisms are important sources of oxygen and another vital nutrient called nitrogen for other marine organisms . Many of these organisms also produce a variety of other chemicals known as “natural products” to help them to survive in their environments . Some of these natural products have shown potential as medicinal drugs . The search for new chemicals with useful medicinal properties has led researchers to collect samples of algae and cyanobacteria from various locations around the world . An approach called mass spectrometry is often used to identify new chemicals because it can provide information about the structure of a molecule based on how much its fragments weigh . Luzzatto-Knaan et al . used mass spectrometry to search for new chemicals in samples of algae and cyanobacteria that had been collected by diving and snorkeling in a wide variety of tropical marine environments over several decades . The experiments reveal that the organisms in these samples produce a diverse range of chemicals , most of which were previously unknown and have not been found in other similar environmental collections . The data were grouped together into eight major collection areas covering different parts of the tropics . The samples from some areas contained a wider variety of chemicals than others . Within each collection area , some molecules were found to be very common whereas others were only present at specific locations . To highlight the distribution of these natural products , Luzzatto-Knaan et al . display the data on a world map . Further experiments used this approach as a guide to extract a previously unknown chemical called yuvalamide A from a marine cyanobacterium . The next challenge would be to associate the geographical patterns of chemicals to their potential ecological roles . This approach offers a new way to explore large-scale collections of environmental samples to discover and study new natural products .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology" ]	2017	Digitizing mass spectrometry data to explore the chemical diversity and distribution of marine cyanobacteria and algae
Oncogenic stress provokes tumor suppression by p53 but the extent to which this regulatory axis is conserved remains unknown . Using a biosensor to visualize p53 action , we find that Drosophila p53 is selectively active in gonadal stem cells after exposure to stressors that destabilize the genome . Similar p53 activity occurred in hyperplastic growths that were triggered either by the RasV12 oncoprotein or by failed differentiation programs . In a model of transient sterility , p53 was required for the recovery of fertility after stress , and entry into the cell cycle was delayed in p53- stem cells . Together , these observations establish that the stem cell compartment of the Drosophila germline is selectively licensed for stress-induced activation of the p53 regulatory network . Furthermore , the findings uncover ancestral links between p53 and aberrant proliferation that are independent of DNA breaks and predate evolution of the ARF/Mdm2 axis . Throughout the animal kingdom , p53 occupies a central position within conserved stress response networks . The protein integrates diverse signals associated with DNA damage and uncontrolled proliferation to govern adaptive downstream responses such as increased DNA repair , arrested cell cycle , and apoptosis ( Vousden and Lane , 2007 ) . Where examined , the genes encoding p53 are not essential for viability but have been implicated as regulators of aging ( Derry et al . , 2007; Donehower et al . , 1992; Lee et al . , 2003; Sogame et al . , 2003 ) . It is now well appreciated that ancestral roles for this gene family must have predated functions in tumor suppression . In support of this , members of the p53 gene family are present in unicellular protists and short-lived multicellular organisms ( Lu et al . , 2009; Mendoza et al . , 2003; Nordstrom and Abrams , 2000 ) . Furthermore , cancer was probably a negligible source of selection pressure during the course of human evolution ( Aranda-Anzaldo and Dent , 2007 ) and the combined removal of canonical p53 effectors ( p21 , Puma , and Noxa ) does not account for tumor suppression in mice ( Valente et al . , 2013 ) . These and other observations suggest that tumor suppressive roles for the p53 family were co-opted from primordial functions , some of which may have been linked to meiotic recombination ( Lu et al . , 2010 ) . In recent years , considerable evidence has surfaced linking p53 action to stem cell biology . For example , in mammary stem cells p53 promotes asymmetric division and cell polarity , thereby helping to limit the population of stem cells in the mammary gland ( Cicalese et al . , 2009 ) . Furthermore , reprogramming of somatic cells into induced pluripotent stem cells ( iPSCs ) is greatly increased in p53 deficient cells , suggesting that p53 may act as a ‘barrier for induced pluripotency’ ( Krizhanovsky and Lowe , 2009 ) . Consistent with this , several labs have shown that p53 induces embryonic stem cell differentiation to maintain genomic stability after DNA damage ( Lin et al . , 2005; Neveu et al . , 2010; Zhao and Xu , 2010 ) . Together with recent studies in planaria , these observations indicate that an ancestral focus of p53 action could operate in stem cells ( Pearson and Sanchez Alvarado , 2009 ) . We directly tested this possibility using a p53 biosensor to visualize Drosophila germline stem cells and their progeny . When DNA breaks were exogenously imposed or intrinsically engineered , Drosophila p53 ( Dp53 ) was activated selectively in germline stem cells ( GSCs ) and their immediate daughters , indicating that these cells are uniquely licensed for p53 action . Furthermore , in various germline tumor models Dp53 was constitutively hyperactivated , suggesting that ancient links between p53 and inappropriate growth predate canonical effectors that connect these regulatory networks ( e . g . , ARF and MDM2 ) . The Drosophila gonad is a classic system for studying the stem cell compartment since stem cells , their immediate daughters , and the surrounding niche are easily identified . In the ovary , germline stem cells ( GSCs ) undergo self-renewing divisions that typically produce a GSC and a cystoblast ( CB ) . These GSCs support egg production throughout the lifespan of female adults ( Figure 1B ) . We used in vivo biosensors ( Lu et al . , 2010; Brodsky et al . , 2000 ) to visualize p53 activity as GSCs responded to various sources of stress ( Figure 1A ) . To exclude technical artifacts , two GFP reporters were used—one localizes to the nucleus ( p53R-GFPnls ) and the other does not ( p53R-GFPcyt ) . As previously described ( Lu et al . , 2010 ) , programed p53 activity triggered by meiosis was only observed in region 2 ( Figure 1B ) . After exposure to ionizing radiation ( IR ) stress , p53 activity was induced in virtually all germaria . However , despite widespread damage to the organ ( Figure 1—figure supplement 1 ) , this unprogrammed response was remarkably restricted to germline stem cells ( GSCs ) and their immediate progeny ( CBs ) ( Figure 1C , E ) . Furthermore , as seen in Figure 1—source data 1A , this response was highly penetrant . Since we rarely observe reporter activation only in CBs , the signal seen in CBs probably reflects GFP perduring from the parental stem cells . Furthermore , post-irradiation levels of GFP were noticeably more robust than the programed activity during meiosis ( compare solid arrows to open arrows in Figure 1C , D ) ( Lu et al . , 2010 ) . As expected , p53 biosensor activity was not observed within the ovary of p53−/− animals and was also absent from ovaries lacking the upstream Chk2 kinase ( Figure 1E , Figure 1—figure supplement 2A′ , A′′ , Figure 1—source data 1A ) . 10 . 7554/eLife . 01530 . 003Figure 1 . Genotoxic stress selectively triggers p53 activity in ovarian stem cells . ( A ) Construction of p53 biosensors . A well-characterized p53 enhancer ( black line ) that contains a p53 consensus binding site ( blue box ) conserved from flies to humans resides upstream of the reaper locus ( gray box ) ( Brodsky et al . , 2000 ) . A 150-bp fragment containing this enhancer was placed upstream of GFP ( p53R-GFP ) . Transgenic fly strains are made with two reporter constructs , one contains a nuclear localization signal for GFP ( p53R-GFPnls ) and the other one does not ( p53R-GFPcyt ) . Stimuli that trigger p53 activation induce GFP expression . These biosensors require wild-type p53 and are effective readouts for p53 function . ( B ) Germline stem cells ( GSCs ) are in contact with cap cells ( in gray ) at the apical tip of the germarium and undergo self-renewing division to produce a GSC and cystoblast ( CBs ) ( Spradling et al . , 2001 ) . In unperturbed ovaries , programed activation of the p53R-GFP biosensor is triggered by meiotic recombination in region 2 of the germarium , marked by open arrowhead in ( C ) and ( D ) ( Lu et al . , 2010 ) . ( C ) After radiation challenge ( IR ) the p53R-GFPcyt biosensor ( green ) is selectively induced in ovarian GSCs and CBs noted by a solid arrowhead . Bracket denotes the germarium . The open arrowhead and dotted line indicates p53 activation in region 2 prompted by meiosis . Insets ( C’ and C’’ ) are confocal images from different irradiated germaria counterstained with DAPI ( blue ) . p53R-GFPcyt induction ( green ) initiates in GSCs that exhibit rounded fusomes ( C’ white arrows ) labeled by α-HTS ( Hu li tai shao , red ) and are in contact with cap cells ( C’’ yellow arrows ) . Cells that activate p53 in ( C’ and C’’ ) were confirmed to be germ cells by α-Vasa staining ( shown in Figure 1—figure supplement 2C–D′′ ) . ( D ) An engineered DNA double-stranded break ( DSB ) mediated by I-SceI ( see texts and ‘Materials and methods’ ) induces the p53R-GFPnls biosensor ( green ) in GSCs/CBs , noted by a solid arrow . Open arrow indicates meiotic p53 . The germarium is counterstained with α-HTS ( red ) and DAPI ( blue ) . ( E ) Quantifies the percentage of germaria activated for the p53 biosensors in GSCs and their immediate progeny . Note that the perturbation-dependent responses reported here are all highly penetrant . Selective activation is IR ( green ) and I-SceI ( blue ) dependent at the 0 . 001 significance level . Note that biosensor activation did not occur in p53−/− ( red ) or chk2−/− ( orange ) mutants ( see Figure 1—figure supplement 2A′ , A′′ ) . Sample sizes are combined from at least two independent trials ( available in Figure 1—source data 1 ) . All scale bars represent 10 μm . In panels C–C’’ the p53R-GFPcyt reporter was used . In panel D , the p53R-GFPnls biosensor was used . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 00310 . 7554/eLife . 01530 . 004Figure 1—source data 1 . Validation of the p53R-GFP biosensors . In A and B reporter activation is quantified as indicated . ( A ) Reporter activation in female GSCs/CBs is p53 dependent and Chk2 dependent but independent of the topoisomerase , Spo11 . Reporter activation in testis is also stimulus dependent and p53 dependent . p53 , Chk2 , or Spo11 status are noted in column 1 . The reporter used ( nuclear or cytoplasmic ) is indicated in column 2 . Column 3 shows unirradiated controls in which the percent reporter activation in GSCs/CBs is provided , as well as the total number of germaria or testis that were assayed . Column 4 shows reporter activation in irradiated tissue at 24 hr post-irradiation with percentage of germaria or testis with GFP positive GSCs/CBs and the number of samples assayed . Quantification of reporter activation is from three independent trials in the ovary and two independent trials in the testis . ( B ) Quantification of p53-GFPnls in region 1 of flies containing I-SceI endonuclease by itself or with the I-SceI cutsite . Reporter activation in I-SceI expressing animals that also have the I-SceI cutsite is comparable to wild-type irradiated flies ( A ) . Quantification of reporter activation is from two independent trials . ( C ) Quantification of p53-GFPnls in GSCs and follicle cells of flies heterozygous ( ATR+/− ) or mutant for ATR ( ATR−/− ) . After irradiation challenge , p53 activation is highly penetrant in both ATR+/− and ATR−/− genotypes . ATR mutants show a robust induction of reporter activation in follicle cells after irradiation . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 00410 . 7554/eLife . 01530 . 005Figure 1—figure supplement 1 . Wide-spread DNA breaks after irradiation . Unirradiated ( A ) and irradiated ( B ) WT germaria were stained for the Drosophila counterpart of H2Ax , designated pH2Av ( green ) . A′ and B′ show the pH2Av channel alone ( white ) from A to B respectively . DNA double-stranded breaks visualized by pH2Av ( green ) appear throughout germarium within 15 min after irradiation . Germaria in A and B are counterstained with α-HTS ( red ) and DAPI ( blue ) . Note that panels A and A′ are the same as in Figure 4—figure supplement 2 A-A′ . Panels B and B′ are the same as in Figure 4—figure supplement 2 C-C′ . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 00510 . 7554/eLife . 01530 . 006Figure 1—figure supplement 2 . Selective p53 action in germline stem cells is detected using a p53 biosensor . Irradiated ovaries in panels ( A–A′′ ) validate the p53R-GFP biosensor , establishing that wild-type p53 and chk2 are required for GFP activation in the germarium ( brackets ) . ( A ) p53R-GFPnls is selectively activated in GSC/CBs ( arrowhead ) upon irradiation . Reporter activation is not seen in p53 mutants ( A′ p53R-GFPnls;p53− ) or in Chk2 mutants ( A′′ Chk2−;p53R-GFPcyt ) . Panel ( B ) is a control for Figure 1D showing that p53R-GFPnls activation is not observed in stem cells ( white arrow ) that express the I-SceI enzyme but lack the I-SceI restriction site . Note that programed meiotic p53 activation is still observed in these ovaries ( open arrowhead ) . See Figure 1E and Figure 1—source data 1 for quantification . In panels ( C–D′′ ) irradiated germarium are stained with α-HTS ( Lin et al . , 1994 ) and α-VASA ( Hay et al . , 1990 ) to confirm GFP expression in the GSCs/CBs . ( C and D ) p53R-GFPcyt activation ( green in C and D , white arrows in C–D′′ ) is restricted to the germline , identified by Vasa staining ( white in C′′ and D′′ ) . p53R-GFP activation is also restricted to GSCs and CBs identified by round fusomes ( HTS staining , red in C and D , white in C′ and D′ ) . Rounded yellow fusomes validate co-incidence of p53R-GFP and rounded HTS . ( E–E′′ ) p53R-GFPcyt activation in GSCs ( green in E ) was seen in cells that contact the cap cells ( yellow arrowheads ) , which are visualized by location and size of DAPI stained nuclei ( blue in E , white in E′′ ) . Note that the sample in C is also shown as the C′ inset in Figure 1C stained with DAPI . Likewise , D is also shown as the C′′ inset in Figure 1C with single channels of α-HTS and DAPI shown in D′ and D′′ respectively . Also , D and E are the same sample but D highlights Vasa positive cells while ( E ) highlights contact with cap cells . In panels A′′ , C , D , and E the p53R-GFPcyt reporter was used and in panels A , A′ , and B the p53R-GFPnls reporter was used . All scale bars are 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 00610 . 7554/eLife . 01530 . 007Figure 1—figure supplement 3 . ATR is not rate limiting for p53 activation in the germline . p53R-GFPnls activation ( green ) was examined in ATR mutants ( A–B′ ) after irradiation . ( A–A′ ) GSCs ( yellow arrowhead ) that are in contact with cap cells ( yellow arrows in A and A′ ) identified by DAPI staining ( white in A , red in A′ ) . These observations show that ATR is not rate limiting for p53 activation in GSCs . Panels ( B–B′ ) show that induction of the p53R-GFPnls biosensor is not selective in ATR mutants ( B′ ) when compared to WT controls ( B ) . Genotype for ( A–A′ ) is mei-41[D3]/[D3] and for ( B′ ) is mei-41[D3]/[RT] . The p53R-GFPnls biosensor was used for panels A–B′ . ( C ) Quantification of p53R-GFPnls reporter in GSCs/CBs and follicle cells in ATR heterozygous controls and ATR mutants with and without irradiation . Both the ATR+/− control and ATR−/− show a robust induction of p53R-GFPnls in GSCs after irradiation . ATR mutants also show a robust induction of reporter activation in follicle cells after irradiation . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 007 Double stranded DNA breaks ( DSBs ) are responsible for many of the biological effects associated with IR ( Ward , 1994 ) . Therefore , to determine whether DSBs are sufficient to induce the p53 reporter , we ubiquitously expressed the I-SceI endonuclease in the germline of flies engineered to harbor a single I-SceI recognition site in each nucleus . As seen with IR exposure , p53 activity occurred only in GSCs/CBs when DSBs were induced ( Figure 1D , E , Figure 1—figure supplement 2B , Figure 1—source data 1B ) . Furthermore , it is notable that a single DSB was sufficient to provoke robust p53 activity in GSCs/CBs . Therefore , whether exogenously imposed or intrinsically engineered , DSBs triggered p53 selective activation that was confined to GSCs and their immediate progeny . Furthermore , this stem cell restricted response is clearly under genetic control . For example , in directed tests of chosen mutants we identified a class of lesions that exhibit non-selective p53 action throughout the ovary only after IR challenge ( see Figure 1—figure supplement 3 , Figure 1—source data 1C ) . Therefore , p53 is present and potentially functional in all cells of the ovary but , under normal conditions , its action is somehow confined to GSCs and their immediate progeny . To ask whether this pattern might reflect a general property of germline stem cells , we similarly examined the male gonad . As seen in the ovary , we observed selective p53 reporter activation in GSCs and their immediate progeny ( gonioblasts ) in irradiated testis ( Figure 2 ) . Likewise , stimulus-dependent activity required p53 and was not seen in unchallenged testis ( Figure 2C , Figure 2—figure supplement 1A , B , Figure 1—source data 1A ) . Occasionally , the biosensor was also present in early spermatogonial cysts , perhaps reflecting perduring GFP and/or independent activation associated with dying cells ( Figure 2D–D′′ , Figure 2—figure supplement 1D-E′′ ) . Collectively , the observations in Figure 1 and Figure 2 demonstrate that selective p53 activation in the stem cell compartment is a general property of germline tissues exposed to genotoxic stress . We note that perturbation-dependent induction of the p53 biosensor in gonadal stem cells was highly penetrant ( Figure 1E , Figure 2C ) . However , like all stress responses , the strength of signal and the number of responding cells were variable from animal to animal ( Figure 1C , D , Figure 2D ) perhaps reflecting distinct cell cycle dynamics occurring in GSCs at the time of challenge . 10 . 7554/eLife . 01530 . 008Figure 2 . Selective p53 activity occurs in male germline stem cells . ( A ) p53R-GFPcyt ( green ) is induced at the apical tip of an irradiated testis ( arrowhead ) , where stem cells are located ( see B ) . α-HTS co-staining ( red ) highlights early stages of germline development . The inset in panel ( A ) shows a higher magnification view from a different irradiated testis . ( B ) Male GSCs are in contact with cap cells ( blue flower pattern ) at the apical tip of the testis and divide to produce a gonioblast daugther ( GB ) . ( C ) Quantifies the percentage of testis activated for the p53 biosensors in GSCs and their immediate progeny . Selective activation is IR ( green ) dependent and conditional upon p53 since p53R-GFP activation did not occur in p53−/− mutants ( red bar ) . ( D–D’’ ) Confocal images from other irradiated testes confirmed that stem cells induced for p53R-GFPcyt ( green , D and D’’ ) are also positive for rounded HTS staining ( red , D and D’’ ) and the germline specific marker Vasa ( white , D’ and D’’ ) as expected . The hub ( dotted line , D ) was routinely identified by the characteristic nuclei pattern as illustrated in B ( blue cells ) and by negative Vasa staining ( D’ and D’’ ) . Asterisks mark p53R-GFP positive cells that are adjacent to the hub and Vasa positive or Vasa positive with rounded fusomes . Also note that the hub was identified by α-Armadillo staining ( Figure 2—figure supplement 1C ) . Open arrowhead in ( D and D’’ ) is likely a dying cyst as indicated by pyknotic and condensing nuclei and irregular HTS ( Figure 2—figure supplement 1D–E′ ) . In panels A , D–D’’ the p53R-GFPcyt reporter was used . All scale bars represent 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 00810 . 7554/eLife . 01530 . 009Figure 2—figure supplement 1 . p53 Reporter activation in the male germline ( seen in Figure 2A ) is conditional upon irradiation ( A ) and is p53 dependent ( B ) . Compare A and B to Figure 2A ) ( C–C′ ) p53R-GFPnls ( green ) activation in testis after irradiation . The hub is identified here by α-Armadillo staining ( red in H , white in H′ ) , which is noted by a dotted line in ( H ) . ( D–D′ ) Image in ( D ) is a different z projection from the same irradiated testis shown in Figure 2D , where HTS ( red ) and DAPI ( blue ) are used to identify cells . The p53R-GFP positive dying cyst ( green ) indicated by the open arrowhead in Figure D–D′′′ exhibits pyknotic nuclei ( D′ arrow ) , condensing nuclei ( D′ open arrowheads ) and irregular HTS . Image in ( D′ ) represents a magnified view of the dashed box in ( D ) . Compare D and D′ to the unirradiated WT testis control in ( E–E′ ) that shows a branched fusome detected by α-HTS ( red in E and E′ ) and nuclei of similar size by DAPI ( blue in E and E′ ) . In panels A , B , and E the p53R-GFPcyt reporter was used . In panels C , the p53R-GFPnls reporter was used . All scale bars are 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 009 We tested whether other genome destabilizing factors elicited similar p53 activity in stem cells . To examine the effect of deregulated retrotransposons , we introduced the p53 biosensor into cutoff or aubergine mutant animals . These genes encode essential components of the piwi-associated RNA ( piRNA ) pathway , acting to silence retrotransposons in the germline ( Chen et al . , 2007 ) . The corresponding mutants exhibit disregulated retrotransposition , reduced fecundity , and egg shell ventralization ( Chen et al . , 2007 ) . Figure 3A shows that in cutoff mutants induction of p53R-GFP occurs exclusively in GSCs and their progeny at a penetrance comparable to irradiated wild-type animals ( Figure 3—source data 1 ) . Frequent p53 activation in the germline was similarly observed in the GSCs of aubergine mutants ( Figure 3B ) and rad54 mutants defective for DNA repair ( Figure 3C ) . However , in contrast to cutoff mutants , the p53 biosensor was not entirely restricted to GSCs/CBs in these mutants ( Figure 3—figure supplement 1B , C , Figure 3—source data 1 ) perhaps reflecting differences in the kinetics of repair that may occur in these different backgrounds ( Klattenhoff et al . , 2007 ) . 10 . 7554/eLife . 01530 . 010Figure 3 . Stem cell associated p53 activity in defective DNA repair and retrotransposon silencing mutants . ( A–B ) Activation of the p53 biosensor ( green ) in the germarium of piRNA mutants , ( A ) cutoff[QQ}/[WM] and ( B ) aubergine[HN]/[QC] . ( C ) Activation of the p53 biosensor in rad54 , a meiotic DNA repair mutant . ( D–F ) Germaria were found to express p53R-GFPcyt in GSCs/CBs with a penetrance of 90% for cutoff mutants ( D , p<0 . 0001 ) , 80% for aubergine mutants ( E , p=0 . 0018 ) , and 33% for rad54 mutants ( F , p=0 . 0039 ) . Asterisks indicate significant differences between heterozygous controls and homozygous mutants . GSCs/CBs were identified by rounded fusomes detected with α-HTS ( red in merge A , B , C and white in A’ , B’ , C’ ) . Arrowheads indicate that p53R-GFP positive cells are also germ cells identified by Vasa staining ( blue in A , B , C and white split channel in A’’ , B’’ , C’’ ) . Note that this particular α-Vasa antibody cross-reacts against the muscle sheath that surrounds each ovariole . If the sheath is not fully dissected and removed , then background staining is evident , as seen in Figure 2B′′ . Control genotypes were cuff[WM]/CyO , aub[HN]/CyO , rad54[AA]/CyO . Note that aub and rad54 mutants occasionally showed p53 activation beyond region 2 of the germarium ( arrow in C ) , quantified in Figure 3—figure supplement 1 , Figure 3—source data 1 . All scale bars represent 10 μm . In panels A , B , and C , the p53R-GFPcyt reporter was used . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 01010 . 7554/eLife . 01530 . 011Figure 3—source data 1 . Quantification of p53 activation in defective DNA repair and retrotransposon silencing mutants . Mutants defective for ( A ) meiotic repair ( rad54 and rad50 ) and ( B ) retrotransposon silencing ( aubergine and cutoff ) have increased spontaneous reporter activation compared to heterozygous controls . The percentage of ovarioles positive for p53R-GFP in the regions indicated ( GSC/CB , region 3 , stage 2–8 egg chamber ) was calculated and the number of ovarioles assayed per region is indicated . Quantification of reporter activation is from three independent trials . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 01110 . 7554/eLife . 01530 . 012Figure 3—figure supplement 1 . Quantification of p53 activation in defective DNA repair and retrotransposon silencing mutants in region 3 and stage 2–8 egg chambers . A , B , and C show quantification of reporter activation in region 3 of the germarium . For rad54 mutants , p53 activation was significantly different from the heterozygous control for region 3 ( C′ , p=0 . 0157 ) . A′ , B′ , C′ show quantification of p53 reporter activation in stage 2–8 egg chambers . For aubergine mutants , p53 activation was significantly different from the heterozygous control for stage 2–8 egg chambers ( B′′ , p=0 . 0437 ) . Control genotypes were cuff[WM]/CyO , aub[HN]/CyO , rad54[AA]/CyO . See Figure 3—source data 1 for number of ovarioles quantified . Samples shown here are combined from three independent trials . The p53R-GFPcyt reporter was used for these studies . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 012 In somatic cells , Dp53 promotes stress-induced apoptosis ( Sogame et al . , 2003 ) . Therefore , we examined the germarium for evidence of cell death by detecting cleaved caspase-3 . In the 24-hr period post challenge , over 90% of GSCs induce the reporter but the average incidence of apoptosis was less than 4% ( Figure 4—figure supplement 1 , Figure 4—source data 1A ) . Furthermore , we did not observe an obvious role for p53 in regulating stem cell numbers in the Drosophila ovary in the presence or absence of stress ( Figure 1—source data 1 , Figure 4—source data 1A ) . We also used α-pH2Av immunostaining , the Drosophila counterpart of mammalian pH2AX ( Mehrotra and McKim , 2006 ) , to follow the repair of DSBs after IR and found that resolution of these lesions was unaffected in the germaria of p53 mutants ( Figure 4—figure supplement 2 ) . Similarly , in BrdU incorporation studies , the rates at which wild-type and p53−/− GSCs/CBs entered proliferative arrest were also indistinguishable ( Figure 4A ) . However , in the post-stress period , we did observe that p53 mutants were significantly delayed for re-entry into the cell cycle ( Figure 4A ) . Furthermore , this defect is reversed in p53 genomic rescue strains confirming an assignment of this phenotype to the p53 locus ( Figure 4—source data 1B ) . 10 . 7554/eLife . 01530 . 013Figure 4 . p53 mutants exhibit impaired fertility and delayed re-entry into the cell cycle after irradiation . ( A ) BrdU incorporation in GSCs after 4 krad of IR . The percentage of germaria containing BrdU positive GSCs/CBs was plotted on the Y axis . WT and p53−/− GSCs arrest with similar kinetics but p53−/− GSCs were significantly delayed for re-entry into the cell cycle . Error bars represent standard deviation from tests of three independent cohorts . WT and two rescue strains are significantly different from p53−/− at the 0 . 05 level at the x24 hr time point . Percentages and number of germaria assayed are included in Figure 4—source data 1B . In panels A and B , p53−/− represents animal transheterozygous for two p53 null alleles , p53ns and p53K1 . ( B ) Fertility in wild-type ( WT ) and p53−/− females was measured after exposure to 11 . 5 krad of IR ( see ‘Materials and methods’ ) , which induces persisting sterility in p53 mutants . WT fertility is significantly different from p53−/− during time points 7–10 , 10–15 , and 15- at the 0 . 05 level ( see ‘Materials and methods’ ) . Two rescue strains showed partial restoration of fertility . Rescue 1A strain showed restored fertility is significantly different from p53−/− at the 0 . 05 level at days 10–15 and 15- . Note that after 15 days post irradiaton , fertility was monitored for at least 9 more days as indicated by 15- . Error bars represent standard deviation from five independent trials . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 01310 . 7554/eLife . 01530 . 014Figure 4—source data 1 . Quantification of proliferative potential and apoptosis of germaria challenged with irradiation . ( A ) Quantification of germaria that have cleaved-caspase3 ( CC3 ) positive GSCs/CBs . Columns indicate the time points after irradiation . Rows indicate the genotype . The percentage of germaria that have CC3 positive GSCs/CBs and the number of germaria is quantified from three independent trials for the x4 hr time point and from two independent trials for the non-irradiated , x2 hr , x8 hr , and x24 hr time points . ( B and C ) Quantification of germaria that have BrdU positive cells in region 1 . The percentage of germaria that have BrdU positive cells in region 1 and the number of germaria assayed are quantified from three independent trials in B and two trials in C per time point . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 01410 . 7554/eLife . 01530 . 015Figure 4—figure supplement 1 . Reporter activation after irradiation does not lead to purging of GSCs through apoptosis . Time course analysis of stainings for cleaved-caspase3 ( CC3 ) in GSCs/CBs after 4 . 0 krad of irradiation . The percent of germaria with CC3 positive GSCs/CBs are plotted on the Y axis . The highest incidence of CC3 was only 8% at 4 hr post irradiation in wild-type flies . This is considerably different from the incidence of p53R-GFP positive stem cells after irradiation ( ∼90% , see Figure 1E ) . The inset is a magnified view of the same graph to better appreciate the error bars . Error bars represent standard deviations from two trials for no irradiation , 2 hr , 8 hr , 24 hr and three trials for 4 hr for both genotypes . Percentages and number of germaria assayed are provided in Figure 4—source data 1B . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 01510 . 7554/eLife . 01530 . 016Figure 4—figure supplement 2 . Radiation-induced DNA double-stranded breaks appear and disappear with similar kinetics in WT and p53−/− GSCs . Time course of α-pH2Av ( green ) clearance after irradiation of wild-type ( A , C , E , G , I ) and p53 mutants ( B , D , F , H , J ) at a dose of 4 krad . Little or no pH2Av staining is observed in unirradiated WT ( A ) or p53−/− stem cells ( B ) . Similar pH2Av staining is observed in WT and p53−/− stem cells 15 min after irradiation ( C and D ) . In both cases damage was generally cleared from GSCs within 24 hr ( I and J ) . Note that many cells are damaged after irradiation ( compare A to G ) yet p53 biosensor activation is restricted to GSCs/CBs ( Figure 1 ) . White circles indicate stem cells . Insets are magnified views of tip of the germarium from the same image for better GSC visualization . HTS ( red ) and DAPI ( blue ) are used to highlight cells in the germarium . Note that the samples in A , A′ , C and C′ are included in Figure 1—figure supplement 1 A-B′ . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 01610 . 7554/eLife . 01530 . 017Figure 4—figure supplement 3 . Fertility recovery correlates with proliferation by GSCs and their progeny . ( A ) Fertility recovery after 9 . 0 krad of IR . A similar pattern is observed as seen at higher doses ( Figure 4B , 11 . 5 krad ) . WT fertility is significantly different from p53−/− at all time points ( p<0 . 05 ) . The A1 Rescue and p53−/− are significantly different at 10–15 and 15- days after IR ( p<0 . 05 ) . Error bars represent standard deviation from four independent trials . Note that fertility was monitored for a total of 25+ days after irradiation as indicated by 15− . To link the fertility defect to cell cycle kinetic differences we observe at lower doses ( Figure 4A , 4 . 0 krad ) , we performed 2 trials where we assayed fertility ( B ) and BrdU incorporation in region 1 of the germarium ( C ) after 9 . 0 krad of irradiation . Panel ( C ) shows that GSCs and CBs in p53−/− flies have a reduced proliferation potential at 2 and 7 days post irradiation ( Figure 4—source data 1C ) . Fertility recovery suggests a radiation sensitivity phenotype since p53−/− flies recover fertility in a dose dependent manner ( compare to Figure 4B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 017 To examine how the action of p53 might coordinate adaptive stress responses in GSCs , we developed a fertility recovery assay . In this study , females were irradiated to induce transient sterility and the recovery of fertility was scored over time ( see ‘Materials and methods’ ) . Figure 4B shows that wild-type females recovered from infertility within 1 week post-exposure to IR at a dose of 11 . 5 krad . In contrast , females lacking p53 remained permanently infertile even when tracked over 2 weeks after IR ( Figure 4B ) . To confirm that p53 gene function is responsible for this phenotype , we tested p53− females carrying a genomic rescue fragment spanning the p53 gene ( see ‘Materials and methods’ ) . We tested two rescue strains and in both cases the sterility defect was reversed ( Figure 4B ) . However , neither rescue strain fully restored fertility to wild type levels , possibly reflecting incomplete restoration of wild type regulation in the transgenes . To test whether we could link the fertility defect ( Figure 4B ) to the cell cycle defects observed at a lower dose ( Figure 4A ) , we examined fertility and cell cycle kinetics at an intermediate dose ( 9 krad ) of IR . After this challenge , p53−/− females exhibit impaired fertility , whereas WT flies remained fertile ( Figure 4—figure supplement 3 ) . We performed BrdU incorporation studies over 7 days with females irradiated at 9 krad and assayed the number of germaria that had BrdU positive cells in region 1 . Under these conditions , we observed persistently reduced proliferative activity in p53−/− stem cells even 7 days after IR ( Figure 4—figure supplement 3C , Figure 4—source data 1C ) . This result is consistent with the possibility that fertility defects seen in p53−/− flies are linked to the impaired cell cycle kinetics found in GSCs . Furthermore , the data in Figure 4B and Figure 4—figure supplement 3A suggest that radiosensitivity associated with the p53−/− genotype , previously been documented for larval stages ( Sogame et al . , 2003 ) , also applies to germline tissue . Oncogenic properties are thought to simulate ‘stemness’ and oncogenic signals frequently result in p53 activation ( Vousden and Lane , 2007 ) . However , it is not known whether this regulatory axis is conserved beyond mammals . To test whether inappropriate growth triggers Drosophila p53 function , we examined the p53 biosensors in various germline tumor models . First , we expressed an oncogenic form of RAS commonly found in human cancers together with the p53 biosensor ( Lee et al . , 1996 ) . Transient expression of the Drosophila RasV12 counterpart provoked robust p53 activation mainly in the GSCs and CBs ( Figure 5B , Figure 5—source data 1 ) . Figure 5E shows that another oncoprotein , Cyclin E , produced similar results . We also examined these biosensors in bam mutants , where a block in differentiation causes extensive hyperplasia ( McKearin and Ohlstein , 1995 ) and in these tumors extensive reporter activity was also seen ( Figure 5C , D ) . Likewise , expanded BMP ( bone morphogenic protein ) signaling ( Chen and McKearin , 2003 ) or reduced Lsd1 ( lysine-specific demethylase 1 ) activity ( Eliazer et al . , 2011 ) in neighboring somatic cells can also cause inappropriate growth and robust p53 activity was similarly observed in these germline tumors as well ( Figure 5F , G , H ) . Therefore , whether caused by forced oncoprotein expression ( panels B , E ) , failed differentiation programs ( panels C , D ) , or expansion of the stem cell niche ( panels F–H ) , inappropriate growth of Drosophila tissues was consistently accompanied by p53 activity . As seen with genotoxic stress , biosensor responses seen in these contexts was somewhat variable , perhaps reflecting complex signaling and/or cell cycle dynamics that occur in these tumor models . Technical sources of variation linked the UAS-GAL4 driver system and/or non-uniform accumulation of the oncogenic product could also contribute to variability in these contexts . 10 . 7554/eLife . 01530 . 018Figure 5 . Deregulated growth in the stem cell compartment provokes p53 action . ( A ) In an unperturbed wild-type ( WT ) germarium , the p53R-GFPnls biosensor is absent from GSCs/CBs , marked here by rounded fusomes stained with α-HTS ( red ) . The modest signal in region 2 reflects meiotic p53 activity ( dotted bracket ) ( Lu et al . , 2010 ) . When perturbed by RasV12 ( B ) the p53 biosensor ( green ) is induced in GSCs/CBs ( solid bracket in B , see Figure 5—source data 1 ) . Perturbation from failed differentiation programs caused by the bam mutation ( C–D ) or Cyclin E over-expression ( E ) provokes similar p53 biosensor activity . Likewise , increased DPP signaling caused by a constitutively active Tkv receptor ( F ) or ectopic DPP ligand expression ( G ) also prompts induction of the p53 reporter . Induction of the p53 reporter is also seen , when the stem cell niche is expanded by silencing of Lsd1 ( H ) ( Eliazer et al . , 2011 ) . Insets in panels E–H are magnified views of tumor cysts showing that p53R-GFP positive cells exhibit stem-like properties with rounded fusomes detected by α-HTS co-staining ( red ) . Note in panels B , E and F , the indicated UAS transgenes were expressed using the germline specific driver , nanos-GAL4VP16 ( Rorth , 1998 ) . For panels G and H , expression was achieved by the driver c587-GAL4 in somatic cells of the ovariole tip ( Song et al . , 2004 ) . All images shown are immunostainings for the p53R-GFPnls biosensor ( green ) , HTS ( red ) , and/or DAPI ( blue ) except for panel D which was co-stained with α-Vasa ( red ) to show that p53 activated cells retain the germline marker in bam mutants . All other panels ( A–C , E–H ) were stained with α-HTS ( red ) . Note that panel D stained with α-Vasa is the same bam ovariole shown in C with α-HTS . Relevant quantification including the nanosGAL4 driver alone is shown in Figure 5—source data 1 . Scale bars = 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 01810 . 7554/eLife . 01530 . 019Figure 5—source data 1 . Quantification of biosensor activity in germline tumors . This table quantifies the number of p53R-GFPnls positive stem-like cells associated with a rounded fusome ( α-HTS ) . Ovaries containing the nanosGAL4 driver alone ( control ) or the UAS oncogene indicated ( Rasv12 or CyclinE ) and the same GAL4 driver were scored . Note that the numbers of stem cells activated for p53 is much greater when either Rasv12 or CyclinE are present when compared to the control alone . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 01910 . 7554/eLife . 01530 . 020Figure 5—figure supplement 1 . Reporter induction during forced proliferation signals is independent of DNA damage . ( A ) Immunostaining for α-pH2Av ( red ) and p53R-GFPnls ( green ) in bamΔ86 ovaries . ( A′ ) shows α-pH2Av channel from ( A ) . Note the incidence of pH2Av ( arrows ) is rare and infrequently colocalizes with GFP . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 020 We considered the possibility that inappropriate growth might indirectly activate p53 by provoking DNA damage . To test this , we stained bamΔ86 ovaries for pH2Av ( Joyce et al . , 2011 ) . We observed very few pH2Av foci in bam tumors and , notably , these foci did not co-localize with p53 biosensor activity ( Figure 5—figure supplement 1 ) . Therefore , p53 activity in these tumors is not triggered by DSBs but instead , appears to be directly triggered by signals associated with hyperplastic growth . As seen in Figure 5 , diverse types of hyperplastic growth triggered constitutive p53 activity . To ask how p53 functions in these tumors , we examined bam−/− ovaries that were either WT or null for p53 . Tumor size was not significantly altered in the absence of p53 , but we did observe dramatically altered cytology in tumors that lacked p53 . As seen in Figure 6A , bam−/− ovarian cysts are typically filled with stem-like cells that exhibit round or dumbbell-shaped fusomes when stained with α-HTS ( Lin et al . , 1994 ) . As documented in Figure 6C , defective fusomes were seen in all bam−/−;p53−/− cysts and , in nearly half of these unusually large nuclei were observed . Though not quantified , micronuclei were also prevalent in these samples . Since defective fusome morphologies and irregular nuclei are consistent with aberrant mitosis , our data suggests a role for p53 in promoting proper cell cycle progression in these stem-like tumors . 10 . 7554/eLife . 01530 . 021Figure 6 . Abnormal fusomes and irregular nuclei are seen in bam−/−p53−/− tumors . ( A–A’ ) Cells in bam−/− tumors have rounded fusomes normally associated with the undifferentiated GSC fate . These are detected by α-HTS staining ( red in B , white in B’ ) . The nuclei of these cells counterstained with DAPI have diameters less than 10 μm ( blue in B ) . ( B–B’ ) bam−/−;p53−/− tumors frequently exhibit disorganized fusomes detected here by α-HTS staining ( red in C , white in C’ , yellow arrowhead ) . These tumors also have many fragmented and enlarged nuclei with a diameter significantly greater than 10 μm ( blue in C , yellow arrow ) . ( C ) Quantification of altered fusome structure and irregular nuclei in bam−/− and bam−/−;p53−/− tumors . Note that in panel C , counts for irregular nuclei do not include micronuclei . A total of 14 cysts were assayed in bam−/−;p53−/− and 8 cysts were assayed for bam−/− . All scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 021 To further examine the functional role of p53 in this context , we examined gene expression profiles of bam−/− ovaries that were WT or null for p53 by microarray . In total , we found that 297 gene transcripts were altered by at least twofold or greater in the absence of p53 . Table 1 lists the top 20 genes that are affected ( upregulated or downregulated ) by p53 in these tumors . Using the Gene Expression Commons ( GEXC ) tool , we compared these gene sets to existing germline , embryonic , and somatic expression profiles . We did not find a coherent pattern among the top 20 genes that are normally upregulated by p53 . However , among the top 20 genes that are normally suppressed by p53 in these germline tumors , we observed a modest enrichment for transcripts that were absent in either the embryonic stages or other somatic tissues ( Table 1—source data 1 ) . These data , together with our histological studies ( Figure 6 ) , establish that p53 exerts functional activities that impact cellular and molecular properties of Drosophila stem cell tumors . 10 . 7554/eLife . 01530 . 022Table 1 . p53 status impacts expression profiles in bam−/− tumorsDOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 02210 . 7554/eLife . 01530 . 023Table 1—source data 1 . Expression features of the top 20 genes suppressed by p53 . The top 20 genes that were suppressed by p53 in bam−/−;p53−/− tumors ( see Table 1 ) were examined using GEXC ( Seita et al . , 2012 ) to identify enriched pathways . Using this collection we observed a mild enrichment for genes that were absent in embryos or absent in adult somatic tissues relative to all genes in the fly genome . DOI: http://dx . doi . org/10 . 7554/eLife . 01530 . 023Downregulated by p53Upregulated by 53Gene symbolFold changeGene symbolFold change1CG316818 . 7CG31809−7 . 22CG51568 . 0CG31810−5 . 63LysX7 . 9CG2177−5 . 24CG319017 . 6CG7106−5 . 15CG167627 . 5CG1504−4 . 56CG322777 . 3CG15614−4 . 37CG172397 . 2unpg−4 . 28CG170127 . 1CG7329−4 . 29CG98977 . 1CG15236−4 . 110Ser126 . 8CG9294−4 . 111CG21916 . 8esg−3 . 912CG332586 . 6Ugt36Ba−3 . 713CG181256 . 5CG14297−3 . 614CG127806 . 4CG17129−3 . 615CG47836 . 3Cyp6a14−3 . 616Cyp6a186 . 3CG5568−3 . 417CG172346 . 2CG1077−3 . 318CG180636 . 2CG11226−3 . 319CG95686 . 1CG33105−3 . 120CG328346 . 0CG3328−3 . 1We performed microarray analysis on bam−/− and bam−/−;p53−/− tumors . The genes that are altered by p53 status in bam−/− tumors are recorded . Listed on the left are the top 20 genes whose abundance is directly or indirectly suppressed by p53 . Listed on the right are the top 20 genes whose abundance is directly or indirectly induced by p53 . The gene symbol is listed on the left and the fold change in gene expression between bam−/− and bam−/−;p53−/− tumors is listed on the right . Many genes listed here are dramatically affected when p53 is absent . We found that adult Drosophila exposed to genotoxic stress or genome destabilizers selectively activated p53 in GSCs and their immediate progeny . This striking specificity was observed despite widespread Dp53 expression ( FBgn0039044 m . ; Jin et al . , 2000; Ollmann et al . , 2000 ) and widespread tissue damage ( Figure 1—figure supplement 1 ) . We note that stem cell specificity was not an artifact intrinsic to the biosensors , since independent reporters behaved similarly in both the female and male germline and required the wild-type Dp53 locus in both cases . Furthermore , in certain mutant backgrounds stress-induced activity restricted to GSCs was lost and non-selective p53 activation was widespread throughout the ovary ( Figure 1—figure supplement 3B′ ) . Therefore , despite the fact that it is present and activatable throughout the gonad , functional p53 is restricted to stem cells and their immediate progeny by specific genetic determinants . Collectively , our work supports previous indications that there is an intimate and ancient link between p53 and stem cell biology ( Pearson and Sanchez Alvarado , 2009 ) . Our findings also offer rare and novel opportunities to operationally mark the stem cells in the fly germline , as visualized by p53R-GFP . This marker is distinct from conventional stem cell labels ( Deng and Lin , 1997 ) since it is not constitutively expressed but , instead , represents a functional output that is conditional upon a perturbation . We further note that like all reporter systems , our p53 biosensors may not reflect the full scope of effector output regulated by this network , and activities visualized here could transmit only subsets of p53-mediated responses . Nevertheless , despite this possible limitation , our results are consistent with suggestions that stem cells may be acutely sensitive to sources of genomic instability with a higher propensity for engaging adaptive responses relative to other cells ( Mandal et al . , 2011; Sperka et al . , 2012 ) . We propose that in reproductive tissues , the p53 regulatory network is either preferentially licensed in stem cells or selectively blocked outside of this compartment . What upstream regulators might specify p53 activation in GSCs/CBs ? Given that stem cells have unique properties , p53 activation in these cells might lie downstream of a novel pathway . Consistent with this idea , ATR expression , was not rate limiting for p53 activation in the germline ( Figure 1—figure supplement 3A′ , Figure 1—source data 1C ) . Furthermore , unlike meiotic induction , p53 induction in GSCs/CBs was independent of the topoisomerase , Spo11 ( Figure 1—source data 1A; Lu et al , 2010 ) . Chk2 could contribute to the selective activation in stem cells seen here , but since Chk2 is also broadly expressed and functionally associated with oocyte development throughout the ovary ( Abdu et al . , 2002; Oishi et al . , 1998 ) any potential role in GSCs must extend beyond a simple presence or absence of this kinase . Our findings also imply stimulus-dependent effectors of p53 in stem cells that are not yet appreciated . For example , within detection limits , we observed no obvious connection between p53 status and apoptosis , DNA double-strand break repair , or cell cycle arrest . However , irradiated p53−/− GSCs were significantly delayed in the re-entry phase for cell cycle . Future studies will explore this defect and also examine progeny derived from stressed GSCs for transgenerational phenotypes that might be adaptive . Our discovery that p53 action is coupled to hyperplasia in a non-vertebrate species was unexpected for two reasons . First , the role of this gene family as a tumor suppressor is thought to be a derived feature that evolved only in vertebrate lineages . Second , the canonical ARF/MDM2 pathway that links aberrant growth to p53 is absent outside of higher vertebrates ( Lu et al . , 2009 ) . Surprisingly , our combined results suggest that ancient pathways linking p53 to aberrant stem cell proliferation may predate the divergence between vertebrates and invertebrates . All fly stocks were maintained at 22–25°C on standard food media . We obtained rad54 , aubergine and cutoff mutants: rad54RU , rad54AA , aubHN , aubQC , cuffWM , and cuffQQ from T Schupbach ( Princeton University , Princeton , NJ , USA ) ; c587-GAL4 , UAS-dpp , UAS- Lsd1KD ( Eliazer et al . , 2011 ) , homozygous viable allele of bamΔ86 ( McKearin and Spradling , 1990; McKearin and Ohlstein , 1995 ) , nanos-GAL4VP16 , and UASp-tkvCA ( Chen and McKearin , 2003 ) have been described previously . All other stocks were obtained from Bloomington Stock Center ( Indiana University , Bloomington , IN , USA ) . The Dp53 rescue strain was engineered by ϕC31 integration of a 20-kb genomic fragment BAC containing the Dp53 locus into an attP site on the X chromosome of the PBac{y + -attP-9A}VK00006 line ( Bloomington #9726 ) . The parent BAC CH322-15D03 was obtained from the P[acman] resource library ( Venken et al . , 2009 ) and Rainbow Transgenic Flies performed the injection and screening for recombinants . The I-SceI endonuclease strain was generated by K Galindo ( Galindo et al . , 2009 ) , which was crossed to p53R-GFPnls ( STI150 ) ; HS- ( 70Flp ) ( 70 I- Sce I ) /TM6 for heat-inducible I-SceI endonuclease expression . Adult females were fattened for 2–3 days after eclosion and then subjected to heat shock in a circulating water bath at 37°C for 90 min and repeated for three consecutive days . 24 hr after the last heat shock , ovaries were dissected for immunostaining . For forced proliferation assays , two GAL4 lines were used: nanos-GAL4VP16 was used to achieve overexpression in the germline with UAS constructs for RasV12 , CyclinE , and Thickveins ( Rorth , 1998 ) . c587-GAL4 was used to achieve overexpression of UAS constructs of Dpp or Lsd1-RNAi in the somatic cells of the ovariole tip ( Song et al . , 2004 ) . For cyclinE overexpression , stocks were maintained at 25°C and female virgins were collected upon eclosion , shifted to 29°C for 4–5 days then subjected to immunostaining . For the RasV12 studies , female virgins were shifted to 29°C for 1 day and then shifted down to 25°C for 3 days prior to immunostaining . The Gal4-UAS system ( adapted from yeast ) often produces optimal expression at temperatures higher than 25°C . Since the UAS-Rasv12 and UAS-CyclinE constructs were not optimized for expression in the germline we applied these temperature shifts to produce more penetrant phenotypes . Well-fed flies were exposed to ionizing radiation using a Cs-137 Mark 1-68A irradiator ( JL Shepherd & Associates , San Ferando , CA , USA ) at a dose of 4 krad unless otherwise noted . When irradiating several genotypes , each genotype was placed in an individual vial , and all vials were exposed to IR at the same time on a rotating turntable inside the irradiator . For visualizing reporter activation after IR , flies were dissected 24 hr post-IR to allow for stable GFP expression . 3- to 5-days-old well-fed females were dissected in PBS and fixed in 4% EM-grade formaldehyde ( Polysciences , Warrington , PA ) diluted in PBS-0 . 1% tween-20 , with three times the volume of heptane . After washing , tissues were blocked in 1 . 5% BSA , then incubated with primary antibodies at 4°C overnight . Antibodies used: rabbit α-GFP ( Invitrogen , Carlsbad , CA ) ; rabbit α-pH2Av ( kindly provided by K McKim with specific staining protocols ) , rabbit α-cleaved caspase 3 ( Asp175 ) ( Cell Signaling , Danvers , MA ) ; mouse α-Armadillo , mouse α-BrdU ( Sigma , St . Louis , MO ) , mouse α-HTS clone 1B1 ( Developmental Studies Hybridoma Bank , Iowa City , IA ) , and rat α-Vasa ( Developmental Studies Hybridoma Bank ) . For fluorescence visualization , Alexa-488 , 568 ( Invitrogen ) , and DyLight 649 ( Jackson ImmunoResearch , West Grove , PA ) secondary antibodies were used and 0 . 1 μg/ml of DAPI ( Invitrogen ) for DNA staining was added in the first wash step . After three washes , ovaries were further hand dissected and mounted in VECTASHIELD ( Vector Laboratories , Burlingame , CA ) for microscopy imaging . For validating stimulus-dependent p53 action as visualized by the reporters , we routinely confirmed absence of GFP expression using flies null for Dp53 . We note that p53R-GFPnls shows constitutive expression independent of p53 in a subset of gut cells and in the region of the testis containing elongated spermatids , reflecting position effects upon this transgene . In fertility assays , two p53 null alleles , 238H ( ns ) and 5A-1-4 ( k1 ) were used in trans-combination to reduce genetic background influences . Two wild-type strains , yw and w1118 were used for comparison . p53 rescue transgenes were tested in a transheterozygous p53−/− background ( A1; ns/k1 and A2; ns/k1 ) to exclude contributions from background modifiers . 5- to 7-day-old females were irradiated at desired doses ( 11 . 5 krad for Figure 4B and 9 krad for Figure 4—figure supplement 3 ) and fertility was tracked over time in groups . Each group contained 10 females and five unirradiated wild-type Canton-S males . The animals were transferred to a new vial at designated time points , and fertility was scored by the presence of larvae 10 days after the parents were removed . Each trial contained 2 to 15 replicates per genotype . For Figure 4B percentages of fertile samples are plotted based on five trials . In the proliferative arrest assay , ovaries were dissected and immersed in Grace’s media containing BrdU ( 10 µM ) for 1 hr at room temperature . After fixation , ovaries were treated with 2N HCl for 30 min then 100 mM of borax was added for 2 min to neutralize the pH . Tissues were then processed for blocking and regular immunostaining . For all statistical analysis , data were placed into GraphPad Prism software . For statistics on the IR and Isce-I reporter activation ( Figure 1G , Figure 1—source data 1 ) , one-way ANOVA test was performed on all genotypes with a Tukey’s Multiple Comparison post-test . Reporter activation in aubergine , cutoff , and rad54 mutants ( Figure 3 , Figure 3—source data 1 ) was analyzed using a two-tailed unpaired t-test comparing the transheterozygous mutant to the heterozygous control . The same analysis was carried out for region 3 and stage 2–8 ( Figure 3—figure supplement 1 , Figure 3—source data 1 ) . For statistical analysis on fertility and BrdU incorporation assays ( Figure 4 ) , one-way ANOVA test was performed for each time point with a Dunnett post-test in which p53−/− data was the control . For cleaved-caspase 3 analysis ( Figure 4—figure supplement 1 ) , the data was analyzed using a two-tailed unpaired t test . In cases where replicates produce identical values incompatible with the prism two-tailed unpaired t-test tool , one value was negligibly revised to enable computation by this software ( e . g . , when both values were 0 , one value was changed to 1 . 0e−12 ) . About 200 ovaries from bam or bamp53 adult females were dissected in batches and pooled together to extract total RNA using Trizol ( Invitrogen ) . After verifying RNA integrity using Bioanalyzer ( 2100; Agilent ) , whole-genome expression of each genotype was analyzed using Affymetrix Drosophila Genome 2 . 0 Array at UTSW Genomics & Microarray core facility . Microarray data sets were uploaded to Gene Expression Commons ( https://gexc . stanford . edu ) and analyzed with 17 other public available data sets . In Gene Expression Commons , raw microarray data is individually normalized against a large-scale common reference ( for Drosophila genome , n = 2687 as of Nov 2013 ) , mapped onto the probeset meta profile . This strategy enables profiling of absolute expression levels of all genes on the microarray , instead of conventional methods where differences in gene expression are compared only between samples within an individual experiment ( Seita et al . , 2012 ) .	The most common genetic change seen in cancer patients produces a faulty version of the p53 protein , which normally restricts tissue growth . This change promotes cancer because cells can now divide faster and fail to die when they should . Much remains to be learned about how p53 functions to restrain growth . As p53 is found in primitive organisms , and cancer is unlikely to have significantly influenced evolution , suppressing tumor formation was almost certainly not the original function of this gene . Furthermore , p53 works in a different way compared to many other tumour suppressors . Therefore , prevention of cancer is likely to have evolved as a side effect derived from more ancient functions . Recently , a link between p53 and stem cells has been uncovered . Stem cells are special because they can develop into many different types of cells , and they are crucial for the growth and repair of tissues . To form a particular type of cell , the stem cell divides to create two daughter cells . Commonly , one daughter cell stays in the stem state , whereas the other becomes a particular type of cell , such as a nerve cell or muscle cell . Because of this special property , scientists hypothesize that stem cells have special mechanisms to protect them from DNA damage that might partially depend on p53 . This would prevent the spread of damaged genomes that would otherwise occur among daughter cells . To learn more about how p53 influences stem cells , Wylie , Lu et al . monitored its activity in the gonads of fruit flies , which are a powerful genetic model . They found that damaging DNA activates p53 in stem cells and their daughter cells , but not in other types of cells that have been damaged . In addition , p53 is activated by the uncontrolled growth and division of stem cells in the gonad , even when DNA is not damaged . This is unexpected since molecules linking inappropriate growth to p53 were thought to be present only in mammals . Therefore , it appears that the tumor-suppressing behavior of p53 in mammals was adapted from its more ancient ability to regulate stem cell growth—an ability that evolved before organisms divided into vertebrates and invertebrates .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2014	p53 activity is selectively licensed in the Drosophila stem cell compartment
The phylum Apicomplexa comprises human pathogens such as Plasmodium but is also an under-explored hotspot of evolutionary diversity central to understanding the origins of parasitism and non-photosynthetic plastids . We generated single-cell transcriptomes for all major apicomplexan groups lacking large-scale sequence data . Phylogenetic analysis reveals that apicomplexan-like parasites are polyphyletic and their similar morphologies emerged convergently at least three times . Gregarines and eugregarines are monophyletic , against most expectations , and rhytidocystids and Eleutheroschizon are sister lineages to medically important taxa . Although previously unrecognized , plastids in deep-branching apicomplexans are common , and they contain some of the most divergent and AT-rich genomes ever found . In eugregarines , however , plastids are either abnormally reduced or absent , thus increasing known plastid losses in eukaryotes from two to four . Environmental sequences of ten novel plastid lineages and structural innovations in plastid proteins confirm that plastids in apicomplexans and their relatives are widespread and share a common , photosynthetic origin . The phylum Apicomplexa is a major group of protistan parasites important in animal disease globally ( we will use the name Apicomplexa hereinafter for the clade of parasites sensu stricto; a synonym of the taxon Sporozoa Leuckart , 1879; Adl et al . , 2019 ) . The group includes the human pathogens Plasmodium ( haemosporidians ) , Toxoplasma ( eucoccidians ) , Babesia ( piroplasms ) , and Cryptosporidium ( cryptosporidians ) , whose cell biology and genomes have been extensively studied . Conversely , most apicomplexans from invertebrates such as eugregarines , archigregarines , blastogregarines , protococcidians , and agamococcidians lack detailed genomic information , in part because they cannot be cultured in laboratory conditions . Because these uncultured groups are also deep-branching lineages with unresolved relationships to the medically important taxa ( Leander and Ramey , 2006; Rueckert et al . , 2011; Simdyanov et al . , 2018; Simdyanov et al . , 2017 ) , the lack of their genomes and cell biology data hinders our understanding of apicomplexan evolution and the origin of parasitism itself . This , in turn , limits insights into infection mechanisms across the group: the structural and molecular make-up of machineries for host cell invasion such as the apical complex , pellicle , and glideosome , and how these relate to parasite life cycles , host preferences , and habitats . Apicomplexans are evolutionarily derived from mixotrophic algae , a realization that first came with sequencing a plastid genome in Plasmodium and localizing it into a cryptic , non-photosynthetic organelle called ‘the apicoplast’ ( Gardner et al . , 1991; McFadden et al . , 1996; Wilson et al . , 1996 ) . The apicoplast is a four-membrane plastid ( a broader term we will use hereinafter to describe the organelle in both parasitic and free-living organisms ) , which is derived from a secondary endosymbiont . Where exactly this plastid endosymbiont came from was settled by data from two newly discovered photosynthetic relatives of Apicomplexa , Chromera velia and Vitrella brassicaformis ( Janouskovec et al . , 2010; Moore et al . , 2008; Oborník et al . , 2012 ) . The photosynthetic plastids in Chromera and Vitrella are also surrounded by four membranes and they share conspicuous similarities with both the apicomplexan plastid and the plastid in peridinin-pigmented dinoflagellates , pointing to their common origin ( Janouskovec et al . , 2010 ) . Chromera and Vitrella belong to a monophyletic group , called the chrompodellids , with heterotrophic colpodellids Alphamonas edax , Voromonas pontica , and 'Colpodella' angusta , which also retain non-photosynthetic plastids ( Gile and Slamovits , 2014; Janouškovec et al . , 2015 ) . Plastids are nevertheless absent in Cryptosporidium ( Abrahamsen et al . , 2004 ) , and they have never been recorded in other deep-branching apicomplexans ( Toso and Omoto , 2007 ) . Lack of plastids in these groups once fueled alternative ideas about independent gains of plastids in apicomplexans , dinoflagellates , and even Chromera ( Bodył , 2005; Bodył et al . , 2009 ) . However , it also brings into question the metabolic importance of the plastid for the apicomplexan cell and its implications for plastid maintenance and loss in eukaryotes more broadly ( Cavalier-Smith , 2013; Janouškovec et al . , 2015 ) . Here , we sought to understand the phylogeny and plastid evolution of Apicomplexa by filling in major gaps in their sequence data . We used individual cells of parasites to generate transcriptomes from all major apicomplexan groups that currently lack laboratory cultures and large-scale transcriptomic or genomic data . By resolving their phylogeny , we observe that parasites with apicomplexan-like morphologies are polyphyletic and originated at least three times independently . We also show that gregarines and eugregarines are monophyletic , and blastogregarines are related to archigregarines , highlighting the importance of several traits uniquely shared among them . Many deep-branching apicomplexans contain plastidial metabolism and divergent , AT-rich plastid genomes , but eugregarines lost plastids at least twice independently . Phylogeny of 16S ribosomal RNA genes and structural novelties in plastid proteins demonstrate that plastids are widespread and ancestral in the group . We generated 13 transcriptomes for 10 parasites , representing six deep apicomplexan lineages with poor presence of sequence data: protococcidians , agamococcidians , blastogregarines , archigregarines , eugregarines ( three different superfamilies ) , and incertae sedis species ( Supplementary file 1 ) . Between 1 and 80 cells per species were isolated from the intestines of marine annelids , molluscs , and barnacles . The parasite cells were washed and preserved for RNA extraction ( Materials and methods ) . Transcriptomes were sequenced from amplified cDNA by pair-end Illumina HiSeq and assembled in Trinity . To resolve deep apicomplexans relationships , we modified a published dataset of slow-evolving nucleus-encoded markers ( Derelle et al . , 2016 ) by including broad , representative sampling of genomes and transcriptomes of apicomplexans and related taxa ( Supplementary file 2 ) . This produced a phylogenetic matrix of 296 protein sequences , which were individually verified for orthology by maximum likelihood phylogenies – this allowed us to unambiguously identify paralogous and contaminant sequences ( Materials and methods ) . Our apicomplexan transcriptomes typically contained a single ortholog per gene; two exceptions to this were suggestive of cryptic species among the collected cells . In Rhytidocystis sp . 1 , the most complete of multiple isoforms was selected , whereas two distinct sequence variants were present in Siedleckia nematoides ( Figure 1—figure supplement 1A ) and eventually merged into a single taxonomic unit ( Figure 1A ) . Three other taxa were merged in the final dataset due to poor sequence representation: Ascogregarina , and two unidentified parasites of hexapods ( Borner and Burmester , 2017 ) ( Figure 1A and Figure 1—figure supplement 1A ) . All three are members of the superfamily Actinocephaloidea based on a consensus of protein and ribosomal RNA gene ( rDNA ) phylogenies ( Materials and methods ) . The final phylogenetic matrix contained 50 species , 99908 amino acid positions , and relatively little ( 10 . 6% ) missing information ( Figure 1—source data 1 ) . Maximum likelihood analysis with the LG+C60+F+G4+PMSF model in IQ-TREE ( non-parametric and UFBoot2 supports ) and PhyloBayes analysis with the CAT+GTR model ( ten independent runs ) produced congruent topologies that were fully resolved at most internal branches ( Figure 1A ) . The protococcidian Eleutheroschizon duboscqi was a sister of eucoccidians , and rhytidocystids branched as basal coccidiomorphs , the group containing most medically important apicomplexans . The blastogregarine Siedleckia was strongly related to the archigregarine Selenidium pygospionis . Gregarine apicomplexans ( sensu lato including blastogregarines but excluding Digyalum; see below ) and eugregarines were both monophyletic , in contrast to most interpretations based on ribosomal RNA genes ( rDNA ) ( Cavalier-Smith , 2014; Leander , 2008; Rueckert et al . , 2011 ) . To test relationships between major apicomplexan lineages , we analyzed datasets in which the two longest branches in the tree , Cephaloidophora and Gregarina , were excluded either individually or both together ( LG+C60+F+G4+PMSF model with UFBoot2 supports ) . The resulting trees had congruent topologies with all internal branches fully supported except for two eugregarine subclades and the position of cryptosporidians ( Figure 1—figure supplement 1B ) . Similarly , seven statistical tests on 105 tree topologies representing all possible relationships between coccidiomorphs , cryptosporidians , eugregarines , archigregarines , and blastogregarines rejected all alternative topologies at p=0 . 01 except those differing in the placement of cryptosporidians ( Supplementary file 3 ) . The relationship of cryptosporidians therefore requires additional support , although their sister position to gregarines , which was unambiguously recovered in all trees including the ten PhyloBayes runs , is a preferred hypothesis . Two apicomplexan-like parasites branched outside the main apicomplexan clade ( Figure 1A , B ) . Digyalum oweni , a formally described archigregarine was fully resolved as a sister lineage to all apicomplexans and chrompodellids . A previously undescribed parasitic symbiont of the annelid Scoloplos with apicomplexan-like traits , named ‘Symbiont X’ , was specifically related to Chromera velia . Thus , apicomplexan parasites in the traditional sense are polyphyletic ( see Discussion ) . We keep using the name ‘Apicomplexa’ for the clade of parasites sensu stricto hereinafter ( Figure 1A ) . We next explored the existence of plastids in gregarines , Digyalum , and other parasites , where none have been known . Searching their sequence data with sequences of known plastid-localized proteins ( Janouškovec et al . , 2015; Ralph et al . , 2004; Seeber and Soldati-Favre , 2010 ) revealed broad presence of plastidial pathways ( Materials and methods; Supplementary file 4 ) . Digyalum , Selenidium , Siedleckia , rhytidocystids , and Eleutheroschizon all contain near-complete plastidial biosynthesis of isoprenoid precursors , heme , and fatty acids , ferredoxin redox system , iron-sulfur cluster synthesis , and plastid genomes ( Figure 2A ) . The eugregarine Lankesteria unusually contains fatty acid biosynthesis as the only plastidial pathway , whereas Symbiont X contains only the isoprenoid pathway , similar to piroplasms ( Lizundia et al . , 2009 ) . Both Lankesteria and Symbiont X appear to lack plastid genomes ( Figure 2A ) . The distribution of control , signature plastid genes involved in polypeptide import , folding , and DNA replication in the plastid ( cpn60 , sDer-1 , PREX ) , matches the presence of plastid metabolism and genomes ( Figure 2A ) . Maximum likelihood phylogenies of all individual proteins allowed us to readily distinguish the apicomplexan sequences from bacteria and other contaminants in the datasets ( Materials and methods ) . In most phylogenies , the apicomplexan sequences cluster with algal plastid forms , confirming that they came from the plastid endosymbiont rather than the eukaryotic host . The phylogeny is different in several genes that are either derived by horizontal gene transfer from bacteria or in fact localize outside of the plastid in Plasmodium ( in heme biosynthesis; see below ) . N-terminal regions of plastid sequences from our new transcriptomes often carry signal peptides typical for targeting to the plastid ( the other have incomplete N-termini or lack targeting signatures by default , such as most triose phosphate translocators ) . The signal peptides are often followed by transit peptide-like regions , although these are more difficult to predict computationally ( Supplementary file 5 ) . In Digyalum , transit peptides have a net positive charge similar to other plastid leaders ( they are low in acidic and high in basic residues; Figure 2—figure supplement 1A ) . Digyalum transit peptides are compositionally similar to transit peptides in Plasmodium , and likewise lack the phenylalanine motif at the first position after the signal peptide cleavage site ( Figure 2—figure supplement 1B ) ( Patron and Waller , 2007 ) . Predicted localizations for plastidial proteins and mitochondrial ALAS correspond closely to experimental evidence in Plasmodium and Toxoplasma ( Figure 2A ) . The only exceptions to this pattern are the last three enzymes in heme biosynthesis , which are predicted to be plastidial in some parasites ( see Discussion ) . The reconstructed plastid pathways also reflect known dependencies between their modules ( Figure 2B ) . Iron-sulphur cluster assembly and ferredoxin system are widely required as co-factors for isoprenoid and fatty acid synthesis , whereas heme biosynthesis can be lost independently of other modules – likely the case in Lankesteria and Symbiont X . The synthesis of 3-phosphoglycerate ( GAPDH-II and PGK-II ) is present selectively . SufA was not identified in the Chromera genome . Pyruvate dehydrogenase and fatty and lipoic acid synthesis protein sequences in Digyalum are notably divergent , including three that are more closely related to bacterial than plastid sequences ( Figure 2A ) . The cytosolic mevalonate pathway for isoprenoid precursor synthesis is absent in all species , but the mitochondrial cysteine desulphurase ( IscS ) and cytosolic fatty acid synthase ( FASI ) and elongase ( ELO ) pathways are present , as expected ( Materials and methods ) ( Dellibovi-Ragheb et al . , 2013; Ramakrishnan et al . , 2012; Zhu et al . , 2004 ) . No plastid genes were identified in five eugregarine lineages other than Lankesteria , including in the draft genome of Gregarina ( Figure 2A; see Discussion about the PREX fragment in Ascogregarina ) . This result is suggestive of at least two losses of plastids in eugregarines , which were independent of the one in Cryptosporidium ( Discussion ) . The discovery of plastids with genomes in deep-branching apicomplexans raises questions about whether they correspond to the undescribed apicomplexan-related lineages ( ARLs ) as defined by plastidial 16S rDNA ( Janouškovec et al . , 2012 ) . Here , we discovered ten novel ARLs among environmental 16S rDNAs in GenBank and clustered 16S rDNAs obtained from the VAMPS database ( Huse et al . , 2014 ) by a phylogenetic sorting approach similar to the one used previously ( Janouškovec et al . , 2012 ) ( Materials and methods ) . We also discontinue the use of two ARL-X and ARL-XI described recently ( Mathur et al . , 2018 ) , which we find to be members of the Vitrella clade ( ARL-I; Figure 3A ) and bacterial contaminants , respectively ( Materials and methods ) . This brings the total number of unidentified ARLs to 15 , in addition to three ARLs represented by Chromera , Vitrella , and corallicolid clades ( see Supplementary file 6 for reference sequences for all ARLs ) . A global , maximum likelihood phylogeny of 16S rDNAs of bacteria , plastids and ARLs ( including representative sequences of known ARLs and all GenBank and VAMPS centroid sequences of novel ARLs ) readily illustrates that many plastids in apicomplexans and their relatives are yet to be discovered ( Figure 3A ) . It also shows that some new ARLs ( ARL-XV , ARL-XVI , ARL-XVII ) are comparatively diversified and abundant . Despite that ARLs have now more than doubled in number , none of them corresponds to the plastid 16S rDNAs of Digyalum , Eleutheroschizon , Siedleckia , Selenidium or Rhytidocystis . Instead , the five genera have some of the fastest-evolving and most AT-rich 16S rDNAs of all apicomplexans , and they cluster with compositionally similar sequences of the more distantly related hematozoans ( compare Figure 1A and Figure 3A ) . Such artificial grouping of highly divergent sequences is a well-known phylogenetic artifact , and deep-level relationships in the tree thus ought to be interpreted with caution . The 16S rDNAs of Digyalum , Eleutheroschizon , Siedleckia , Selenidium or Rhytidocystis were recovered among a set of AT-rich transcriptomic contigs , which encode other genes typical of apicomplexan plastid genomes ( Supplementary file 7; Materials and methods ) . In Rhytidocystis species 1 and 2 , the AT content of 16S rDNAs ( 84% and 86% , respectively ) and plastidial transcripts ( 88% and 91% , respectively ) is among the highest of all plastid genomes described to date ( Figure 3B ) ( Su et al . , 2019 ) . Digyalum , Siedleckia , Selenidium , and Rhytidocystis sp . one use a non-canonical genetic code in which UGA encodes for tryptophan; UGA is absent in the fragmentary plastid DNA of Rhytidocystis sp . two and encodes for STOP codons in Eleutheroschizon ( Figure 3B ) . The Digyalum plastid encodes six genes never identified in apicomplexan plastids and lacks nine genes they do contain; only one gene ( rps18 ) was relocated to the nucleus in parallel in both lineages ( Figure 3C ) . To test if the plastids in Digyalum , chrompodellids and apicomplexans are likely derived from a common source we searched for shared innovations in their plastid genes . We observed that the unusual plastid DNA replication and repair complex ( PREX ) ( Seow et al . , 2005 ) is found not only in apicomplexans and chrompodellids ( Janouškovec et al . , 2015 ) but also in Digyalum ( Figure 2A and Figure 4 ) . PREX protein contains N-terminal primase and helicase domains , which are homologous to the mitochondrial primase-helicase Twinkle and fused with an Aquifex-type exonuclease-polymerase downstream . Phylogeny of the polymerase unit confirms that it was acquired from an unknown bacterial source related to Aquifex before the Digyalum-apicomplexan split ( Figure 4A , C ) . The gene subsequently fused with the Twinkle gene and the product became targeted to the plastid ( the N-terminus of PREX is incomplete in most species including Digyalum but contains a signal peptide in Chromera; Supplementary file 5 ) . Another unusual fusion took place in 4-hydroxy-3-methylbut-2-enyl diphosphate reductase ( IspH ) , the last enzyme in the plastid isoprenoid precursor biosynthesis . A canonical plastid ispH gene of a cyanobacterial origin , which is also present in dinoflagellates and Perkinsus , was replaced by a Chlamydiae-like variant in the ancestor of Digyalum and Apicomplexa and fused with the plastid gene for sedoheptulose-1 , 7-bisphosphatase form 3 ( SBP3 ) ( Figure 4B , C ) . The fusion is absent in Toxoplasma , Plasmodium and piroplasms , which lack SBP3 , and has been interpreted as a derived characteristic in Chromera ( Petersen et al . , 2014 ) , but instead we find it is broadly distributed in apicomplexans and their relatives . Another unique evolutionary event in apicomplexan plastids is a fission in a non-conserved region of the plastid-encoded rpoC2 gene . The unprecedented split was once interpreted as a read-through frame shift ( in Plasmodium ) or read-through STOP codons ( in Toxoplasma and Eimeria ) in a continuous rpoC2 ( Cai et al . , 2003; Wilson et al . , 1996 ) . We instead find that all apicomplexan rpoC2 genes are split within the same region , including those in the deep-branching Selenidium and Siedleckia ( Figure 4C and Figure 4—figure supplement 1 ) . The two rpoC2 moieties are found in different reading frames in most species , but two different in-frame STOP codons ( UAA and UAG ) separate them in Toxoplasma . Both frame shifting and STOP codon read-through would be required for continuous rpoC2 translation in Eleutheroschizon ( Figure 4—figure supplement 1 ) . Such variable gene arrangements are incongruent with the expression of the apicomplexan rpoC2 as a single protein . Messenger RNA editing likewise does not correct the rpoC2 reading frame in Plasmodium ( Nisbet et al . , 2016 ) . Indeed , the downstream rpoC2 moiety almost unequivocally possesses an ATG start codon near the split site allowing it to be translated independently ( Figure 4—figure supplement 1 ) . The evidence altogether points to a fission event in rpoC2 , which is absent in all other plastids , including those of Digyalum and chrompodellids , and thus represents a defining ancestral characteristic of the apicomplexan plastid . Two additional plastid-associated proteins in apicomplexan and chrompodellid plastids derive from horizontally acquired genes . The first is Alphaproteobacteria-like ferrochelatase ( HemH ) , which is localized to mitochondria in Plasmodium but likely targeted to plastids in Chromera and possibly in some apicomplexans ( see Discussion ) ( Koreny et al . , 2011; Sato and Wilson , 2003; Varadharajan et al . , 2004 ) , and the other is Verrucomicrobia-like beta-ketoacyl-acyl carrier protein reductase ( FabG ) ( Janouškovec et al . , 2015 ) . Digyalum encodes the same alphaproteobacterial HemH but it has an unusual FabG , which is related to other bacteria ( Figure 4C , Figure 4—figure supplement 2 and Figure 4—figure supplement 3 ) . Finally , genes of two well-known cytosolic proteins ( Huang et al . , 2004; Zhu and Keithly , 2002 ) were also acquired from bacteria in the Digyalum-Apicomplexa ancestor: lactate/malate dehydrogenase and glutamine synthase type I ( Figure 4C ) . Generating transcriptomes from uncultured apicomplexans across their evolutionary diversity provides the first comprehensive insights into relationships between major apicomplexan groups . Two species with apicomplexan-like morphology either described as apicomplexans ( Digyalum ) or yet unclassified ( Symbiont X ) are not members of Apicomplexa sensu stricto . This shows that apicomplexan-like parasites are polyphyletic and evolved at least three times independently . Specifically , large trophont stages attached to intestines of marine invertebrates by specialized apical structures are products of convergent evolution . The trophonts of Digyalum , for example , parasitize the gut epithelium of Littorina snails and their attachment structure contains an apical complex with a protruded polar ring , which provides a gateway for rhoptry-mediated secretion – a combination of traits typical for gregarine apicomplexans ( Dyson et al . , 1994; Dyson et al . , 1993 ) . Symbiont X is known only from light microscopy data but would be also readily classified as an apicomplexan based on crude characteristics: it parasitizes the gut of Scoloplos armiger polychaetes being attached to the host epithelium by its apical end ( unpublished data ) . Tracing the evolution of such parasite characteristics , however , indicates that the basis for convergence lies in evolution acting on similar preconditions . Apical complexes with rhoptries , micronemes and pseudoconoids are present in free-living relatives of apicomplexans , such as in the predatory Colpodella and Psammosa and photosynthetic Chromera , and in more distantly related parasites such as Perkinsus and Parvilucifera ( Foissner and Foissner , 1984; Norén et al . , 1999; Oborník et al . , 2011; Okamoto and Keeling , 2014; Perkins , 1996 ) . Such broad distribution points to a single origin of the apical complex in the ancestor of apicomplexans and dinoflagellates in a non-parasitic context ( Figure 1A ) . Because the apical complex mediates secretion often associated with cell-to-cell interactions , it is likely an important precondition in multiple origins of parasitism in both apicomplexans and dinoflagellates . In similar host environments the structure may have also promoted convergent parasite morphologies . The use of the apical complex in extracellular attachment and secretion in the gut epithelium of animal hosts , for example , may have triggered convergent expansion in the cell size of gregarine , Digyalum and Symbiont X trophonts . Unsurprisingly , convergent similarities in the three lineages are accompanied by considerable differences in detailed morphology: Digyalum and Symbiont X do not glide or twist but they pulsate , and detailed ultrastructure of their apical complex and pellicle ( Dyson et al . , 1994; Dyson et al . , 1993 ) ( unpublished data ) is distinct from that in gregarines ( Kováčiková et al . , 2017; Paskerova et al . , 2018; Valigurová et al . , 2017 ) . Similar divergence characterizes their molecular make-ups: apicomplexans are well-known auxotrophs for purines , but Digyalum contains a pathway for their synthesis ( data not shown ) and , despite that both lineages lost photosynthesis , their plastid genomes have been reduced in different ways ( Figure 3C ) . The convergent morphologies of Digyalum , Symbiont X and apicomplexans are therefore rather superficial similarities , but the possibility that they were driven by the presence of shared ancestral traits ( such as the apical complex ) in similar host habitats highlights the importance of preconditions in the origin of parasites ( Janouskovec and Keeling , 2016 ) . Key findings of multiprotein phylogenies are that eugregarines and gregarines are unequivocally monophyletic ( Figure 1 and Figure 1—figure supplement 1 ) . This broadly supports traditional morphological classifications ( Grassé , 1953 ) in contrast to many recent proposals based on rDNA phylogenies , which regard both groups as polyphyletic ( Cavalier-Smith , 2013 ) . Indeed , protein sequences allow for building larger phylogenetic matrices and have more even substitution rates than rDNAs , some of which are notoriously divergent and phylogenetically unstable . Although the sampling of eugregarine diversity is incomplete , our phylogeny contains six of their seven main lineages at the superfamily level ( Simdyanov et al . , 2017 ) – one of them being Polyrhabdina , which is apparently not a lecudinoid ( Figure 1A and unpublished data ) . The eugregarine monophyly provides the first unambiguous phylogenetic support for their two candidate synapomorphies: the ultrastructure of the epimerite , and the ultrastructure of epicytic crests ( Simdyanov et al . , 2017 ) . It also partially resolves the little understood relationships between eugregarine superfamilies ( Cavalier-Smith , 2014; Simdyanov et al . , 2017; Simdyanov et al . , 2015 ) . The blastogregarine Siedleckia groups strongly with the archigregarine Selenidium . The sampling of both groups is incomplete , especially in the archigregarines , but they do share a combination of characteristics that are rare or absent in other apicomplexans , namely active bending and twisting movement in trophozoites , pellicular folds running along the body length in many species , and one or more layers of longitudinal microtubules underlying the pellicle ( Schrével et al . , 2016; Simdyanov et al . , 2018 ) . Both lineages also feed by myzocytosis , which is known in some relatives of apicomplexans ( Foissner and Foissner , 1984; Mylnikov and Mylnikova , 2008 ) but not in eugregarines , cryptosporidians or coccidiomorphs . Because the blastogregarine ultrastructure and life cycle share additional similarities with coccidians ( Simdyanov et al . , 2018 ) , confirming their phylogenetic position is important for reconstructing character evolution in Apicomplexa as a whole . The monophyly of gregarines sensu lato in our trees ( Figure 1 and Figure 1—figure supplement 1 ) provides phylogenetic support for a tentative synapomorphy of this group , the forming of a gametocyst during the life cycle ( Simdyanov et al . , 2017 ) , although this characteristic is absent in blastogregarines . The position of cryptosporidians is still not fully resolved , but the group is consistently recovered next to gregarines ( Figure 1 and Figure 1—figure supplement 1 ) . Eleutheroschizon is clearly important to understanding the origin of eucoccidians ( Figure 1A ) ( Valigurová et al . , 2015 ) , although its classification as a protococcidian is apparently incorrect ( unpublished data ) . Two Rhytidocystis species are positioned as a basal group in the class Coccidiomorpha ( Figure 1A ) , which supports traditional views on their classification closer to coccidians ( Levine , 1979; Porchet Hennere , 1972 ) rather than to gregarines ( Cavalier-Smith , 2014 ) . The bigger group that includes rhytidocystids and related parasites of annelids and molluscs ( Kristmundsson et al . , 2011 ) therefore provides a stepping stone for understanding the evolution of coccidiomorphs , including the majority of medically important apicomplexans . Evidence for endosymbiotic genes with plastid targeting signals and plastid genomes shows that plastids are common in deep-branching apicomplexans , despite not having been previously recognized ( Figure 2 , Figure 3 and Supplementary file 5 ) . Their plastid metabolic networks are similar to those in Toxoplasma and Plasmodium and generally serve to produce three essential metabolites: isoprenoid precursors for the synthesis of carotenoids , ubiquinone and prenyls; heme as a protein co-factor; and fatty acids as constituents of cellular lipids ( Imlay and Odom , 2014; Ramakrishnan et al . , 2012; van Dooren et al . , 2012 ) . Heme biosynthesis is partially localized in the cytosol and mitochondria , and fatty acids can be extended by non-plastidial elongases ( Figure 2A ) ( Ramakrishnan et al . , 2012; Zhu et al . , 2004 ) , but de novo biosynthesis of all three metabolites outside of the plastid appears to be absent . This creates dependence on the plastid for the production of isoprenoid precursors , heme , and fatty acids , which can be overcome only by metabolite uptake from the environment . Salvage of host isoprenoids , heme , and fatty acids is common in parasites , but obtaining them in sufficient amounts across the life cycle is apparently challenging . Most parasites retain all three plastidial pathways , and only two lineages have lost plastid organelles altogether ( Figure 5 ) : cryptosporidians ( Abrahamsen et al . , 2004 ) , and the dinoflagellates Hematodinium and Amoebophrya ( Gornik et al . , 2015; John et al . , 2019 ) . In free-living organisms , plastids are always retained , perhaps because replacing their metabolism by salvage is even more difficult ( Janouškovec et al . , 2017 ) . Most consistently required is the biosynthesis of isoprenoid precursors ( Figure 5 ) , which is the only indispensable plastidial pathway in piroplasms , the Plasmodium blood stage , and possibly in Symbiont X ( Figure 2A ) ( Lizundia et al . , 2009; Yeh and DeRisi , 2011 ) . A fully unprecedented situation exists in the eugregarine Lankesteria , where fatty acid biosynthesis appears to be the only retained plastidial anabolic pathway ( the transcriptome is fragmentary but enzymes for the biosynthesis of isoprenoid precursors and heme are absent ) . Interestingly , the existence of plastids in Selenidium and Rhytidocystis is also consistent with ultrastructural reports of multimembrane organelles in some of their species , which correspond to non-photosynthetic plastids in other apicomplexans by size , appearance , and their position near the nucleus ( Figure 5 ) ( Leander and Ramey , 2006; Porchet Hennere , 1972; Schrével , 1971; Schrével et al . , 2016; Wakeman et al . , 2014 ) . Five eugregarines lack any evidence for plastid presence; these include Gregarina where a draft genome is available ( Figure 2A ) . Since Lankesteria is phylogenetically nested among them , eugregarines must have lost plastids at least twice ( Figure 5 ) . Ascogregarina contains a fragment of PREX ( Figure 2A and Figure 4A ) , but because the species lacks evidence for other endosymbiont-derived genes we conservatively consider it encodes for an ex-plastidial protein or is a contaminant . Our results indicate that the two plastid losses in eugregarines are independent of the one in cryptosporidians ( Figure 1A ) , unlike previous assumptions ( Cavalier-Smith , 2014; Toso and Omoto , 2007 ) . This increases total known losses of plastid organelles in apicomplexans to three and in all eukaryotes to four ( Figure 5 ) . Plastids in Chromera and Vitrella share several unique characteristics with the apicomplexan plastid ( Janouskovec et al . , 2010 ) but their common origin was doubted in the past ( Bodył et al . , 2009 ) and other early endosymbiotic events could underlie plastid evolution in the broader group including dinoflagellates ( Waller and Kořený , 2017 ) . Testing whether plastids in apicomplexans , chrompodellids and Digyalum derive from the same endosymbiont is therefore relevant to understanding plastid evolution and frequency of plastid losses in general . We find three lines of evidence that support a common origin . Firstly , phylogenies of individual plastid genes repeatedly group Digyalum with apicomplexans and chrompodellids , and their topologies either reiterate nuclear phylogenies directly or are unresolved without consistently supporting alternatives ( Figure 4A , B , Figure 4—figure supplement 2 and Figure 4—figure supplement 3 ) . Secondly , unidentified lineages that carry plastid-encoded 16S rDNAs related to apicomplexans and chrompodellids ( ARLs ) are growing in number and diversity ( Figure 3A ) . Some are apicomplexans ( Janouškovec et al . , 2013; Kwong et al . , 2019 ) and others may be free-living , but they altogether support the idea that plastids in the lineage are widespread ( Janouškovec et al . , 2012 ) . Finally , unique evolutionary innovations in the plastidial PREX , ispH , and hemH link together the plastids in apicomplexans , chrompodellids , and Digyalum ( bacterial fabG and split rpoC2 further link plastids in some of them ) . It is difficult to see how endosymbiosis would move plastids in or out of the lineage without changing the distribution of these genes . Altogether , the evidence suggests that plastids in apicomplexans , chrompodellids , and Digyalum were inherited vertically from a common ancestor , are widely distributed in the group , and are most likely retained by default , particularly in free-living representatives ( Janouškovec et al . , 2015; Janouskovec et al . , 2010 ) . Core plastid metabolism in Digyalum , Symbiont X , and apicomplexans has been remarkably conserved across long time scales and reveals only a few outstanding variations . In Plasmodium falciparum , triose phosphate sugars are imported by two triose phosphate translocators ( TPTs ) residing in the outermost ( PfoTPT ) and innermost ( PfiTPT ) plastid membranes , respectively ( Mullin et al . , 2006 ) . Toxoplasma contains only one TPT phylogenetically corresponding to PfoTPT . Orthologs of the PfoTPT lack N-terminal targeting signatures and are ubiquitous in apicomplexans , chrompodellids , and Digyalum ( Figure 2—figure supplement 2 ) . PfiTPT orthologs are less common ( Gile and Slamovits , 2014 ) , but they possess N-terminal signal peptides in Chromera , Vitrella , and Rhytidocystis sp . two compatible with their targeting to the inner plastid membrane ( Figure 2A and Figure 2—figure supplement 2 ) . The split between the two forms therefore likely predated Apicomplexa and the loss of the PfiTPT form in the lineage leading to Toxoplasma is apparently a derived evolutionary state . One enzyme that processes the TPT substrate dihydroxyacetone phosphate ( DHAP ) is triose phosphate isomerase ( TPI-II ) , and the failure to identify TPI-II in piroplasms once led to the proposition that their plastids import glyceraldehyde-3-phosphate , the TPI-II product ( Fleige et al . , 2010 ) . We find that TPI-II , although highly divergent in piroplasms , is present in all apicomplexans with plastids , and it frequently possesses N-terminal signal peptides for plastid targeting ( Figure 2A ) . This suggests that the import and conversion of DHAP is conserved across apicomplexan plastids . Analysis of heme biosynthesis enzymes suggests that the pathway consistently starts in the mitochondrion , as in heterotrophic eukaryotes – the algal C5-pathway must have been lost prior to the Digyalum-Apicomplexa divergence . Delta-aminolevulinic acid is likely imported to the plastid and processed to coproporphyrinogen III ( Ralph et al . , 2004; Sato et al . , 2004 ) . The last three enzymes in some apicomplexans sequenced by us are unexpectedly predicted to be also plastidial , similarly to Chromera and Vitrella ( Koreny et al . , 2011 ) . In Plasmodium , and perhaps also in Toxoplasma and Digyalum , the last three heme biosynthesis enzymes localize in the cytosol and mitochondria , where heme is most needed ( Varadharajan et al . , 2004 ) . It would be interesting to explore these contrasting localization predictions experimentally , including the possibility that some enzymes are dually targeted , or their isoforms ( where present; Figure 2A ) are differentially targeted to different cellular compartments . Unlike metabolism in the plastid , plastid genome structure shows unexpected variations across Apicomplexa . Plastid genomes in some of the newly sequenced taxa , as partially reconstructed from transcripts , have unusually AT-rich and fast-evolving sequences ( Figure 3A ) . Completing the fragmentary plastid genome of Rhytidocystis species 2 ( Supplementary file 7 ) is of particular interest because it has the most AT-rich 16S rDNA , and potentially the most AT-rich genome , ever recorded among plastids ( Figure 3B ) ( Su et al . , 2019 ) . Plastid genome reduction in Digyalum and Apicomplexa from their common ancestor is primarily explained by the loss of photosynthesis ( Figure 3C ) . This was accompanied by relatively modest transfer of genes to the nucleus , which involved remarkably different gene sets in the two lineages ( only rps18 was relocated in both in parallel ) . Underlying the very existence of plastid DNA in Plasmodium and Toxoplasma is sufB , which encodes one of only two broadly conserved , plastid-encoded proteins with function other than transcribing and translating the genome itself . Of the newly sequenced species with plastids , only Symbiont X lacks any evidence for the sufB gene or plastid genome ( Figure 3A ) . It would be expected that Symbiont X sufB is encoded in the nucleus and was relocated there in its common ancestor with Chromera , Voromonas , Colpodella , and Alphamonas ( Janouškovec et al . , 2015 ) , thus allowing heterotrophs in this lineage to lose plastid genomes at least three times independently ( Figure 5 ) . We provide a strongly resolved phylogeny and large-scale sequence data for major apicomplexan groups , but the sampling of Apicomplexa is far from complete . Expanding the phylogenetic dataset with new species and genes will allow for testing key conclusions of our study ( e . g . , monophyly of gregarines and eugregarines ) and understanding the relationships of taxa that remain poorly sampled ( archigregarines , blastogregarines , protococcidians ) or lack genome-level sequences ( corralicolids , adeleid and aggregatid coccidians , and other incertae sedis taxa ) . Parasites with characteristics similar to apicomplexans could provide a study system for morphological and molecular convergence and insights into the transition from free-living species to obligate symbionts . Plastids in apicomplexans and their relatives are apparently ancestral and widespread , and more are likely to be discovered . The plastid function in apicomplexans , chrompodellids and Digyalum rarely includes photosynthesis , but it always involves synthesis of one or more indispensable metabolites ( Figure 5 ) . Losses of plastid organelles and their genomes are infrequent but did occur several times in the broader group , and are likely to provide an unparalleled model for understanding factors that mediate plastid maintenance and loss in eukaryotes as a whole . Interestingly , two key conclusions of our study are independently reinforced by a bioRxiv preprint ( Mathur et al . , 2019 ) , which describes a complementary set of parasite transcriptomes from the same group . Firstly , the manuscript reports two parasites with apicomplexan-like morphologies that likewise branch outside Apicomplexa . The early-branching Platyproteum , a representative of ‘squirmids’ ( Cavalier-Smith , 2014 ) , is the sister group of Digyalum , as has been apparent to us from 18S rDNA phylogenies ( unpublished data ) . Piridium then represents a sister taxon to Vitrella and therefore a fourth independent emergence of apicomplexan-like parasitism in the lineage . Secondly , plastids appear to be consistently absent in most eugregarine superfamilies except for the Lecudinoidea ( Lankesteria in our study and Lecudina and Pterospora in the other study ) , where fatty acid biosynthesis is the only core plastidial pathway . The presence of two additional eugregarine superfamilies in our trees ( Ancoroidea and Polyrhabdina ) points to an extra case of plastid loss ( Figure 5 ) , but the relationships of eugregarine superfamilies which are present in both studies are fully congruent . Integrating sequence datasets from both studies will be a first step in creating a phylogenetic framework for apicomplexan evolution . This framework will likely be useful in illuminating steps in the emergence of parasitism and in predicting cell biology of less known parasites by the methods of comparative genomics . Parasite cells ( 1 to approximately 70 individuals ) were isolated from marine annelid , mollusc and barnacle hosts collected at the White Sea Biological Station of Moscow State University during August of 2016 ( Supplementary file 1 ) . Cells were hand-picked by using a glass micropipette ( 30% ethanol was used to detach the cells of Digyalum and Eleutheroschizon ) , washed 1 to 4 times in clean seawater and transferred into a clean tube in the final volume of 2–5 ul of seawater ( excess seawater removed if necessary after a brief spin ) . Care was taken to avoid contamination by animal host cells and gut contents when isolating parasite cells but we were not able to fully prevent it ( host contaminants were observed in the sequence data but they were clearly distinguishable from the parasites; see below ) . A 100 ul of Lysis buffer ( RNAqueous-Micro Total RNA Isolation Kit; Ambion/ThermoFisher , cat no . AM1931 ) was added into the tube with parasite cells and the samples were stored at −80C for several weeks . Total RNA was extracted from samples by RNAqueous-Micro Total RNA Isolation Kit according to manufacturer instructions but without the DNase I digest step . In Digyalum and Eleutheroschizon , RNA from two independent cell isolations was combined before further processing ( Supplementary file 1 ) . Cells of Digyalum oweni WS3-2017 was stored in RNALater and used for reverse-transcription without extracting RNA . RNA was reverse-transcribed by using SMART-Seq v4 Ultra Low Input RNA Kit ( Takara; cat no . 634888 ) , however20-30 amplification cycles; optimization was done as in SMARTer Pico PCR cDNA Synthesis Kit ) . Indexed TruSeq libraries were built at Edinburgh Genomics and sequenced as paired-end 150 bp reads in two multiplexed lanes on the Illumina HiSeq 4000 machine . For the Digyalum oweni WS3-2017 sample , paired-end 150 bp HiSeq 4000 reads were produced independently of other samples . Demultiplexed reads were processed by cutadapt v1 . 8 . 3 to remove low quality nucleotides ( -q 20 setting ) , polyA tails , SMARTer adapters and leftover Illumina barcodes ( on 3' ends of shorter sequence fragments ) . Reads of minimum length of 100 nucleotides were assembled in Trinity v2 . 4 . 0 by using the default settings . Predicted proteomes were generated as six-frame translations of transcriptomic contigs . Analysis of assembled reads revealed minor cross-contamination between samples due to adapter swapping , but contaminants were well distinguishable by read coverage . Raw sequence reads and our assemblies were deposited in NCBI Bioprojects PRJNA557242 and PRJNA556465 . Photographs of studied parasites ( Figure 1B ) were made using Leica DM 2500 microscopes equipped with DIC optics and a digital camera ( Leica , Germany ) at the White Sea Biological Station of Moscow State University and the Marine Biological Station of St . Petersburg State University . We built our dataset from 339 protein sequence alignments previously used in stramenopile phylogenies and tested for orthology ( Derelle et al . , 2016 ) . Firstly , slow-evolving stramenopile sequence query for each alignment ( mostly the Phytophthora sequence ) was used to retrieve five best BLASTP hits ( e-value cutoff of 1e-5 ) and select the closest ortholog in translated alveolate , stramenopile , and rhizarian genomes ( Supplementary file 2 ) . Secondly , a new alveolate protein sequence query ( primarily from Vitrella ) was used in the same way to expand the dataset with our newly generated and existing transcriptomes . Closest hits from animal , fungal , microsporidian and bodonid genomes were also included to distinguish sequences derived from animal host and other contaminants in the samples . Orthologous sequences were identified by multiple rounds of alignment clean-up as guided by Maximum Likelihood phylogenies . Initial rounds of phylogenies were focused on removing out-paralogs and contaminants by using default alignment in MAFFT v7 . 402 ( Katoh and Standley , 2013 ) , alignment trimming in BMGE v1 . 12 ( Criscuolo and Gribaldo , 2010 ) with the -b 3 g 0 . 4 setting and phylogeny in Fasttree v2 . 1 . 10 ( Price et al . , 2010 ) with the -lg -gamma setting . Later rounds were focused on resolving difficult cases of paralogy , horizontal gene transfer , and the orthology of fast-evolving sequences: localpair , --linsi , alignment in MAFFT; -b 4 g 0 . 4 settings in BMGE; and IQ-TREE v1 . 6 . 5 ( Nguyen et al . , 2015 ) phylogeny with the LG+I+G4+F model and 1000 UFBoot2 supports . Of the original 339 single protein sequence alignments ( see Supplementary file 1 in Derelle et al . , 2016 ) , 43 were excluded in the process: 10 were absent in apicomplexans and dinoflagellates ( alignments #72 , #290 , #291 , #292 , #294 , #295 , #300 , #301 , #304 and #315 ) and 33 were sparsely sampled or strongly incongruent with known organismal relationships ( alignments #29 , #57 , #70 , #73 , #82 , #149 , #151 , #155 , #157 , #159 , #163 , #164 , #167 , #171 , #183 , #184 , #191 , #193 , #194 , #198 , #241 , #246 , #260 , #265 , #270 , #283 , #285 , #289 , #302 , #303 , #305 , #330 , #331 ) . In the remaining 296 alignments , paralogs were reduced to the most slowly evolving sequence with non-conflicting phylogenetic placement ( else both were removed ) and multiple gene isoforms were reduced to the most complete sequence . Cross-contaminant sequences were identified based on read coverage and excluded . Divergent regions of sequences resulting from frame shift errors or imperfect gene models were removed from alignments manually . Adjacent protein sequence fragments that were apparently derived from the same gene fragmented by wrong genome annotation were merged . The final 296 protein sequence datasets were realigned by the localpair algorithm in MAFFT , trimmed by using the -b 4 g 0 . 4 settings in BMGE and concatenated in Scafos v1 . 25 ( Roure et al . , 2007 ) . During the latter step , sequences derived from different strains of the same species ( in Colpodella angusta , Sarcocystis neurona , Voromonas pontica , and unidentified actinocephaloid parasites of Helicoverpa armigera and Helicoverpa assulta ) were merged into single operational taxonomic units ( Supplementary file 2 ) . The initial phylogenetic matrix contained 54 species , 99948 sites and 14 . 4% missing data ( Figure 1—figure supplement 1A ) . To further reduce missing data we merged sequences of two Oxyrrhis marina strains and of two distinct variants found in Siedleckia nematoides transcriptomes; the latter are possibly derived from different cryptic species ( variant one was preferentially identified and/or found to be more complete in the transcriptome of the WS1 strain; variant two in the transcriptome of the WS2 strain ) . We also merged sequences of three representatives of the superfamily Actinocephaloidea with low sequence presence: Ascogregarina and two unidentified parasites of the insects Teleopsis and Helicoverpa , which contaminate the host transcriptomes ( Borner and Burmester , 2017 ) . The 18S rDNA of the Teleopsis parasite branches among other actinocephaloids ( data not shown ) and both unidentified group with Ascogregarina in the multiprotein phylogeny ( Figure 1—figure supplement 1A ) . Merging all these taxa produced the main phylogenetic matrix with 50 species , 99908 positions and 10 . 6% missing data ( Figure 1—source data 1 ) . Three additional matrices were created by excluding all sequences of Gregarina , Cephaloidophora , or both species ( Figure 1—figure supplement 1B ) . Final matrices were first analyzed in IQ-TREE by using the LG+I+G4+F model and the best tree was used as a guide tree in a more thorough analysis with the LG+G4+F+C60+PMSF model: 1000 UFBoot2 replicates were computed for all datasets ( Figure 1A and Figure 1—figure supplement 1 ) and 100 non-parametric bootstrap were computed for the main matrix ( Figure 1A ) . Seven statistical tests of 105 tree topologies corresponding to all possible relationships between coccidiomorphs , cryptosporidians , eugregarines , archigregarines , and blastogregarines in Figure 1A were calculated in IQ-TREE v1 . 6 . 5 with the LG+I+G4+F model and 10000 replicates ( Supplementary file 3 ) . The main phylogenetic matrix ( Figure 1A ) was also analyzed by 10 independent PhyloBayes runs ( either standard version 4 . 1 c or MPI version 1 . 7b ) ( Lartillot et al . , 2009 ) with the GTR+CAT model and constant sites removed ( -dc setting ) . Each chain was run for 1000 cycles , of which the initial 250 were discarded and the remaining ( 10 × 750 ) were combined to compute a consensus tree ( maxdiff = 0 . 23 , meandiff = 0 . 00057 ) . We next searched the orthologs of apicomplexan plastid proteins with focus on core pathways ( Figure 2A ) in transcriptomes and genomes of apicomplexans and their relatives ( including Perkinsus and dinoflagellates ) by using similar approach as for the nuclear genes above . Five best BLASTP hits at the e-value threshold of 1e-5 were retrieved for each species by using a comparatively slow evolving sequence query ( typically Vitrella or Chromera ) . These hits were included in plastid protein alignments created previously ( Janouškovec et al . , 2017; Janouškovec et al . , 2015 ) , or in new alignments created here by retrieving representative outgroup sequences from GenBank and the local database by BLASTP searches with the same query sequences ( Supplementary file 4 ) . This process included non-plastidial genes of the heme biosynthesis pathway , and three genes representing controls for non-plastidial pathways: iscS ( mitochondrial iron-sulfur biosynthesis ) , enoyl reductase domain ( FASI , cytosolic fatty acid synthesis ) , and ELO ( endoplasmic reticulum-localized fatty acid elongation ) . Alignments were reduced by an iteration process analogous to that used for nuclear genes . Final phylogenetic matrices were prepared by localpair alignment in MAFFT and -b 4 g 0 . 4 trimming in BMGE , and analyzed in IQ-TREE by using the built-in ModelFinder to select the best model with the preset LG substitution matrix ( for example , LG+F+R7 model for TPT in Figure 2—figure supplement 2 ) . This approach allowed to distinguish true apicomplexan sequences from contaminants in our transcriptomes and confirmed their origins to be in the plastid endosymbiont ( grouping with other plastidial sequences ) or eukaryotic host ( grouping with eukaryotic cytosolic or mitochondrial sequences; Figure 2A ) . Protein phylogenies of six horizontally acquired genes ( Figure 4A , B , Figure 4—figure supplement 2 and Figure 4—figure supplement 3 ) were built by expanding previous datasets ( Janouškovec et al . , 2015 ) and computed by using the same method as plastid phylogenies above . Proteins of core plastidial pathways ( Figure 2A ) derived from our transcriptomes and those of Plasmodium falciparum , Babesia microti and Toxoplasma gondii were scanned for the presence of identifiable N-terminal signal peptides in SignalP v4 . 1 ( Petersen et al . , 2011 ) . All methionines downstream of the predicted protein start were also tested for targeting signals . Positive sequences were checked for the presence of N-terminal extensions – truncated proteins were filtered out as false positives . The remaining signal peptide-positive proteins were recorded in Figure 2A and further checked for identifiable transit peptides in ChloroP 1 . 1 ( Emanuelsson et al . , 1999 ) . The SignalP and ChloroP statistics were listed in Supplementary file 5 . Apicomplexan proteins that have known experimental localization ( as described in primary literature and Apiloc3 ) or that were classified as high-confidence plastid-localized proteins in Plasmodium falciparum by BioID ( Boucher et al . , 2018 ) , were recorded in Figure 2A and listed in Supplementary file 4 . To characterize transit peptides in Digyalum oweni , we expanded its identified plastid protein set . Additional plastidial proteins in the WS1+two isolates were searched by BLASTP with apicomplexan queries , comprising known Toxoplasma and Plasmodium plastid proteins primarily involved in transcript and protein processing ( e . g . , stromal processing peptidase , clpP chaperone , histone-like protein HU , tRNA-Met ligase , etc . ) . Matched sequences were verified by reverse BLASTP searches on KEGG and NCBI BLAST websites ( aided by distance tree phylogenies at the latter site ) . Pooling positive hits with those identified previously ( Figure 2A ) and filtering for sequences with N-terminal extension carrying a signal peptide ( as identified by SignalP v4 . 1 ) produced 41 plastid proteins . The first 14 transit peptide residues downstream of the signal peptide cleavage site ( an approach to allow comparison with results in Patron and Waller , 2007 were merged and analyzed altogether for their amino acid composition by ‘Protein Stats’ script at the Sequence Manipulation Suite website . Composition of the 41 mature proteins ( without transit peptides ) was also analyzed – because transit peptide cleavage sites are difficult to predict in silico the initial 50 residues were removed arbitrarily ( Figure 2—figure supplement 1A ) . The amino acid frequency at the first 20 positions across the 41 Digyalum transit peptides was analyzed at the WebLogo3 website ( Figure 2—figure supplement 1B ) . Plastids transcripts were identified by three search strategies: BLASTN searches with plastidial 16S rDNA; BLASTP searches with sequences of plastid-encoded protein sequences of apicomplexans or chrompodellids;and searches for high AT contigs ( typically contigs > 70% AT and 1–3 kb in length ) . Hits from the former search were examined in a 16S rDNA tree . Hits those from the latter two search strategies were combined and reversely compared by BLASTX against the NCBI nr database to search if they match genes in apicomplexan , chrompodellids and algal plastid genomes . Positive hits were limited to representative , non-redundant contigs of 1 kb or longer ( Supplementary file 7 ) . Identification of individual genes was based on NCBI BLAST searches and tRNAscan-SE , and it was further aided by Artemis 17 . 0 . 1 and MFannot website . AT content was examined in 16S rDNAs and in whole plastid genomes or combined plastid transcripts by the ‘DNA Stats’ script on the Sequence Manipulation Suite website ( Figure 3B and Supplementary file 7 ) . The phylogeny of 16S rDNA was based on an earlier dataset with representatives of ARL-I to ARL-VIII clades ( Janouškovec et al . , 2012 ) . We also requested sequences of ARL-X and ARL-XI as reported in Mathur et al . ( 2018 ) from the authors of the study . Of the 34 sequences of clustered centroids we received , only one centroid was named ‘ARL-XI’ and this was a Pelagibacter-like bacterial contaminant; we therefore consider ARL-XI to be invalid . The remaining 33 centroids were named ‘ARL-X’: nine of them were bacterial contaminants and the remaining formed two non-overlapping groups of 3 and 21 sequences . We included in our phylogeny all three sequences from the first group and four slowly evolving sequences from the second group , but they all branched within ARL-I ( Figure 3A ) , which leads us to synonymize ARL-X with ARL-I . Finally , we included in the phylogeny 10 new ARLs identified by phylogenetic sorting of the VAMPS database ( Huse et al . , 2014 ) . Briefly , we obtained all VAMPS reads annotated as ‘Organelle’ or ‘Unknown’ , selected those being 350 bp or longer , and clustered each group in Usearch at 97% identity . Centroids were individually classified by maximum likelihood phylogenies in a dataset containing plastids , bacteria and mitochondria by an approach used previously ( Janouškovec et al . , 2012 ) . The initial round of phylogenies was computed by using Fasttree2 ( -lg -gamma setting ) and later rounds by using IQ-TREE ( LG+I+G4+F model; additional rounds with modified taxon sampling were used for sequences that were difficult to classify ) . Candidate ARL sequences were used to retrieve additional ARLs from centroid databases by BLASTN searches ( i . e . , those that had been misplaced by Fasttree2 ) . The apicomplexan affiliation of all ARL sequences was verified in the Figure 1A dataset . Representative sequences for all known and novel ARLs were listed in Supplementary file 6 . The final 16S rDNA phylogeny was based on a localpair MAFFT alignment trimmed in BMGE ( -h 0 . 4 g 0 . 65 settings ) and computed by using IQ-TREE with ModelFinder selection of the best fit model and 10000 UFBoot2 supports ( Figure 3A ) .	Microscopic parasites known collectively as apicomplexans are responsible for several infectious diseases in humans including malaria and toxoplasmosis . The cells of the malaria parasite and many other apicomplexans contain compartments known as cryptic chloroplasts that produce molecules the parasites need to survive . Cryptic chloroplasts are similar to the chloroplasts found in plant cells , but unlike plants the compartments in apicomplexans are unable to harvest energy from sunlight . Since the cells of humans and other animals do not contain chloroplasts , cryptic chloroplasts are a potential target for new drugs to treat diseases caused by apicomplexans . However , it remains unclear how widespread cryptic chloroplasts are in these parasites , largely because few apicomplexans have been successfully grown in the laboratory . To address this question , Janouškovec et al . used an approach called single-cell transcriptomics to study ten different apicomplexans . This provided new data about the genetic make-up of each parasite that the team analysed to find out how they are related to one another . The analysis revealed that , unexpectedly , apicomplexan parasites do not share a close common ancestor and are therefore not a natural grouping from an evolutionary perspective . Instead , their similar physical appearances and lifestyles evolved independently on at least three separate occasions . Further analysis demonstrated that cryptic chloroplasts are common in apicomplexan parasites , including in lineages where they were not previously known to exist . However , at least three lineages of apicomplexans have independently lost their cryptic chloroplasts . The findings of Janouškovec et al . shed new light on the importance of chloroplasts in the evolution of life and may help develop new treatments for diseases caused by apicomplexan parasites . Several drugs targeting the cryptic chloroplasts in malaria parasites are currently in clinical trials , and this work suggests that these drugs may also have the potential to be used against other apicomplexan parasites in the future .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "microbiology", "and", "infectious", "disease" ]	2019	Apicomplexan-like parasites are polyphyletic and widely but selectively dependent on cryptic plastid organelles
The target of rapamycin complex I ( TORC1 ) regulates cell growth and metabolism in eukaryotes . Previous studies have shown that nitrogen and amino acid signals activate TORC1 via the small GTPases , Gtr1/2 . However , little is known about the way that other nutrient signals are transmitted to TORC1 . Here we report that glucose starvation triggers disassembly of TORC1 , and movement of the key TORC1 component Kog1/Raptor to a single body near the edge of the vacuole . These events are driven by Snf1/AMPK-dependent phosphorylation of Kog1 at Ser 491/494 and two nearby prion-like motifs . Kog1-bodies then serve to increase the threshold for TORC1 activation in cells that have been starved for a significant period of time . Together , our data show that Kog1-bodies create hysteresis ( memory ) in the TORC1 pathway and help ensure that cells remain committed to a quiescent state under suboptimal conditions . We suggest that other protein bodies formed in starvation conditions have a similar function . The Target Of Rapamycin Complex I ( TORC1 ) is made up of three essential proteins , called mTOR , Raptor and mLST8 in humans and Tor1 , Kog1 and Lst8 in yeast ( Kim et al . , 2002; Loewith et al . , 2002 ) . Biochemical and structural studies show that these proteins form a stable ring structure , containing two copies of each subunit ( Adami et al . , 2007; Yip et al . , 2010 ) . Kog1/Raptor is known to recruit substrates to the TOR kinase ( Tor1 ) and is required for regulation of Tor1 activity ( Hara et al . , 2002; Kim et al . , 2002 ) . Lst8 , on the other hand , binds directly to the kinase domain in Tor1 ( Yip et al . , 2010 ) and may help stabilize the interaction between Kog1 and Tor1 ( Kim et al . , 2002 ) . In yeast , TORC1 also includes the non-essential ( and poorly characterized ) protein Tco89 ( Reinke et al . , 2004 ) . In the presence of pro-growth hormones and the appropriate nutrients , TORC1 is active and drives protein , lipid , and nucleotide synthesis by phosphorylating a wide range of proteins ( Bodenmiller et al . , 2010; Hsu et al . , 2011; Laplante and Sabatini , 2012; Loewith and Hall , 2011; Robitaille et al . , 2013 ) . In contrast , when hormone , nutrient or energy levels drop below a critical level , TORC1 is inhibited , causing the cell to switch from anabolic to catabolic metabolism and eventually enter a quiescent state ( Barbet et al . , 1996; Duvel et al . , 2010; Loewith and Hall , 2011 ) . In higher eukaryotes , hormone signals are transmitted to TORC1 through the Phosphatidylinositol 3-kinase to AKT , which in turn phosphorylates the tuberous sclerosis complex ( Inoki et al . , 2002; Manning et al . , 2002 ) . The tuberous sclerosis complex ( TSC1 , TSC2 and TBC1D7 ) then dissociates from the lysosomal surface , triggering activation of the small GTPase , Rheb , and subsequently TORC1 ( Dibble and Manning , 2013; Menon et al . , 2014 ) . However , these pro-growth signals are blocked in energy starvation conditions due to stimulation of the AMP activated protein kinase ( AMPK ) , which phosphorylates and hyperactivates TSC1-TSC2 to inhibit Rheb and TORC1 ( Inoki et al . , 2003a; Inoki et al . , 2003b ) . AMPK also inhibits TORC1 by phosphorylating Kog1/Raptor at two sites ( Gwinn et al . , 2008 ) . One of these sites is highly conserved in eukaryotes ( Ser 959 in S . cerevisiae ) but does not appear to influence TORC1 activity in budding yeast ( Kawai et al . , 2011 ) . In both higher and lower eukaryotes , amino acid/nitrogen signals are transmitted to TORC1 through another class of small GTPases , known as Gtr1/2 in yeast and the Rags in humans ( Binda et al . , 2009; Kim et al . , 2008; Sancak et al . , 2008 ) . Gtr1/2 are tethered to the vacuolar/lysosomal membrane via the EGO complex ( Ragulator in humans ) and the vacuolar ATPase , where they respond to amino acid signals by binding and activating TORC1 ( Bar-Peled et al . , 2012; Binda et al . , 2009; Panchaud et al . , 2013; Sancak et al . , 2010; Zoncu et al . , 2011 ) . Outside of nitrogen/amino acid starvation conditions , little is known about the way nutrient and stress signals trigger changes in TORC1 activity , particularly in organisms such as S . cerevisiae that are missing TSC1/2 and where there is no clear link between AMPK and TORC1 activity . For example , studies in yeast have shown that glucose starvation completely blocks TORC1 signaling , but only around 20% of this inhibition depends on Gtr1/2 and the other known TORC1 regulator in yeast , Rho1 ( Hughes Hallett et al . , 2014 ) . Here , to learn more about TORC1 regulation , we examine the localization of Tor1 and Kog1 in budding yeast , in glucose starvation and other conditions . Surprisingly , we find that glucose starvation leads to disassembly of the TOR complex and movement of Kog1 from the vacuolar membrane to a single body on the edge of the vacuole , while glucose repletion leads to a reversal of this process . Following up on these observations we show that the AMPK , Snf1 , drives Kog1 into bodies by phosphorylating , or triggering the phosphorylation of , Kog1 at Ser 491 and 494 . Furthermore , we find that these phosphorylation sites are located in one of two , glutamine-rich prion-like motifs in Kog1 , and that these motifs are required for the formation of Kog1-bodies . Finally , by measuring phosphorylation of the key TORC1 substrate , Sch9 , in a variety of mutants , we show that Kog1 agglomeration increases the threshold for TORC1 activation . Taken together , our results reveal a novel mechanism of TORC1 regulation and show that protein body formation can block the activation of a pathway by weak stimuli . The numerous other protein bodies formed in yeast , and other organisms , may have a similar function—creating hysteresis ( memory ) in key signaling and metabolic pathways to ensure that once cells commit to a stress or starvation state , they only exit when conditions are optimal . To determine if TORC1 localization is altered in glucose starvation conditions , we subjected yeast carrying Kog1-YFP , Tco89-YFP ( see Material and methods ) , or Tor1 with a triple GFP insertion ( Binda et al . , 2009; Sturgill et al . , 2008 ) to acute glucose starvation and acquired 3D images using a fluorescence microscope . Surprisingly , we found that Kog1 and Tco89 move from their known location on the vacuolar membrane ( marked with Vph1-mCherry ) , to a single body near the edge of the vacuole , within 60 min in most cells ( Figure 1A and Figure 1—figure supplement 1 ) . In contrast , Tor1 remained associated with the vacuolar membrane and/or moved to the cytoplasm during the same time-period ( Figure 1B ) . 10 . 7554/eLife . 09181 . 003Figure 1 . Kog1-YFP moves from the vacuolar membrane to a single body during glucose and nitrogen starvation . ( A ) Localization of Kog1-YFP and Vph1-mCherry , before ( SD ) and after ( -Glu ) glucose withdrawal ( 60 min ) . ( B ) Localization of Tor1-3xGFP and Vph1-mCherry before ( SD ) and after ( -Glu ) glucose withdrawal ( 60 min ) . ( C ) Time-course data showing the fraction of cells that contain Kog1-bodies after transfer to synthetic medium with 2% glucose ( SD ) , SD + 0 . 4 M KCl ( + KCl ) , SD -glucose ( -Glu ) , SD-glucose and nitrogen ( -Glu + N ) , and SD-nitrogen ( -N ) medium . Each time-point shows the average and standard deviation from three independent experiments ( performed on different days with >200 cells per time-point , per replicate ) . Solid lines show the best fit to a single exponential equation ( -Glu , -N , and –Glu + N ) or a line ( SD and + KCl ) . ( D ) Time-course data showing the fraction of cells that contain Kog1-bodies after adding glucose , or glucose and cycloheximide ( 0 . 02% Glu , 0 . 1% Glu , 2% Glu or 2% Glu + 100 ug/ml cycloheximide ) back to cells that have been in SD–glucose for 60 min . Each datapoint in 0 . 1% glucose and 2% + 100 ug/ml cycloheximide shows the average and standard deviation from three independent experiments ( performed on different days with >200 cells per time-point , per replicate ) . The solid lines show the best fit to a single exponential equation . DOI: http://dx . doi . org/10 . 7554/eLife . 09181 . 00310 . 7554/eLife . 09181 . 004Figure 1—figure supplement 1 . The TORC1 component Tco89 associates with Kog1-bodies . ( A ) Localization of Tco89-YFP in wild type and Kog1 PrDM1 + 2 backgrounds before ( SD ) and after ( -Glu ) glucose withdrawal ( 60 min ) . ( B ) Time-course data showing the fraction of cells with Tco89 foci after transfer to synthetic medium with 2% glucose ( SD ) and SD -glucose ( -Glu ) . Each time-point shows the average and standard deviation from three independent experiments ( performed on different days with >200 cells per time-point , per replicate ) . Solid lines show the best fit to a single exponential equation ( -Glu ) or a line ( SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09181 . 00410 . 7554/eLife . 09181 . 005Figure 1—figure supplement 2 . Kog1-body formation is pervasive during extended periods of nutrient starvation . Mean and standard deviation of two independent experiments is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 09181 . 00510 . 7554/eLife . 09181 . 006Figure 1—figure supplement 3 . Kog1 forms one body per cell . Frequency distribution of Kog1-YFP foci , among cells containing bodies , after 60 min glucose starvation . DOI: http://dx . doi . org/10 . 7554/eLife . 09181 . 00610 . 7554/eLife . 09181 . 007Figure 1—figure supplement 4 . Kog1-body formation is pervasive following exit of exponential growth . Cultures of Kog1-YFP wildtype , Kog1S491/4A PrDm1 mutant , and PrDm1 + 2 mutant were grown for 24hrs to saturation in synthetic media containing 2% glucose ( SD ) . Mean and standard deviation of three independent experiments is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 09181 . 00710 . 7554/eLife . 09181 . 008Figure 1—figure supplement 5 . Kog1-YFP dissociates from perivacuolar bodies and relocalizes to the vacuolar membrane upon glucose repletion . Localization of Kog1-YFP in cells co-expressing the vacuolar membrane marker Vph1-mCherry after 60 min of glucose starvation followed by; 60 min of glucose starvation ( -Glu ) ; 60 min of glucose repletion and 100 ug/ml cycloheximide treatment ( SD + CHX ) ; 60 min of glucose starvation and 100 ug/ml cycloheximide treatment ( -Glu + CHX ) . Addition of cycloheximide alone , a TORC1 activator , does not restore Kog1 localization to the vacuolar membrane indicating glucose specific signals are required for the dissociation of glucose starvation induced Kog1 bodies ( -Glu + CHX ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09181 . 008 To study Kog1-body formation in more detail , we next followed Kog1-YFP localization in cells transferred from rich medium ( SD ) to glucose starvation , nitrogen starvation , glucose + nitrogen starvation , and osmotic stress conditions . In glucose starvation and nitrogen + glucose starvation we saw rapid formation of Kog1-bodies ( τ = 11 . 0 ± 2 . 8 and 13 . 6 ± 2 . 4 min ) in 57 ± 5% and 73 ± 4% of cells , respectively ( Figure 1C ) . Kog1-bodies also formed in nitrogen starvation conditions , albeit at a much slower rate ( τ = 153 ± 25 min; Figure 1C ) . However , we did not see a significant increase in Kog1-body formation in osmotic stress or when cells were simply transferred into fresh SD medium ( Figure 1C ) . In separate experiments , we also followed Kog1-body formation during long periods of glucose and/or nitrogen starvation ( 2-–6 hrs ) . In all cases , we saw slow accumulation of Kog1-bodies , beyond the levels found at one hour , until 80-85% of cells had a least one , but on occasion 2 or 3 , Kog1 bodies ( Figure 1—figure supplement 2 , 3 ) . Cultures grown to saturation in SD medium also formed Kog1-bodies in ~∼85% of cells ( Figure 1—figure supplement 4 ) . Finally , to determine if Kog1-body formation is a reversible process , we exposed cells to glucose starvation conditions for 60 min and then followed Kog1-YFP localization after adding 0 . 02% , 0 . 1% , or 2% glucose back to the medium . In both 0 . 1% and 2% glucose , we saw Kog1 move out of the bodies and back onto the vacuolar membrane over an approximately one-hour time-period ( τ = 48 ± 8 and 45 ± 10 min , respectively; Figure 1D ) . This relocalization also occurred in the presence of cycloheximide , demonstrating that existing Kog1-bodies dissociate during glucose repletion ( Figure 1D and Figure 1—figure supplement 5 ) . Putting these data together we conclude that Kog1-body formation is a rapid ( τ = ∼10 min ) and reversible process , triggered by glucose and nitrogen starvation . Previous work has shown that TORC1 can localize to stress granules in both yeast and human cells ( Takahara and Maeda , 2012; Thedieck et al . , 2013; Wippich et al . , 2013 ) . To determine if Kog1 associates with stress granules ( or related structures known as P-bodies [Parker , 2012] ) in glucose starvation conditions , we transfected cells carrying Kog1-YFP with mCherry-tagged versions of the stress granule marker Pbp1 or the P-body marker Edc3 , and followed YFP and RFP localization . We found that stress granules start to form after 60 min in glucose starvation conditions , and become pervasive in cells grown to saturation in SD medium , but only co-localize with Kog1-bodies in 2 . 6% of cells ( 0% co-localization after 60 min , n = 94; 2 . 6% co-localization after 24 hrs , n = 190; Figure 2A ) . Similarly , P-bodies form in >80% cells after 60 min of glucose starvation , but only co-localize with Kog1-bodies in 3 . 4% of cells ( n = 404; Figure 2B ) . Therefore , it appears that the Kog1-bodies formed in starvation conditions are distinct from both stress granules and P-bodies . 10 . 7554/eLife . 09181 . 009Figure 2 . Kog1-YFP does not associate with stress granules or P-bodies in glucose starvation conditions . ( A ) Localization of Kog1-YFP after 60 min or 24 hrs of glucose starvation in cells expressing the stress granule marker Pbp1-mCherry . ( B ) Localization of Kog1-YFP after 60 min of glucose starvation in cells expressing the P-Body marker Edc3-mCherry . The images in ( A ) were deconvolved using Deltavision software to ensure we could distinguish Pbp1 granules from cytoplasmic Pbp1 . DOI: http://dx . doi . org/10 . 7554/eLife . 09181 . 009 Once we determined that Kog1-body formation occurs independently of stress granule and P-body formation , we sought to identify the signal that triggers Kog1 agglomeration in glucose starvation conditions . In a previous study we showed that the AMP activated protein kinase , Snf1 , is required for TORC1 pathway inhibition during glucose starvation ( Hughes Hallett et al . , 2014 ) . We therefore wondered if Snf1 regulates TORC1 , at least in part , by driving Kog1 agglomeration . To test this hypothesis we monitored Kog1-YFP localization in snf1Δ cells . These experiments revealed that deletion of Snf1 causes a dramatic , 20-fold , increase in the time-constant for Kog1-body formation ( τ = 219 ± 21 min in snf1Δ cells; Figure 3A ) . 10 . 7554/eLife . 09181 . 010Figure 3 . Snf1 Dependent Phosphorylation of Kog1 drives Kog1-body formation . ( A-––B ) Time-course data showing the fraction of ( A ) snf1Δ and ( B ) Kog1S491A/S494A cells that contain Kog1-bodies in SD medium ( 2% Glu ) and at various time-points after glucose withdrawal ( 0% Glu ) . The blue points show the average and standard deviation from three independent experiments ( performed on different days with >200 cells per time-point , per replicate ) , while the blue lines show the best fit to a single exponential equation ( 0% Glu ) or a line ( 2% Glu ) . The green and orange lines show the best fit to the wild-type data ( from Figure 1C ) for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 09181 . 01010 . 7554/eLife . 09181 . 011Figure 3—figure supplement 1 . Mass spectrometry generated peptide map for Kog1 . Trypsin digested and trypsin-thermolysin digested Kog1 purified from glucose staved cells were subjected to mass spectrometry-based peptide mapping . ( A ) Amino acid sequence colored in green have a FDR of 1% or less , sequence colored in yellow have a FDR of 5% or less , and uncolored sequence were not identified in the experiment . Labels of “‘C’ above the amino acid sequence show locations of carbamidomethyl modification ( a result of chemical alkylation directed at cysteines during sample preparation ) , ‘O’ above the amino acid sequence show locations of oxidation ( spontaneous oxidation of methionine ) , and ‘P’ above the amino acid sequence show locations of phosphorylation . The green coloring of the ‘P’ labels represents the site localization probability for phosphorylation at the site , as determined by PhosphoRS , as greater than 99% . ( B ) Filtered protein identification results indicating amino acids S491 , T492 , S494 and S1045 of Kog1 are potentially phosphorylated . Filtering was performed to a 5% false discovery rate . DOI: http://dx . doi . org/10 . 7554/eLife . 09181 . 01110 . 7554/eLife . 09181 . 012Figure 3—figure supplement 2 . Protein identification result identifying S491 and S494 as sites of phosphorylation . Liquid chromatography coupled to tandem mass spectrometry ( LC-MS/MS ) was used to analyze proteolytic digestion products of immunoprecipitation-purified Kog1 from glucose staved cells . The resulting mass spectrometry data were analyzed by the protein database search algorithm SEQUEST-HT to identify mass spectra to peptides . ( A ) Kog1 peptide 485-FAVANL ( pS ) TM ( pS ) LVNNPALQSR-504 showing all theoretical fragment ions for a triply-charged ion species for the peptide . Fragment masses in red and blue indicate b-series ions and y-series ions , respectively , detected in the tandem mass spectrum ( shown in panel B ) . ( B ) Tandem mass spectrum identifying Kog1 peptide 485-FAVANL ( pS ) TM ( pS ) LVNNPALQSR-504 . Peaks in red and blue indicate b-series and y-series ions that were matched to the theoretical peptide fragment masses in panel A . DOI: http://dx . doi . org/10 . 7554/eLife . 09181 . 01210 . 7554/eLife . 09181 . 013Figure 3—figure supplement 3 . Strains carrying Kog1S491D/S494D and Kog1S491E/S494E at the native loci are inviable . To create the Kog1 variants Kog1S491A/S494A , Kog1S491D/S494D , and Kog1S491E/S494E we cloned Kog1 ( with Ser 491 and 494 mutated to Ala , Asp , or Glu ) into a plasmid carrying URA3 under an ADH1 promoter ( pRS306 ) . The plasmids were then cut using PflF1 , and transformed into a wild type train ( ACY044 ) . In all cases this led to insertion of the plasmid at the Kog1 locus , creating a wild-type copy of Kog1 , followed by the URA3 gene , followed by a mutant copy of Kog1 ( AA , DD , or EE; as confirmed by sequencing ) . We then isolated colonies that excised the plasmid by selection on 5FOA . In this final step , we expected to find colonies that retain the wild-type copy of Kog1 as well as colonies that retain the mutated copy of Kog1 ( depending on the site of recombination during the loop out event ) . In line with our expectations , we found 7/10 of the colonies created by looping in , and looping out , Kog1S491A/S494A had the AA mutation ( the other 3 were wild-type ) . In contrast , we found 0/20 colonies had the DD mutation and 0/12 strains had the EE mutation . Specifically , we can calculate the probability of finding 0/20 DD and 0/12 EE mutations by chance , given that we found 7/10 AA mutations , at p<0 . 0001 and p<0 . 001 , respectively , using Fisher’s exact test . The probability of finding 0/32 DD and EE mutants by chance , given that we found 7/10 AA mutants , is less than 1 x 10-–5 . This strongly suggests that strains carrying Kog1 with phosphomimetic substitutions at Ser 491 and 494 are inviable . DOI: http://dx . doi . org/10 . 7554/eLife . 09181 . 013 Next , to determine how Snf1 activates Kog1-body formation , we purified TORC1 from cells: ( 1 ) growing in SD medium , ( 2 ) exposed to glucose starvation conditions for 5 min , and ( 3 ) exposed to osmotic stress conditions for 5 min ( see Material and methods ) . We then used mass spectrometry to identify the phosphorylation sites on Tor1 and Kog1 in each condition . These data showed that Kog1 is phosphorylated on Ser 491 and 494 in glucose starvation conditions , but not in osmotic stress or SD medium ( Figure 3—figure supplements 1 , 2 ) . A short time later , Young and coworkers showed that Ser 491 and 494 on Kog1 are Snf1-dependent phosphorylation sites ( Braun et al . , 2014 ) . Therefore , to test if Snf1 drives Kog1-body formation by promoting phosphorylation of Ser 491 and 494 , we constructed a strain carrying Kog1S491A/S494A-YFP at the native locus and monitored Kog1 localization in glucose starvation conditions . We found that Kog1S491A/S494A forms bodies slowly ( ττ = 309 ± 48 min; Figure 3B ) , and at a rate similar to that found in the snf1Δ strain ( τ = 219 ± 21 min; Figure 3A ) . Thus , Snf1 increases the rate of Kog1 body formation by phosphorylating , or triggering phosphorylation , of Kog1 in glucose starvation conditions . Kog1S491A/S494A cells also showed a reduced level of Kog1-body formation in SD medium ( 2 . 3 ± 0 . 1% bodies in Kog1S491A/S494A cells versus 8 . 6 ± 2 . 8% in wild type cells; Figure 3B ) . This was not seen in the snf1Δ strain ( 8 . 3 ± 0 . 8% bodies in SD; Figure 3A ) , suggesting that either an additional kinase phosphorylates Kog1 at positions 491 and 494 in log growth conditions or that mutation of Ser 491 and 494 to alanine also inhibits Kog1 agglomeration by altering the physical properties of Kog1 . Finally , to test the impact that constitutive phosphorylation of Kog1 has on Kog1-body formation; we attempted to create strains carrying the phosphomimetic variants Kog1S491D/S494D and Kog1S491E/S494E . However , these mutants were not viable , indicating that Kog1-body formation ( or Kog1 phosphorylation ) inhibits cell growth and/or division ( Figure 3—figure supplement 3 ) . Protein body formation can be driven by called prion-like motifs or PriLMs ( Alberti et al . , 2009; Decker et al . , 2007; Gilks et al . , 2004; Han et al . , 2012 ) . These motifs tend to have long stretches of glutamine and/or asparagine residues and a low number of hydrophobic and/or charged residues ( Alberti et al . , 2009 ) . In examining the Kog1 sequence we identified two such motifs , both containing long stretches of glutamine ( Figure 4A ) . These regions were also identified as PriLMs using a hidden Markov Model trained on known prion sequences ( Alberti et al . , 2009 ) . Interestingly , the first of these PriLMs includes the Snf1 dependent phosphorylation sites at Ser 491 and 494 , while the second , smaller , PriLM is located approximately 300 amino acids away ( Figure 4A ) . 10 . 7554/eLife . 09181 . 014Figure 4 . Kog1-body Formation Depends on Prion-Like Domains in Kog1 . ( A ) Sequence of two prion-like motifs in Kog1 ( red letters ) as defined by Alberti and coworkers ( Alberti et al . , 2009 ) . The Snf1 dependent phosphorylation sites at Ser 491 and Ser 494 are shown in blue . ( B ) Time-course data showing the fraction of cells carrying Prion Domain mutation 1 ( PrDm1 ) , Prion Domain mutation 2 ( PrDm2 ) , or both mutations ( PrDm1 + 2 ) , that have Kog1-bodies in SD medium ( 0% Glu ) , and after glucose withdrawal ( 2% Glu ) . The blue points show the average and standard deviation from three independent experiments ( performed on different days with >200 cells per time-point , per replicate ) while the blue lines show the best fit of each dataset to a single exponential equation ( 0% Glu for PrDm1 and PrDm2 ) or a line . The green and orange lines show the best fit to the wild-type data ( from Figure 1C ) for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 09181 . 014 To determine if the PriLMs in Kog1 are involved in Kog1-body formation we mutated a stretch of glutamines in each motif , to form a stretch of alanines ( Figure 4A ) , and measured the impact on Kog1-YFP localization . Disruption of prion-like motif 1 ( PrDm1 ) caused a >2-fold decrease Kog1-body formation , both in SD medium ( 2 . 76 ± 0 . 6% in PrDm1 versus 8 . 6 ± 2 . 8% in wt; Figure 4B ) and in glucose starvation conditions ( 26 ± 2 . 7% in PrDm1 versus 57 ± 5% in wt; Figure 4B ) . Disruption of prion-like motif 2 ( PrDm2 ) also caused a ( small ) decrease Kog1-body formation in glucose starvation conditions ( 45 ± 4% in PrDm2 versus 57 ± 5% bodies in wt; Figure 4B ) . More remarkably , however , disruption of PriLM1 and PriLM2 completely blocked Kog1-body formation in both SD medium and glucose starvation conditions ( 1 . 0 ± 2 . 1% cells with bodies; Figure 4B ) . Thus , Kog1-body formation depends on the two prion-like motifs in Kog1 . During log growth , TORC1 phosphorylates and activates the S6 kinase , Sch9 , to drive protein and ribosome synthesis ( Bodenmiller et al . , 2010; Huber et al . , 2011; Urban et al . , 2007 ) . This key signaling event is readily measured using a band-shift assay developed by Loewith and coworkers ( Urban et al . , 2007 ) . Using this assay , we asked if and how Kog1-bodies influence TORC1 signaling by following Sch9 phosphorylation in five strains that have a defect in Kog1-body formation ( Kog1S491A/S494A , snf1Δ , PrDm1 , PrDm2 , PrDm1 + 2 ) . We found that Sch9 is phosphorylated at or near wild-type levels during log phase growth in all of our mutant strains ( Figures 5A , B ) . Sch9 is then rapidly ( ≤2 . 5 min ) and completely dephosphorylated during glucose starvation in the wild-type , Kog1S491A/S494A , PrDm1 , PrDm2 and PrDm1 + 2 strains ( Figures 5A , B ) . Thus , Kog1-body formation does not appear to play a significant role in TORC1-Sch9 signaling during acute glucose starvation since ( 1 ) Sch9 dephosphorylation occurs much faster than the accumulation of Kog1-bodies ( ττ = <2 . 5 min versus 11 ± 3 min ) and ( 2 ) most of the mutants that have a defect in Kog1-body formation behave like the wild-type strain in the Sch9 phosphorylation assay . The only exception is the snf1Δ strain , where we see a clear defect in Sch9 dephosphorylation after 5 and 10 min of glucose starvation ( Figure 5A ) . However , it is very unlikely that this defect is due to inhibition of Kog1-body formation , since the Kog1S491A/S494A strain does not have a defect in the Sch9 assay ( Figure 5A ) . Instead , it appears that Snf1 regulates the TORC1 pathway by triggering Kog1 phosphorylation at Ser 491 and 494 , and through an additional , unknown , mechanism . 10 . 7554/eLife . 09181 . 015Figure 5 . Impact of Kog1-bodies on TORC1 Signaling . ( A ) Bandshift assays measuring Sch9 phosphorylation in SD medium ( 0 min; 2% glucose ) , at various time-points after glucose withdrawal ( 2 . 5-–60min; 0% glucose ) , and then at various time-points after adding 2% glucose back to the cultures ( 1 , 5 , 10 , and 20 min; 2% glucose ) . ( B ) Quantitation of the band-shift data from ( A ) . The data are normalized to set the wild-type value to 1 . 0 in 2% glucose , and 0 . 0 in 0% glucose ( see Methods ) . ( C-–D ) Sch9 phosphorylation data for wild-type , snf1Δ , Kog1S491A/S494A , and PrDm1 + 2 cells , grown in SD medium ( 2% glucose ) , after transfer into synthetic medium - glucose for 60 min ( 0% glucose ) , and then at various time-points after adding 0 . 02% ( C ) or 0 . 1% glucose ( D ) back to the culture . Quantitation in ( C ) and ( D ) was performed as described in ( B ) but here the graphs show the average and standard deviation from three separate experiments . Data for PrDm1 and PrDm2 are shown in Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 09181 . 01510 . 7554/eLife . 09181 . 016Figure 5—figure supplement 1 . Impact of Kog1-bodies on TORC1 signaling . Sch9 phosphorylation data for PrDm1 and PrDm2 cells grown in synthetic medium containing 2% glucose ( SD; 0min ) , after transfer into synthetic medium lacking glucose ( -G; 60 min ) , and then at 1 , 5 , 10 , and 20 min time-points after adding back 0 . 02% ( A ) or 0 . 1% ( B ) glucose . Quantitation of bandshift data , as described in Figure 5 , showing the mean and standard deviation from three separate experiments is displayed in the histograms . DOI: http://dx . doi . org/10 . 7554/eLife . 09181 . 01610 . 7554/eLife . 09181 . 017Figure 5—figure supplement 2 . Sch9 reactivation depends upon TORC1 activity . Sch9 phosphorylation data for wt and PrDm1 + 2 cells grown in synthetic medium containing 2% glucose ( SD; 0min ) , after transfer into synthetic medium lacking glucose ( -G; 60 min ) , and then at 1 , 5 , 10 , and 20 min time-points after adding back 0 . 02% ( A ) or 0 . 1% ( B ) glucose in the presence of rapamycin ( 200 ng/ml ) . Quantitation of bandshift data , as described in Figure 5 is displayed in the histograms . DOI: http://dx . doi . org/10 . 7554/eLife . 09181 . 017 After discovering that TORC1/Sch9 inhibition precedes Kog1-body formation , we reasoned that Kog1 agglomeration might act as a slow step to lock TORC1 in an inactive state . To test this hypothesis we subjected wild-type and mutant strains to acute glucose starvation for 60 min , added 2% glucose back to the medium , and then followed Sch9 phosphorylation ( Figure 5A ) . To our surprise , we saw complete , or near complete , Sch9 phosphorylation in all of our strains ( Figures 5A , B ) . However , when we repeated this experiment , but only added 0 . 02% or 0 . 1% glucose back to the medium , we saw significantly more Sch9 phosphorylation in the mutant cells than in wild-type cells ( Figures 5C , D and Figure 5—figure supplement 1 and 2 ) . In fact , 0 . 1% glucose triggered ∼50% Sch9 phosphorylation in the wild-type strain compared with ∼100% Sch9 phosphorylation in our strongest mutant , PrDm1 + 2 ( Figure 5D ) . These results are remarkable since only 60% of wild-type cells have Kog1-bodies in our starting conditions ( Figure 1C ) and as a result , it should only be possible to see a 60% increase in Sch9 phosphorylation in the mutant strains ( an increase from 40% activity in wild-type cells to 100% activity in mutant cells ) . Therefore , putting all of our band-shift data together , we conclude that Kog1-body formation acts to increase the threshold for TORC1 activation in cells subjected to medium or long-term glucose starvation . In this study we show that glucose starvation leads to rapid inactivation of TORC1/Sch9 signaling ( τ<2 . 5 min , Figure 5 ) , dissociation of the Kog1-Tor1 complex , and the formation of Kog1-bodies ( τ = 11 ± 3 min ) . Kog1-bodies then act to increase the threshold for TORC1 activation--probably by limiting the number of intact TORC1 molecules in the cell . In other words , Kog1-body formation helps ensure that cells commit to a starvation or quiescent state , once they have been in glucose starvation conditions for a significant period of time . This is likely important for survival in natural conditions where yeast cells are exposed to a complex and fluctuating environment , and may need to remain in a quiescent state for years ( Loewith and Hall , 2011 ) . We also show that Kog1-body formation is driven by the AMPK , Snf1 . This kinase is inactive during log growth and then activated upon glucose starvation , where it triggers massive changes in transcription and metabolism ( Braun et al . , 2014; Hedbacker and Carlson , 2008; Usaite et al . , 2009 ) . As part of this program , Snf1 activates phosphorylation of Kog1 at Ser 491 and 494 . It is important to note , however , that Snf1-dependent phosphorylation of Kog1 is probably indirect since the sequence around Ser 491 and 494 ( AVANLSTMSLVN ) does not match known AMPK targets sites such as LRRVxSxxNL and MKKSxSxxDV ( Gwinn et al . , 2008 ) . Further work is therefore needed to identify the kinase that acts downstream of Snf1 to phosphorylate Kog1 . Snf1-dependent phosphorylation of Kog1 occurs in the middle of a glutamine rich , prion-like motif . We show that this prion-like motif and another similar motif , located 300 amino acids away , are essential for Kog1-body formation . Interestingly , the Snf1-dependent phosphorylation sites , and both prion-like domains found in S . cerevisae , are conserved in numerous other yeast species including the pathogen Candida glabrata ( Figure 6 ) . However , we could not find prion-like domains in Kog1 from Schizosaccharomyces pombe , Neurospora Crassa , and related species ( Figure 6 ) . Strikingly , most ( 5/6 ) of the yeast species missing prion-like domains in Kog1 have genes encoding the tuberous sclerosis complex ( TSC1 and TSC2 ) , while all ( 17/17 ) of the yeast species with prion-like domains in Kog1 are missing the tuberous sclerosis complex ( TSC1 or TSC1 and TSC2 genes; Table 1 ) . 10 . 7554/eLife . 09181 . 018Figure 6 . Partial alignment of Kog1 sequences from S . cerevisiae and 22 fungal orthologs . ( A ) PriLM1 and ( B ) PriLM2 ( shown in bold ) were defined using a hidden Markov Model trained on known yeast prion domains ( Alberti et al . , 2009 ) . The Glutamine ( Q ) and Asparagine ( N ) residues in each PriLM are highlighted in red , while the Snf1 targets sites ( Ser 491 and 494 ) are highlighted in blue . PriLMs 1 and 2 are both insertions between highly conserved domains in Kog1 . DOI: http://dx . doi . org/10 . 7554/eLife . 09181 . 01810 . 7554/eLife . 09181 . 019Figure 6—figure supplement 1 . Partial alignment of Kog1 from S . cerevisiae with orthologous genes in higher eukaryotes . ( A ) Elegans group nematodes show a loosely conserved expansion of Glutamine ( Q ) and Asparagine ( N ) residues in-between conserved domains of Kog1/Raptor . Q and N residues within this region are highlighted in red text . ( B ) A Glutamine and Asparagine rich domain in Kog1/Raptor is anti-correlated with the presence of Tsc1 and Tsc2 genes ( TSC complex ) in higher eukaryotes . The table shows the number , and ( percent ) of Q and N residues within a loosely conserved Q/N rich domain in Kog1/Raptor . Elegans group nematodes , lacking Tsc1 and Tsc2 genes , and the tuberous sclerosis complex ( TSC ) signaling pathway , carry a Q/N rich domain within the Kog1/Raptor ortholog daf-15 . Genebank references to Tsc1 and Tsc2 orthologs are provided . DOI: http://dx . doi . org/10 . 7554/eLife . 09181 . 01910 . 7554/eLife . 09181 . 020Table 1 . PriLM1 and PriLM2 are anti-correlated with the presence of Tsc1 and Tsc2 genes in fungal genomes . Quantitation , number and ( percentage ) , of Glutamine ( Q ) and Asparagine ( N ) residues within PriLM1 and PriLM2 domains of Kog1 and othologs ( Figure 6 ) reveals three clusters of hosts . ( 1 ) Species carrying Q/N rich PriLM1 and PriLM2 domains lack both Tsc1 and Tsc2 genes and the tuberous sclerosis complex ( TSC ) signaling pathway but show conservation of S491/494 phosphorylation motif . ( 2 ) Species carrying PriLM1 of intermediate Q/N richness lack Tsc1 but carry the Tsc2 gene . The influence on TSC pathway signaling is unclear . ( 3 ) Species lacking PriLM1 and PriLM2 carry both Tsc1 and Tsc2 genes and an operational TSC signaling pathway . Genbank references to Tsc1 and Tsc2 orthologs are provided . DOI: http://dx . doi . org/10 . 7554/eLife . 09181 . 020PriLM1PriLM2Q/N ( %QN ) Q/N ( %QN ) Tsc1Tsc2S491/494S . cerevisiae 28 ( 35 . 90 ) 20 ( 58 . 82 ) +S . kluyveri 49 ( 55 . 68 ) 17 ( 65 . 38 ) +S . paradoxus 29 ( 36 . 71 ) 20 ( 58 . 82 ) +S . mikatae 28 ( 36 . 36 ) 20 ( 58 . 82 ) +S . castellii 32 ( 37 . 21 ) 15 ( 55 . 56 ) +S . bayanus 24 ( 32 . 43 ) 21 ( 60 . 00 ) +C . glabrata 21 ( 26 . 92 ) 17 ( 50 . 00 ) +A . gossypii 19 ( 26 . 39 ) 18 ( 56 . 25 ) +K . waltii 24 ( 34 . 29 ) 11 ( 52 . 38 ) +L . elongosporus 31 ( 40 . 26 ) 1 ( 16 . 67 ) K . lactis 17 ( 27 . 42 ) 24 ( 63 . 16 ) CAH03101 . 1+C . parapsilosis 17 ( 27 . 42 ) 0 ( 0 . 00 ) CCE43871 . 1C . lusitaniae 1 5 ( 28 . 30 ) 0 ( 0 . 00 ) EEQ40372D . hansenii 14 ( 23 . 73 ) 1 ( 50 . 00 ) CAG86052 . 2C . albicans 14 ( 28 . 00 ) 0 ( 0 . 00 ) EAK92235 . 1C . guilliermondii 11 ( 24 . 44 ) 0 ( 0 . 00 ) EDK41275 . 2C . tropicalis 11 ( 16 . 42 ) 0 ( 0 . 00 ) EER32197 . 1A . nidulans 3 ( 20 . 00 ) 0 ( 0 . 00 ) EAA58119 . 1Y . lipolytica 3 ( 21 . 43 ) 0 ( 0 . 00 ) CAG79453 . 1CAG79234 . 1N . crassa 2 ( 13 . 33 ) 0 ( 0 . 00 ) ESA42867 . 1ESA42646 . 1S . japonicus 1 ( 6 . 67 ) 0 ( 0 . 00 ) EEB05964 . 1EEB09703 . 2S . octosporus 1 ( 6 . 67 ) 0 ( 0 . 00 ) EPX4426 . 1EPX71342 . 1S . pombe 1 ( 6 . 67 ) 0 ( 0 . 00 ) CAA91078 . 1CAB52735 . 1 We also find prion-like domains in Kog1 from C . elegans and other related species , but not in flies , mice or humans ( Figure 6—figure supplement 1 ) . Again , the organisms with prion-like domains in Kog1 are missing TSC1 and TSC2 ( Figure 6—figure supplement 1 ) . The observation that most organisms either have prion-like domains in Kog1 , or have genes encoding the tuberous sclerosis complex , is especially interesting when you consider the similarities between TSC1-TSC2 and Kog1-body function . Studies in human cells have shown that the TSC1-TSC2 complex serves to inhibit TORC1 in the absence of pro-growth hormones and/or the presence of AMPK activity ( Huang and Manning , 2008; Inoki et al . , 2002; Inoki et al . , 2003b ) , while we show that Kog1-body formation serves to inhibit TORC1 in the absence of glucose and the presence of AMPK activity . It therefore appears that the mechanisms underlying TORC1 regulation diverged in early eukaryotic evolution , so that most simple eukaryotes use Kog1-body formation to regulate TORC1 signaling , while higher eukaryotes ( like flies , mice and humans ) use the TSC1-2 complex . Going forward , it will be important to learn more about the differences between these two regulatory mechanisms , especially as drugs targeting Kog1-bodies may be able to block the growth of fungi and worms without affecting TORC1 signaling in humans and other higher eukaryotes . Beyond establishing a role for Kog1-body formation in TORC1 regulation , the data presented here shed light on the role that protein body formation plays in a cell . Studies in yeast , and other organisms , have shown that hundreds of proteins move into bodies , when cells are exposed to stress or starvation conditions ( Narayanaswamy et al . , 2009; O'Connell et al . , 2014; O'Connell et al . , 2012; Shah et al . , 2014 ) . For example , more than 30 proteins involved in mRNA decay and translation move into stress granules and P-bodies during long-term starvation ( Decker and Parker , 2012 ) . However , to date , it has been difficult to pinpoint the function of RNA granules and other bodies ( Decker and Parker , 2012; O'Connell et al . , 2012 ) . We show that the movement of Kog1 into bodies acts to increase the threshold for TORC1 activation . This creates hysteresis in the TORC1 pathway and likely helps ensure that cells exposed to low levels of glucose do not try to re-enter a rapid growth state . We suggest that other protein bodies formed during stress and starvation conditions function in a similar way; setting new activation thresholds for key signaling and metabolic pathways . Further work will be needed to test this idea , and to determine if there are differences between the hysteretic behavior created by reversible protein agglomeration and other mechanisms such as feedback loops ( Ferrell , 2002 ) .	In humans , yeast and other eukaryotes , a group of proteins called the Target of Rapamycin Complex I ( TORC1 ) promote cell growth and increase metabolic activity when nutrients are plentiful . Previous studies have shown how molecules that contain the nutrient nitrogen – which is needed to make proteins – activate TORC1 . However , it is not clear how other nutrients regulate this complex . Bakers yeast is a simple , single celled organism that researchers often use as a model to study how cells work . The yeast TORC1 is made up of three core proteins , including Kog1 and Tor1 . Kog1 selectively recruits proteins to the complex , where they are modified by Tor1 to regulate their activity . Here , Hughes Hallett et al . used microscopy to study what effect sugar starvation has on the complex . In the experiments , yeast cells were genetically engineered so that Kog1 and Tor1 appeared fluorescent under the microscope . The experiments reveal that , when sugar is in short supply , Kog1 breaks away from the rest of the TORC1 and moves to another part of the cell where it accumulates to form a cluster called a “body” . This movement is driven by a “kinase” enzyme that adds chemical groups called phosphates to Kog1 , and by regions within the Kog1 protein known as prion like domains . When sugar first becomes available again , Kog1 is still in the body so Tor1 cannot immediately trigger cell growth . However , once a steady supply of sugar resumes , Kog1 rejoins the rest of the complex and the cells start to grow . Together , Hughes Hallett et al . ’s findings reveal that the formation of Kog1 bodies during sugar starvation creates a “memory” that prevents TORC1 from reactivating cell growth if sugar is only temporarily available . Humans have over 100 proteins that contain prion like domains . Therefore a future challenge is to find out whether any of these proteins form similar bodies that enable our cells to remember past events .	[ "Abstract", "Introduction", "Results", "Discussion" ]	[ "biochemistry", "and", "chemical", "biology", "computational", "and", "systems", "biology" ]	2015	Snf1/AMPK promotes the formation of Kog1/Raptor-bodies to increase the activation threshold of TORC1 in budding yeast
Macrophage activation/polarization to distinct functional states is critically supported by metabolic shifts . How polarizing signals coordinate metabolic and functional reprogramming , and the potential implications for control of macrophage activation , remains poorly understood . Here we show that IL-4 signaling co-opts the Akt-mTORC1 pathway to regulate Acly , a key enzyme in Ac-CoA synthesis , leading to increased histone acetylation and M2 gene induction . Only a subset of M2 genes is controlled in this way , including those regulating cellular proliferation and chemokine production . Moreover , metabolic signals impinge on the Akt-mTORC1 axis for such control of M2 activation . We propose that Akt-mTORC1 signaling calibrates metabolic state to energetically demanding aspects of M2 activation , which may define a new role for metabolism in supporting macrophage activation . Macrophages are pleiotropic cells that assume a variety of functions depending on tissue of residence and tissue state . Their ability to acquire diverse , context-dependent activities requires activation ( or polarization ) to distinct functional states , triggered by various factors including microbial products , cytokines , and growth factors ( Davies et al . , 2013; Murray and Wynn , 2011 ) . M1 or classical activation is triggered during infection by microbial products including LPS , leading to the transcriptional upregulation of genes encoding antimicrobial activities and inflammatory cytokines . M2 or alternative activation is triggered by IL-4 and IL-13 produced during parasite infections , and activates the transcription factor Stat6 to induce a transcriptional program that coordinates fibrosis , tissue remodeling , and Type 2 inflammation ( Davies et al . , 2013; Murray and Wynn , 2011 ) . Therefore , the induction of multi-component transcriptional programs underpins macrophage activation . While macrophage activation is relatively well-understood at the level of signal transduction , transcriptional regulation , and acquisition of new effector activities , the metabolic underpinnings remain less clear . An emerging view is that macrophage activation to particular states is associated with distinct metabolic shifts ( Pearce and Everts , 2015; Galván-Peña and O'Neill , 2014; Biswas and Mantovani , 2012 ) . For example , M1 macrophages upregulate glucose and glutamine utilization ( Tannahill et al . , 2013; Cramer et al . , 2003 ) , while M2 macrophages augment β-oxidation and glutamine consumption ( Vats et al . , 2006; Jha et al . , 2015 ) . Importantly , such metabolic shifts critically support macrophage activation . Increased glycolytic flux in M1 macrophages is coupled to de novo lipogenesis , which enables ER and Golgi expansion and production of high levels of inflammatory cytokines ( Everts et al . , 2014 ) . Another consequence of enhanced glycolysis is accumulation of the TCA cycle metabolite succinate , leading to stabilization of the transcription factor HIF-1α and transcriptional induction of Il1b and other target genes in the M1 macrophage ( Tannahill et al . , 2013 ) . How oxidative metabolism boosts M2 activation is not clear , but glutamine metabolism fuels production of UDP-GlcNAC , an important modification of multiple M2 markers ( Jha et al . , 2015 ) . Consistent with the idea that macrophage activation is supported by metabolic shifts , recent studies indicate that macrophage polarizing signals impinge on metabolic signaling pathways . Polarizing signals like LPS and IL-4 regulate the activity of Akt , mTORC1 , and AMPK ( Everts et al . , 2014; Byles et al . , 2013; Cheng et al . , 2014; Weichhart et al . , 2008 ) , presumably to coordinate metabolic processes that critically underlie macrophage polarization . Limited studies indicate that perturbing the activity of these metabolic regulators impairs macrophage metabolism and activation ( Everts et al . , 2014; Cheng et al . , 2014 ) . For example , Akt mediates enhanced glycolysis to support lipid synthesis and inflammatory cytokine secretion in M1 macrophages ( Everts et al . , 2014 ) . Akt similarly stimulates glucose-fueled lipid synthesis in growing and proliferating cells , where lipids are used to build cellular membranes ( Robey and Hay , 2009 ) . Therefore , M1 macrophages co-opt a metabolic process ( Akt-dependent lipogenesis ) in order to coordinate a macrophage-specific function ( inflammatory cytokine secretion ) . In general , however , how polarizing signals control metabolic shifts , and the full implications of this for control of macrophage activation , remains poorly understood . Here we show that integration of the Akt-mTORC1 pathway into IL-4 signaling allows for selective control of some M2 responses . Control is exerted at the level of Acly , a key enzyme in Ac-CoA production , thereby modulating histone acetylation and transcriptional induction of a subset of M2 genes . Consistent with its role as an important metabolic sensor , the Akt-mTORC1 pathway couples metabolic input to such gene-specific control . Our findings also reveal subsets of the M2 response , including chemokine production and cellular proliferation , that are linked to metabolic state by Akt-mTORC1 signaling . Akt is a major metabolic regulator implicated in M2 activation ( Byles et al . , 2013; Ruckerl et al . , 2012 ) , but the underlying mechanisms remain poorly characterized . To begin to address this question , we employed unbiased metabolic profiling of M2 macrophages , using LC/MS-based metabolomics and a platform that measures ~290 small metabolites representative of all major pathways of intermediary metabolism ( Ben-Sahra et al . , 2013 ) . Top enriched pathways include urea cycle and arginine and proline metabolism , consistent with previous studies indicating upregulation of arginine metabolism in M2 macrophages ( Van Dyken and Locksley , 2013 ) , as well as amino acid utilization and metabolism and nucleotide metabolism ( Figure 1A , Supplementary file 1 ) . Other top enriched pathways include glycolysis , amino sugar metabolism , and glycine , serine , and threonine metabolism , suggesting altered flux through glycolysis and glycolytic shunts ( Figure 1A , Supplementary file 1 ) . 10 . 7554/eLife . 11612 . 003Figure 1 . Akt regulates enhanced glucose utilization in M2 macrophages . ( A ) Top metabolic pathways enriched in macrophages stimulated for 12 hr with IL-4 ( relative to unstimulated macrophages ) as identified by LC/MS-based metabolomics profiling . ( B ) M2 macrophages increase glucose uptake in an Akt-dependent manner . BMDMs were treated with IL-4 for the indicated time periods ( left ) or 16 hr +/- the Akt inhibitor MK2206 ( Akti ) ( right ) , followed by analysis of uptake of 3H-deoxy-D-glucose . ( C ) Increased glucose utilization in M2 macrophages is associated with enhanced oxidative metabolism and glycolysis . BMDMs were treated with IL-4 for 20 hr +/- Akt inhibitor , followed by analysis of spare respiratory capacity ( SRC ) and aerobic glycolysis ( ECAR ) in extracellular flux analyses . ( D ) M2 gene induction is sensitive to the glycolysis inhibitor 2-deoxyglucose ( 2-DG ) . BMDMs were treated with IL-4 for 16 hr +/- 2-DG or the β-oxidation inhibitor etomoxir pretreatment , followed by analysis of M2 gene induction by qPCR . ( E ) Akt does not regulate β-oxidation in M2 macrophages . BMDMs stimulated for 36 hr with IL-4 +/- Akt inhibitor pretreatment were incubated for 3 hr with 3H-palmitate for analysis of β-oxidation . The student’s t-test was used to determine statistical significance , defined as P<0 . 05 , P<0 . 01 , and *P<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11612 . 003 As M2 activation is thought to be sustained by fatty acid rather than glucose utilization ( Cramer et al . , 2003; Vats et al . , 2006 ) , we decided to re-examine the role of glycolysis in M2 macrophages . We found that BMDMs increased glucose uptake in a time-dependent manner in response to IL-4 treatment . Such increase was reduced by cotreatment with the Akt inhibitor MK2206 ( Figure 1B ) , indicating control by Akt and consistent with a role for Akt in regulating glycolysis in many settings ( Robey and Hay , 2009 ) . Moreover , enhanced glucose consumption in M2 macrophages was associated with an Akt-dependent increase in both glycolysis and oxidative metabolism , as indicated by extracellular flux assays ( Figure 1C ) . Importantly , glycolytic flux was needed for optimal implementation of the M2 program . Similar to the β-oxidation inhibitor etomoxir , the glycolysis inhibitor 2-DG reduced IL-4-mediated induction of some M2 genes ( Figure 1D ) . Therefore , Akt mediates enhanced glucose consumption in M2 macrophages , and this contributes to induction of M2 gene expression . Such glucose consumption may also fuel production of UDP-Glc-NAc , the substrate for glycosylation of some M2 markers ( Jha et al . , 2015 ) . In contrast , Akt does not control β-oxidation in M2 macrophages ( Figure 1E ) . Because the increase in glucose utilization was relatively modest , we considered that Akt could play additional roles in control of M2 activation and turned to an analysis of M2 gene regulation . We examined induction of Retnla , Arg1 , Mgl2 , Chi3l3 , Cd36 , and Fabp4 , “hallmark” M2 genes commonly used in studies of M2 activation ( Van Dyken and Locksley , 2013 ) . Consistent with the role of Stat6 as a transcriptional master regulator of M2 activation ( Odegaard and Chawla , 2011 ) , induction of these M2 genes was ablated in Stat6 KO BMDMs ( Figure 2—figure supplement 1A ) . Importantly and as reported ( Byles et al . , 2013; Ruckerl et al . , 2012 ) , Akt activity controlled the induction of a subset of M2 genes . In the presence of the Akt inhibitor MK2206 , induction of Arg1 , Retnla , and Mgl2 was reduced ~40–80% , while Chi3l3 , Cd36 , and Fabp4 were not affected ( or even super-inducible ) ( Figure 2A ) . Use of a structurally distinct Akt inhibitor , Aktviii , yielded similar results , suggesting specificity in inhibition ( data not shown ) . Below , these two groups of genes will be referred to as Akt-dependent and Akt-independent M2 genes , respectively . 10 . 7554/eLife . 11612 . 004Figure 2 . Akt regulates inducible histone acetylation at some M2 genes . ( A ) Akt activity stimulates induction of a subset of M2 genes . BMDMs were stimulated with IL-4 for 16 hr +/- the Akt inhibitor MK2206 ( Akti ) pretreatment , followed by analysis of M2 gene induction by qPCR . ( B ) The Jak-Stat and Akt-mTORC1 pathways are activated independently downstream of the IL-4R . WT and Stat6 KO BMDMs were stimulated with IL-4 +/- Akt inhibitor as indicated . Analysis of Stat6 , Akt , and mTORC1 activation was assessed by western blotting . ( C ) IL-4 induces a global increase in histone H3 acetylation . BMDMs were stimulated with IL-4 over the time course indicated , followed by analysis of histone H3 acetylation by western blotting . Bottom , quantitation of acetylated H3 over total H3 . ( D , E ) Akt regulates inducible H3 ( D ) and H4 ( E ) acetylation at some M2 genes . BMDMs stimulated with IL-4 for 16 hr +/- Akt inhibitor pretreatment were subject to ChIP analysis using antibodies to acetylated H3 or acetylated H4 . Enrichment of the indicated M2 gene promoters was assessed by qRT-PCR . The student’s t-test was used to determine statistical significance , defined as P<0 . 05 , P<0 . 01 , and P<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11612 . 00410 . 7554/eLife . 11612 . 005Figure 2—figure supplement 1 . Stat6 and Akt-mTORC1 pathways are independent signaling branches downstream of the IL-4R . ( A ) M2 gene induction is dependent on Stat6 . WT and Stat6 KO BMDMs were stimulated with IL-4 followed by analysis of M2 gene induction by qPCR . ( B ) IL-4 signaling activates the Jak-Stat and Akt-mTORC1 pathways . Receptor ligation activates the latent activity of Jak1 and Jak3 kinases , leading to phosphorylation and activation of Stat6 . Jak-mediated phosphorylation of the IL-4R also allows engagement of the adaptor protein IRS2 . IRS2 recruits PI3K , which generates PIP3 from PIP2 at the plasma membrane , followed by recruitment , phosphorylation , and activation of Akt . Activated Akt phosphorylates and inactivates the TSC complex , a negative regulator of mTORC1 , to activate mTORC1 . ( C ) Stat6 transcriptional activity is not affected by block of Akt activity . BMDMs were transfected with a Stat6-dependent luciferase reporter , followed by 16 hr IL-4 stimulation +/- Akt inhibitor pretreatment . Stat6 activity was assessed by a luciferase assay . The student’s t-test was used to determine statistical significance , defined as P<0 . 05 , P<0 . 01 , and P<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11612 . 00510 . 7554/eLife . 11612 . 006Figure 2—figure supplement 2 . Akt regulates inducible histone acetylation at some M2 genes . ( A ) IL-4 increases global histone acetylation in an Akt-dependent manner . BMDMs stimulated as indicated were analyzed by western blotting for H3 and H4 acetylation . ( B ) Quantitation of the western blotting data in A . ( C ) IL-4 induces an increase in H3 , H3K27 , and H4 acetylation at promoters of M2 genes . BMDMs were stimulated with IL-4 over the time course indicated , followed by ChIP analysis to examine H3 , H3K27 , and H4 acetylation at M2 gene promoters . ( D ) Akt regulates Pol II recruitment at some M2 genes . BMDMs stimulated for 16 hr with IL-4 +/- Akt inhibitor pretreatment were subject to ChIP analysis using antibodies to Pol II . Enrichment of the indicated M2 gene promoters was assessed by qRT-PCR . The student’s t-test was used to determine statistical significance , defined as P<0 . 05 , P<0 . 01 , and P<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11612 . 006 The IL-4R activates Jak-Stat signaling as well as Akt-mTORC1 signaling in macrophages ( Byles et al . , 2013 ) ( Figure 2—figure supplement 1B ) . Receptor ligation activates the latent activity of Jak1 and Jak3 kinases , leading to phosphorylation and activation of Stat6 , as well as engagement of the adaptor protein IRS2 . IRS2 recruits PI3K , which generates PIP3 from PIP2 leading to phosphorylation and activation of Akt . Activated Akt phosphorylates and inactivates the TSC complex , a negative regulator of mTORC1 , to activate mTORC1 . While the precise relationship between Jak-Stat and Akt-mTORC1 signaling remains unclear , the data in Figure 2A and Figure 2—figure supplement 1A suggest that they may operate in parallel and independently downstream of the IL-4R . Indeed , IL-4-mediated increases in Stat6 activation , as indicated by phosphorylation on Y641 , was not affected in the presence of an Akt inhibitor ( Figure 2B ) . Stat6 activity as measured by a Stat6-dependent luciferase reporter was also not impaired by inhibition of Akt activity ( Figure 2—figure supplement 1C ) . Conversely , WT and Stat6 KO BMDMs could similarly activate Akt , as indicated by phosphorylation on S473 , as well as mTORC1 , as indicated by phosphorylation of the mTORC1 target S6K , in response to IL-4 ( Figure 2B ) . These findings support the idea that the Jak-Stat and Akt-mTORC1 pathways are independent signaling branches downstream of the IL-4R , and suggest a basis by which all M2 genes are controlled by Stat6 while a subset receives additional inputs from the Akt-mTORC1 pathway . How might Akt signaling regulate a subset of M2 genes ? A seminal study from Wellen and colleagues indicated that in cancer cells and differentiating adipocytes , metabolic state is linked to gene expression via effects on histone acetylation ( Wellen et al . , 2009 ) , thus we hypothesized that Akt may control histone acetylation to regulate M2 gene expression . Indeed , IL-4-treatment of BMDMs enhanced global acetylation of H3 and H4 histones , as indicated by western blot of whole cell lysates ( Figure 2C , Figure 2—figure supplement 2A , B ) . Importantly , IL-4-inducible increases in global H3 and H4 acetylation were reduced by cotreatment with an Akt inhibitor , indicating at least partial dependence on Akt ( Figure 2—figure supplement 2A , B ) . In contrast , tubulin acetylation was not modulated by IL-4 treatment ( Figure 2—figure supplement 2A , B ) . We next examined gene-specific patterns of H3 and H4 acetylation by chromatin immunoprecipitation ( ChIP ) experiments . IL-4 treatment increased H3 and H4 acetylation at promoters of M2 genes ( Figure 2D , E , Figure 2—figure supplement 2C ) , with the degree of inducible acetylation correlating fairly well with the degree of gene induction ( Figure 2A ) . Interestingly , such increases in H3 and H4 acetylation were reduced by an Akt inhibitor at M2 genes induced in an Akt-dependent manner ( Arg1 , Retnla , Mgl2 ) , but not at M2 genes induced independently of Akt ( Chi3l3 , Cd36 , Fabp4 ) ( Figure 2D , E ) . Pol II recruitment to M2 gene promoters paralleled H3 and H4 acetylation , and was controlled by Akt at M2 genes induced in an Akt-dependent manner ( Figure 2—figure supplement 2D ) . Together , these findings support the hypothesis that Akt regulates histone acetylation and Pol II recruitment at a subset of M2 genes . How might Akt regulate increased histone acetylation in M2 macrophages ? We hypothesized that Akt may control production of Ac-CoA , the metabolic substrate for histone acetylation . Using quantitative stable isotope dilution-LC-MS , we found that IL-4 treatment led to a maximal increase in Ac-CoA levels of ~40–75% ( Figure 3A , C ) . A key regulator of Ac-CoA production is the enzyme Acly , which cleaves cytosolic citrate to produce a nuclear-cytoplasmic pool of Ac-CoA ( Wellen et al . , 2009 ) . Akt has been shown to phosphorylate and activate Acly ( Berwick et al . , 2002; Lee et al . , 2014 ) , and we found that in M2 macrophages , IL-4 treatment stimulated the activating phosphorylation of Acly in an Akt-dependent manner ( Figure 3B , Figure 3—figure supplement 1A ) . Use of lysates from MEFs transfected with ACLY siRNA confirmed specificity in detection of phosphorylated and total Acly ( Figure 3—figure supplement 1B ) . Importantly , cotreatment with Akt or Acly inhibitors blocked the IL-4-mediated increases in Ac-CoA levels ( Figure 3C ) , indicating Akt- and Acly-mediated control of Ac-CoA production in M2 macrophages . Conversely , citrate , the substrate for the Acly reaction , accumulated in the presence of the inhibitors ( Figure 3—figure supplement 1C ) . 10 . 7554/eLife . 11612 . 007Figure 3 . The Akt-Acly axis regulates inducible Ac-CoA production in M2 macrophages . ( A ) IL-4 treatment increases Ac-CoA production . BMDMs were stimulated for the indicated time periods with IL-4 , followed by analysis of Ac-CoA levels by LC-MS . ( B ) Akt regulates IL-4-inducible Acly phosphorylation . BMDMs were stimulated as indicated , followed by analysis of Acly phosphorylation by western blotting . Arrow indicates phospho-Acly . ( C ) Akt and Acly regulate IL-4-inducible production of Ac-CoA . BMDMs stimulated for 16 hr with IL-4 +/- inhibitor pretreatment were analyzed for levels of Ac-CoA by LC-MS . ( D ) BMDMs were stimulated or not for 12 hr with IL-4 , followed by a 2 hr incubation with 13C6-glucose , 13C16-palmitate , or 13C5-glutamine . Carbon tracing into Ac-CoA was assessed by LC-MS . Data shows arbitrary units of labeled 13C ( M+2 ) in the different conditions . The student’s t-test was used to determine statistical significance , defined as P<0 . 05 , P<0 . 01 , and P<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11612 . 00710 . 7554/eLife . 11612 . 008Figure 3—figure supplement 1 . Akt regulates Acly to control inducible Ac-CoA production in M2 macrophages . ( A ) Quantitation of the Acly phosphorylation from Figure 3B . ( B ) Knockdown of Acly in MEFs . Lysates from MEFs transfected with control siRNA ( siCT ) or siRNA to Acly ( siAcly ) were run next to BMDM lysates to unequivocally identify the bands corresponding to p-Acly and total Acly . ( C ) Citrate accumulates upon block of Akt or Acly activity . Citrate levels in BMDMs from the steady state metabolomics experiment in Figure 1A . The student’s t-test was used to determine statistical significance , defined as P<0 . 05 , P<0 . 01 , and P<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11612 . 008 Next , we asked about the carbon source of the Ac-CoA that supports optimal M2 gene induction . Untreated or IL-4-treated BMDMs were incubated with 13C6-glucose , 13C16-palmitate , 13C5-glutamine , followed by carbon tracing into Ac-CoA as assessed by LC-MS ( Figure 3D ) . IL-4 treatment enhanced 13C ( M+2 ) Ac-CoA labeling regardless of the tracer , indicating that all three metabolic fuels contributed to the elevated Ac-CoA pool . The highest labeling was observed in BMDMs fed palmitate . While LC-MS does not specifically measure the nuclear-cytosolic pool of Ac-CoA , these data suggests that palmitate may be the major carbon source for histone acetylation in M2 macrophages ( Figure 3D ) . These data prompted us to investigate a role for Acly in M2 activation . Indeed , the Acly inhibitor SB-204990 reduced IL-4-mediated induction of Akt-dependent M2 genes ( Arg1 , Retnla , Mgl2 ) but not Akt-independent M2 genes ( Chi3l3 , Fabp4 , Cd36 ) ( Figure 4A ) . The structurally distinct Acly inhibitor MEDICA 16 had similar effects , indicating specificity in inhibition ( data not shown ) . Moreover , SB-204990 treatment attenuated IL-4-mediated increases in H3 and H4 acetylation at promoters of Akt-dependent M2 genes , but not Akt-independent M2 genes ( Figure 4B , Figure 4—figure supplement 1A ) . Likewise , SB-204990 treatment diminished Pol II recruitment at Akt-dependent M2 genes ( Figure 4—figure supplement 1B ) . 10 . 7554/eLife . 11612 . 009Figure 4 . Acly controls inducible histone acetylation at some M2 genes . ( A ) Acly regulates induction of some M2 genes . BMDMs stimulated for 16 hr with IL-4 +/- Acly inhibitor pretreatment were analyzed for M2 gene induction by qRT-PCR . ( B ) Acly regulates inducible H3 acetylation at some M2 genes . BMDMs stimulated for 16 hr with IL-4 +/- Acly inhibitor pretreatment were subject to ChIP analysis using antibodies to acetylated H3 . Enrichment of the indicated M2 gene promoters was assessed by qRT-PCR . ( C ) The p300 inhibitor C646 reduces induction of some M2 genes . BMDMs stimulated for 16 hr with IL-4 +/- C646 pretreatment were analyzed for M2 gene induction by qRT-PCR . ( D ) Akt and Acly control IL-4-inducible arginase activity . BMDMs were stimulated for IL-4 for 24 hr +/- inhibitor pretreatment , followed by analysis of arginase activity in cellular lysates as assessed by urea production . The student’s t-test was used to determine statistical significance , defined as P<0 . 05 , P<0 . 01 , and P<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11612 . 00910 . 7554/eLife . 11612 . 010Figure 4—figure supplement 1 . Acly controls inducible histone acetylation at some M2 genes . ( A ) Acly regulates inducible H4 acetylation at some M2 genes . BMDMs stimulated for 16 hr with IL-4 +/- Acly inhibitor pretreatment were subject to ChIP analysis using antibodies to acetylated H4 . Enrichment of the indicated M2 gene promoters was assessed by qRT-PCR . ( B ) Acly regulates Pol II recruitment at some M2 genes . BMDMs stimulated for 16 hr with IL-4 +/- Acly inhibitor pretreatment were subject to ChIP analysis using antibodies to Pol II . Enrichment of the indicated M2 gene promoters was assessed by qRT-PCR . The student’s t-test was used to determine statistical significance , defined as P<0 . 05 , P<0 . 01 , and P<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11612 . 01010 . 7554/eLife . 11612 . 011Figure 4—figure supplement 2 . Akt-Acly signaling regulates M2 activation in peritoneal macrophages . Elicited peritoneal macrophages were treated for 16 hr with IL-4 +/- the indicated inhibitors , followed by analysis of M2 gene induction by qRT-PCR . DOI: http://dx . doi . org/10 . 7554/eLife . 11612 . 011 Because Akt and Acly regulate a global increase in Ac-CoA levels ( Figure 3C ) but control inducible histone acetylation only at some M2 gene promoters ( Figure 2D , E , 4B , and Figure 4—figure supplement 1A ) , Ac-CoA production is necessary but not sufficient for stimulating gene-specific increases in histone acetylation , which must be conferred by specific transcription factors and histone acetyltransferases ( HATs ) . The activity of some HATs , including p300 , is regulated by Ac-CoA levels and metabolic status ( Mariño et al . , 2014; Pietrocola et al . , 2015 ) . Interestingly , the p300 inhibitor C646 reduced induction of Akt-dependent but not Akt-independent M2 genes ( Figure 4C ) . Therefore , p300 may link the Akt/Acly-dependent rise in Ac-CoA levels to increased histone acetylation and gene induction at some Akt-dependent M2 genes , while distinct HATs at Akt-independent genes are insensitive to such modulation of Ac-CoA levels . Arginase activity is a hallmark feature of M2 activation that supports collagen production and polyamine synthesis ( Van Dyken and Locksley , 2013 ) . Consistent with effects on induction of Arg1 , arginase activity was regulated by Acly and Akt ( Figure 4D ) . Additionally , Akt and Acly inhibitors reduced induction of Akt-dependent M2 genes in peritoneal-elicited macrophages , indicating that control of M2 activation by the Akt-Acly axis may be applicable to multiple macrophage populations ( Figure 4—figure supplement 2 ) . Finally , induction of M2 gene expression by IL-13 , a cytokine closely related to IL-4 that also triggers M2 activation ( Van Dyken and Locksley , 2013 ) , was also dependent on Akt and Acly ( data not shown ) . Our findings that Akt regulates Acly activity to control Ac-CoA production and M2 activation led us to consider a role for mTORC1 in this process . mTORC1 is a key downstream effector of Akt signaling and their activities are intricately linked in many settings ( [Dibble and Manning , 2013; Pollizzi and Powell , 2014; Laplante and Sabatini , 2012] and Figure 2—figure supplement 1B ) . Indeed , we found that induction of Akt-dependent M2 genes was deficient in BMDMs lacking Raptor , a defining subunit of the mTORC1 complex ( Dibble and Manning , 2013 ) . In contrast , induction of Akt-independent M2 genes was not reduced ( Figure 5A ) . mTORC1 is known to stimulate Acly expression ( Porstmann et al . , 2008; Düvel et al . , 2010 ) , and we found that Raptor-deficient BMDMs expressed lower levels of Acly protein ( Figure 5B ) . Conversely , BMDMs with constitutive mTORC1 activity resulting from deletion of Tsc1 ( Byles et al . , 2013 ) , a negative regulator of mTORC1 ( Dibble and Manning , 2013 ) , displayed elevated Acly levels that were reduced by treatment with the mTORC1 inhibitor rapamycin ( Figure 5—figure supplement 1 ) . Additionally , we noted that IL-4-inducible Acly phosphorylation was reduced in Raptor-deficient BMDMs ( Figure 5B ) . This raises the possibility that mTORC1 could also regulate Acly activating phosphorylation , through mechanisms that remain to be clarified in future studies . Taken together , these data indicate that the Akt-mTORC1 axis controls Acly activating phosphorylation and protein levels , likely contributing to its control of M2 activation . 10 . 7554/eLife . 11612 . 012Figure 5 . mTORC1 controls Acly protein levels to regulate M2 activation . ( A ) mTORC1 regulates M2 activation . M2 gene expression in Raptorfl/fl and Raptor△/△ BMDMs stimulated with IL-4 for 16 hr as assessed by qRT-PCR . ( B ) mTORC1 regulates Acly protein levels . Acly protein expression in Raptorfl/fl and Raptor△/△ BMDMs stimulated as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 11612 . 01210 . 7554/eLife . 11612 . 013Figure 5—figure supplement 1 . mTORC1 activity regulates levels of Acly . Tsc1fl/fl and Tsc1△/△ BMDMs were treated with rapamycin or not for the indicated time periods , followed by analysis of levels of Acly by western blotting . DOI: http://dx . doi . org/10 . 7554/eLife . 11612 . 013 The Akt-mTORC1 pathway is a major metabolic sensor , and mTORC1 activity in particular is controlled by amino acid levels , ADP/ATP levels , and other metabolic inputs ( Dibble and Manning , 2013; Laplante and Sabatini , 2012 ) . Therefore , we considered that incorporation of the Akt-mTORC1 pathway into IL-4 signaling , parallel to canonical Jak-Stat signaling , may allow particular subsets of the M2 transcriptional program to integrate signals reflecting the cellular metabolic state ( Figure 6A ) . Amino acids directly and potently regulate mTORC1 activity independent of the TSC complex ( Dibble and Manning , 2013; Laplante and Sabatini , 2012 ) and can also activate Akt in some contexts ( Tato et al . , 2011; Novellasdemunt et al . , 2013 ) , hence we varied amino acid concentrations as a way to modulate Akt-mTORC1 activity . As expected , mTORC1 activity , as assessed by phosphorylation of its downstream target S6K , was greatly reduced in amino acid deficient media and intermediate in media containing low levels of amino acids ( Figure 6B ) . In line with ( Tato et al . , 2011; Novellasdemunt et al . , 2013 ) , increasing amino acid levels also augmented Akt activation , as indicated by enhanced phosphorylation on two critical residues , T308 and S473 ( Figure 6B ) . Titrating amino acids had no effect on Stat6 phosphorylation and activation ( Figure 6B ) , validating the use of this experimental model to modulate the Akt-mTORC1 axis independent of canonical Stat6 signaling . Consistent with effects on mTORC1 and Akt activity , amino acid levels dose dependently increased Acly phosphorylation and protein levels ( Figure 6B ) as well as Ac-CoA production ( Figure 6C ) . Importantly , amino acids potentiated induction of Akt-dependent but not Akt-independent M2 genes ( Figure 6D ) . This effect of amino acids was at least partially Raptor-dependent , indicating a critical role for mTORC1 in this process ( Figure 6—figure supplement 1 ) . 10 . 7554/eLife . 11612 . 014Figure 6 . The Akt-mTORC1-Acly axis links metabolic input to control of M2 activation . ( A ) Proposed model for how Akt-mTORC1-Acly signaling exerts gene-specific control of M2 activation . Akt-TORC1-Acly signaling integrates metabolic input to control levels of Ac-CoA production , which modulates histone acetylation and gene induction at some M2 genes by HATs such as p300 . ( B ) Amino acid levels modulate the activity of the Akt-mTORC1-Acly axis . BMDMs cultured in media containing varying levels of amino acids ( normal , low , or no ) were stimulated with IL-4 for the indicated time periods , followed by analysis of Akt , mTORC1 , and Acly activity by western blotting . ( C ) Amino acid levels modulate Ac-CoA production . BMDMs stimulated as in B . were harvested for LC-MS analysis of Ac-CoA levels after 12 hr IL-4 stimulation . ( D ) Amino acid levels modulate induction of some M2 genes . BMDMs stimulated as in B . were harvested for qRT-PCR analysis of M2 gene induction after 9 hr IL-4 stimulation . ( E ) Leucine deficiency attenuates the activity of the Akt-mTORC1-Acly axis . BMDMs cultured in leucine-replete or leucine-deficient media were stimulated with IL-4 for the indicated time periods , followed by analysis of Akt , mTORC1 , and Acly activity by western blotting . Right , quantitation of Acly phosphorylation . ( F ) Leucine deficiency reduces induction of some M2 genes . BMDMs stimulated as in E . were harvested for qRT-PCR analysis of M2 gene induction after 16 hr IL-4 stimulation . The student’s t-test was used to determine statistical significance , defined as P<0 . 05 , P<0 . 01 , and P<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11612 . 01410 . 7554/eLife . 11612 . 015Figure 6—figure supplement 1 . Amino acid levels modulate M2 gene expression in part through Raptor . Raptorfl/fl and Raptor△/△ BMDMs were stimulated as in the experiment shown in Figure 6D , followed by analysis of M2 gene expression by qRT-PCR . IL-4-stimulated gene expression in the absence of amino acids is arbitrary set at 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11612 . 01510 . 7554/eLife . 11612 . 016Figure 6—figure supplement 2 . Feeding and fasting regulate M2 polarization of adipose tissue macrophages . Analysis of ATM M2 activation in fed and fasted mice . Mice ( n=4/group ) were allowed to feed ad-lib or fasted O/N prior to their sacrifice . The ATM-containing stromal vascular fraction of perigonadal white adipose tissue was obtained for analysis of Akt activity by western blotting ( A ) ; H3 acetylation by western blotting ( A , B ) ; expression of M2 genes by qRT-PCR ( C ) ; and expression of IL-13 by qRT-PCR ( D ) . The student’s t-test was used to determine statistical significance , defined as P<0 . 05 , P<0 . 01 , and P<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11612 . 016 We also examined M2 activation using the complementary model of leucine deprivation , since leucine is particularly critical in regulation of mTORC1 activity ( Hara et al . , 1998 ) . Here comparisons were made between culture conditions that differed only in the presence or absence of one amino acid , without significant effects on total levels of amino acids . Culture in leucine-deficient media attenuated IL-4-inducible mTORC1 and Akt activity and Acly phosphorylation , but not Stat6 phosphorylation ( Figure 6E ) . Importantly , leucine deficiency selectively reduced expression of Akt-dependent M2 genes ( Figure 6F ) . Taken together , these results indicate that amino acids and likely other metabolic inputs feed into the Akt-mTORC1 axis to calibrate M2 activation to the metabolic state ( Figure 6A ) . Finally , we found that physiological changes to nutrient levels can modulate M2 activation in adipose tissue macrophages ( ATMs ) . ATM M2 polarization is thought to critically maintain insulin sensitivity in white adipose tissue , so such feeding-induced increases in M2 activation may coordinate responses to nutrient influx to mediate metabolic homeostasis in the postprandial state ( Odegaard and Chawla , 2011 ) . Specifically , we found that Akt activation was increased in the fed state compared to the fasted state in the ATM-containing stromal vascular ( SVF ) fraction of the white adipose tissue ( Figure 6—figure supplement 2A ) . Although we were unable to reliably detect pAcly or Acly in the SVF for technical reasons , global H3 acetylation ( Figure 6—figure supplement 2A–B ) and M2 gene expression ( Figure 6—figure supplement 2C ) followed a similar pattern and were elevated in the fed state . Expression of all M2 genes was elevated in the fed state ( Figure 6—figure supplement 2C ) , consistent with an important role for IL-13 , a critical regulator of ATM M2 polarization ( Odegaard and Chawla , 2011 ) that is increased in the fed state ( Figure 6—figure supplement 2D ) , in feeding-induced ATM polarization , although postprandial elevations in nutrients like amino acids and glucose may also contribute . Therefore , feeding-inducible Akt activity correlated with increases in histone acetylation and M2 activation in ATMs . We employed genome wide transcriptional profiling to obtain a comprehensive view of regulation of M2 activation by the Akt-Acly pathway . BMDMs were treated for 16 hr with IL-4 with or without Akt or Acly inhibitors , followed by RNA seq ( Figure 7 ) or microarray analysis ( data not shown ) . In the RNA seq analysis , 758 genes were induced >2 . 0 fold by IL-4 , of which 91were downregulated >30% by both Akt and Acly inhibitors ( including Arg1 , Retnla , and Mgl2 ) , confirming critical roles for Akt and Acly in control of M2 activation as well as substantial overlap in the activities of the two proteins ( Figure 7A , B ) . A subset of Akt inhibitor sensitive genes was sensitive to Acly inhibitor ( 91/327 ) , in line with a broader role for Akt in control of cell physiology . In contrast , most genes sensitive to Acly inhibitor were sensitive to Akt inhibitor ( 91/118 ) . This indicates that in the context of M2 activation , Acly is a major target of Akt and is critically controlled by Akt activity , likely in regulation of Ac-CoA production and histone acetylation at M2 genes ( Figure 7A , B ) . 10 . 7554/eLife . 11612 . 017Figure 7 . The Akt-Acly axis controls functional subsets of the M2 program . ( A ) Venn diagram depicting the number of IL-4-inducible genes regulated by Akt and/or Acly signaling . ( B ) Heatmap of normalized rank ordered Log2 RPKM values of top 50 IL-4 response genes co-regulated by Akt and Acly . ( C ) Heatmap of enriched KEGG pathways within the cohort of IL-4-inducible genes . ( D ) Heatmap of enriched Gene Ontology terms within the cohort of IL-4-inducible genes . ( E ) qPCR analysis validates regulation of chemokine and Ear genes by Akt-Acly signaling . BMDMs were stimulated with IL-4 for 16 hr +/- Akt or Acly inhibitor . ( F ) ELISA analysis indicates that Akt-Acly signaling regulates production of CCL17 and CCL24 . BMDMs were stimulated with IL-4 for 36 hr +/- Akt or Acly inhibitor . The student’s t-test was used to determine statistical significance , defined as P<0 . 05 , P<0 . 01 , and P<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11612 . 01710 . 7554/eLife . 11612 . 018Figure 7—figure supplement 1 . Working models . ( A ) The Akt-mTORC1-Acly pathway couples metabolic input to control of particular subsets of the M2 program , including chemokines and cellular proliferation . ( B ) p300 links metabolic status and Akt-mTORC1 activity in the form of Ac-CoA levels to M2 gene induction . p300 has a high Km and is responsive to changes in the levels of its substrate , allowing it to couple increased Ac-CoA levels to enhanced histone acetylation at some Akt-dependent M2 genes . In contrast , HATs at Akt-independent M2 genes ( not shown ) have low Km and are relatively insensitive to metabolic status and changing Ac-CoA levels . DOI: http://dx . doi . org/10 . 7554/eLife . 11612 . 018 Gene enrichment analysis of the 91 Akt- and Acly-coregulated genes identified preferential enrichment of several pathways , including cell cycle and DNA replication ( Figure 7C , D ) . IL-4 triggered BrdU labeling of a subset of BMDMs in vitro ( data not shown ) and proliferation of macrophages in vivo ( Ruckerl et al . , 2012 ) , thus IL-4 may stimulate macrophage proliferation in an Akt- and Acly-dependent manner . Consistently , metabolic processes underlying cellular proliferation were among the top enriched pathways in our metabolomics analysis , including nucleotide metabolism and protein biosynthesis ( Figure 1A ) . Interestingly , chemokines were also enriched in Akt- and Acly-coregulated genes ( Figure 7C–D ) , including Ccl2 , Ccl7 , Ccl17 , and Ccl24 . Akt- and Acly-dependent induction of CCL17 and CCL24 was confirmed by qRT-PCR and ELISA ( Figure 7E–F ) . Finally , genes in the eosinophil associated ribonucleases ( Ear ) family were found to be regulated by Akt and Acly . While barely missing the stringent cutoffs that we set for the RNA-seq analysis , qPCR analysis confirmed coregulation of Ear2 , Ear11 , and Ear12 genes by the Akt-Acly pathway ( Figure 7E ) . Ear genes are of interest because Ear2 and Ear11 are thought to have chemoattractant activity for dendritic cells and macrophages and are known to be highly induced in settings of Type 2 inflammation ( Cormier et al . , 2002; Yamada et al . , 2015 ) . Therefore , the transcriptional profiling analysis indicated that the Akt-Acly pathway controls selective subsets of the M2 program to allow their modulation by metabolic input ( Figure 7—figure supplement 1A ) . As M2 macrophages play a key role in metabolic homeostasis , parasite infection , allergic diseases , and wound healing and tissue repair ( Van Dyken and Locksley , 2013; Odegaard and Chawla , 2011 ) , these findings are relevant for metabolic control of macrophage function in diverse contexts . The Akt-mTORC1 pathway has a well-established role in promoting anabolic metabolism in growing/proliferating cells , tumor cells , and metabolic tissues . In the context of cellular proliferation , for example , Akt-mTORC1 activity couples growth factor signaling and nutrient availability to the synthesis of proteins , lipids , and nucleotides ( Dibble and Manning , 2013 ) . In contrast , the role of the Akt-mTORC1 pathway in macrophages is much less intuitive . What is the teleological rationale for control of macrophage activation by Akt-mTORC1 signaling ( and metabolism more generally ) ? Here we propose that IL-4 signaling co-opted the Akt-mTORC1 pathway to couple metabolic input to regulation of certain components of the M2 response , including chemokines and cellular proliferation ( Figure 7—figure supplement 1A ) . This is supported by our findings that IL-4 signaling leads to parallel and independent activation of the Akt-mTORC1 pathway and the canonical Jak-Stat pathway , allowing the Akt-mTORC1 axis to regulate a subset of M2 genes through control of Acly activity/expression , Ac-CoA production , and histone acetylation . Why should some but not other components of the M2 response be regulated in this way ? Control of cellular proliferation is intuitive , since Akt-mTORC1 signaling acts as a metabolic checkpoint in the context of cellular division to allow growth and proliferation only when nutrients are abundant . What about chemokines ? We propose that chemokines may be controlled by the Akt-mTORC1 pathway because of their key role in amplifying energetically costly immune responses ( Hotamisligil and Erbay , 2008 ) . This allows metabolic status to calibrate immune responses such that inflammation is amplified and sustained only under metabolically favorable conditions . Interestingly , previous studies have shown that a critical role for Akt-mTORC1 signaling in activated CD8 T cells is to support their migration to sites of inflammation ( Finlay and Cantrell , 2011 ) . Therefore , Akt-mTORC1 signaling regulates both facets of immune response amplification , i . e . , the ability of tissue-resident sentinel cells to mobilize activated leukocytes and of activated T cells to be recruited . Together these findings add another dimension to our emerging understanding of how metabolism supports leukocyte activation and immune responses . As discussed above , the Akt/Acly-dependent rise in Ac-CoA production is necessary but not sufficient for stimulating gene-specific increases in histone acetylation . Such specificity is most likely conferred by HATs with distinct Km ( Pietrocola et al . , 2015 ) . Indeed , our analysis suggests that p300 may preferentially regulate at least a subset of the Akt-dependent M2 genes ( Figure 4C ) . Its high Km ( Mariño et al . , 2014; Pietrocola et al . , 2015 ) may allow p300 to link metabolic status and Akt/mTORC1 activity , in the form of Ac-CoA levels , to histone acetylation and transcriptional induction at some M2 genes ( Figure 7—figure supplement 1B ) . In contrast , HATs at Akt-independent M2 genes may have a low Km and are thus insensitive to such modulation of Ac-CoA levels . Presumably , differential HAT recruitment is mediated by distinct transcription factors at Akt-dependent and independent M2 genes , which would be important to address in future studies . Although Akt activity has been linked to M2 activation , ( Byles et al . , 2013; Ruckerl et al . , 2012 ) , the role of mTORC1 remained unclear . Here , we use Raptor△/△ BMDMs to show that mTORC1 activity stimulates M2 activation ( Figure 5A ) . Furthermore , amino acids modulate mTORC1 activity ( Figure 6B ) to potentiate M2 gene induction in a Raptor-dependent manner ( Figure 6D , Figure 6—figure supplement 1 ) . Together these findings indicate that the Akt-mTORC1 signaling module supports M2 activation . Acly appears to be a key target , with its expression levels and activating phosphorylation controlled by mTORC1 and Akt respectively . In seeming contrast to these data indicating that mTORC1 supports M2 activation , we and others have shown that aberrantly increased mTORC1 activity in Tsc1-deficient BMDMs attenuates M2 activation ( Byles et al . , 2013; Zhu et al . , 2014 ) . We hypothesize that the difference between the two models reflects divergent control of M2 activation by physiological and pathophysiological mTORC1 activity respectively . Downstream of the insulin receptor , such context-dependent roles of mTORC1 are well-established . In lean/healthy animals , mTORC1 critically mediates insulin signaling in metabolic tissues ( to coordinate postprandial nutrient storage ) , but in obesity , chronic nutrient excess leads to an aberrant increase in mTORC1 activity that contributes directly to insulin resistance and metabolic dysregulation ( Laplante and Sabatini , 2012 ) . Similarly , while physiological mTORC1 activity couples metabolic input to M2 activation , pathophysiological mTORC1 activation during chronic nutrient excess may impair M2 activation . It would be interesting to see if the latter is true in adipose tissue macrophages in the context of diet-induced obesity , and if so , the consequences for tissue inflammation and metabolic homeostasis . Interestingly , while inducible Akt phosphorylation occurred within minutes of IL-4 stimulation ( Figure 2B ) , inducible Acly phosphorylation was detected with slightly delayed kinetics ( ~2 h , Figure 5B and data not shown ) . Such delay may reflect a need for other inputs that facilitate Akt-mediated Acly phosphorylation , or the reduced sensitivity and dynamic range of the pAcly antibody compared to the pAkt antibodies . Once pAcly is detectable at ~2 h , Akt and Acly phosphorylation nicely parallel and steadily increase up to ( and perhaps beyond ) 8 hr ( Figure 5B ) . As expected , inducible Akt and Acly phosphorylation precede increases in global histone acetylation , which is observed starting only at 4 hr ( Figure 2C ) . However , Ac-CoA levels increase only 8 hr after IL-4 stimulation ( Figure 3A ) . One possibility , supported by the increase in global histone acetylation at 4 h , is that diversion of Ac-CoA into acetylated histones diminishes the free Ac-CoA pool . Another possibility is that because the LC-MS analysis measures bulk Ac-CoA rather than the nuclear-cytoplasmic pool relevant for histone acetylation , changes in mitochondrial Ac-CoA levels could be confounding . Again , global histone acetylation , which may more accurately reflect nuclear-cytoplasmic pools of Ac-CoA , increases 4 hr after IL-4 treatment ( Figure 2C ) , as does gene-specific increases in histone acetylation at Akt-dependent M2 genes ( Figure 2—figure supplement 2C ) . Therefore , we believe that the preponderance of the data support our model that IL-4 triggers Ac-CoA production and histone acetylation as a consequence of Akt-mediated Acly activation . Metabolic status has long been proposed to modulate epigenetic control of gene expression ( Teperino et al . , 2010; Kaelin and McKnight , 2013; Gut and Verdin , 2013 ) , but only recently have a handful of studies linked physiological changes in metabolite levels to chromatin regulation of gene expression ( Wellen et al . , 2009; Lee et al . , 2014; Shimazu et al . , 2013; Carey et al . , 2015 ) . Here we show how the Akt-mTORC1 axis couples metabolic input in the form of Ac-CoA levels to histone acetylation and gene regulation , and importantly , to control specific subsets of the M2 program . In addition to a recent study ( Lee et al . , 2014 ) , this is only the second example of how Akt-Acly signaling controls gene regulation through histone acetylation . Other macrophage polarizing signals and common gamma chain cytokines ( γc ) ( e . g . IL-2 , IL-15 ) engage the Akt-mTORC1 axis , thus our findings may have implications for multiple programs of macrophage polarization and leukocyte activation . Canonical signaling downstream of the polarizing signal or γc specifies which genes are induced , while regulation of Ac-CoA levels and histone acetylation by the Akt-mTORC1-Acly pathway allows metabolic input to calibrate genes encoding energetically demanding processes; it would be informative in future studies to determine the nature of these processes . Alternatively , Ac-CoA can be synthesized independently of the Akt-mTORC1-Acly axis by AceCS1 ( Hallows et al . , 2006 ) or nuclear pyruvate dehydrogenase ( Sutendra et al . , 2014 ) to mediate histone acetylation . AceCS1 activity is controlled by SIRT1 , thus providing a means for Ac-CoA production and histone acetylation in conditions of low energy or nutrients ( Hallows et al . , 2006 ) . Therefore , future studies to determine how gene-specific histone acetylation is regulated during different macrophage activation programs are warranted . These studies could pave the way towards new therapeutic approaches of modulating macrophage function in diverse contexts , including Type 2 inflammation , metabolic homeostasis , and antimicrobial immunity . BMDM cultures were established as described ( Byles et al . , 2013 ) . For stimulations , BMDMs were pretreated for 1 hr with inhibitors followed by addition of 10 ng/ml IL-4 for 16 hr unless otherwise indicated . Inhibitors were used as follows: AKT inhibitor MK-2206 , 2–5 μM ( Selleck , Houston , TX ) ; ACLY inhibitor SB-204990 , 40 μM ( Tocris , United Kingdom ) ; p300 inhibitor C646 , 10 μM; etomoxir , 200 μM ( Sigma , St . Louis , MO ) , and 2-deoxy-glucose , 1 mM ( Sigma ) . For amino acid titration experiments , BMDMs were plated in DMEM containing low levels of amino acids for 6 hr ( to deplete cellular amino acid pools ) prior to changing the media to DMEM with varying levels of amino acids ( no , low , or normal ) +/- IL-4 for 16 hr . Normal is normal tissue culture media , while low indicates media containing 5% of the normal levels of amino acids ( obtained by mixing normal media and media lacking amino acids ) . In experiments with leucine free media , BMDMs were stimulated in complete DMEM or –Leu complete DMEM ( Crystalgen , Commack , NY ) +/- IL-4 for 16 hr . Tsc1△/△ BMDMs were described previously ( Byles et al . , 2013 ) . BMDMs from UbiquitinC-CreERT2 Raptorfl/fl mice were treated with tamoxifen to delete Raptor; parallel treatment of Raptorfl/fl BMDMs were used as controls . C57BL/6 mice were used for in vivo studies and as a source of BMDMs . Mice were maintained at Harvard Medical School and all procedures were performed in accordance with the guidelines set forth by the Institutional Animal Care and Use Committees at the institution . To generate UbiquitinC-CreERT2 Raptorfl/fl mice , previously described Raptorfl/fl mice ( Sengupta et al . , 2010 ) were crossed with UbiquitinC-CreERT2 mice ( The Jackson Laboratory , Bar Harbor , ME ) in David Sabatini’s laboratory at the Whitehead Institute in Cambridge , Massachusetts , in accordance with the guidelines set forth by the Institutional Animal Care and Use Committee at the institution . Cells were lysed directly in 6X SDS loading buffer ( histone western blots ) or in 1% NP-40 buffer ( all other western blots ) . Protein concentration was determined using the Bradford method . Primary antibodies were purchased from Cell Signaling except for α-Tubulin ( Sigma ) , acetylated Tubulin ( Sigma ) , acetylated H3 ( Millipore , Germany ) , acetylated H4 ( Millipore ) , and total H4 ( Abcam , Cambridge , MA ) . Arginase assay was done as described ( Byles et al . , 2013 ) . Oxygen consumption and extracellular acidification rates were measured with a XF96 extracellular flux analyzer ( Seahorse Bioscience , North Billerica , MA ) . Seahorse assay media containing 11 mM glucose or plain assay media was used for the mitochondrial and glycolysis stress tests respectively . OCR measurements were taken before and after the sequential addition of 1 μM oligomycin , 1 . 5 μM FCCP and 2 μM antimycin/rotenone ( Sigma ) . ECAR measurements were taken before and after the sequential addition of 11 mM glucose , 1 μM oligomycin and 0 . 5 M 2-DG ( Sigma ) . Values were normalized with Hoechst 33342 staining ( Life Technologies , Carlsbad , CA ) . BMDMs were washed with Krebs-Ringer bicarbonate HEPES ( KRBH ) buffer once , followed by addition of 400 μl KRBH buffer . 100 μl loading buffer ( KRBH buffer with 0 . 5 mM 2-deoxy-D-glucose ( Sigma ) and 1 μCi/well 3H-deoxy-D-glucose ( 2-3H[G] ) ( PerkinElmer , Waltham , MA , 1 mCi/ml in EtOH:water [9:1] ) was added and incubated at 37°C for exactly 15 min . 20 μl stop solution ( 1 . 5 mM cytochalasin B ( Sigma ) in DMSO ) was added , and the cells were washed with KRBH buffer before lysis in 0 . 1 N NaOH . The glucose uptake rate was determined by normalizing cellular 3H-deoxy-D-glucose count to protein concentrations . Fatty acid oxidation was done as described ( Byles et al . , 2013 ) . ChIP was done as described ( Byles et al . , 2013 ) , using acetylated H3 ( Millipore 06–599 ) , acetylated H4 ( Millipore 06–866 ) , or IgG ( Santa Cruz , Dallas , TX , SC-2027 ) antibodies . Fold enrichment was calculated as ChIP signals normalized to input . ChIP primer sequences as well as position relative to transcription start site ( TSS ) are provided in Supplementary file 2 . RNA was isolated using RNA-Bee ( Tel-Test , Friendswood , TX ) per manufacturers protocol . cDNA synthesis was done using High Capacity cDNA Reverse Transcription Kit ( Applied Biosystems , Foster City , CA ) . A Bio-Rad C1000 Thermocycler was used for qPCR , and data was analyzed by means of the CFX Manger Software ( Bio-Rad , Hercules , CA ) using the delta/delta CT method . BMDM samples were normalized to hypoxanthine phosphoribosyltransferase while ex vivo samples were normalized to the macrophage marker CD68 . BMDMs were electroporated using mouse macrophage nucleofector kit ( Lonza , Hopkinton , MA ) and the Amaxa machine with STAT6-Firefly luciferase ( Addgene , Cambridge , MA , plasmid #35554 ) along with Renilla–Luciferase plasmid as a transfection control . BMDMs were stimulated with or without 10 ng/ml IL-4 4 hr post electroporation for another 24 hr . Cell lysates were collected and analyzed using the Promega Dual-Luciferase Reporter Assay System . BMDMs were lysed in 800 μl ice cold 10% TCA ( Tricholoracetic acid ) . Sc5-sulfosalicylic acid ( SSA ) , ammonium formate , [13C6]-glucose , sodium [13C16]-palmitate , and analytical standards for acyl-CoAs were from Sigma-Aldrich ( St . Louis , MO ) . Optima LC-MS grade methanol , ammonium acetate , acetonitrile ( ACN ) and water were purchased from Fisher Scientific ( Pittsburgh , PA ) . Calcium [13C315N1]-pantothenate was purchased from Isosciences ( King of Prussia , PA ) . [13C315N1]-acyl-CoA internal standards for quantitation were generated by pan6 deficient yeast culture as previously described ( Snyder et al . , 2015 ) , with 100 µL of extract spiked into samples before extraction . Standard curves were prepared using the same batch of internal standard , and all samples were extracted by solid phase extraction as previously described ( Basu and Blair , 2012 ) . Acyl-CoAs were analyzed as previously described for quantitation ( Basu et al . , 2011 ) and for isotopolog analysis ( Worth et al . , 2014 ) by liquid chromatography-tandem mass spectrometry on an Agilent 1200 coupled to an API4000 in the positive ion mode monitoring the acyl-CoA specific neutral loss of 507 amu from each acyl-CoA , internal standard and isotopolog . For carbon tracing experiments , BMDMs were treated with 10 ng/ml IL-4 for 12 hr before the addition of tracers ( 2g/L 13C6-glucose , 50 μM 13C16-palmitate , or 2 mM 13C5-glutamine ) for another 2 hr . BMDMs were stimulated for 10 hr with IL-4 before media was refreshed by addition of complete RPMI with IL-4 for another 2 hr . Preparation of cellular extracts was done as described ( Ben-Sahra et al . , 2013 ) . Steady state metabolomics was done at Beth Israel Deaconess Medical Center Mass Spectrometry Facility . Data analysis was performed as described ( Ben-Sahra et al . , 2013 ) . Strand-specific libraries were generated using 500ng RNA input using TruSeq library preparation kit ( Illumina , San Diego , CA ) . cDNA libraries were multiplexed using specific unique adaptors and sequenced using Illumina NextSeq 500 under single end 75bp read length parameters . Reads were aligned to the mouse mm10 reference genome using TopHat using default settings ( Langmead et al . , 2009 ) . Alignments were restricted to uniquely mapping reads , with up to 2 mismatches permitted . RPKM was calculated as described for mm10 Refseq genes by counting exonic reads and dividing by mRNA length ( Mortazavi et al . , 2008 ) . Coexpressed gene classes were generated with Cluster3 by applying k-means clustering to mean-centered log2 ( FPKM ) expression values . Differential analyses was performed using DEseq ( Anders and Huber , 2010 ) using default parameters for the indicated comparisons . Cohort of IL-4 inducible genes was defined by following: >2 RPKM , Log2fold>1 . 0 , DESeq P-adj<0 . 05 yielding 758 IL-4 inducible genes . Inhibition by AKT or ACLY inhibitors defined as 30% reduction in RPKM and DESeq P-adj <0 . 05 . Enrichment of KEGG pathways and Gene Ontology ( GO ) terms analysis performed using DAVID ( Huang et al . , 2008 ) . 8–10 week old C57BL/6 mice were fasted overnight or allowed to feed ad-libitum . Mice were sacrificed the next morning and the perigonadal adipose tissue was excised . A small section of whole adipose tissue ( WAT ) was homogenized in RNA-Bee for analysis of gene expression in unfractionated WAT . The remaining adipose tissue was minced and digested in 5 ml Krebs ringer buffer ( KRBH ) containing 2% fatty acid free BSA and 2 mg/ml collagenase ( Sigma , C2674 ) for 20 min at 37°C . The resulting cell suspension was filtered through a 250 mm nylon mesh and centrifuged at 1200 RPM to obtain a cell pellet corresponding to the stromal vascular fraction ( SVF ) , which was lysed for RNA extraction or western blotting . Statistical analysis was carried out using Prism ( GraphPad ) software . The student’s t-test was used to determine statistical significance , defined as P<0 . 05 , P<0 . 01 , and *P<0 . 001 .	Macrophages are immune cells that are found in most of the tissues of the body . Exactly what the macrophages do depends on which tissue they are in , and the state of the tissue . For example , M2 macrophages can multiply in numbers , heal wounds or help to fight off parasites depending on the signals they receive from their environment . Conversely , when macrophages sense pathogens such as bacteria they can also become M1 macrophages , which produce inflammatory molecules that help kill the invading bacteria . As a macrophage transforms into a more specialized state , its metabolism – the set of chemical reactions the cell performs in order to survive and thrive – also changes . This shift appears to play an important role in activating the macrophages and determining how they’ll specialize . However , little is known about how metabolism exerts this control . The metabolism of a cell can be investigated in part by studying the molecules , or “metabolites” , that the cell produces . Covarrubias et al . studied what happens when unspecialized macrophages from mice were activated by a signaling molecule called IL-4 . This signaling molecule causes the cells to become M2 macrophages , and the experiments revealed that IL-4 signaling controls the amount of a metabolite called acetyl-CoA in the cells . Acetyl-CoA can influence how the DNA of a gene is packaged in a cell , and thus affect whether a gene is switched on and “expressed” or not . Covarrubias et al . therefore also analyzed a major metabolic sensing pathway – the Akt-mTORC1 pathway – and showed how this pathway was able to act as a nutrient sensor for the macrophage and control the enzyme responsible for making acetyl-CoA . Therefore , the Akt-mTORC1 pathway can control the level of gene expression changes in the macrophages as a result of IL-4 signaling . The analysis showed that the increase in acetyl-CoA levels increases the expression of some of the genes that cause the M2 macrophages to change state and develop their specialist behaviors . However , only a subset of these genes – those that encode metabolically demanding activities such as immune cell trafficking – have their expression controlled in this way . Further studies are now needed to investigate whether other macrophage types use the same pathways to control their responses .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "immunology", "and", "inflammation" ]	2016	Akt-mTORC1 signaling regulates Acly to integrate metabolic input to control of macrophage activation
The zebrafish was used to assess the impact of social isolation on behaviour and brain function . As in humans and other social species , early social deprivation reduced social preference in juvenile zebrafish . Whole-brain functional maps of anti-social isolated ( lonely ) fish were distinct from anti-social ( loner ) fish found in the normal population . These isolation-induced activity changes revealed profound disruption of neural activity in brain areas linked to social behaviour , social cue processing , and anxiety/stress . Several of the affected regions are modulated by serotonin , and we found that social preference in isolated fish could be rescued by acutely reducing serotonin levels . Social preference behaviour , the drive for individuals to identify and approach members of their own species ( Rogers-Carter et al . , 2018; Winslow , 2003 ) , is a fundamental component of all social behaviour . We previously found that most zebrafish develop a strong social preference by 2–3 weeks of age ( Dreosti et al . , 2015 ) , yet we also found a small number ( ~10% ) of ‘loner’ fish that were averse to social cues . A similar diversity of individual social preferences has been found in many species , including humans ( Sloan Wilson et al . , 1994 ) . Loneliness , undesired isolation from social interaction , has been linked to a reduction in social preference ( Engeszer et al . , 2004; Shams et al . , 2018 ) . We therefore asked whether the socially-averse loner fish found in the normal population would show a similar behavioural phenotype and neuronal activity to socially-averse lonely fish raised in isolation . To answer this question , we compared the behavioural and functional responses of isolated fish to controls during viewing of conspecifics . This comparison found that isolation induces patterns of brain activity that are not present in the normal population . We then asked if we could rescue the aversive behaviour of isolated fish . Since some of the highly activated areas in isolated fish are serotoninergic , we used Buspirone , a 5HT1A receptor agonist . These findings will have important implications for how we understand and treat the impact of social isolation . Prolonged periods of social isolation are particularly detrimental to humans during early development . However , even brief periods of social isolation have been shown to impact mental and physical health . We therefore tested two models of social isolation , Full ( fish raised completely without social interaction ) and Partial ( fish isolated for 48 hr prior to behavioural testing ) . Each experiment comprised two sessions , 15 min of acclimation to the chamber followed by 15 min of exposure to two size matched sibling fish that were not isolated . To quantify social preference , we calculated a visual preference index ( VPI ) that compares the amount of time fish spend in the chamber nearest the conspecifics versus the opposite chamber where they are visually isolated from social cues ( see Materials and methods ) . Full social isolation ( Fi ) caused a significant decrease in social preference relative to normally raised sibling controls ( C ) ( Figure 1A , left and middle panel: C vs Fi , p=8 . 3e−8 , Mann-Whitney ) . Specifically , there was an increase in the number of individuals that had a large negative VPI . We therefore decided to divide the fish into three sociality groups: a ) anti-social ( -S ) fish with VPIs below −0 . 5; b ) pro-social ( +S ) fish with VPIs above +0 . 5; c ) non-social fish with −0 . 5 < VPI < +0 . 5 . Fish that underwent Partial isolation ( Pi ) , exhibited an intermediate , yet highly significant , change in social preference ( Figure 1A , right panel: C vs Pi , p=2 . 5e−8 , Mann-Whitney ) . As previously reported ( Zellner et al . , 2011 ) , we found that fish raised in isolation were significantly less active than their normally raised siblings during the acclimation period ( Figure 1B: C vs Fi , p=9 . 0e−6; C vs Pi , p=2 . 8e-9 Mann-Whitney ) and during the social viewing session ( Figure 1—figure supplement 1A: left C vs Fi , p=0 . 0001; C vs Pi , p=0 . 004 Mann-Whitney ) . We then divided fish into groups based on their social preference . Interestingly , anti-social fully and partially isolated fish showed very similar movement activity compared to anti-social controls during the acclimation ( Figure 1C left: C ( -S ) vs Fi ( -S ) , p=0 . 17 Mann-Whitney; C ( -S ) vs Pi ( -S ) p=0 . 23 Mann-Whitney ) and during the social viewing session ( Figure 1—figure supplement 1B left: C ( -S ) vs Fi ( -S ) , p=0 . 48 Mann-Whitney; C ( -S ) vs Pi ( -S ) p=0 . 10 Mann-Whitney ) . The pro-social isolated fish , which also exhibited a reduction in activity relative to controls during the acclimation session ( Figure 1C right: C ( -S ) vs Fi ( -S ) , p=8 . 0e−5 Mann-Whitney; C ( -S ) vs Pi ( -S ) p=1 . 0e−7 Mann-Whitney ) , instead showed similar activity relative to controls during social viewing ( Figure 1—figure supplement 1B right: C ( -S ) vs Fi ( -S ) , p=0 . 02 Mann-Whitney; C ( -S ) vs Pi ( -S ) p=0 . 14 Mann-Whitney ) . In addition , we noticed that all isolated fish behaved qualitatively differently , exhibiting prolonged periods of quiescence ( freezing ) even when observing conspecifics ( Figure 1D and Video 1 ) . Freezing is a hallmark of anxiety-like behaviour observed in many species , and reported in zebrafish exposed to stressors ( Giacomini et al . , 2015; Shams et al . , 2018 ) , including periods of social isolation ( Egan et al . , 2009; Shams et al . , 2017 ) . In order to quantify freezing behaviour , we measured the percentage of time spent in continuous periods ( >3 s ) without motion ( Figure 1—figure supplement 1C-D ) . We found that both fully and partially isolated fish exhibited significantly more freezing than controls during the acclimation period ( Figure 1—figure supplement 1C left: C vs Fi , p=3 . 4e−16 Mann-Whitney; C vs Pi , p=2 . 8e−5 Mann-Whitney ) , and that this increase relative to controls persisted for fully isolated fish during social viewing , but was reduced in partially isolated fish , perhaps representing some recovery during the 15 min of social interaction ( Figure 1—figure supplement 1D left: C vs Fi , p=6 . 3e−13 Mann-Whitney; C vs Pi , p=0 . 03 Mann-Whitney ) . When we compared freezing behaviour of groups with similar social preference , we found , as expected , that anti-social fish exhibited increased freezing during social viewing regardless of rearing condition . However , pro-social fully isolated fish also showed increased freezing during social viewing , suggesting that they were not engaged in typical social interaction , but rather remained immobile on the side with the conspecifics ( Figure 1—figure supplement 1D right ) . The behavioural similarities between anti-social isolated ( lonely ) and anti-social control ( loner ) fish led us to hypothesize that isolation might simply predispose fish to the same anti-social state found in the normal population . If this is the case , neural activity of anti-social isolated and anti-social control fish should be similar when presented with social cues . To test this hypothesis , we performed whole-brain two-photon imaging of c-fos expression , an immediate early gene whose expression is associated with increased neural activity ( Herrera and Robertson , 1996 ) , in juvenile brains following testing in the social preference assay . Dissected brains were imaged with the dorsal surface down ( bottom-up ) to achieve clear views of the ventral brain structures that have been previously implicated in the social brain network ( Figure 2A , also see Materials and methods ) . Volumes of 1 . 5 mm x 1 . 5 mm x 700 µm , with a voxel size of 1 × 1×3 µm , were acquired from 135 zebrafish brains across all experimental groups and registered to a reference brain ( Marquart et al . , 2017 ) . These c-fos whole-brain functional maps were first normalised to a background intensity level ( see Materials and methods ) and then used to compare the neural activity patterns of different test groups . We compared the average activity map for each rearing/sociality condition with the average map acquired from similarly raised sibling fish that were placed in the behavioural assay for 30 min without any social cues ( nsc , no social-cue ) . The resulting normalised difference stacks ( e . g . ( +S - nsc ) /nsc ) allowed us to identify changes in neural activity associated with exposure to a visual social cue ( Figure 2A ) . Several brain areas showed strong activation or inhibition in normally raised fish upon social cue exposure . We focused on areas that have been reported as social brain areas ( O'Connell and Hofmann , 2011 ) and show differences between our experimental groups ( Figure 2B: C ( +S and -S ) ) . The caudal hypothalamus was differentially activated in pro- vs . anti-social control fish . A dorsal sub-region was significantly activated in pro-social controls ( Figure 2B and D: dHc - C ( +S ) vs C ( nsc ) , p=0 . 007 , Mann-Whitney ) , whereas it was inhibited , along with the adjacent ventral sub-region , in anti-social controls ( Figure 2B and D: vHc - C ( -S ) vs C ( nsc ) , p=0 . 003 , Mann-Whitney ) . The caudal hypothalamus is known to express high levels of serotonin and dopamine , as well as glutamate and histamine ( Filippi et al . , 2010; Kaslin and Panula , 2001 ) . Furthermore , a segregation into distinct dorsal and ventral areas of the caudal hypothalamus has already been shown for some of these markers , such as tyrosine hydroxylase 1 and 2 , ( Th1 and Th2 ) ( Yamamoto et al . , 2010 ) and we confirmed these previous results with immunostaining ( Figure 2C left ) , as well as for the dopamine and serotonin transporters , DAT and slc6a4b ( Figure 2C right ) ( Filippi et al . , 2010; Lillesaar , 2011 ) . Changes in serotonin and dopamine levels have been widely documented in response to social interaction ( Scerbina et al . , 2012 ) , viewing social cues ( Saif et al . , 2013 ) , and social isolation ( Huang et al . , 2015; Shams et al . , 2018; Shams et al . , 2015 ) . While the serotoninergic system has been linked to stress and arousal ( Backström and Winberg , 2017 ) , the dopamine circuitry has been shown to regulate the reward system underlying social behaviour ( Teles et al . , 2013 ) . Since the caudal hypothalamus expresses both of these neurotransmitters , and our data demonstrate a pattern of activation/inhibition that is distinct for pro- and anti-social fish , then this area could be crucial in regulating social preference . The second social brain area we investigated was the preoptic area . Our data showed a similar activation pattern for anti-social and pro-social fish characterised by a small increase in the dorsal preoptic area ( dPa ) and a small decrease in the ventral preoptic area ( vPa ) . However , only anti-social control fish showed a significant change in the ventral area ( Figure 2B and D: C ( -S ) vs C ( nsc ) , vPa p=0 . 003 , Mann-Whitney ) . The activation of the preoptic area during social behaviour is consistent with previous literature in a number of species ( O'Connell and Hofmann , 2011 ) . This area has been shown to express several neuropeptides involved in social behaviour such as arginine/vasotocin and oxytocin ( Heinrichs et al . , 2009; Herget and Ryu , 2015 ) . It was recently shown that oxytocin does not seem to be responsible for social interaction ( Ribeiro et al . , 2019 ) as mutants for oxytocin receptors shows no alteration in social preference , but rather reduced social recognition . Furthermore , injections of oxytocin do not have any effect on shoaling and interaction ( Langen et al . , 2015 ) . The neuropeptide vasotocin , instead , has been shown to have a specific effect on reducing social interaction ( Langen et al . , 2015 ) and not shoaling behaviour . This neuropeptide has also been shown to be involved in aggression ( Teles et al . , 2016 ) and stress by stimulating cortisol release . We then compared the brain activity maps of anti- and pro-social control fish with fully and partially isolated fish . As described previously , anti-social control ( loner ) fish showed a behavioural phenotype very similar to anti-social isolated ( lonely ) fish . Therefore , we investigated whether their brain activity maps were also similar following the presentation of a social cue . Contrary to our hypothesis , c-fos functional maps of anti-social fully isolated fish ( Figure 2B: Fi ( -S ) ) revealed a completely different activity profile than their anti-social sibling controls ( Figure 2B: C ( -S ) ) . The ventral sub-region of the caudal hypothalamus ( vHc ) of Fi ( -S ) fish was not inactivated , while the preoptic area was strongly activated in both the dorsal ( dPa ) and the ventral ( vPa ) regions , but significantly only in the dorsal ( Figure 2B and D: Fi ( -S ) vs Fi ( nsc ) , p=0 . 006 dPa; p=0 . 07 vPa , Mann-Whitney ) . Furthermore , the pro-social fully isolated fish ( Figure 2B: Fi ( +S ) ) , who exhibited an increase of freezes and reduced motility compared to control fish when viewing conspecifics , showed a similar activation to pro-social controls in the caudal hypothalamus , but increased activity in the dorsal preoptic area . Interestingly , the preoptic area was activated differently in pro-social and anti-social isolated fish , with only the dorsal preoptic area strongly activated in the pro-social group ( Figure 2B and D: Fi ( +S ) vs Fi , p=0 . 04 vPa , p=0 . 002 dPa , Man-Whitney ) . These data suggest that long social isolation causes abnormal neural responses during viewing of social cues . Furthermore , anti- and pro-social fish exposed to a brief isolation for only 48 hr prior to testing , showed similar functional activity changes to fully isolated fish , albeit less strong ( Figure 2B and D: Pi ( -S ) vs Pi ( nsc ) , p=0 . 18 dHc; p=0 . 28 vHc; p=0 . 04 vPa; p=0 . 04 dPa , Mann-Whitney; Figure 2B and D: Pi ( +S ) vs Pi ( nsc ) , p=0 . 17 dHc; p=0 . 05 vHc; p=0 . 007 vPa; p=0 . 006 dPa ) . Together with the behavioural data , this finding supports the idea that short term isolation is enough to induce brain activity changes similar to those observed following complete isolation , and strikingly different than those observed in anti-social controls . We were next interested in understanding why social isolation promotes social aversion instead of increasing the drive for social interaction . An important clue was found in the pattern of brain activity changes that were unique to isolated fish . When we directly compared the normalised c-fos functional brain maps of isolated and control fish that were not exposed to social cues during the assay ( Figure 3A ) , we found a significant increase in two interesting areas; one associated with visual processing , the optic tectum , [McDowell et al . , 2004] ) , and one involved in stress responses , the posterior tuberal nucleus ( Ziv et al . , 2013 ) . In pro-social control fish , viewing social cues resulted in a significant increase of neuronal activity in the optic tectum ( Figure 3B top: C ( +S ) vs C ( nsc ) , p=0 . 004 Mann-Whitney ) . However , in fully isolated fish , there was already increased neuronal activity in the optic tectum in the absence of social cues ( Figure 3B top: Fi ( nsc ) vs C ( nsc ) , p=0 . 0004 , Mann-Whitney ) , suggesting that isolation increases visual sensitivity , as previously reported in humans ( Cacioppo et al . , 2015 ) . This increased sensitivity of fully isolated fish not presented with social cues was weaker in partially isolated fish ( Figure 3B top: Pi ( nsc ) vs C ( nsc ) , p=0 . 03 , Mann-Whitney ) . However , a much larger increase in tectal activity was observed when pro-social partially isolated fish viewed conspecifics , revealing that some visual sensitization had occurred ( Figure 3B top: Pi ( +S ) vs C ( +S ) , p=0 . 0002 , Mann-Whitney ) . In addition , increased tectal activity was also present in both fully and partially isolated anti-social fish ( Figure 3B top: Fi ( -S ) vs C ( -S ) , p=0 . 048; Pi ( -S ) vs C ( -S ) , p=0 . 005 , Mann-Whitney ) , even though these fish largely avoided the chamber with visual access to conspecifics . We also observed isolation-related activity increases in the posterior tuberal nucleus , an area associated with stress responses in zebrafish ( Wee et al . , 2019; Ziv et al . , 2013 ) . Full isolation caused a significant increase in posterior tuberal nucleus activity in the absence of social cues ( Figure 3B bottom: Fi ( nsc ) vs C ( nsc ) , p=0 . 015 , Mann-Whitney ) and in both anti-social and pro-social fish exposed to social cues ( Figure 3B bottom: Fi ( +S ) vs C ( +S ) , p=0 . 003; Fi ( -S ) vs C ( -S ) , p=0 . 016 , Mann-Whitney ) . Following partial isolation , posterior tuberal nucleus activity was not increased in the absence of social cues ( Figure 3B bottom: Pi ( nsc ) vs C ( nsc ) , p=0 . 29 , Mann-Whitney ) , only slightly in pro-social fish ( Figure 3B bottom: Pi ( +S ) vs C ( +S ) , p=0 . 018 ) , but significantly so in anti-social fish ( Figure 3B bottom: Pi ( -S ) vs C ( -S ) , p=0 . 0005 ) . Given these results from the optic tectum and posterior tuberal nucleus , we propose that isolation initially heightens sensitivity to social stimuli . However , when prolonged , this heightened sensitivity results in an increase of stress and anxiety levels during social viewing that leads to an aversion for social stimuli . To test our hypothesis that reducing anxiety could reverse the anti-social behaviour observed in isolated zebrafish , we acutely treated control and partially isolated fish with Buspirone , an agonist of the auto-inhibitory 5HT1A receptor . The choice of isolation duration was motivated by the intermediate behavioural and functional phenotype of partial isolation relative to normal-rearing and full isolation , which would allow us to more easily detect both positive and negative impacts of treatment on sociality . The choice of Buspirone was supported by the changes in activity observed in the caudal hypothalamus of isolated fish , and by the fact that the caudal hypothalamus and the preoptic area strongly express Htr1ab receptors , one of the two orthologues of the 5HT1A receptor ( Norton et al . , 2008 ) . Buspirone has been shown to reduce anxiety in humans , mice , and zebrafish ( Bencan et al . , 2009; Lalonde and Strazielle , 2010; Lau et al . , 2011; Patel and Hillard , 2006 ) . While it is not fully understood how Buspirone reduces anxiety , it has been shown to enhance social interaction of rats ( File and Seth , 2003; Gould et al . , 2011 ) , sociability of zebrafish ( Barba-Escobedo and Gould , 2012 ) , and reduce social phobia in humans ( Schneier et al . , 1993; van Vliet et al . , 1997 ) . Its ability to counter the effects of social isolation in zebrafish has not been investigated . We first tested the effects of acute exposure to Buspirone in control fish , and , as expected , we observed a small significant increase in social preference relative to untreated controls , however , a population of ~10% anti-social fish remained ( Figure 4—figure supplement 1; C ( no drug ) vs C ( 30 µM ) , p=0 . 01 , Mann-Whitney ) . We then treated partially isolated fish with 30 µM and 50 µM ( Figure 4—figure supplement 1 , n = 46 , n = 72 fish ) of Buspirone . Remarkably , the acute drug treatment was sufficient in both concentrations to reverse the anti-social phenotype caused by isolation ( Figure 4A; Pi vs Pi ( Buspirone 30 µM and 50 µM combined ) , p=2 . 56 e-05 , Mann-Whitney ) . When we then compared the time course of this phenotype reversal by computing the VPIs for each minute throughout the 15 min of the behavioural experiment ( Figure 4B ) . We found that the isolated fish treated with Buspirone , while initially anti-social , would rapidly recover normal social preference behaviour within the first 5 min of exposure to social cues ( Figure 4B: C vs Pi ( Buspirone ) , p=0 . 016 , first minute; p=0 . 37 , fourth minute , Mann-Whitney ) . In contrast , the VPIs of untreated isolated fish remained significantly lower than controls throughout the entire session . We next compared the time course of movement activity ( Figure 4C ) , and found that it generally increased quickly throughout the first 5 min of the social viewing session . Notably , the activity of isolated fish treated with Buspirone was already at the level of controls from the start of the social viewing session ( Figure 4B: C vs Pi ( Buspirone ) , p=0 . 31 , first minute , Mann-Whitney ) , which suggests that the recovery of normal movement activity , possibly as a result of reduced anxiety , precedes the recovery of normal social preference . Therefore , Buspirone’s impact on the rate of recovery of social preference indicates that it may do so by reducing anxiety , perhaps at the level of the preoptic and/or caudal hypothalamic area , allowing circuit plasticity to down-regulate the hypersensitivity to social stimuli acquired during the isolation period . In summary , our results demonstrate that lonely fish , which have been isolated from social cues and show anti-social behaviour , have a completely different functional response to social stimuli than loner fish , anti-social fish found in the normal population . In addition , the functional changes caused by social deprivation are consistent with an increase in anxiety resulting from hyper-sensitization to social stimuli , similar to the effects of isolation on humans . We could reverse isolation’s effects in zebrafish with an existing anxiolytic drug that acts on the monoaminergic system . Zebrafish will thus provide a powerful new tool for studying the impact of loneliness ( isolation ) on brain function and exploring different strategies for reducing , or even reversing , its harm . AB strain zebrafish maintenance and breeding was performed at 28 . 5C with a 14 hr:10 hr light-dark cycle . Isolated fish were housed in custom chambers ( length = 15 cm , width = 5 cm , height = 10 cm ) made of opaque white acrylic with translucent lids , either from fertilization ( full isolation ) or for 48 hr prior to the behavioural experiment ( partial isolation ) . All experiments were performed according to protocols approved by local ethical committee ( AWERB Bloomsbury Campus UCL ) and the UK Home Office . Experimental details and image acquisition were performed as described previously ( Dreosti et al . , 2015 ) . Fish were positioned in custom-built behavioural arenas ( Figure 1D ) made of white acrylic , and illuminated with visible light using a laser light projector ( Microvision , ShowwX+ , US ) . The videography system comprised a high-speed camera ( Flea3 , PointGrey , CA ) , an infrared light ( Advanced Illumination , US , 880 nm ) , an IR filter ( R70 , Hoya , JP ) , and a vari-focal lens ( Fujinon , JP ) . Experiments were recorded using custom written workflows in Bonsai ( Langen et al . , 2015 ) . Test fish were positioned in the main C-shape compartment of the arena by pipetting , and left for 15 min to acclimate . A social cue , two fish of same age and similar size , was then introduced into one of the two adjacent chambers randomly . Test fish could see the social cue through a glass window . Each fish was run only once in the behavioural assay . Images were analysed using custom written computer vision scripts in Python based on OpenCV ( https://www . dreo-sci . com/resources/ ) . Each frame was cropped , background subtracted , and thresholded . The centroid , position , orientation , and per frame motions of the test fish were identified , and stored in a CSV file . All videos have been saved with H . 264 compression for subsequent offline analysis , and are available upon request . The source code can be downloaded at http://www . dreo-sci . com/resources/ . The visual preference index ( VPI ) was calculated by subtracting the number of frames in which the fish was located on the side of the arena nearest the social stimulus ( Social cue ( SC ) side in Figure 1B ) by the number of frames located on the opposite side of the arena ( nsc ( No SC ) side ) . This difference was then divided by the total number of frames recorded [VPI = ( SC – No SC ) /Total frames] . The percent time moving was calculated by counting each frame with detectable changes in the fish image relative to the previous frame ( i . e . motion ) , and dividing by the total number of frames . The percent time freezing was calculated by detecting contiguous sequences without motion longer than 3 s , counting all frames that are part of such sequences , and dividing by the total number of frames . Fluorescent in situ hybridizations using digoxigenin-labelled c-fos were performed on dissected juvenile zebrafish with few modification to the original method ( Brend and Holley , 2009 ) . After overnight fixation in 4% PFA , protein K treatment ( 2 mg/ml 20 min of incubation ) , inactivation of endogenous peroxidase with H2O2 ( 22% v/v for 30 min at room temperature ) , additional fixation ( 30 min at room temperature ) and 3 hr of incubation with the hybridisation buffer , fish were incubated with the c-fos probe ( courtesy from Ricardo N . Silva ( Forward CCGATACACTGCAAGCTGAA and Reverse ATTGCAGGGCTATGGAAGTG ) , or with dopamine transporter ( DAT ) , tyrosine hydroxylase 1 ( Th1 ) , tyrosine hydroxylase ( Th2 ) ( Filippi et al . , 2010 ) , or the 5‐HT transporter , solute carrier family 6 member 4b ( Slc6a4b ) probes ( Norton et al . , 2008 ) . C-fos , DAT and Slc6a4b probes were detected with anti-Digoxigenin-POD , Fab fragments ( Roche , 1:3000 ) and TSA Plus Cyanine 3 System ( Perkin Elmer , 1:50 ) . Nuclear staining was obtained using DAPI ( Sigma-Aldrich , 1: 500 ) . Fish were then mounted for imaging in low melting point agarose ( 2 . 5% Agarose , 0 . 8% glycerol , PBS-Tween ) and imaged . A custom built two-photon microscope ( INSS ) was used for image acquisition of whole-brain in situs . Both DAPI and Cy3 Images were collected with a 10x objective ( Olympus , W Plan-Apochromat 10x/0 . 5 M27 75 mm ) using a ‘Chameleon’ titanium–sapphire laser tuned to 1030 nm ( Coherent Inc , Santa Clara , CA , US ) and controlled using custom written software in LabView . Registration of in-situ images was performed using ANTs ( Advanced Normalisation Tools ) version 2 . 1 . 0 running on the UCL Legion compute cluster . Images were down-sampled to 512512 and parameters were slightly modified from Marquart et al . ( 2017 ) fixed registration:antsRegistration -d 3 --float 1 -o [Registered_Image_ , Registered_Image _warped . nii . gz] --interpolation WelchWindowedSinc --use-histogram-matching 0 r [reference_Image , Registered_Image , 1] -t rigid[0 . 1] -m MI[reference_Image , Registered_Image _0 . nii , 1 , 32 , Regular , 0 . 25] -c [1000 × 500×250 × 100 , 1e-8 , 10] --shrink-factors 12 × 8×4 × 2 s 4 × 3×2 × 1 t Affine[0 . 1] -m MI[reference_Image , Registered_Image , 1 , 32 , Regular , 0 . 25] -c [1000 × 500×250 × 100 , 1e-8 , 10] --shrink-factors 12 × 8×4 × 2 s 4 × 3×2 × 1 t SyN[0 . 1 , 6 , 0] -m CC[reference_Image , Registered_Image _0 . nii , 1 , 2] -c [1000 × 500×500x250 × 100 , 1e-7 , 10] --shrink-factors 12 × 8×4x2 × 1 s 4 × 3×2x1 × 0 antsApplyTransforms -d 3 v 0 --float -n WelchWindowedSinc -i Registered_Image _1 . nii -r reference_Image -o Registered_Image _warped_red . nii . gz -t Registered_Image _1Warp . nii . gz -t Registered_Image _0GenericAffine . mat The registered image stacks were then normalised to adjust for intensity variations between imaging sessions caused by a variety of sources ( staining efficiency , laser power fluctuations , light detector sensitivity , etc . ) . Normalisation was accomplished by computing an intensity histogram for each fish brain’s volume ( with 10000 discrete intensity bins spanning the range −4000 . 0 to 70000 . 0 ) for all 512512*273 voxels . The minimum value bin ( with at least 100 voxels ) was used as the bias offset , and subtracted from all voxel values . The mode value , minus the bias , provided a robust estimate of the background/baseline fluorescence and was thus used to normalise voxel values for the entire volume . Therefore , after normalisation , an intensity value of 1 reflected the background level while two indicates fluorescence level that is twice the background , and so on . Histogram normalisation was performed for each individual fish’s brain volume prior to any region or voxel-based analysis . Figures 2B and 3A Reconstruction of cross section images were obtained by using the Fiji ‘Volume viewer’ plugin . Schematics of cross- and horizontal-section were obtained by using the ‘Neuroanatomy of the zebrafish brain’ . Figures 2D Percentages of c-fos activation were calculated for each of the six different areas highlighted in Figures 2B and 3A , using custom written Python functions , in the following way . A 3D mask for each area was generated by using the ‘Segmentation Editor’ plugin Fiji ( https://imagej . net/Segmentation_Editor ) . C-fos percentage values for each condition ( C ( +S ) , C ( -S ) , Fi ( -S ) , Pi ( -S ) ) were obtained by subtracting and then dividing each c-fos average value of the mask by the basal c-fos average value calculated in control fish No Social Cue . Statistical analysis was performed using Python scipy stats libraries . Since VPI , percent time moving/freezing , and c-fos activity distributions were generally not normally distributed , we used the non-parametric Mann-Whitney U-test of independent samples for hypothesis testing throughout the manuscript . Juvenile fish were treated with 30 µM or 50 µM Buspirone ( Buspirone HCl , Sigma ) for 10 min prior the experiment . After washing , fish were run through the behavioural assay . Each fish was used only once . All the images , video , protocols , analysis scripts , and data that support the findings of this study are available from this website ( http://www . dreo-sci . com/resources/ ) , or our GitHub repository ( https://github . com/Dreosti-Lab/Lonely_Fish_2020; Dreosti , 2020; copy archived at https://github . com/elifesciences-publications/Lonely_Fish_2020 ) , or from the corresponding author upon request .	Socialising is good for people’s mental health and wellbeing . The connections and relationships that we form can make us more resilient and healthier . Researchers also know that prolonged periods of social isolation , and feeling lonely , can be detrimental to our health , especially in early childhood . The paradox is that loneliness often results in an even lower desire for social contact , leading to further isolation . But not everyone craves social contact . Some people prefer to be alone and feel more comfortable avoiding social interaction . Zebrafish display the same social preferences . This , along with their transparent brains , makes them a useful model to study the links between social behaviour and brain activity . Like humans , zebrafish are social animals , with most fish taking a strong liking to social interactions by the time they are a few weeks old . A small number of ‘loner’ fish , however , prefer to avoid interacting with their siblings or tank mates . And so , if loneliness quells the desire for more social contact , the question becomes , does isolation turn otherwise social fish into loners ? Here , Tunbak et al . use zebrafish to study how social isolation changes brain activity and behaviour . Social fish were isolated from others in the tank for a few days . These so-called ‘lonely fish’ were then allowed back in contact with the other fish . This revealed that , after isolation , previously social fish did avoid interacting with others . With this experimental set-up , Tunbak et al . also compared the brains of lonely and loner fish . When fish that prefer social interaction were deprived of social contact , they had increased activity in areas of the brain related to stress and anxiety . These lonely fish became anxious and very sensitive to stimuli; and their brain activity suggested that social interaction became overwhelming rather than rewarding . Positively , the lonely fish quickly recovered their normal , social behaviour when given a drug that reduces anxiety . This work provides a glimpse into how human behaviour could be affected by lengthy periods in isolation . These results suggest that humans could feel anxious upon returning to normal life after spending a long time alone . Moreover , the findings show the impact that social interaction and isolation can have on the young , developing brain .	[ "Abstract", "Introduction", "Materials", "and", "methods" ]	[ "short", "report", "neuroscience" ]	2020	Whole-brain mapping of socially isolated zebrafish reveals that lonely fish are not loners
Many studies of visual processing are conducted in constrained conditions such as head- and gaze-fixation , and therefore less is known about how animals actively acquire visual information in natural contexts . To determine how mice target their gaze during natural behavior , we measured head and bilateral eye movements in mice performing prey capture , an ethological behavior that engages vision . We found that the majority of eye movements are compensatory for head movements , thereby serving to stabilize the visual scene . During movement , however , periods of stabilization are interspersed with non-compensatory saccades that abruptly shift gaze position . Notably , these saccades do not preferentially target the prey location . Rather , orienting movements are driven by the head , with the eyes following in coordination to sequentially stabilize and recenter the gaze . These findings relate eye movements in the mouse to other species , and provide a foundation for studying active vision during ethological behaviors in the mouse . Across animal species , eye movements are used to acquire information about the world and vary based on the particular goal ( Yarbus , 1967 ) . Mice , a common model system to study visual processing due to their genetic accessibility , depend on visual cues to successfully achieve goal-directed tasks in both artificial and ethological freely-moving behavioral paradigms , such as the Morris water maze , nest building , and prey capture; ( Morris , 1981; Clark et al . , 2006; Hoy et al . , 2016 ) . It is unclear , however , how mice regulate their gaze to accomplish these visually mediated goals . Previous studies in both freely moving rats and mice have shown that eye movements largely serve to compensate for head movements ( Wallace et al . , 2013; Payne and Raymond , 2017; Meyer et al . , 2018; Meyer et al . , 2020 ) , consistent with the vestibulo-ocular reflex ( VOR ) present in nearly all species ( Straka et al . , 2016 ) . While such compensation can serve to stabilize the visual scene during movement , it is not clear how this stabilization is integrated with the potential need to shift the gaze for behavioral goals , particularly because mice lack a specialized central fovea in the retina , and also have laterally facing eyes resulting in a relatively limited binocular field ( roughly 40° as opposed to 135° in humans [Dräger , 1978] ) . In addition , because eye movements are altered in head-fixed configurations due to the lack of head movement ( Payne and Raymond , 2017; Meyer et al . , 2020 ) , understanding the mechanisms of gaze control and active visual search benefits from studies in freely moving behaviors . Prey capture can serve as a useful paradigm for investigating visually guided behavior . Recent studies have shown that mice use vision to accurately orient towards and pursue cricket prey ( Hoy et al . , 2016 ) , and have begun to uncover neural circuit mechanisms that mediate both the associated sensory processing and motor output ( Hoy et al . , 2019; Shang et al . , 2019; Zhao et al . , 2019; Han et al . , 2017 ) . Importantly , prey capture also provides a context to investigate how mice actively acquire visual information , as it entails identifying and tracking a localized and ethological sensory input during freely moving behavior . Here , we asked whether mice utilize specific eye movement strategies , such as regulating their gaze to maximize binocular overlap , or actively targeting and tracking prey . Alternatively , or in addition , mice may use directed head movements to target prey , with eye movements primarily serving a compensatory role to stabilize the visual scene . Predators typically have front-facing eyes which create a wide binocular field through the overlap of the two monocular fields , allowing for depth perception and accurate estimation of prey location ( Cartmill , 1974 ) . Prey species , in contrast , typically have laterally facing eyes , and as a result , have large monocular fields spanning the periphery , which allow for reliable detection of approaching predators . Though mice possess these characteristics of prey animals , they also act as predators in pursuing cricket prey ( Hoy et al . , 2016 ) . How then do animals with laterally placed eyes target prey directly in front of them , especially when these targets can rapidly move in and out of the narrow binocular field ? This could require the modulation of the amount of binocular overlap , through directed lateral eye movements , to generate a wider binocular field , such as in the case of cuttlefish ( Feord et al . , 2020 ) , fish ( Bianco et al . , 2011 ) , many birds ( Martin , 2009 ) , and chameleons ( Katz et al . , 2015 ) . In fact , these animals specifically rotate their eyes nasally before striking prey , thereby creating a larger binocular zone . However , it is unknown whether mice use a similar strategy during prey capture . Alternatively , they may use coordinated head and eye movements to stabilize a fixed size binocular field over the visual target . Foveate species make eye movements that center objects of interest over the retinal fovea , in order to use high acuity vision for complex visual search functions including identifying and analyzing behaviorally relevant stimuli ( Hayhoe and Ballard , 2005 ) . Importantly , afoveate animals ( those lacking a fovea ) represent a majority of vertebrate species , with only some species of birds , reptiles , and fish possessing distinct foveae ( Harkness and Bennet-Clark , 1978 ) , and among mammals , only simian primates possessing foveae ( Walls , 1942 ) . It remains unclear whether mice , an afoveate mammalian species , actively control their gaze to target and track moving visual targets using directed eye movements , or whether object localization is driven by head movements . We therefore aimed to determine the oculomotor strategies that allow for effective targeting of a discrete object , cricket prey , within the context of a natural behavior . Recent studies have demonstrated the use of head-mounted cameras to measure eye movements in freely moving rodents ( Wallace et al . , 2013; Meyer et al . , 2018; Meyer et al . , 2020 ) . Here , we used miniature cameras and an inertial measurement unit ( IMU ) to record head and bilateral eye movements while unrestrained mice performed a visually guided and goal-directed task , approach and capture of live insect prey . We compared the coordination of eye and head movements , as well as measurements of gaze relative to the cricket prey during approach and non-approach epochs , to determine the oculomotor strategies that mice use when localizing moving prey . Food-restricted mice were habituated to hunt crickets in an experimental arena , following the paradigm of Hoy et al . , 2016 . To measure eye and head rotations in all dimensions , mice were equipped with two reversibly attached , lateral head-mounted cameras and an inertial measurement unit ( IMU ) board with an integrated 3-dimensional accelerometer and gyroscope ( Figure 1A and B; Video 1 ) . The estimated error in measurement of head and eye angle were both less than one degree ( see Materials and methods ) . In addition , we recorded the behavior of experimental animals and the cricket prey with an overhead camera to compute the relative position of the mouse and cricket , as well as orientation of the head relative to the cricket . Following our previous studies ( Hoy et al . , 2016; Hoy et al . , 2019 ) , we defined approaches based on the kinematic criteria that the mouse was oriented towards the cricket and actively moving towards it ( see Materials and methods ) . Together , these recordings and analyses allowed us to synchronously measure eye and head rotations along with cricket and mouse kinematics throughout prey capture behavior ( Figure 1C; Video 1 ) . The cameras and IMU did not affect overall mouse locomotor speed in the arena or total number of crickets caught per 10 min session ( paired t-test , p=0 . 075; Figure 1D/E ) , suggesting that placement of the cameras and IMU did not significantly impede movement or occlude segments of the visual field required for successful prey capture behavior . To determine whether mice make convergent eye movements to enhance binocular overlap during approaches , we first characterized the coordination of bilateral eye movements . We defined central eye position , that is 0° , as the average pupil location for each eye , across the recording duration . Measurement of eye position revealed that freely moving mice nearly constantly move their eyes , typically within a ± 20 degree range ( Figures 1C , 2A and B ) , as shown previously ( Meyer et al . , 2020; Sakatani and Isa , 2007 ) . Figure 2C shows example traces of the horizontal position of the two eyes ( top ) , along with running speed of the mouse ( bottom ) . As described previously ( Wallace et al . , 2013; Payne and Raymond , 2017; Meyer et al . , 2018 ) and analyzed below ( Figure 3D ) , the eyes are generally stable when the mouse is not moving . In addition , the raw traces reveal a pattern of eye movement wherein rapid correlated movements of the two eyes are superimposed on slower anti-correlated movements . The pattern of rapid congruent movements and slower incongruent movements was also reflected in the time-lagged cross-correlation of the change in horizontal position across the two eyes ( Figure 2E ) , which was positive at short time lags and negative at longer time lags . We next calculated the vergence angle , which is the difference in the horizontal position of the two eyes ( Figure 2D ) . The range of vergence angles was broadly distributed across negative ( converged ) and positive ( diverged ) values during non-approach periods , but became more closely distributed around zero ( neutral vergence ) during approaches ( Figure 2F; paired t-test , p=0 . 024 ) . This can be observed in the individual trace of eye movements before , during , and after an approach ( Figure 2G , top ) , showing that while the eyes converge and diverge outside of approach periods , they move in a more coordinated fashion during the approaches . Thus , mice do not converge their eyes nasally to create a wider binocular field during approaches; rather the eyes are more tightly aligned , but at a neutral vergence , during approaches relative to non-approach periods . Previous studies have demonstrated that eye vergence varies with head pitch ( Wallace et al . , 2013; Meyer et al . , 2018; Meyer et al . , 2020 ) . As the head tilts downwards , the eyes move outwards; based on the lateral position of the eyes , this serves to vertically stabilize the visual scene relative to changes in head pitch ( Wallace et al . , 2013 ) . We therefore sought to determine whether the stabilization of horizontal eye vergence we observed during approaches reflects corresponding changes in head pitch . Consistent with previous studies , we also found eye vergence to covary with head pitch ( Figure 2H ) , such that when the head was vertically centered , the eyes no longer converged or diverged , but were aligned at a neutral vergence ( i . e . , no difference between the angular positions across the two eyes , see schematic in Figure 2D ) . Strikingly , we found that while the relationship between head pitch and vergence was maintained during approaches ( Figure 2H ) , the distribution of head pitch was more centered during approach periods ( Figure 2H and I; paired t-test , p=0 . 0250 ) , indicating a stabilization of the head in the vertical dimension . This can also be seen in the example trace in Figure 2G , where the head pitch becomes maintained around zero during approach . These data show that the increased alignment of the two eyes observed during approaches largely represents the stabilization of up/down head rotation , consequently reducing the need for compensatory vergence movements . Next , we aimed to understand the relationship between horizontal head movements ( yaw ) and horizontal eye movements during approach behavior . In order to isolate the coordinated movement of the two eyes , removing the compensatory changes in vergence described above , we averaged the horizontal position of the two eyes for the remaining analyses ( Figure 3A ) . Changes in head yaw and mean horizontal eye position were strongly negatively correlated at zero time lag ( Figure 3B ) , suggesting rapid compensation of head movements by eye movements , as expected for VOR-stabilization of the visual scene . The distribution of head and eye movements at zero lag ( Figure 3C ) shows that indeed changes in head yaw were generally accompanied by opposing changes in horizontal eye position , represented by the points along the negative diagonal axis . However , there was also a distinct distribution of off-axis points , representing a proportion of non-compensatory eye movements in which the eyes and head moved in the same direction ( Figure 3C ) . Many studies have reported a limited range of infrequent eye movements in head restrained mice ( Payne and Raymond , 2017; Niell and Stryker , 2010; Samonds et al . , 2018; Stahl , 2004 ) , consistent with the idea that eye movements are generally driven by head movement . Correspondingly in the freely moving context of the prey capture paradigm , we found greatly reduced eye movements when the animals were stationary versus when the animals were running ( Figure 3D; Kolmogorov-Smirnov test , p=0 . 032 ) . We next compared the distribution of mean eye position during approaches and non-approach periods . In contrast to the stabilization of head pitch described above , the distribution of head yaw velocities was not reduced during approaches as shown ( Figure 3E; paired t-test p=0 . 938 ) , consistent with the fact that mice must move their heads horizontally as they continuously orient to pursue prey . For both non-approach and approach periods , eye position generally remained within a range less than the size of the binocular zone ( ±20 degrees; Figure 3F , paired t-test , p=0 . 156 ) , suggesting that the magnitude of eye movements would not shift the binocular zone to an entirely new location . Comparison of horizontal eye velocity between non-approach and approach epochs revealed that the eyes move with similar dynamics across both behavioral periods ( Figure 3G , panel 1; paired t-test , p=0 . 155 ) . Additionally , at times when head yaw was not changing , horizontal eye position also did not change ( Figure 3G , panel 2; paired t-test , p=0 . 229 ) . Together , these observations suggest that most coordinated eye movements in the horizontal axis correspond to changes in head yaw , and that the eyes do not scan the visual environment independent of head movements or when stationary . Gaze position - the location the eyes are looking in the world - is a combination of the position of the eyes and the orientation of the head . Compensatory eye movements serve to prevent a shift in gaze , whereas non-compensatory eye movements ( i . e . , saccades ) shift gaze to a new position . Although the vast majority of eye movements are compensatory for head movements , as demonstrated by strong negative correlation in Figure 3B/C , a significant number of movements are not compensatory , as seen by the distribution of off-axis points in Figure 3C . These eye movements will therefore shift the direction of the animal’s gaze relative to the environment . We next examined how eye movements , and particularly non-compensatory movements , contribute to the direction of gaze during free exploration and prey capture . In particular , are these gaze shifts directed at the target prey ? We segregated eye movements into compensatory versus gaze-shifting by setting a fixed gaze velocity threshold of ±180 °/sec , based on the gaze velocity distribution ( Figure 4A ) , which shows a transition between a large distribution around zero ( stabilized gaze ) and a long tail of higher velocities ( rapid gaze shifts ) . This also provides a clear segregation in the joint distribution of eye and head velocity ( Figure 4B ) , with a large number of compensatory gaze-stabilizing movements ( black points ) where eye and head motion are anti-correlated , and much smaller population of gaze shifts ( red ) . This classification approach provides an alternative to standard primate saccade detection ( Andersson et al . , 2017; Stahl , 2004; Matthis et al . , 2018 ) , which is often based on eye velocity rather than gaze velocity , since in the freely moving condition , particularly in afoveate species , rapid gaze shifts ( saccades ) often result from a combination of head and non-compensatory eye movements , rather than eye movements alone ( Land , 2006 ) . We next determined how compensatory and non-compensatory eye movements contribute to the dynamics of gaze during ongoing behavior , by computing the direction of gaze as the sum of eye position and head position . Strikingly , the combination of compensatory and non-compensatory eye movements ( Figure 4C , top ) with continuous change in head orientation ( Figure 4C , middle ) results in a series of stable gaze positions interspersed with abrupt shifts ( Figure 4C , bottom ) . This pattern of gaze stabilization interspersed with rapid gaze-shifting movements , known as ‘saccade-and-fixate , ’ is present across the animal kingdom and likely represents a fundamental mechanism to facilitate visual processing during movement ( Land , 1999 ) . These results demonstrate that the mouse oculomotor system also engages this fundamental mechanism . Durations of fixations between saccades showed wide variation , with a median of 220 ms ( Figure 4D ) . To quantify the degree of stabilization achieved , we compared the root mean square ( RMS ) deviation of gaze position and head yaw during stabilization periods ( Figure 4E ) . This revealed that the gaze is nearly three times less variable than the head ( Figure 4F; median head = 3 . 87 deg; median gaze = 1 . 58 deg; p=0 ) , resulting in stabilization to within nearly one degree over extended periods , even during active approach toward the cricket . Saccade-and-fixate serves as an oculomotor strategy to sample and stabilize the visual world during free movement . In primates , saccades are directed towards specific targets of interest in the visual field . Is this true of the non-compensatory movements in the mouse ? In other words , do saccades directly target the cricket ? To address this , we next analyzed the dynamics of head and gaze movements relative to the cricket position during hunting periods , to compare how accurately the direction of the gaze and the head targeted the cricket during saccades . Figure 5A shows example traces of head and eye dynamics across an approach period ( see also Video 2 ) . Immediately before approaching the cricket , the animal begins a large head turn towards the target , thereby reducing the azimuth angle ( center of the head relative to cricket ) . This head turn is accompanied by a non-compensatory eye movement in the same direction ( Figure 5A , 3rd panel , see mean trace in black ) that accelerates the shift in gaze . Then during the approach , the eyes convert the continuous tracking of the head into a series of stable locations of the gaze ( black sections in Figure 5A , bottom ) . Note also the locking of the relative position of the two eyes ( Figure 5A , 3rd panel , blue and purple ) , as described above in Figure 2 . To determine how head and eye movements target the prey , we computed absolute value traces of head and gaze angle relative to cricket ( head and gaze azimuth ) , and aligned these to the onset of each non-compensatory saccadic eye movement . The average of all traces during approaches revealed that saccades are associated with a head turn towards the cricket , as shown by a decrease in the azimuth angle ( Figure 5B ) . Immediately preceding a saccade , the gaze is stabilized while the head turns , and the saccade then abruptly shifts the gaze . Notably , following the saccade , the azimuth of gaze is the same as the azimuth of the head , suggesting that eye movements are not targeting the cricket more precisely , but simply ‘catching up’ with the head , by re-centering following a period of stabilization . To further quantify this , we assessed the accuracy of the head and gaze at targeting cricket position before and after saccades . Preceding saccades , the distribution of head angles was centered around the cricket , while the gaze less accurately targeted and was offset from the cricket to the left or right ( Figure 5C/5D top; paired t-test; p=8 . 48×10−9 p=2×10−5 ) , due to compensatory stabilization . After the saccade , however , gaze and head were equally targeted towards the cricket ( Figure 5C/D bottom; p=0 . 979p=0 . 4 ) , as the saccade recentered the eyes relative to the head and thereby the cricket . This pattern of stabilizing the gaze and then saccading to recenter the gaze repeats whenever the head turns until capture is successful ( see Video 2 ) . Further supporting a strategy where the head guides targeting , with the eyes following to compensate , we examined how both head and eye movements are correlated with the cricket’s position . At short latencies , the change in head angle relative to the location of the cricket was highly correlated ( Figure 5E ) , indicating that during approach the animal is rapidly reorienting its head towards the cricket . However , the change in gaze with the azimuth instead showed only a weak correlation because the eyes themselves are not always aligned with the azimuth due to stabilization periods ( Figure 5F ) . Together , these results suggest that in mice , tracking of visual objects in freely moving contexts is mediated through directed head movements , and corresponding eye movements that stabilize the gaze and periodically update to recenter with the head as it turns . Here we investigated the coordination of eye and head movements in mice during a visually guided ethological behavior , prey capture , that requires the localization of a specific point in the visual field . This work demonstrates that general principles of coordinated eye and head movements , observed across species , are present in the mouse . Additionally , we address the potential targeting of eye movements towards behaviorally relevant visual stimuli , specifically the moving cricket prey . We find that tracking is achieved through directed head movements that accurately target the cricket prey , rather than directed , independent eye movements . Together , these findings define how mice move their eyes to achieve an ethological behavior and provide a foundation for studying active visually-guided behaviors in the mouse . One potential limitation of our eye tracking system is the 60 Hz framerate of the miniature cameras . This temporal resolution is significantly lower than traditional eye tracking paradigms using videography or eye-coil systems in head-restrained humans , non-human primates , and rodents ( Payne and Raymond , 2017; Sakatani and Isa , 2007 ) , though similar to recent video-based tracking in freely moving rodents ( Meyer et al . , 2018; Meyer et al . , 2020 ) and humans ( Matthis et al . , 2018 ) . We do not expect that this would significantly alter our findings , as the basic parameters of eye movements ( amplitude and speed ) that we found ( Figures 2B , 3C and F ) were similar to measurements made in both head-fixed mice with high-speed videography ( Sakatani and Isa , 2007 ) and freely moving mice with a magnetic sensor ( Payne and Raymond , 2017 ) . However , although we are able to detect peak velocities over 300°/sec , we may still be under-estimating the peak velocity during saccades . Therefore increasing the temporal resolution further could lead to more robust detection of rapid gaze shifts and would potentially enhance classification of saccadic eye movements . We found a pattern of gaze stabilization interspersed with abrupt , gaze-shifting saccades during both non-approach and approach epochs . This oculomotor strategy has been termed ‘saccade-and-fixate’ ( reviewed in Land , 1999 ) , and is present in most visual animal species , from insects to primates , and was recently demonstrated in mice ( Meyer et al . , 2020 ) . In primates , gaze shifts can be purely driven by eye movements , but in other species saccades generally correspond to non-compensatory eye movements during head rotation , suggesting transient disengagement of VOR mechanisms . These saccadic movements are present in invertebrates and both foveate and non-foveate vertebrates ( reviewed in Land , 1999 ) , and work to both recenter the eyes and relocate the position of gaze as animals turn . We found that these brief congruent head and eye movements are interspersed with longer duration ( median ~200 ms ) periods of compensatory movements , which stabilize the gaze to within nearly 1° as the head continues to rotate . Together these eye movements function to create a stable series of images on the retina even during rapid tracking of prey . However , the saccade-and-fixate strategy raises the question of whether mice actively target a specific relevant location with saccadic eye movements . We examined this during periods of active approach toward the cricket to determine whether the eyes specifically target the cricket , relative to head orientation . During approaches , most saccades occur during corrective head turns toward the cricket location . While saccades do bring the gaze closer to the cricket , they do not do so more accurately than the head direction . In fact , prior to the saccade , mice sacrifice centering of the gaze on the target to instead achieve visual scene stability . The eyes then ‘catch up’ to the head as it is rotating ( Figure 5B/C ) . Thus , these eye movements serve to reset eye position with the head , rather than targeting the cricket specifically . Combined with the fact that mice do not make significant eye movements in the absence of head movements ( Figure 3F ) , this suggests that mice do not perform either directed eye saccades or smooth pursuit , which are prominent features of primate vision . On the other hand , the fact that mice use a saccade-and-fixate strategy makes it clear that they are still actively controlling their retinal input , despite low visual acuity . Indeed , the saccade-and-fixate strategy makes mouse vision consistent with the vast majority of species across the animal kingdom . We also examined whether mice make specific vergence eye movements that could serve to modulate the binocular zone , as in some other species with eyes located laterally on the head . We find that rather than moving the eyes nasally to expand the binocular zone , during approach toward the cricket the two eyes become stably aligned , but at a neutral vergence angle that is neither converged or diverged ( Figure 2E ) . While several species with laterally-placed eyes use convergent eye movements during prey capture to create a wider binocular field ( Feord et al . , 2020; Bianco et al . , 2011; Martin , 2009; Katz et al . , 2015 ) , our results show that mice do not utilize this strategy during prey capture . However , vergence eye movements in rodents have previously been shown to compensate for head tilt ( Wallace et al . , 2013 ) , and correspondingly we find that during approach periods mice stabilize head tilt . Thus , the stable relative alignment of the two eyes during approach likely reflects stabilization of the head itself . These results suggest that the 40 degree binocular zone is sufficient for tracking centrally located objects , as the eyes to not move to expand this during approaches . This is consistent with previous work showing that during active approach the mouse’s head is oriented within ±15 degrees relative to the cricket ( Hoy et al . , 2016 ) , meaning that even the resting binocular zone would encompass the cricket . However , it remains to be determined whether mice actually use binocular disparity for depth estimation during prey capture . A recent study demonstrated that mouse V1 encodes binocular disparities spanning a range of 3–25 cm from the mouse’s head ( Land , 2018 ) , suggesting that disparity cues are available at the typical distances during approach ( interquartile range 14 . 6 cm to 27 . 6 cm ) . Alternatively , mice may use retinal image size or other distance cues , or may simply orient to the azimuthal position of the cricket regardless of distance . The finding that mice do not specifically move their eyes to target a location does not preclude the possibility that different regions of retinal space are specialized for certain processing . In fact , as a result of targeting head movements , the cricket prey is generally within the binocular zone during approach , so any mechanisms of enhanced processing in the binocular zone or lateral retina would still be behaviorally relevant . Anatomically , there is a gradient in density of different retinal ganglion cell types from medial to lateral retina ( Bleckert et al . , 2014 ) . Likewise behavioral studies have shown enhanced contrast detection when visual stimuli are located in the binocular field , rather than the monocular fields ( Speed et al . , 2019 ) . Based on the results presented here , in mice these specializations are likely to be engaged by head movements that localize stimuli in the binocular zone in front of the head , as opposed to primates , which make directed eye movements to localize stimuli on the fovea . Together , the present findings suggest that orienting relative to visual cues is driven by head movements rather than eye movements in the mouse . This is consistent with the general finding that for animals with small heads it is more efficient to move the head , whereas animals with large heads have adapted eye movements for rapid shifts to overcome the inertia of the head ( Land , 2018 ) . From the experimental perspective , this suggests that head angle alone is an appropriate measure to determine which visual cues are important during study of visually guided , goal-directed behaviors in the mouse . However , measurements of eye movements will be essential for computing the precise visual input animals receive ( i . e . , the retinal image ) during ongoing freely moving behaviors , and how this visual input is processed within visual areas of the brain . The saccade-and-fixate strategy generates a series of stable visual images separated by abrupt changes in gaze that shift the visual scene and location of objects on the retina . How then are these images , interleaved with periods of motion blur , converted into a continuous coherent percept that allows successful natural behaviors to occur ? Anticipatory shifts in receptive field location during saccades , as well as gaze position-tuned neural populations , have been proposed as mechanisms in primates to maintain coherent percepts during saccades , while corollary discharge , saccadic suppression , and visual masking have been proposed to inhibit perception of motion blur during rapid eye movements ( Higgins and Rayner , 2015; Wurtz , 2008 ) . However , the mechanisms that might mediate these , at the level of specific cell types and neural circuits , are poorly understood . Studying these processes in the mouse will allow for investigation of the neural circuit basis of these perceptual mechanisms through the application of genetic tools and other circuit dissection approaches ( Huberman and Niell , 2011; Luo et al . , 2008 ) . Importantly , most of our visual perception occurs during active exploration of the environment , where the combined dynamics of head and eye movements create a dramatically different image processing challenge than typical studies in stationary subjects viewing stimuli on a computer monitor . Examination of these neural mechanisms will extend our understanding of how the brain performs sensory processing in real-world conditions . All procedures were conducted in accordance with the guidelines of the National Institutes of Health and were approved by the University of Oregon Institutional Animal Care and Use Committee ( protocol number 17–27 ) . Animals used for this study were wild-type ( C57 Bl/6J ) males and females ( 3 males and four females ) aged 2–6 months . Prey capture experiments were performed following the general paradigm of Hoy et al . , 2016 . Mice readily catch crickets in the homecage without any training or habituation , even on the first exposure to crickets . However , we perform a standard habituation process to acclimate the mice to being handled by the experimenters , hunting within the experimental arena , and wearing cameras and an IMU while hunting . Following six 3 min sessions ( over 1–2 days ) of handling , the animals were placed in the prey capture arena to explore with their cagemates . The duration of this group habituation was at least six 10 min sessions over 1–2 days . One cricket ( Rainbow mealworms , 5 week old ) per mouse was placed in the arena with the mice for the last half of the habituation sessions . For the subsequent habituation step , the mice were placed in the arena alone with one cricket for 7–10 min . This step was repeated for 2–3 training days ( 6–9 sessions ) until most mice successfully caught crickets within the 10 min period . Animals were then habituated to head-fixation above a spherical Styrofoam treadmill ( Dombeck et al . , 2007 ) . Head fixation was only used to fit , calibrate , and attach cameras before experiments . Cameras were then fitted to each mouse ( described below ) and mice were habituated to wearing the cameras while walking freely in the arena , which took 1–2 sessions lasting 10 min . After the animals were comfortable with free locomotion with cameras , they were habituated to hunting with cameras attached . This took roughly one to two e hunting sessions of 10 min duration for each mouse . The animals were then food deprived for a period of ~12–18 hr and then run in the prey capture assay for three 10 min sessions per data collection day . Although animals will hunt crickets without food restriction , this allowed for more trials within a defined experimental period . The rectangular prey capture arena was a white arena of dimensions 38 × 45×30 cm ( Hoy et al . , 2016 ) . The arena was illuminated with a 15 Watt , 100 lumen incandescent light bulb placed roughly one meter above the center of the arena to mimic lux during dawn and dusk , times at which mice naturally hunt ( Bailey and Sperry , 1929 ) . Video signal was recorded from above the arena using a CMOS camera ( Basler Ace , acA2000–165 umNIR , 30 Hz acquisition ) . Following the habituation process , cameras were attached and mice were placed in the prey capture arena with one cricket . Experimental animals captured and consumed the cricket before a new cricket was placed in the arena . The experimenters removed any residual cricket pieces in the arena before the addition of the next cricket . A typical mouse catches and consumes between 3–5 crickets per 10 min session . Control experiments were performed using the same methods , but with no cameras or IMU attached . To allow for head-fixation during initial eye camera alignment , before the habituation process mice were surgically implanted with a steel headplate , following Niell and Stryker , 2010 . Animals were anesthetized with isoflurane ( 3% induction , 1 . 5–2% maintenance , in O2 ) and body temperature was maintained at 37 . 5°C using a feedback-controlled heating pad . Fascia was cleared from the surface of the skull following scalp incision and a custom steel headplate was attached to the skull using Vetbond ( 3M ) and dental acrylic . The headplate was placed near the back of the skull , roughly 1 mm anterior of Lambda . A flat layer of dental acrylic was placed in front of the headplate to allow for attachment of the camera connectors . Carprofen ( 10 mg/kg ) and lactated Ringer’s solution were administered subcutaneously and animals were monitored for three days following surgery . To measure eye position , we used miniature cameras that could be reversibly attached to the mouse’s head via a chronically implanted Millmax connector . The cameras ( 1000 TVL Mini CCTV Camera; iSecurity101 ) were 5 × 6 × 6mm with a resolution of 480 × 640 pixels and a 78 degree viewing angle , and images were acquired at 30 Hz . Some of the cameras were supplied with a built in NIR blocking filter . For these cameras , the lens was unscrewed and the glass IR filter removed with fine forceps . A 200 Ohm resistor and 3 mm IR LED were integrated onto the cameras for uniform illumination of the eyes . Power , ground , and video cables were soldered with lightweight 36 gauge FEP hookup wire ( Cooner Wire; CZ 1174 ) . A 6 mm diameter collimating lens with a focal distance of 12 mm ( Lilly Electronics ) was inserted into custom 3D printed housing and the cameras were then inserted and glued behind this ( see Figure 1 for schematic of design ) . The inner side of the arm of the camera holder housed a male Mill-Max connector ( Mill-Max Manufacturing Corp . 853-93-100-10-001000 ) cut to 5 mm ( 2 rows of 4 columns ) , used for reversible attachment of the cameras to the implants of experimental animals . A custom IMU board with integrated 3-dimensional accelerometer and gyroscopes ( Rosco Technologies ) was attached to the top of one of the camera holders ( see Figure 1B ) . The total weight of the two cameras together , with the lenses , connectors , 3D printed holders , and IMU was 2 . 6 grams . Camera assemblies were fitted onto the head by attaching them to corresponding female Mill-Max connectors . Cameras were located in the far lateral periphery of the mouse’s visual field , roughly 100° lateral of the head midline and 40 degrees above the horizontal axis , and covered roughly 25 × 25° of the visual field . When the camera was appropriately focused on the eye , the female connectors were glued onto the acrylic implant using cyanoacrylate adhesive ( Loctite ) . Because the connectors were each positioned during this initial procedure and permanently fixed in place , no adjustment of camera alignment was needed for subsequent experimental days . With this system , the average magnitude of camera shake jitter across experiments was 0 . 49 + / - 0 . 33 pixels ( mean + / - s . d . , N = 7 animals ) , as measured by computing the RMS frame-to-frame jitter of stationary points on the animal’s head ( base of the implant ) in the recorded videos . Video data with timestamps for the overhead camera were acquired at 30 frames per second ( fps ) using Bonsai ( Lopes et al . , 2015 ) . We used DeepLabCut ( Mathis et al . , 2018 ) for markerless estimation of mouse and cricket position from overhead videos . For network training , we selected eight points on the mouse head ( nose , two camera connectors , two IR LEDs , two ears , and center of the head between the two ears ) , and two points for the cricket ( head and body ) . Following estimation of the selected points , analysis was performed with custom MATLAB scripts , available at Michaiel et al . , 2020 . To determine periods when the animal was moving versus stationary , head movement speed was median filtered across a window of 500 ms and a threshold of 1 cm/sec was applied . Position and angle of the head were computed by fitting the eight measured points on the head for each video frame to a defined mean geometry plus an x-y translation and horizontal rotation . The head direction was defined as the angle of this rotation , referenced to the line between the nose and center of the head . We also used this head-centered reference to compute the azimuth , which is the angle of the mouse relative to the cricket . Following Hoy et al . , 2016 , we defined approaches as times at which the velocity of the mouse was greater than 1 cm/sec , the azimuth of the mouse was between −45 and 45 degrees relative to cricket location , and the distance to the cricket was decreasing at a rate greater than 10 cm/sec . Although mice eventually catch and consume the cricket in each trial , and are motivated to hunt due to food restriction , we cannot rule out the possibility that some approach periods may represent tracking or chasing without the intent to capture . Analog voltage signals from the IMU were recorded using a LabJack U6 at 50 Hz sampling rate . Voltages from the accelerometer channels were median filtered with a window of 266 . 7 ms to remove rapid transients and converted to m/sec2 , providing angular head orientation . Voltages from the gyroscope channels were converted to radians/sec without filtering , providing head rotation velocity . Video data with timestamps for the two eyes were acquired at 30fps using Bonsai . The video data are delivered by the camera in NTSC format , an interlaced video format in which two sequential images ( acquired at 60fps ) are interdigitated into each frame on alternate horizontal lines . We therefore de-interlaced the video in order to restore the native 60fps resolution by separating out alternate lines of each image . We then linearly downsampled the resolution along the horizontal axis by a factor of two , to match spatial resolution in horizontal and vertical dimensions . To track eye position , we used DeepLabCut ( Mathis et al . , 2018 ) to track eight points along the edge of the pupil . The eight points were then fit to an ellipse using the least-squares criterion . In order to convert pupil displacement into angular rotation , which cannot be calibrated by directed fixation as in primates , we followed the methods used in Wallace et al . , 2013 . This approach is based on the principle that when the eye is pointed directly at the camera axis , the pupil is circular , and as the eye rotates , the circular shape flattens into an ellipse depending on the direction and degree of angular rotation from the center of the camera axis . To calculate the transformation of a circle along the camera axis to the ellipse fit , two pieces of information are needed: the camera axis center position and the scale factor relating pixels of displacement to angular rotation . To find the camera axis , we used the constraint that the major axis of the pupil ellipse is perpendicular to the vector from the pupil center to the camera axis center . This defines a set of linear equations for all of the pupil observations with significant ellipticity , which are solved directly with a least-squares solution . Next , the scale factor was estimated based on the equation defining how the ellipticity of the pupil changes with the corresponding shift from the camera center in each video frame . Based on the camera center and scale factor for each video , we calculated the affine transformation needed to transform the circle to the ellipse fit of the pupil in each frame , and the angular displacement from the camera axis was then used for subsequent analyses . Mathematical details of this method are presented in Wallace et al . , 2013 . Following computation of kinematic variables ( mouse , cricket , and eye position/rotation ) , these values were linearly interpolated to a standard 60 Hz timestamp to account for differences in acquisition timing across the multiple cameras and the IMU . To characterize the robustness of the tracking system , we estimated the error in eye and head position measurements . As there is no ground truth measurement for eye position to compare to , we estimated an upper bound based on the stability of the eye when the head was stationary , as this is when eye movements are expected to be minimal . Specifically , we computed the standard deviation of horizontal eye position between frames during times when mouse speed was less than 1 cm/sec and head rotation <1 degree . We computed the error to be 0 . 51 + / - 0 . 25 degrees ( mean + / - s . d . , n = 105 trials ) . Similarly , to estimate the error of head angle measurements , we compared the independent estimates of head yaw rotation between frames as measured by both the IMU and DeepLabCut tracking . These measures have an RMS difference of 0 . 95 + / - 0 . 25 ( mean+/-s . d . , n = 105 trials ) , which represents an upper bound as it is based on the combined error of these two measurements separately . Thus , we infer that errors in estimating eye and head position are both less than one degree . Two-tailed paired t-tests or Wilcoxon Rank sum tests were used to compare data between non-approach and approach epochs . For comparisons between experimental and control groups , two-sample tests ( Kolmogorov-Smirnov or two-sample two-tailed t-test ) were used . Significance was defined as p<0 . 05 , although p-values are presented throughout . In all figures , error bars represent ±the standard error of the mean or median , as appropriate .	As you read this sentence , your eyes will move automatically from one word to the next , while your head remains still . Moving your eyes enables you to view each word using your central – as opposed to peripheral – vision . Central vision allows you to see objects in fine detail . It relies on a specialized area of the retina called the fovea . When you move your eyes across a page , you keep the images of the words you are currently reading on the fovea . This provides the detailed vision required for reading . The same process works for tracking moving objects . When watching a bird fly across the sky , you can track its progress by moving your eyes to keep the bird in the center of your visual field , over the fovea . But the majority of mammals do not have a fovea , and yet are still able to track moving targets . Think of a lion hunting a gazelle , for instance , or a cat stalking a mouse . Even mice themselves can track and capture insect prey such as crickets , despite not having a fovea . And yet , exactly how they do this is unknown . This is particularly surprising given that mice have long been used to study the neural basis of vision . By fitting mice with miniature head-mounted cameras , Michaiel et al . now reveal how the rodents track and capture moving crickets . It turns out that unlike animals with a fovea , mice do not use eye movements to track moving objects . Instead , when a mouse wants to look at something new , it moves its head to point at the target . The eyes then follow and ‘land’ on the target . In essence , head movements lead the way and the eyes catch up afterwards . These findings are consistent with the idea that mammals with large heads evolved eye movements to overcome the energy costs of turning the head whenever they want to look at something new . For small animals , moving the head is less energetically expensive . As a result , being able to move the eyes independent of the head is unnecessary . Future work could use a combination of behavioral experiments and brain recordings to reveal how visual areas of the brain process what an animal is seeing in real time .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2020	Dynamics of gaze control during prey capture in freely moving mice
Prdm14 is a sequence-specific transcriptional regulator of embryonic stem cell ( ESC ) pluripotency and primordial germ cell ( PGC ) formation . It exerts its function , at least in part , through repressing genes associated with epigenetic modification and cell differentiation . Here , we show that this repressive function is mediated through an ETO-family co-repressor Mtgr1 , which tightly binds to the pre-SET/SET domains of Prdm14 and co-occupies its genomic targets in mouse ESCs . We generated two monobodies , synthetic binding proteins , targeting the Prdm14 SET domain and demonstrate their utility , respectively , in facilitating crystallization and structure determination of the Prdm14-Mtgr1 complex , or as genetically encoded inhibitor of the Prdm14-Mtgr1 interaction . Structure-guided point mutants and the monobody abrogated the Prdm14-Mtgr1 association and disrupted Prdm14's function in mESC gene expression and PGC formation in vitro . Altogether , our work uncovers the molecular mechanism underlying Prdm14-mediated repression and provides renewable reagents for studying and controlling Prdm14 functions . Prdm14 is a sequence-specific transcriptional regulator that plays key roles in promoting primordial germ cell ( PGC ) specification and safeguarding pluripotency of embryonic stem cells ( ESCs ) ( Nakaki and Saitou , 2014 ) . During mouse embryogenesis , Prdm14 is expressed in preimplantation embryos , where its asymmetric expression promotes allocation of cells toward the pluripotent inner cell mass ( ICM ) fate ( Burton et al . , 2013; Nakaki and Saitou , 2014 ) . Prdm14 expression ceases in postimplantation epiblast cells and their differentiated progeny . However , during PGC specification from the epiblast , cells reacquire many transcriptional and epigenetic characteristics of the preimplantation state , and Prdm14 is re-expressed along with several other pluripotency-associated factors ( reviewed in [Magnúsdóttir and Surani , 2014; Saitou et al . , 2012] ) . The loss of Prdm14 in mice results in sterility associated with early germ cell deficiency , as cells fated to become PGCs fail to reacquire expression of key pluripotency factors and undergo epigenetic reprogramming ( Yamaji et al . , 2008 ) . Furthermore , overexpression of Prdm14 in epiblast-like cells ( EpiLCs ) is sufficient to induce PGCs in vitro ( albeit with low frequency ) , suggesting a central role of Prdm14 in the mouse PGC regulatory network ( Magnúsdóttir et al . , 2013; Nakaki et al . , 2013 ) . Furthermore , Prdm14 is repressed in normal somatic tissues but is aberrantly reactivated in human malignancies of various tissue origin , including leukemias and lymphomas , breast , testicular , and lung cancers ( Carofino et al . , 2013; Dettman et al . , 2011; Nishikawa et al . , 2007; Ruark et al . , 2013; Zhang et al . , 2013 ) . Given the poor accessibility and transient nature of preimplantation embryo cells and PGCs in vivo , mechanistic studies of Prdm14 function in early development have been chiefly conducted in the context of mouse ESCs ( mESCs ) . These cells represent a so-called 'naïve' pluripotent state , thought to resemble preimplantation embryo ICM and serve as a useful system for understanding early cell fate decisions ( Nichols and Smith , 2009 ) . Loss of Prdm14 destabilizes mESCs and sensitizes them to differentiation stimuli , leading to acquisition of alternative embryonic states , such as the postimplantation epiblast state or extraembryonic endoderm state , and eventual depletion of the naïve cell subpopulation ( Ma et al . , 2011; Yamaji et al . , 2013 ) . The differentiation in Prdm14−/− cells is thought to result from upregulation of signaling pathways such as the fibroblast growth factor receptor ( FGFR ) pathway and by widespread DNA hypermethylation ( Grabole et al . , 2013; Hackett et al . , 2013; Leitch et al . , 2013 ) . Indeed , genome-wide Prdm14 occupancy studies by chromatin immunoprecipitation with sequencing ( ChIP-seq ) suggest that these are direct effects of the loss of Prdm14 , as Prdm14 occupies and represses the regulatory elements of genes involved in FGFR signaling and de novo DNA methylation ( Leitch et al . , 2013; Ma et al . , 2011; Magnúsdóttir et al . , 2013; Yamaji et al . , 2013 ) . Nonetheless , Prdm14−/− mESCs can be maintained indefinitely under 2i conditions ( Grabole et al . , 2013; Payer et al . , 2013; Yamaji et al . , 2013 ) , in which differentiation stimuli , including FGFR signaling , are chemically inhibited , providing an opportunity to study the early effects of Prdm14 deficiency upon release from such inhibition . Although the cellular and molecular phenotypes associated with loss of Prdm14 in mESCs have been well characterized , much less is known about molecular mechanisms and partners through which Prdm14 acts . As a member of the PRDM family , Prdm14 contains both a zinc-finger array , responsible for sequence-dependent DNA binding ( Ma et al . , 2011 ) , and a PR domain that is related to the SET domain ( Su ( var ) 3–9 , Enhancer-of-zeste and Trithorax ) ( Fog et al . , 2012; Hohenauer and Moore , 2012 ) . Many SET domains harbor methyltransferase activity for either histone or non-histone substrates ( Del Rizzo and Trievel , 2011 ) . However , to date , no enzymatic activity has been reported for Prdm14 , and interestingly multiple members of the PRDM family appear to be catalytically inactive . Instead , candidate-based co-immunoprecipitation studies implicated Polycomb complex PRC2 as a mediator of Prdm14-dependent repression ( Chan et al . , 2013; Yamaji et al . , 2013 ) . Nonetheless , it remains unclear whether PRC2 is a major or auxiliary partner of Prdm14 , and what other molecular players are important for Prdm14's function . To address these questions , we used an unbiased biochemical approach to uncover major Prdm14-associated proteins in mESCs . We identified an ETO-family corepressor , myeloid translocation gene related 1 ( Mtgr1 , a . k . a . Mtg8r , Cbfa2t2 , and Zmynd3 ) , as a direct , stoichiometric partner of Prdm14 . We demonstrate that Mtgr1 co-occupies Prdm14 target loci , and its deletion in mESCs results in phenotypes and gene expression defects similar to those observed upon loss of Prdm14 . Moreover , Mtgr1 knockout cells show impaired induction of PGC-like cells in vitro . To further facilitate studies of the Prdm14-Mtgr1 complex , we mapped interaction domains and developed multiple synthetic binding proteins , termed monobodies , that specifically recognize the SET domain of Prdm14 in a manner independent of , or alternatively , competitive with Mtgr1 . Taking advantage of the stabilizing effect of one such monobody , we obtained a crystal structure of the Prdm14-Mtgr1 complex , revealing an extensive interface and electrostatic interactions mediating the association of the two proteins . Furthermore , structure-guided mutagenesis of the interface and the use of an inhibitory monobody demonstrated the function of the complex in safeguarding pluripotency and PGC-like cell induction . Altogether , we report a multi-disciplinary study that advances our understanding of Prdm14 function , identifying Mtgr1 as the major partner of Prdm14 in its roles in pluripotency and PGC induction , and providing the community with renewable , genetically encoded reagents that can be used both in vitro and in vivo to study and control Prdm14 function in development and malignancy . To identify Prdm14 partners in an unbiased manner , we employed a two-step immunopurification strategy ( FLAG followed by HA [FH] ) from a previously described clonal mESC transgenic line stably expressing tagged FH-Prdm14 ( Ma et al . , 2011 ) . Examination of recovered proteins by sodium dodecyl sulfate polyacrylamide gel electrophoresis ( SDS-PAGE ) and silver staining revealed two major polypeptides that were present in similar quantities in the FH-Prdm14 purifications , but not in the control immunoprecipitates ( Figure 1A ) . These polypeptides were subsequently identified by mass spectrometry as Prdm14 and Mtgr1 ( Figure 1A , Figure 1—source data 1 ) , the latter of which is one of the three members of the ETO family of co-repressors ( Davis et al . , 2003 ) . Mass spectrometry analysis also identified additional polypeptides enriched uniquely in the FH-Prdm14 purifications , including the other two ETO proteins Mtg8 and Mtg16 , their known repressive complex partners Tbl1/Tblr1 and histone deacetylases ( HDACs ) , as well as Brg1 complex components and Oct4 , among others ( Figure 1—source data 1 ) . Of note , we did not detect components of the Polycomb complex PRC2 ( Chan et al . , 2013; Yamaji et al . , 2013 ) . 10 . 7554/eLife . 10150 . 003Figure 1 . Prdm14 directly binds to the ETO family protein , Mtgr1 . ( A ) Two-step immunoaffinity purification of Prdm14-associated proteins . FLAG–HA immunoprecipitations were performed from wild-type ( wt ) or FH-Prdm14 mESC extracts , followed by visualization of polypeptides by SDS-PAGE-silver stain and mass spectrometry identification . Polypeptides corresponding to Mtgr1 and Prdm14 are highlighted . ( B ) Reciprocal Mtgr1 and Prdm14 co-immunoprecipitations from FH-Prdm14 mESCs . ( C , D ) Identification of the Prdm14-Mtgr1 interaction regions . Co-immunoprecipitations were performed in HEK293 cells transfected with full-length V5-tagged Mtgr1 and distinct FH-Prdm14 constructs ( C ) or full-length FH-Prdm14 and distinct V5-Mtgr1 constructs ( D ) , as indicated in the top diagrams; co-immunoprecipitated proteins were visualized by immunoblotting with α-HA ( Prdm14 ) and α-V5 ( Mtgr1 ) antibodies ( top left and right panels , respectively ) . Tagged Prdm14 and Mtgr1 levels in the input extracts are shown in the bottom panels . ( E ) Recombinant biotinylated Prdm14 ( pre-SET+SET ) was immobilized on streptavidin-coated beads and incubated with recombinant Mtgr1 ( NHR1 ) , and changes in fluorescence ( SAV-Dylight650 ) were measured . ( F , G ) Recombinant biotinylated Mtgr1 ( NHR1 ) was immobilized on streptavidin-coated beads and incubated with ( F ) Prdm14 ( pre-SET+SET ) or ( G ) Prdm14 ( SET ) and changes in fluorescence ( SAV-Dylight650 ) were measured . The error bars and the errors for the KD values are the standard deviation ( n = 3 ) . The curves show the best fit of the 1:1 binding model using the GraphPad software . * indicates non-specific bands . HEK , human embryonic kidney; Mtgr1 , myeloid translocation gene related 1; MW , molecular weight marker; NHR1 , nervy homology region 1 . N . D . , not determined . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 00310 . 7554/eLife . 10150 . 004Figure 1—source data 1 . List of proteins recovered from the Prdm14 IP-MS experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 00410 . 7554/eLife . 10150 . 005Figure 1—figure supplement 1 . Expression levels of mRNAs encoding ETO proteins . Plotted are RPKM values for Mtgr1 , Mtg1 and Mtg16 obtained in RNA-seq experiments from mESC grown under serum+LIF or 2i+LIF conditions or from EpiLCs . EpiLCs , epiblast-like cells; LIF , leukemia inhibitory factor; mESC , mouse embryonic stem cell; Mtgr1 , myeloid translocation gene related 1; RNA-seq , RNA sequencing; RPKM , reads per kilobase of exon per million reads mapped . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 005 Mtgr1 was the major polypeptide identified in our analysis and its stoichiometric recovery in our purifications indicated it might represent a strong and direct partner of Prdm14 . To confirm this , we first verified Prdm14 and Mtgr1 association using reciprocal co-immunoprecipitations from mESC nuclear extracts ( Figure 1B ) . Next , we mapped the minimal regions within Prdm14 and Mtgr1 that were required for the interaction by overexpressing differentially-tagged proteins ( V5-Mtgr1 and FH-Prdm14 ) in HEK293 cells , followed by IP-Western analysis ( Figure 1C , D ) . This strategy revealed that the nervy homology region 1 ( NHR1 ) domain of Mtgr1 was necessary and sufficient for the interaction with Prdm14 , whereas both Prdm14 SET domain and the region directly preceding it ( pre-SET ) were important for efficient binding to Mtgr1 ( Figure 1C , D ) . To quantify the strength of the Prdm14–Mtgr1 interaction , we next expressed and purified recombinant proteins corresponding to the NHR1 domain of Mtgr1 ( residues 98–206 ) and pre-SET+SET domains of Prdm14 ( residues 184–373 ) , and performed a bead-based binding assay in reciprocal orientations ( Nishikori et al . , 2012 ) . The obtained binding measurements yielded a dissociation constant ( KD ) in the low nanomolar range ( Figure 1E and F ) , which is consistent with a robust , direct interaction between the two proteins . On the other hand , the binding of the Prdm14 SET domain alone ( residues 232–373 ) to the Mtgr1 NHR1 domain was barely detectable ( Figure 1G ) , further supporting that both pre-SET and SET domains are required for the high affinity interaction with Mtgr1 . Altogether , our approach identified an ETO protein Mtgr1 as a novel , direct partner of Prdm14 in mESCs . While the ETO proteins , especially Mtg8 ( a . k . a . ETO ) , have been studied in the context of acute myeloid leukemias ( AML ) ( reviewed in Hatlen et al . , 2012 ) , their function in ESCs and during early embryogenesis has not been explored . Notably , all three ETO family members have the capacity to interact with Prdm14 ( not shown ) , but the high expression of Mtgr1 in mESCs compared with Mtg8 and Mtg16 ( Figure 1—figure supplement 1 ) likely accounts for the preferential recovery of Mtgr1 in our experiments and suggests that this family member may be most relevant in the context of mESCs . We therefore proceeded to explore the functional significance of the Prdm14–Mtgr1 interaction in mESC biology . Prdm14 is a sequence-dependent DNA-binding protein that binds many genomic loci in mESCs , corresponding primarily to distal regulatory elements , whereas ETO proteins do not contain domains implicated in direct DNA sequence recognition ( Rossetti et al . , 2004 , 2008 ) . To examine whether Mtgr1 is brought to genomic targets occupied by Prdm14 , we performed Mtgr1 ChIP coupled with high-throughput DNA sequencing ( ChIP-seq ) from wt mESCs , FH-Prdm14 overexpressing mESCs , and as a control for antibody specificity , Mtgr1−/− mESCs ( generation of which is described in more detail later ) , cultured for 5 days under serum+leukemia inhibitory factor ( LIF ) conditions . In parallel , we profiled Prdm14 occupancy by performing ChIP-seq analysis from FH-Prdm14 cells , using an anti-HA antibody due to the unavailability of ChIP-grade Prdm14 antibodies . Overall , we identified ~ 8000 Mtgr1 peaks present in both FH-Prdm14 and wt mESCs , but absent in Mtgr1−/− mESCs . These bound sites include loci known to be occupied and repressed by Prdm14 ( e . g . near Prdm14 , Dnmt3b , Wnt8a , Peg10 , and targets of the FGFR pathway Fgfr2 and Shc1; Figure 2A ) . 10 . 7554/eLife . 10150 . 006Figure 2 . Prdm14 and Mtgr1 co-occupy genomic targets . ( A ) Prdm14 and Mtgr1 ChIP-seq enrichments at selected gene loci . Tracks represent sequence tag enrichments as determined by Quest software . ( B ) Scatter plot of Prdm14 and Mtgr1 genomic occupancies in FH-Prdm14 mESC line . ( C ) The top sequence motif recovered in Mtgr1 ChIP-seq corresponds to the Prdm14 motif , as defined previously ( Ma et al . 2011 ) . Logos for the consensus motifs were generated using SeqPos . ( D ) Scatter plot of Mtgr1 genomic occupancy in wt and FH-Prdm14 mESC lines . The plot is colored based on the presence of Prdm14 motif ( red , motif is present , p-value <10-3; black , motif is absent ) . ChIP-seq , chromatin immunoprecipitation with sequencing; mESC , mouse embryonic stem cell; Mtgr1 , myeloid translocation gene related 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 00610 . 7554/eLife . 10150 . 007Figure 2—figure supplement 1 . Genomic occupancy of Mtgr1 . ( A ) Correlation coefficient , R , between Mtgr1 and Prdm14 ( anti-HA ) ChIP-seq genomic occupancy levels in two different clonal FH-Prdm14 mESC lines . ( B ) Functional annotation categories of the Mtgr1 binding sites in FH-Prdm14 mESCs as determined by GREAT . ChIP-seq , chromatin immunoprecipitation with sequencing; GREAT , genomic regions enrichment of annotations tool; mESCs , mouse embryonic stem cell; Mtgr1 , myeloid translocation gene related 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 00710 . 7554/eLife . 10150 . 008Figure 2—figure supplement 2 . Genomic occupancy of Mtgr1 at Prdm14 motif-lacking sites . ( A ) Genomic tracks at select Prdm14 motif-lacking Mtgr1 binding sites . Note that these sites lack Prdm14 binding and Mtgr1 peaks are present in wt cells , but not in the FH-Prdm14-overexpressing line . ( B ) Aggregate plots of Mtgr1 ChIP-seq enrichments at Prdm14 motif-containing ( left ) or Prdm14 motif-lacking ( right ) Mtgr1 sites from indicated ESC cell lines . ( C ) Top five motifs recovered from the Mtgr1-binding sites where Prdm14 motif was absent . ChIP-seq , chromatin immunoprecipitation with sequencing; ESC , embryonic stem cell; Mtgr1 , myeloid translocation gene related 1; wt , wild type . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 008 Generally , the genomic occupancies of Mtgr1 and Prdm14 were well correlated ( correlation coefficient ~0 . 9 , Figure 2—figure supplement 1A ) , and we have not been able to detect a substantial class of Prdm14-bound sites devoid of Mtgr1 occupancy ( Figure 2B ) . Not surprisingly , the Prdm14 and Mtgr1 sites shared common functional ontologies , with enrichment for processes involved in embryonic development and cell differentiation ( Figure 2—figure supplement 1B ) . Furthermore , the most highly enriched DNA sequence motif at Mtgr1-bound sites corresponded to the previously identified Prdm14 motif ( Figure 2C ) . Interestingly , we noted that , at many targets , Mtgr1 binding was enhanced by Prdm14 overexpression ( see tracks in Figure 2A , compare wt and FH-Prdm14 ESC ) . This observation prompted us to quantitatively compare Mtgr1 ChIP-seq enrichments in wt ESCs and FH-Prdm14 cells that are characterized by ~5-fold overexpression of Prdm14 . We observed that Mtgr1 enrichments were higher in FH-Prdm14 than in wt ESCs at most target sites , consistent with Prdm14-mediated recruitment of Mtgr1 to chromatin ( Figure 2D ) . However , we also noticed that a subset of Mtgr1 sites was bound more weakly in FH-Prdm14 cells than in wt ESCs ( Figure 2D , examples shown in Figure 2—figure supplement 2A ) . The major distinction between these two populations was the presence of the Prdm14 sequence motif and Prdm14 occupancy at the sites where Mtgr1 binding was enhanced by Prdm14 overexpression , and lack of the Prdm14 sequence motif with low/no Prdm14 occupancy at the sites where Mtgr1 binding was diminished by Prdm14 overexpression ( Figure 2D ) . Of note , at the Prdm14 motif-lacking sites , the most enriched sequence motifs corresponded to helix-loop-helix transcription factor recognition sites , suggesting that a TF from this family may be involved in mediating Mtgr1 binding at these sites ( Figure 2—figure supplement 2C ) . Regardless , our results indicate that Prdm14 is sufficient to augment interaction of Mtgr1 with chromatin at its cognate binding sites and , at high levels , redirect it away from the motif-lacking sites . Thus , Prdm14 might be a limiting factor for Mtgr1 recruitment to chromatin . To test this notion further , we performed Mtgr1 ChIP-seq analysis from Prdm14−/− ESCs and generated average signal profiles at Prdm14 motif-containing and Prdm14 motif-lacking sites across all our Mtgr1 ChIP-seq datasets . We observed that at Prdm14 motif-containing sites , Mtgr1 binding is increased in FH-Prdm14 overexpressing cells and diminished ( but not completely abrogated ) in Prdm14−/− cells ( Figure 2—figure supplement 2B , left panel ) . On the other hand , at Prdm14 motif-lacking sites , Mtgr1 binding is depleted by FH-Prdm14 overexpression , but it is also moderately affected in Prdm14−/− cells despite low/no Prdm14 binding at these sites , suggesting an indirect effect ( Figure 2—figure supplement 2B , right panel ) . Altogether , these results are consistent with the Mtgr1 genomic occupancy being sensitive to the Prdm14 dosage ( either loss or gain ) at the Prdm14-motif containing sites . However , these results also demonstrate that even in the absence of Prdm14 , some Mtgr1 binding remains at the motif-containing sites , suggesting partial redundancies in the recruitment mechanisms . Prdm14 has well-characterized roles in pluripotency and PGC formation , and if Mtgr1 is a key mediator of Prdm14's functions then the loss of Mtgr1 should impact these processes in a similar manner . To test this hypothesis , we used CRISPR-Cas9 with a guide RNA targeting the third exon of the Mtgr1 gene to generate Mtgr1−/− mESCs , and verified the presence of the homozygous deletions and loss of the Mtgr1 protein in the three clonal lines selected for further analysis ( Figure 3—figure supplement 1 ) . As a reference for comparison , we also isolated and characterized two Prdm14−/− mESC lines by targeting the second exon of the Prdm14 gene ( Figure 3—figure supplement 2 ) . Moreover , we reconstituted each of the Mtgr1−/− and Prdm14−/− cell lines with FH-Mtgr1 or FH-Prdm14 complementary DNA ( cDNA ) , respectively , to generate 'rescue' cell lines and ensure specificity of the observed phenotypes . All aforementioned cell lines were isolated and maintained under the serum-free 2i+LIF conditions in which the major differentiation cues are inhibited and that support self-renewal even in the absence of Prdm14 ( Grabole et al . , 2013; Yamaji et al . , 2013 ) . After being transferred into standard serum+LIF growth conditions , the Mtgr1−/− lines exhibited changes in morphological appearance with less compact colonies , diminished cell–cell interactions and cell flattening , as previously reported for loss of Prdm14 in mESCs and reproduced here with our Prdm14−/− lines ( Ma et al . , 2011; Yamaji et al . , 2013 ) ( Figure 3—figure supplement 3 ) . These features were not observed in wt mESCs or after rescue with the respective protein constructs ( Figure 3—figure supplement 3 ) . Loss of Prdm14 has been shown to sensitize mESC to differentiation stimuli , resulting in upregulation of genes associated with epiblast and extraembryonic endoderm fates ( Ma et al . , 2011; Yamaji et al . , 2013 ) . To examine whether these molecular phenotypes are also observed upon loss of Mtgr1 , we conducted RNA sequencing ( RNA-seq ) transcriptome analyses from wt mESCs , Prdm14−/− and Mtgr1−/− cell lines , and their respective rescue lines after transfer from 2i+LIF to serum+LIF conditions . As seen in Prdm14−/− mESCs , Mtgr1−/− mESCs showed upregulation of epiblast ( e . g . Fgf5 , Dnmt3b , Oct6 , Wnt8a ) and extraembryonic endoderm ( e . g . Krt19 , Sparc , H19 , Fgfr2 ) markers , and downregulation of naïve pluripotency genes ( e . g . Esrrb , Zfp42 , Tbx3 , Tet2 ) , compared with either wt or FH-Mtgr1 rescue mESCs ( Figure 3A , Figure 3—figure supplement 4A ) . 10 . 7554/eLife . 10150 . 009Figure 3 . Loss of Mtgr1 phenocopies requirement for Prdm14 in safeguarding pluripotency and PGC induction . ( A ) RNA-seq from Prdm14−/− cells or Mtgr1−/− cells ( y-axis ) were compared to wt cells ( x-axis ) and expression values ( RPKM ) of all significantly changed transcripts were plotted . Select transcripts corresponding to those enriched in the post-implantation epiblast , extraembryonic endoderm or naïve pluripotent mESC are highlighted in red , green or blue , respectively; shaded colors indicate no significant difference . ( B ) Heatmap displaying top 100 variable genes between wt mESCs grown under naïve 2i+LIF or serum+LIF conditions , Prdm14−/− ( 2 clones ) or Mtgr1−/− ( 3 clones ) cells , and their respective rescue lines . Clustering represents sample divergence . ( C ) Principal component analysis on the same populations as in B . ( D ) mESC to mEpiLC transition followed by PGC-LC induction using defined media in cells containing Stella:GFP reporter . Schematic of the Stella:GFP transgene reporter that contains a 10kb 5’ upstream sequence and includes exon 1 and part of exon 2 fused in-frame with eGFP , followed by the SV40 polyadenylation sequence ( Payer et al . 2006 ) . The reporter is active in mPGC-LCs when Stella expression is activated . ( E ) Quantification of the GFP signal in wt cells during the mESC to mEpiLC and further to mPGC-LC transition . ( F ) FACS plots and gated quantification of GFP signal as a measure of mPGC-LC induction from wt cells , Prdm14−/− cells or Prdm14 rescue clones . ( G ) FACS plots and quantification of GFP signal as a measure of PGC-LC induction from wt cells , Mtgr1−/− cells or Mtgr1 rescue clones . FACS , fluorescence-activated cell sorting; GFP , green fluorescent protein; LIF , leukemia inhibitory factorm; EpiLCs , mouse epiblast-like cells; mESCs , mouse embryonic stem cell; mPGC-LCs , mouse primordial germ cell-like cells; Mtgr1 , myeloid translocation gene related 1; RPKM , reads per kilobase of exon per million reads mapped; wt , wild-type . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 00910 . 7554/eLife . 10150 . 010Figure 3—figure supplement 1 . Generation of Mtgr1-null line in Stella:GFP mESCs using CRISPR-Cas9 system . ( A ) Schematic of the Mtgr1 gene and sequence of the guide RNA ( http://crispr . mit . edu/ ) . ( B ) Genomic and cDNA were extracted and chromatograms of the wt ( top ) , mutant A3 line ( clone 1 ) , mutant C1 line ( clone 2 ) , and mutant C3 line ( clone 3 ) are shown . ( C ) Indel spectrum centered around the PAM sequence was performed using software TIDE ( http://tide . nki . nl/ ) . Clones A3 , C1 , and C3 are shown , they all produce aberrant transcript . ( D ) Western blotting of the whole cell lysate from the wt , A3 , C1 and C3 mESC Stella:GFP line . CRISPR , clustered regularly interspaced short palindromic repeat; GFP , green fluorescent protein; PAM , protospacer adjacent motif; mESCs , mouse embryonic stem cells; Mtgr1 , myeloid translocation gene related 1; wt , wild type . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 01010 . 7554/eLife . 10150 . 011Figure 3—figure supplement 2 . Generation of Prdm14-null line in Stella:GFP mESCs using CRISPR-Cas9 system . ( A ) Schematic of the Prdm14 gene and sequence of the guide RNA ( http://crispr . mit . edu/ ) . ( B ) Genomic and cDNA were extracted and chromatograms of the wt ( top ) , mutant G6 line ( clone 1 ) , and mutant F10 line ( clone 2 ) are shown . ( C ) Indel spectrum centered around the PAM sequence was performed using software TIDE ( http://tide . nki . nl/ ) . Clone G6 is shown on top and clone F10 below . Clones G6 and F10 produce aberrant transcript . cDNA , complementary DNA; CRISPR , clustered regularly interspaced short palindromic repeat; GFP , green fluorescent protein; mESCs , mouse embryonic stem cells; wt , wild type . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 01110 . 7554/eLife . 10150 . 012Figure 3—figure supplement 3 . Morphological changes associated with loss of Prdm14 or Mtgr1 . Brighfield images of cell morphology from wt ESCs , Prdm14−/− or Mtgr1−/−ESCs or those reconstituted with wt FH-Prdm14 or FH-Mtgr1 , respectively , all grown under serum+LIF conditions; wt ESC maintained in 2i+LIF are included for relative comparisons ( upper left image ) . ESC , embryonic stem cell; LIF , leukemia inhibitory factor; Mtgr1 , myeloid translocation gene related 1; wt , wild type . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 01210 . 7554/eLife . 10150 . 013Figure 3—figure supplement 4 . Additional analyses of RNA-seq datasets . ( A ) RNA-seq from Prdm14−/− cells or Mtgr1−/− cells ( y axis ) were compared with respective rescue lines ( x axis ) and expression values ( RPKM ) of all significantly changed transcripts were plotted . The transcripts of specific genes are highlighted in red , green or blue as indicated; shaded colors indicate no significant difference . ( B ) Scatter plot of genes upregulated at least twofold upon loss of either Prdm14 or Mtgr1 . Highlighted in blue are genes significant for Mtgr1-null cells only , red – Prdm14-null only , magenta – genes significantly upregulated in both cell lines . Mtgr1 , myeloid translocation gene related 1; RNA-seq , RNA sequencing; RPKM , reads per kilobase of exon per million reads mapped . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 01310 . 7554/eLife . 10150 . 014Figure 3—figure supplement 5 . Expression of differentiation markers in embryoid bodies derived from wt or Mtgr1−/− ESCs . ( A ) RT-qPCRs analyses of indicated markers in EBs induced from wt or Mtgr1−/− ESCs and cultured for four days in suspension after LIF withdrawal; shown relative to wt mESCs cultured under serum+LIF conditions . ( B ) RT-qPCRs analyses of select markers in EBs induced from wt or Mtgr1−/− ESCs and cultured for eight days in suspension after LIF withdrawal; shown relative to wt mESCs cultured under serum+LIF conditions . ESCs , embryonic stem cells; LIF , leukemia inhibitory factor; Mtgr1 , myeloid translocation gene related 1; RT-qPCRs , quantitative reverse transcription-polymerase chain reaction; wt , wild type . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 01410 . 7554/eLife . 10150 . 015Figure 3—figure supplement 6 . Loss of Mtgr1 results in defect in PGC-LC induction . ( A ) Imaging of fixed mPGC-LCs on day 6 of differentiation using bright field , DAPI ( stained with Hoechst ) and GFP channels . Imaging was done using z-stacks and maximum projection was taken afterwards . ( B ) FACS on the Stella:GFP line in ES state or PGC-LC state as indicated . Wt cells or Prdm14-null , Mtgr1-null or respective rescue lines were used . DAPI , 4' , 6-diamidino-2-phenylindole; GFP , green fluorescent protein; FACS , fluorescence-activated cell sorting; mPGC-LCs , mouse primordial germ cell-like cells; wt , wild type; Mtgr1 , myeloid translocation gene related 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 015 Next , we identified genes showing the most variable expression across our datasets and visualized their expression changes in each of our RNA-seq datasets as a heatmap ( Figure 3B ) . Most of the differentially expressed genes were concordantly upregulated in Prdm14−/− and Mtgr1−/− cells , compared with wt mESCs , in agreement with the proposed function of the Prdm14–Mtgr1 complex in gene repression . A more systematic comparison of all genes upregulated at least twofold upon loss of either Prdm14 or Mtgr1 revealed that while indeed , the majority of genes that are upregulated in either knockout are upregulated in both ( Figure 3—figure supplement 4B , purple dots ) , a subset of transcripts is preferentially affected only in one of the knockouts ( red or blue dots ) . Importantly , in the FH-Prdm14 or FH-Mtgr1 reconstituted knockout cells the derepression defects were rescued ( Figure 3B ) . Interestingly , while FH-Mtgr1 cells showed expression patterns highly similar to that of wt mESCs cultured in serum , FH-Prdm14 cells were more similar to the mESCs grown under 2i +LIF , despite being cultured in serum at the time of analysis ( Figure 3B ) . Indeed , many of the expression differences observed between wt mESC grown in 2i versus serum were recapitulated in FH-Prdm14 mESCs ( Figure 3B ) . Of note , a subset of transcripts was upregulated only in FH-Prdm14 cells; many of those genes represent markers of the so-called 2-cell ( 2C ) state ( Amano et al . , 2013; Dan et al . , 2013; Macfarlan et al . , 2012 ) . In our analysis of variably expressed genes , the Prdm14−/− and Mtgr1−/−cell lines clustered together , but separately from the respective rescue lines and wt mESCs ( Figure 3B ) . These observations were further confirmed by the global comparisons of transcriptomes with principal component analysis ( PCA ) , in which the Prdm14−/− and Mtgr1−/−cells were closest to each other and clustered away from the remaining cell lines ( Figure 3C ) . Additionally , the PCA analysis corroborated higher similarity of FH-Mtgr1 cells to wt mESCs grown in serum+LIF , and FH-Prdm14 cells to wt mESCs grown under 2i+LIF conditions ( Figure 3C ) . Given that: ( i ) the FH-Prdm14 cell lines in our study express Prdm14 at levels ~5–6-fold higher than wt ESCs , ( ii ) Prdm14 has an autonomous DNA-binding activity , and ( iii ) Prdm14 overexpression can augment Mtgr1 recruitment to the target genes ( as shown in Figure 2 ) , we propose that expression changes observed in FH-Prdm14 cell lines are associated with more robust repression of differentiation programs by Prdm14-Mtgr1 complex compared with wt cells and consequently , with the stabilization of the naïve pluripotency program . Similar gain-of-function effects are not observed in FH-Mtgr1 cells , likely because Mtgr1 lacks the autonomous ability to access its genomic targets and , at least in mESCs , Prdm14 is limiting for its chromatin association . Altogether , our data uncover the function of Mtgr1 in safeguarding mESC pluripotency and demonstrate that loss of Mtgr1 phenocopies gene expression defects associated with Prdm14 deletion . Prdm14 is critical for the specification of PGCs from the post-implantation epiblast cells ( Magnúsdóttir and Surani , 2014; Yamaji et al . , 2008 ) . To address whether Mtgr1 also plays a role in PGC formation , we used a previously established in vitro model in which naïve mESCs are first differentiated to a primed , post-implantation epiblast-like state ( mEpiLCs ) from which mouse primordial germ cell-like cells ( mPGC-LCs ) are then induced via addition of various cytokines ( Hayashi and Saitou , 2013; Hayashi et al . , 2011 ) . The mPGC-LCs formation is monitored with the fluorescent reporter , Stella:GFP and quantified by fluorescence-activated cell sorting ( FACS ) analysis ( Figure 3D ) . In vitro derived mPGC-LCs have been shown to be competent to differentiate to sperm/oocytes upon transplantation and produce viable , fertile offspring and this simple in vitro differentiation system is therefore considered a useful tool to study mechanisms underlying PGC specification ( Hayashi et al . , 2011; Nakaki et al . , 2013 ) . To examine the role of Prdm14 and Mtgr1 in the context of this model , we derived Prdm14−/− and Mtgr1−/− mESCs in the Stella:GFP reporter background; this reporter recapitulates endogenous Stella induction that occurs during the PGC formation ( Payer et al . , 2006 ) . In agreement with previous reports , differentiation of Stella:GFP mESCs consistently produced mPGC-LCs with an efficiency of ~7–8% on day 6 of differentiation , while low levels ( ~1% ) of GFP-positive cells were detected in mESCs and further diminished upon mEpiLC formation ( Figure 3E ) . In contrast , both Prdm14−/− and Mtgr1−/− mESCs showed significantly decreased efficiency of mPGC-LCs formation ( Figure 3F and G , Figure 3—figure supplement 5 ) . These defects were rescued by the re-introduction of FH-Prdm14 or FH-Mtgr1 , respectively ( Figure 3F and G , Figure 3—figure supplement 5 ) . These data suggest that Mtgr1 , like Prdm14 , is important for mouse PGC establishment . Significantly , loss of Mtgr1 does not result in general block in differentiation , as germ layer markers are expressed at comparable levels in embryoid bodies induced from Mtgr1−/− versus wt ESCs ( Figure 3—figure supplement 5 ) . To develop new tools for understanding the function of Prdm14 and to aid structure determination , we generated designer binding proteins termed ‘monobodies’ recognizing the SET domain of human and mouse Prdm14 , as a part of a larger project aimed at developing new reagents for controlling epigenetic regulatory proteins . Monobodies are small binding proteins ( ~10 kDa ) generated from combinatorial phage-display libraries built on the antibody-like scaffold of the tenth human fibronectin type III domain ( FN3; Figure 4A ) ( Koide et al . , 1998 , 2012b ) . Monobodies can recognize their targets with high affinity and specificity and have a strong tendency to recognize functional binding sites in their target molecules including clefts and planar surfaces , and thus they often are potent inhibitors ( Koide et al . , 2012a; Sha et al . , 2013; Wojcik et al . , 2010 ) . Furthermore , unlike antibodies whose folding depends on disulfide bond formation , monobodies are cysteine-free and thus functional when expressed under reducing environments such as the nucleus and cytoplasm . These attributes make monobodies particularly attractive as genetically encoded intracellular inhibitors . 10 . 7554/eLife . 10150 . 016Figure 4 . Generation of PRDM14-binding monobodies . ( A ) Schematic of the monobody scaffold . The β strands and loops are labeled and the diversified residues are marked as red spheres . The amino acid sequences of the monobody library and monobody clones . In the library designs , ‘X’ denotes a mixture of 30% Tyr , 15% Ser , 10% Gly , 5% Phe , 5% Trp , and 2 . 5% each of all the other amino acids except for Cys; ‘O’ , a mixture of Asn , Asp , His , Ile , Leu , Phe , Tyr , and Val; ‘U’ , a mixture of His , Leu , Phe , and Tyr; ‘Z’ , a mixture of Ala , Glu , Lys , and Thr ( Koide et al . 2012 ) . A hyphen indicates a deletion . ( B ) Titration curves of Mb ( hPRDM14_S4 ) and Mb ( hPRDM14_S14 ) to human PRDM14 and mouse Prdm14 . The error bars are the standard deviation ( n = 3 ) . The curves show the best fit of the 1:1 binding model . ( C ) Binding of Mb ( S4 ) and Mb ( S14 ) expressed on yeast surface to 50 nM of hPRDM14 and its homologues , mouse Prdm14 , human PRDM12 and human PRDM6 . ( D ) Competitive binding assay for Mtgr1 and monobodies . Binding of 10 nM Mtgr1 to biotinylated Prdm14 immobilized on streptavidin coated M280 beads in the absence and presence of 500 nM purified monobodies , Mb ( S4 ) and Mb ( S14 ) . ( E ) Co-immunoprecipitation of FLAG–HA tagged Prdm14 expressed in mESC using Mb ( S4 ) , Mb ( S14 ) , α-FLAG M2 antibody or a negative control antibody ( ‘IgG’ ) . Antibodies used for Western blotting are indicated with the blots . E , elution; FT , flow through; Mtgr1 , myeloid translocation gene related 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 01610 . 7554/eLife . 10150 . 017Figure 4—figure supplement 1 . Sequences of monobodies selected against human PRDM14 and their KD values to hPRDM14 measured in yeast display format . Two types of libraries were used , loop and side library . In the library designs , 'X' denotes a mixture of 30% Tyr , 15% Ser , 10% Gly , 5% Phe , 5% Trp , and 2 . 5% each of all the other amino acids except for Cys , 'B' , a mixture of Gly , Ser and Tyr , 'J' , a mixture of Ser and Tyr , 'O' , a mixture of Asn , Asp , His , Ile , Leu , Phe , Tyr , and Val , 'U' , a mixture of His , Leu , Phe and Tyr , 'Z' , a mixture of Ala , Glu , Lys and Thr ( Koide et al . 2012 ) . A hypen indicates a deletion . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 01710 . 7554/eLife . 10150 . 018Figure 4—figure supplement 2 . Human PRDM14 can substitute for the mouse Prdm14 in mESCs . ( A ) Morphological changes in Prdm14−/− mESCs and those reconstituted with mouse or human FH-PRDM14 cDNA , respectively . Note that all lines were grown under serum+LIF conditions . ( B ) Western blot shows that both human and mouse Prdm14 proteins were expressed at similar levels in the reconstituted lines . ( C ) RT-qPCRs measuring changes in gene expression in lines reconstituted with human or mouse Prdm14 relative to the Prdm14−/− line . Genes marked with an asterisk denote direct mouse Prdm14 targets from ChIP-seq . ChIP-seq , chromatin immunoprecipitation with sequencing; LIF , leukemia inhibitory factor; RT-qPCRs , quantitative reverse transcription-polymerase chain reaction . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 01810 . 7554/eLife . 10150 . 019Figure 4—figure supplement 3 . Monobody affinity pulldown of the endogenous Prdm14 protein . ( A ) Co-precipitation of Mtgr1 with endogenous Prdm14 from wt mESC extracts via affinity pulldown with Mb ( S4 ) , Mb ( S14 ) or , as a negative control , beads only followed by detection of Mtgr1 by Western blotting . ( B ) Silver staining of polypeptides recovered in the affinity pulldown with Mb ( S4 ) from wt mESC extracts . ( C ) Mass spectrometry analysis of proteins recovered in the affinity pulldown with Mb ( S4 ) from wt mESC extracts . Results of the top recovered and correctly matched proteins ( prob >5 . 00; prob – log probability of the correct fit to the data ) are shown . Also shown are total intensity , number of spectra and number of unique peptides recovered , percent coverage and number of amino acids in the recovered protein . E , elution; FT , flow through; mESC , mouse embryonic stem cell; Mtgr1 , myeloid translocation gene related 1; W , wash; wt , wild type . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 019 We isolated Prdm14-binding monobodies from two combinatorial phage display libraries termed the ‘loop’ library and the ‘side’ library ( Koide et al . , 2012a ) . Following phage display selection , we performed gene shuffling for affinity maturation and further selection in the yeast display format to isolate clones with high affinity . We identified a total of 12 clones that bound to human PRDM14 ( hPRDM14 ) with KD <100 nM as measured by yeast surface display ( Figure 4A; Figure 4—figure supplement 1 ) . Among them , we identified two clones , Mb ( hPRDM14_S4 ) and Mb ( hPRDM14_S14 ) , that bound hPRDM14 ( residues 238–487 ) at least five-fold stronger than the closest homologues , hPRDM6 ( residues 194–405 ) and hPRDM12 ( residues 60–229 ) ( Figure 4C ) . We will use shorthand names , Mb ( S4 ) and Mb ( S14 ) , for referring to them hereafter for brevity . Additionally , these two monobodies showed comparable binding to the mouse homologue of hPRDM14 , Prdm14 . Of note , hPRDM14 was able to substitute for the mouse Prdm14 in rescue of the Prdm14-/- ESC defects , suggesting both biochemical and biological conservation of function between mouse and humans ( Figure 4—figure supplement 2 ) . We produced these two monobodies as purified proteins for further characterization . Consistent with analysis using yeast surface display , these purified monobodies showed high affinity with KD <50 nM to both hPRDM14 and Prdm14 in bead-based assays ( Figure 4B ) . We then examined whether these monobodies inhibited the interaction of Prdm14 with Mtgr1 . Mb ( S14 ) potently competed against the binding of Mtgr1 to Prdm14 but Mb ( S4 ) did not ( Figure 4D ) . This result suggests that the two monobodies bind to distinct surfaces of Prdm14 and the epitope for Mb ( S14 ) overlaps with and therefore occludes the Mtgr1-binding surface ( Figure 4D ) . We next tested whetherif these monobodies can immunoprecipitate Prdm14 from mESC lysates . Mb ( S4 ) captured Prdm14 and co-immunoprecipitated vast majority of Mtgr1 from the FH-Prdm14 cell extracts ( Figure 4E ) . On the other hand , Mb ( S14 ) captured lower levels of Prdm14 , in agreement with its competition for the same binding surface as Mtgr1 ( Figure 4E ) . Since PRC2 complex has been previously reported to associate with Prdm14 ( Chan et al . , 2013; Payer et al . , 2013; Yamaji et al . , 2013 ) , we also looked for the presence of Suz12 , a PRC2 component . We did not detect immunoprecipitated Suz12 in the elution fraction for either of the two monobodies . Next , we used monobodies to precipitate endogenous Prdm14 from wt ESCs . Immunoblot analysis with α-Mtgr1 antibody showed that Mb ( S4 ) monobody , which does not disrupt Prdm14-Mtgr1 interaction , recovered endogenous Mtgr1 ( and was able to deplete most of it from the extract , Figure 4—figure supplement 3A and B ) . In addition , we performed Prdm14 Mb ( S4 ) -precipitation/mass spec analysis from wt ESCs , which readily detected Prdm14- and Mtgr1-originating peptides , but did not recover any other PRDM proteins confirming a high specificity of this monobody ( Figure 4—figure supplement 3B and C ) . Common to monobodies generated to recognize native , folded proteins , neither Mb ( S4 ) nor Mb ( S14 ) detected denatured Prdm14 in immunoblotting ( not shown ) . Overall , we report here the generation of the first recombinant affinity reagents targeting two distinct sites of Prdm14 SET domain . To understand how Prdm14 and Mtgr1 interact at the atomic level , we attempted crystallization of the Prdm14-Mtgr1 complex . However , aggregation of both proteins resulted in low yields of the complex suitable for crystallization . To overcome this problem , we designed a fusion construct of Prdm14 and Mtgr1 in which the two proteins were linked with a ten-residue linker ( GSSGSSGS ) , a common strategy for stabilizing heterodimers ( Ernst et al . , 2014; Kobe et al . , 2015; Reddy Chichili et al . , 2013; Zhou et al . , 2015 ) . To confirm that the linker did not distort the Prdm14–Mtgr1 complex , we performed a series of experiments . The fusion protein had the same retention time on size-exclusion chromatography as the unlinked complex ( Figure 5—figure supplement 1A ) , indicating that the linker did not alter the stoichiometry of the complex . We then compared the fusion and the unlinked complex using solution nuclear magnetic resonance ( NMR ) spectroscopy . Most of the cross peaks in the heteronuclear single quantum coherence ( HSQC ) spectrum of 15N-Prdm14 in complex with unlabeled Mtgr1 ( where we observe signals only from 15N-Prdm14 ) overlapped with those in the HSQC spectrum of 15N-labeled Prdm14-linker-Mtgr1 ( where we observe signals from the entire fusion protein including both Prdm14 and Mtgr1 ) ( Figure 5—figure supplement 1B ) . The large number of overlapping peaks in the two spectra strongly suggests that the Prdm14 protein takes on nearly identical average conformation in the unlinked complex and the fusion protein ( Figure 5—figure supplement 1B ) . Furthermore , the fusion protein and the unlinked complex had the same affinity to Mb ( S4 ) , indicating that the linker did not distort the Prdm14 epitope for the monobody ( Figure 5—figure supplement 1C ) . This fusion construct allowed us to overcome the aggregation problem , but it still yielded no crystals in crystallization trials using over 500 conditions . We then used Mb ( S4 ) , the monobody that did not inhibit the Prdm14-Mtgr1 interaction , as a crystallization chaperone . Monobodies , like antibody fragments , often facilitate the crystallization of otherwise recalcitrant systems ( Koide , 2009; Stockbridge et al . , 2015 ) . The addition of Mb ( S4 ) readily led to crystallization of Prdm14-linker-Mtgr1 , and we determined its structure to a resolution of 3 . 06 Å through single wavelength anomalous diffraction ( SAD ) phasing using selenomethionine-labeled crystals ( Figure 5A; Table 1; Figure 5—figure supplement 2A ) . The crystallized complex had two Prdm14-linker-Mtgr1/Mb ( S4 ) complexes in the asymmetric unit . As expected , the monobody bound exclusively to Prdm14 , burying 604 Å2 surface areas , a similar interface size to other monobody/target complexes ( Figure 5—figure supplement 2B , E ) ( Gilbreth et al . , 2008; Wojcik et al . , 2010 ) . In the crystal , monobody–monobody interactions facilitated crystal contacts via face-to-face interactions of the β-sheet surfaces ( not involved in Prdm14 interaction ) , illustrating the importance of Mb ( S4 ) as a crystallization chaperone for this complex ( Figure 5—figure supplement 2D ) . 10 . 7554/eLife . 10150 . 020Figure 5 . Crystal structure of the Prdm14–Mtgr1 complex . ( A ) The overall structure of the Prdm14-linker–Mtgr1/Mb ( S4 ) complex . Missing residues are shown with dotted lines . The pre-SET region ( in orange ) , SET domain ( in yellow ) and post-SET region ( in green ) of Prdm14 are shown . The Mtgr1 helices are marked for clarity ( αA–αD ) . ( B ) Superposition of the Prdm14 crystal structure with the crystal structure of hPRDM12 ( PDB ID 3EP0 ) . ( C ) Superposition of the Mtgr1 crystal structure with the NMR structure of the MTG8 NHR1 ( eTAFH ) domain ( PDB ID 2KNH ) . ( D ) Prdm14–Mtgr1 interface . Prdm14 is shown in white surface representation with the interacting residues in yellow ( Top ) . Mtgr1 is shown in cartoon representation in pink color . In the detailed view , Prdm14 residues are marked in red and Mtgr1 residues are marked in black . Salt bridges and hydrogen bonds between Prdm14 and Mtgr1 are shown in dotted lines . Residues that were mutated based on the structure are underlined . ( E ) Binding of wt , E294K and Y339R Prdm14 ( residues 184–373 ) to immobilized wt Mtgr1 ( top ) and Mtgr1 ( K109E ) ( bottom ) in a bead-based assay . ( F ) Mtgr1 is shown in white surface representation with interacting residues in pink ( left ) . Non-identical residues between Mtgr1 and the other ETO proteins are shown in blue ( left ) . The detailed view shows the interaction of Prdm14 residues N-terminal to the SET domain ( pre-SET ) with Mtgr1 ( right ) . Prdm14 residues are labeled in red and Mtgr1 residues in black . Mtgr1 , myeloid translocation gene related 1; NMR , nuclear magnetic resonance . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 02010 . 7554/eLife . 10150 . 021Figure 5—figure supplement 1 . Inclusion of a linker does not affect the Prdm14–Mtgr1 interaction . ( A ) Elution profiles of Prdm14-linker–Mtgr1 ( in black ) and Prdm14 mixed with Mtgr1 in a 1:2 molar ratio ( in red ) from a Superdex75 size-exclusion column . ( B ) Overlay of the 15N-1H HSQC spectrum of 15N-Prdm14-linker-Mtgr1 complex ( black ) with that of 15N-labeled Prdm14 in complex with unlabeled Mtgr1 ( red ) . ( C ) Titration curves for binding of Prdm14/Mtgr1 complex and Prdm14-linker-Mtgr1 to Mb ( S4 ) . The error bars shown on each data points are the standard deviation from triplicate measurements . The curves show the best fit of the 1:1 binding model . The errors for the KD values are the standard deviations from triplicate measurements . Mtgr1 , myeloid translocation gene related 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 02110 . 7554/eLife . 10150 . 022Figure 5—figure supplement 2 . Structural features of the Prdm14-linker–Mtgr1/Mb ( S4 ) complex . ( A ) Crystals of the Prdm14-linker–Mtgr1/Mb ( S4 ) complex obtained in 16% PEG3350 , 8% Tascimate pH 5 . 3 . ( B ) The Prdm14-monobody interface . Prdm14 is shown in white surface representation with the interface residues on Prdm14 shown in yellow and are labeled in black . Mb ( S4 ) is shown in cartoon representation in cyan and are labeled in red . The upper box illustrates prominent features in the interface involving residues in the FG loop of the monobody . Trp81 of Mb ( S4 ) is in the pocket formed by Pro250 , Phe270 , Val268 , Asp262 and Ala265 of Prdm14 . Gln75 , Tyr77 and Trp81 of Mb ( S4 ) form hydrogen bonds with Prdm14 Arg269 , Asp262 and Glu251 , respectively . Hydrogen bonds and salt bridges are shown in black dotted lines . The bottom box shows a salt bridge formed between Prdm14 Glu340 and Lys47 in the β-strand C of Mb ( S4 ) . ( C ) Unbiased feature-enhanced map showing electron density at the 1 sigma level for residues involved in Prdm14–Mtgr1 interaction . ( D ) Crystal contacts between monobodies and Prdm14 from neighboring molecules . The symmetry related molecules are shown with Prdm14-linker–Mtgr1 in yellow and monobody in cyan . ( E ) Binding of Mb ( S4 ) and its point mutant , Trp81 to Ala , to biotinylated Prdm14 immobilized on streptavidin M280 beads , as measured using the bead-based binding assay . Mutation of Trp81 results in substantial loss of Prdm14 binding , confirming the interface between Prdm14 and Mb ( S4 ) shown in panel B . Mtgr1 , myeloid translocation gene related 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 02210 . 7554/eLife . 10150 . 023Figure 5—figure supplement 3 . Comparison of the structure of the Prdm14–linker-Mtgr1 complex with that of the Prdm9- histone H3 peptide-S-adenosyl-L-homocysteine ( AdoHcy ) complex . ( A ) Left , the crystal structure of the Prdm9 complex ( PDB ID 4C1Q ) . Prdm9 is shown in cartoon representation in gray and the histone peptide is shown in sticks in blue . AdoHcy is shown in green sticks . Right , cartoon representation of Prdm14 in gray with Mtgr1 shown in pink ribbon representation . The residues in pre-SET region of Prdm14 that occupy the region equivalent to the AdoHcy binding site in Prdm9 are shown in orange . ( B ) Left , surface representation of Prdm9 in which the surfaces for the atoms within 4 . 5 Å from the peptide are marked in blue . Right , Mtgr1-interacting residues of Prdm14 are shown in pink . ( C ) . Left , residues in the Prdm9 catalytic site ( labeled in black ) and the Lys4me2 of the histone peptide ( labeled in red ) , shown in sticks . Tyr357 is directly involved in the catalytic activity of Prdm9 . Right , the same region of Prdm14 . Phe273 and Tyr339 in Prdm14 are respectively equivalent to Tyr276 and Tyr341 in Prdm9 . Lys109 of Mtgr1 present in proximity to the Prdm14-Tyr355 ( equivalent to Tyr357 in Prdm9 ) is shown . Also shown are Pro366 and Glu294 that have atoms within 4 . 5 Å of Mtgr1-Lys109 . Prdm14 residues are labeled in black . K109 of Mtgr1 is labeled in red . Mtgr1 , myeloid translocation gene related 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 023 The Prdm14 SET domain ( residues 240–356 ) in the crystal structure is flanked by pre-SET and post-SET regions . The crystallization construct includes residues 184–239 that precede the SET domain ( pre-SET ) and residues 357–373 following the SET domain ( post-SET ) . Unlike many SET domain-containing proteins , Prdm14 does not have cysteine-rich Zn finger domains adjacent to the SET domain , commonly termed pre-SET and post-SET domains . Thus , in the absence of well-defined domains , we refer to these adjacent segments as pre-SET and post-SET regions . The crystal structure of the mouse Prdm14 SET domain is very similar to that of the hPRDM12 SET domain , the closest human homologue of Prdm14 ( PDB ID 3EP0; Cα RMSD=0 . 99; Figure 5B ) . The Prdm14 SET domain in our structure has a total of nine β-strands ( β1-β9 ) arranged in three antiparallel β-sheets with a short 310 helix ( η1 ) inserted between β6 and β7 . The pre-SET region in our construct has a short helix at the N-terminus followed by a long structurally disordered region a part of which ( residues 217–239 ) has no detectable electron density even at 0 . 5 σ contour levels ( 2Fo-Fc ) . The residues that constitute the post-SET region at the C-terminus to the SET domain are arranged in an antiparallel beta sheet ( β10 and β11 ) . Overall , the structural features of the PR/SET domain in Prdm14 show no major differences with other PR/SET domains in Prdm proteins . 10 . 7554/eLife . 10150 . 024Table 1 . Data collection , phasing and refinement statistics for Prdm14-linker-Mtgr1/Mb ( S4 ) complex crystals . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 024NativeSeMet SADData collection BeamlineAPS 19IDAPS 19IDSpace groupP43212P43212Cell dimensions a , b , c ( Å ) 106 . 8 , 106 . 8 , 180 . 7106 . 9 , 106 . 9 , 180 . 9a , b , g ( ° ) 90 , 90 , 9090 , 90 , 90Peak Wavelength0 . 97918 Å0 . 97918 ÅResolution ( Å ) 37 . 7–3 . 05 ( 3 . 16–3 . 05 ) 50–3 . 18 ( 3 . 23–3 . 18 ) Rpim 0 . 024 ( 0 . 482 ) 0 . 022 ( 0 . 315 ) I / σI 30 . 0 ( 1 . 4 ) 52 . 4 ( 2 . 0 ) Completeness ( % ) 100 ( 100 ) 100 ( 100 ) Redundancy20 . 5 ( 19 . 5 ) 86 . 2 ( 45 . 2 ) Refinement Resolution ( Å ) 37 . 7–3 . 05 ( 3 . 16–3 . 05 ) No . of unique reflections20631 ( 2013 ) Rwork / Rfree 0 . 189/0 . 250No . atoms5476Protein5476Ligand/ion0Water0B-factors114 . 3Protein114 . 3Ligand/ion0Water0R . m . s deviationsBond lengths ( Å ) 0 . 006Bond angles ( ° ) 1 . 13Ramachandran statistics Favorable 95 . 8Allowed 4 . 1Outliers 0 . 1 The Mtgr1 NHR1 domain ( also called the TAFH domain ) is highly conserved in the ETO family . The Mtgr1 NHR1 domain in the crystal structure contains four well-defined α-helices arranged in a bundle ( αA-αD; Figure 5C ) . Currently , three NMR structures for the NHR1 domain of human MTG8 , another member of the ETO family , are available ( Park et al . , 2009; Plevin et al . , 2006; Wei et al . , 2007 ) . Mtgr1 in the crystal structure has almost identical topology as the average solution NMR structure of the NHR1 domain from MTG8 ( ETO ) in complex with a stabilizing peptide ( PDB ID 2KNH; Cα RMSD=1 . 77 Å; Figure 5C ) . Similar structural features and high sequence identity for the NHR1 domain in ETO proteins explain the pull-down of Mtg8 and Mtg16 in the FH-Prdm14 purifications ( Figure 5F ( left ) ; Figure 1A; Figure 1—source data 1 ) . The Mtgr1-Prdm14 interaction interface buries 2180 Å2 , a large interface but still within the observed range for high-affinity protein–protein interaction interfaces with a low nanomolar KD value ( Lo Conte et al . , 1999 ) . We observe that Mtgr1 interacts with both SET domain and pre-SET region of Prdm14 ( Figure 5A ) , as was predicted from our earlier interaction mapping co-immunoprecipitation experiments ( Figure 1C ) . The interactions between the Prdm14 SET domain and Mtgr1 , mediated primarily by αA and αD , contribute 65% of the total buried surface area . The substantial interaction interface between the pre-SET region and Mtgr1 ( 35% of the interface ) rationalizes the importance of the pre-SET region in the Prdm14-Mtgr1 interaction ( Figure 1C ) . Several features in the interface are notable . Arg105 located at the N-terminal end of Mtgr1 αA sits in a pocket formed by Ser290 , Met292 , Cys319 and Tyr339 of Prdm14 and its guanidinium moiety makes hydrogen bonds with the side chains of all of these residues ( Figure 5D , Figure 5—figure supplement 2C ) . Mutating Mtgr1 Arg105 to Asp resulted in a loss of detectable binding ( Figure 5E ) , and replacing Prdm14 Tyr339 of the pocket with Arg substantially reduced binding ( Figure 5E ) . In addition , Lys109 in Mtgr1 αA interacts with Glu294 , Tyr355 and Pro366 of Prdm14 SET domain . Lys109 in Mtgr1 forms a salt bridge with Glu294 from β5 of the Prdm14 SET domain ( Figure 5D ) . Mutating either of these residues led to a loss of detectable binding ( Figure 5E ) . Remarkably , charge reversal of this ionic interaction , that is , the combination of Mtgr1 K109E and Prdm14 E294K restored binding , in agreement with the salt bridge formation across the binding interface in solution ( Figure 5E middle ) . These mutation experiments together supports the authenticity of the binding interface observed in the crystal structure and identified critical electrostatic interactions . The Mtgr1 interaction with Prdm14 pre-SET region involves residues from Mtgr1 helix αA , αB and αD ( Figure 5A ) . Several Leu and Val residues from the helix in the pre-SET region contribute to hydrophobic interactions with residues in Mtgr1 helices ( Figure 5F ) . In addition , Mtgr1 Asn141 forms hydrogen bonds with residues Phe185 and Phe187 from the helix in the pre-SET region . These results further reinforce the observation that Prdm14 utilizes both pre-SET region and SET domain to interact with Mtgr1 . Given strong association between Prdm14 and Mtgr1 , as well as phenotypic similarities upon loss of either protein , we next examined whether the lack of Prdm14–Mtgr1 interaction would have a biological effect on mESC gene expression and the efficiency of PGC-LC induction . To this end , we first confirmed that the point mutations designed based on the crystal structure ( Figure 5D and E ) also disrupted the interaction between the full-length proteins in cells , and that the combined charge reversal mutations restored the interaction ( Figure 6A ) . Then , we reconstituted Stella:GFP Prdm14−/− mESCs with cDNAs encoding either wt or mutant ( E294K or Y339R ) FH-Prdm14 and confirmed that all three proteins were expressed at similar levels ( Figure 6B ) . Next , we used RNA-seq to compare gene expression patterns of these mESCs after transition from 2i+LIF to serum+LIF culture ( Figure 6C ) . In the Prdm14−/− cells reconstituted with mutant Prdm14 , we noted elevated expression of genes associated with differentiation to epiblast and extraembryonic endoderm , and diminished expression of naïve pluripotency genes ( with an exception of Tet2 ) , similar to attenuated expression patterns ( albeit not to the same degree ) as we observed upon loss of Prdm14 ( Figure 6C , compare to Figure 3A ) . Principal component and clustering based on differential gene expression further support the notion that Prdm14 E294K lines show hypomorphic expression profile , which falls in between the Prdm14−/− ESCs and those reconstituted with wt Prdm14 ( Figure 6—figure supplement 1 ) . Finally , the efficiency of mPGC-LCs formation from mESC reconstituted with Prdm14 mutants ( E294K , Y339R ) was significantly diminished compared to cells rescued with the wt Prdm14 to almost the same degree as in Prdm14−/− cells ( Figure 6D ) . 10 . 7554/eLife . 10150 . 025Figure 6 . Inhibition of Prdm14–Mtgr1 interaction affects stem cell maintenance and PGC-LC induction . ( A ) Single amino-acid substitutions at the interaction surface abrogate Prdm14–Mtgr1 association in cells . Indicated full length V5-Mtgr1 wt or single point mutant proteins were introduced to HEK293 cells and co-immunoprecipitated with full length FH-Prdm14 wt or single mutant protein as indicated in the diagram . Please note rescue of the association when the combination of Mtgr1 K109E and Prdm14 E294K mutants is tested . ( B ) Western blot showing protein expression levels of wt , E294K and Y339R FH-Prdm14 protein in lysates from Prdm14−/− mESCs cells reconstituted with the respective transgenes . ( C ) RNA-seq from Prdm14−/− cells reconstituted with wt Prdm14 protein ( x axis ) were compared to Prdm14−/− cells reconstituted with E294K Prdm14 mutant ( y axis ) and expression values ( RPKM ) of all significantly changed transcripts were plotted . The transcripts of specific genes are highlighted in red , green , blue or black as indicated; shaded colors indicate no significant difference . ( D ) Quantification of GFP signal as a measure of mPGC-LC induction from Prdm14−/− cells and Prdm14−/− cells reconstituted with transgenes encoding wt , E294K , or Y339R Prdm14 protein . ( E ) Schematics of the piggyBac transposon-based reporter system used to create dual reporter lines . mESC line was transfected with either dox-inducible mCherry construct or dox-inducible mCherry-Mb ( S14 ) fusion protein . The lines were selected using blasticidin and three populations were tested further for their competency to form mPGC-LCs . mESC to mEpiLC transition followed by mPGC-LC transition using defined media in cells containing Stella:GFP reporter ( lower panel ) . Doxycycline was added after mEpiLC stage . ( F ) Quantification of GFP signal as a measure of mPGC-LC induction from mCherry population of cells and mCherry-Mb ( S14 ) population of cells with and without addition of doxycycline . mEpiLCs , mouse epiblast-like cells; mESCs , mouse embryonic stem cell; mPGC-LCs , mouse primordial germ cell-like cells; GFP , green fluorescent protein; Mtgr1 , myeloid translocation gene related 1; RNA-seq , RNA sequencing; RPKM , reads per kilobase of exon per million reads mapped; WCL , whole cell lysates; wt , wild type . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 02510 . 7554/eLife . 10150 . 026Figure 6—figure supplement 1 . Gene expression analyses of Prdm14−/− cells reconstituted with wt or E294K Prdm14 . ( A ) PCA of RNA-seq expression comparisons of the top 898 genes with the highest median average deviation in indicated ESC lines . ( B ) Heatmap displaying top 50 variable genes between Mtgr1−/− ( three clones ) , Prdm14−/− ( 2 clones ) mESCs or Prdm14−/− clones reconstituted with wt or mutant ( E294K ) Prdm14 protein . ESC , embryonic stem cell; PCA , principal component analysis; mESCs , mouse embryonic stem cell; Mtgr1 , myeloid translocation gene related 1; RNA-seq , RNA sequencing . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 02610 . 7554/eLife . 10150 . 027Figure 6—figure supplement 2 . Inhibition of Prdm14–Mtgr1 interaction affects germ cell development . ( A ) FACS plots of the gated quantification of GFP signal shown in Figure 6D as a measure of mPGC-LC induction from wt cells , Prdm14−/− cells or Prdm14−/− clones reconstituted with wt or mutant ( E294K , Y339R ) Prdm14 . ( B ) FACS plots of the gated quantification of GFP signal shown in Figure 6F as a measure of mPGC-LC induction from three independent populations of mCherry cells or mCherry-Mb ( S14 ) cells with and without addition of doxycycline . FACS , fluorescence-activated cell sorting; GFP , green fluorescent protein; wt , wild type . DOI: http://dx . doi . org/10 . 7554/eLife . 10150 . 027 Our results thus demonstrate that association of Prdm14 with Mtgr1 is required for mediating its functions in pluripotency and germ cell formation . Given that one of the monobodies we developed , Mb ( S14 ) , binds to Prdm14 competitively with Mtgr1 , we hypothesized that this reagent can be utilized to inhibit Prdm14 function in living cells or organisms in a highly controlled manner . To provide a proof-of-principle for such strategy , we engineered piggyBac doxycycline-inducible constructs encoding mCherry-Mb ( S14 ) fusion or , as a control , mCherry alone , and introduced them into Stella:GFP reporter mESCs ( Figure 6E ) . Next , we induced mPGC-LC formation from mESCs in the absence or presence of doxycycline ( added during the induction of mPGC-LCs from mEpiLCs ) to activate monobody expression . We observed consistent reduction in PGC-LC formation efficiency in cells expressing the mCherry-Mb ( S14 ) fusion protein , as compared with the same cell population without addition of doxycycline or to cells expressing mCherry alone ( Figure 6F , Figure 6—figure supplement 1 ) . A more moderate effect of the monobody ( compared with the Prdm14 E294K mutation , Figure 6D ) is probably due to the fact that this monobody needs to compete against the high-affinity interaction between Prdm14 and Mtgr1 . In addition , we noticed a short half-life of the mCherry-Mb ( S14 ) fusion protein ( data not shown ) , which may contribute to the moderate effect but could also facilitate inhibition with high temporal resolution in the future . Thus , Mb ( S14 ) represents a novel tool that can be utilized to perturb Prdm14 function in living cells during a dynamic biological process such as the PGC induction and can be further modified in the future for addressing particular questions using standard protein-engineering technologies . Our study identified the ETO family co-repressor Mtgr1 as a new regulator of mESC identity , which facilitates molecular functions of Prdm14 through direct binding to its pre-SET/SET region . Therefore , although no evidence has been previously found for the catalytic activity of the Prdm14 SET domain , our data demonstrate that this domain is nonetheless essential for Prdm14 function by mediating the interaction with its key partner , Mtgr1 . Prdm14 and Mtgr1 co-occupy distal regulatory regions of many target genes linked to differentiation , DNA methylation and chromatin modification , consistent with their tight interaction . Loss of either protein results in upregulation of a subset of these target genes and a gradual loss of mESC self-renewal . However , similar to what has been observed previously for Prdm14 , Mtgr1 is not required for maintenance of mESC under 2i+LIF . This can be explained by the fact that the Fgf/Erk pathway , a major signaling cascade driving both epiblast and extraembryonic endoderm differentiation ( to which loss of Prdm14/Mtgr1 sensitizes cells ) , is inhibited under these conditions ( Nichols and Smith , 2009 ) . Concordant upregulation of gene transcripts in Prdm14−/− or Mtgr1−/− cells compared with wt mESCs supports function of the Prdm14–Mtgr1 complex in gene silencing and agrees with the reported role of Mtgr1 as an HDAC-recruiting co-repressor ( Rossetti et al . , 2004 ) . Notably , we detected HDACs 1–3 in our Prdm14 immunoprecipitates , and HDAC3 and other NCoR1 repressive complex components in our purifications with the Mb ( S4 ) monobody , suggesting that the Prdm14–Mtgr1-dependent repression may indeed be facilitated by histone deacetylation and that Mtgr1 is required for recruiting HDACs to Prdm14-binding loci . Furthermore , although a subset of genomic sites occupied by Mtgr1 occurs at Prdm14 motif-lacking sites , the high similarity of transcriptional changes observed upon loss of either protein suggests that the major impact of Mtgr1 on gene expression and cell identity of mESCs is in the context of its association with Prdm14 . Moreover , our data suggest that Prdm14 can not only guide , but through change in its levels , quantitatively tune the degree of interaction of ubiquitously expressed Mtgr1 with chromatin . Our results from the in vitro PGC-LC formation model strongly suggest that in addition to its role in mESCs , Mtgr1 is also a critical mediator of the Prdm14 function in germline development . Three lines of evidence support this notion: ( i ) Mtgr1 deletion; ( ii ) single point mutations in Prdm14 disrupting association with Mtgr1; and ( iii ) expression of a monobody targeting the Prdm14–Mtgr1 interaction surface , all of which hinder PGC-LC induction in vitro . It is therefore surprising that a Mtgr1 mouse knockout strain has been reported as viable and fertile ( Amann et al . , 2005 ) . However , it remains unclear whether this strain represents a true loss-of-function , because in the mouse targeting strategy , the first six exons are preserved ( encoding amino acids 1–316 , which span the Prdm14 interaction region and NHR2 domain involved in dimerization and association with mSin3/HDAC complex ) ( Rossetti et al . , 2004 ) . In light of our findings and aforementioned caveats , we suggest that the in vivo function of Mtgr1 in the germline should be revisited . It is also possible , however , that requirement for Mtgr1 in germ cell development in vivo can be fully or partially compensated by the other ETO proteins . Because residues that form the Prdm14 binding interface in Mtgr1 are conserved among the ETO family members , we expect that the other ETO proteins would bind directly to the Prdm14 pre-SET/SET regions and could therefore compensate for the absence of Mtgr1 function . Our study provides novel insights into how pre-SET and SET regions might mediate high affinity protein–protein interactions . While many structures of the catalytic SET domains have been obtained with their substrates ( typically , histone tails ) , to the best of our knowledge , our study represents the first structural analysis of the SET domain acting as a module for a high affinity protein–protein interaction . Interestingly , a comparison of our structure with that of the Prdm9–AdoHcy–histone peptide complex indicates that the surface of the Prdm14 pre-SET and SET regions engaged in interaction with Mtgr1 overlaps with surfaces other Prdm proteins use for binding to their histone peptide substrate ( Figure 5—figure supplement 3B; [Wu et al . , 2013] ) . Indeed , Mtgr1 Lys109 is in close proximity to Tyr355 that corresponds to the catalytic Tyr based on the consensus SET domain sequence ( Smith and Denu , 2009 ) . In Prdm9 , the catalytic tyrosine Tyr356 along with Tyr276 and Tyr341 form the Lys4me2 binding pocket and are critical for catalytic activity ( Wu et al . , 2013 ) . In Prdm14 , the side chain of Tyr355 flips over ( with respect to the conformation of Tyr357 in Prdm9 ) and interacts with K109 of Mtgr1 . The location of K109 is distinct from that of Lys4me2 in Prdm9 ( Figure 5—figure supplement 3C ) . In addition , the high affinity of the Prdm14–Mtgr1 interaction strongly suggests that Mtgr1 would be a poor substrate , given that a substrate needs to be released after catalysis for efficient enzyme reaction . We also note that His211 and Ala212 in the pre-SET region of Prdm14 occupy the S-adenosyl-L-methionine ( SAM ) -binding site ( Figure 5—figure supplement 3A ) and thus , at least in the context of the presented structure , preclude the binding of this cofactor necessary for methylation . Taken together , our results suggest that Mtgr1 Lys109 is unlikely to be an actual substrate for Prdm14-mediated methylation . Consistent with this notion , radioactive in vitro methyltransferase assays with recombinant Prdm14 and Mtgr1 proteins , their respective interaction mutants , as well as histone substrates , all failed to yield methyltransferase activity ( not shown ) . However , we cannot exclude the possibility that under presently unknown conditions , such activity can ultimately be found . Regardless , our comparisons suggest that substrate-binding surfaces can , in some SET domain proteins , be co-opted for mediating high affinity protein–protein interactions , which may provide a molecular explanation as to why such surfaces are typically highly conserved even in SET domain proteins with no apparent catalytic activity . Lastly , aberrant reactivation of the PRDM14 locus is associated with a variety of human cancers , and mice overexpressing Prdm14 in blood cells develop early-onset T-cell acute lymphoblastic leukemia ( T-ALL ) ( Carofino et al . , 2013 ) . We speculate that the oncogenic function of Prdm14 may be mediated by the formation of a complex with Mtgr1 , which is broadly expressed and thus readily available for association in many different organ systems , or perhaps with other ETO proteins using the same interface . If this proves to be the case , inhibition of the Mtgr1-interaction surface on Prdm14 could be an attractive target for therapy as Prdm14 expression is restricted under non-pathological conditions to the preimplantation embryo and PGCs , reducing the risk of off-target effects on normal somatic tissues . Thus , the monobodies directed to Prdm14 generated in this study will be powerful tools for testing the 'druggability' of the Prdm14–Mtgr1 interaction . Indeed , we have introduced monobodies into CML cells and demonstrated the potential druggability of a domain interface in Bcr-Abl ( Grebien et al . , 2011 ) . In addition , the crystal structure shows that Mb ( S4 ) binds to a distinct surface of Prdm14 . Because monobodies usually bind to functional sites on target proteins , we hypothesize that the Mb ( S4 ) epitope may also be important for Prdm14 function , which will be a subject of future research . We emphasize that the genetically encoded monobodies are portable tools , as they can be readily introduced into different cells via transfection or viral transduction . This attribute should facilitate the investigation of Prdm14 functions in diverse contexts . Stable lines expressing tagged Prdm14 were established from single colonies by transducing LF2 mESCs with FH-Prdm14 pTrip lentivirus for 24 hr , followed by selection with neomycin as described previously ( Ma et al . , 2011 ) ; these lines were used for immunoprecipitations followed by mass spectrometry experiments . Stella:GFP cells were a gift from A . Surani and were used for all other experiments and further genetic manipulations , unless indicated otherwise . This line contains a transgene spanning 10 kb upstream of the Stella transcriptional start site , exon1 , intron1 , and part of exon2 , followed by eGFP fused in-frame and SV40 polyadenylation sequence . Stella:GFP line acts as a transcriptional reporter and has been previously shown to faithfully recapitulate endogenous Stella expression in mouse and mark PGCs as early as E7 . 5 ( Payer et al . , 2006 ) . Stable double reporter lines ( Stella:GFP and mCherry ) were created by transfecting piggyBac Tet-On expression plasmid controlled by rtTA and doxycycline ( mCherry alone or mCherry fused to Mb ( S14 ) ) with transposase and selected with blasticidin for 7 days . Single clones were picked and expanded further . For maintenance , all mouse ESC lines were grown in so-called ‘2i+LIF’ medium that is serum-free N2B27-based medium supplemented with MEK inhibitor PD0325901 ( 0 . 8 μM ) and GSK3β inhibitor CHIR99021 ( 3 . 3 μM ) in tissue culture ( TC ) dishes pretreated with 7 . 5 μg/ml polyl-ornithine ( Sigma ) and 5 μg/ml laminine ( BD ) ( Hayashi and Saitou , 2013; Hayashi et al . , 2011 ) . For ChIP-seq , RNA-seq and immunoprecipitation experiments , Stella:GFP mESC lines and FH-Prdm14 derivatives were cultured for 5 days in feeder-free conditions in Dulbecco's Modified Eagle Medium ( DMEM ) -high glucose medium ( DMEM/high glucose; HyClone ) containing 15% ( v/v ) fetal bovine serum , 1 mM glutamine , penicillin/streptomycin , 0 . 1mM nonessential amino acids , 0 . 1 mM 2-mercaptoethanol and supplemented with LIF ( serum+LIF conditions ) . To form embryoid bodies , Stella:GFP mESCs line of interest was maintained in a 15 cm Petri dish with serum culture medium without LIF , to allow aggregation as a hanging drop with 400 cells per drop ( 360 drops per mESC line ) . Cells were cultured for 4 days and half of them were collected for further analysis by RT-qPCR , while the other half was transferred into non-adherent Petri dish and allowed to grow for another 4 days ( 8 days total ) before RT-qPCR analysis . To induce mEpiLC differentiation , mESC were washed with phosphate-buffered saline , trypsinized , and strained . A total of about 100 , 000 cells per one well of 12-well plate were plated on TC dishes pretreated with 5 μg/ml fibronectin ( Millipore ) in N2B27-based medium supplemented with 1% KSR ( Invitrogen ) and 12 μg/ml bFGF ( Peprotech ) . The mPGC-LCs were induced similarly to what has been described previously for 6 days ( Hayashi and Saitou , 2013; Hayashi et al . , 2011 ) . Specifically , 1000–2000 mEpiLC cells were aggregated in a hanging drop in a serum-free medium ( GMEM , Invitrogen ) supplemented with 15% KSR , 0 . 1 mM NEAA , 1 mM sodium pyruvate , 0 . 1 mM 2-mercaptoethanol , penicillin/streptomycin , 1 mM glutamine and cytokines 500 ng/ml BMP4 ( R&D Systems ) , 500 ng/ml BMP8a ( R&D Systems ) , 100 ng/ml SCF ( R&D systems ) , 50 ng/ml EGF ( R&D systems ) , LIF . Where indicated doxycycline was added at the point of mPGC-LCs induction from mEpiLCs . Total RNA was isolated with Trizol and afterwards treated with turbo DNase . For reverse transcription of mRNAs , we used 1 μg of DNAse digested RNA , random hexamer primers ( 5×TransAmp Buffer , Bioline ) and reverse transcriptase ( Bioline ) in 20-μl reaction volume . qPCR analyses were carried out with SensiFAST SYBR No-ROX kit ( Bioline ) on LightCycler 480 II qPCR machine ( Roche ) . RNAs from at least two independent biological replicates of indicated cell lines were extracted with Trizol ( Invitrogen ) , following the manufacturer’s recommendations . Ten micrograms of total RNA were subjected to two rounds purification using Dynaloligo-dT beads ( Invitrogen ) . Purified RNA was fragmented with 10× fragmentation buffer ( Ambion ) and used for first-strand cDNA synthesis , using random hexamer primers ( Invitrogen ) and SuperScript II enzyme ( Invitrogen ) . Second strand cDNA was obtained by adding RNaseH ( Invitrogen ) and DNA Pol I ( New England BioLabs ) . The resulting double-stranded cDNA was used for Illumina library preparation and sequenced with Illumina Genome Analyzer . Following library preparation , samples were pooled and sequenced on an Illumina NextSeq instrument using 76 base-pair single-end reads on a NextSeq high output kit ( Illumina ) or HiSeq instrument using 51 base-pair single-end reads . ChIP assays were performed from 107 mESC per experiment , according to previously described protocol with slight modification ( Rada-Iglesias et al . , 2011 ) . Briefly , cells were crosslinked with 1% formaldehyde for 10 min at room temperature and the reaction was quenched by glycine at a final concentration of 0 . 125 M . Chromatin was sonicated to an average size of 0 . 5–2 kb , using Bioruptor ( Diagenode ) . A total of 5 μg of antibody was added to the sonicated chromatin and incubated overnight at 4°C . Subsequently , 50 μl of protein G Dynal magnetic beads were added to the ChIP reactions and incubated for ~4 hr at 4°C . Magnetic beads were washed and chromatin eluted , followed by reversal of crosslinks and DNA purification . ChIP DNA was dissolved in water . ChIP-seq and input libraries were prepared according to Illumina protocol and sequenced using Illumina Genome Analyzer . Following library preparation , samples were pooled and sequenced on an Illumina NextSeq instrument using 76 base-pair single-end reads on a NextSeq high output kit ( Illumina ) . Quality of FASTQ files was assessed using FastQC software . Raw sequencing reads were aligned using Tophat against mm9 genomic index and with refseq gene models as available at illumina . com ftp site . Aligned reads were converted to counts for every gene using HTSeq and gene counts were further analyzed using R and DESeq2 package ( Love et al . , 2014 ) . For the scatter plot we identified significantly affected transcripts and plotted RPKMs of transcripts significantly ( q<0 . 05 ) affected by Prdm14 or Mtgr1 loss . We compiled from published literature a set of official gene symbols for representative marker genes characteristic for epiblast , extraembryonic endoderm and naïve lineages . The heatmap was created by looking at top 100 genes with the highest variance across samples ( topVarGenes ) . We looked at the amount by which each gene deviates in a specific sample from the gene’s average across all samples; thus , we centered and scaled each gene’s values across samples and then plotted a heatmap . To visualize sample-to-sample distances between different lines we used PCA , plotPCA function within DESeq2 package in R on the rlog-transformed counts . Quality of FASTQ files was assessed using FastQC software . ChIPseq peak calls were done with MACS2 callpeak with default settings ( https://pypi . python . org/pypi/MACS2 ) . Superset of intervals was created by merging summits from all calls using mean shift algorithm with 300 bp bandwidth . The modal peaks were extended ± 300 bp and read coverage was calculated with bedtools . Regions with outlier counts in negative controls were excluded from further analysis . Motifs enriched in Mtgr1 ChIP peaks were obtained with SeqPos ( He et al . , 2010 ) , using a set of 1884 coordinates for top Mtgr1 peaks with signal higher in Prdm14 overexpressing cells than in wt cells ( Prdm14-dependent ) and 1721 Mtgr1 peaks with signal higher in wt cell than in the Prdm14 overexpression background ( Prdm14-independent ) . ChIP regions containing the MTGR1/Prdm14 motif were identified with FIMO with P value cutoff set to 0 . 001 . In wt mESCs 64% of Prdm14-dependent sites contain the Prdm14 motif and 16% of Prdm14-independent sites contain the Prdm14 motif ( p-value cutoff at 0 . 0003 ) . This corresponds to odds ratio 9 . 84 and p<<10–16 in Fisher’s exact test . Wt Cas9 plasmid pX330 was obtained from Addgene . The sgRNAs sequences were designed using Zhang Lab website ( http://crispr . mit . edu/ ) . Guide for Prdm14 was in exon 2 ( CGCCGCCGAGGACCAAATTTTGG , score 95 ) and guide for Mtgr1 was in exon 3 ( GACTCTCGTTCTAGCCTTGGTGG , score 78 ) . Note that guides against exons 1 , 2 , and 3 within Mtgr1 were designed as well as nickase version of Cas9 was used , but only aforementioned guide within exon3 produced mutations that resulted in the loss of protein . Stella:GFP mESC line was transfected with the desired sgRNA in pX330 plasmid together with piggyBac mCherry ( transient transfection ) using Lipofectamine 2000 ( Life Technologies ) according to the manufacturer’s instruction manual . Forty-eight hours post-transfection , we did single-cell sorting into 96-well plates on mCherry-positive cells , assuming that these cells got transfected with both mCherry and pX330 plasmids . The colonies that arose from single cells were screened for the presence of the deletion . The target sequence was amplified by PCR with specific primers from genomic DNA . We then picked a restriction enzyme close to the PAM sequence that upon mutation of the sequence would not be able to cut . For Prdm14 , we used PflMI restriction enzyme , and for Mtgr1 , we used StyI restriction enzyme . Clones that could not be digested were further analyzed by doing PCR with specific primers from cDNA and subsequently Sanger sequencing . The results were analyzed with Sequencher 5 . 1 software and TIDE ( Brinkman et al . , 2014 ) . Clones that were confirmed to have a mutation in cDNA were further validated for the presence of protein using Western blotting . Cells carrying Stella:GFP reporter were used to monitor the efficiency of mPGC-LCs formation after 6 days of differentiation . Cells carrying dual reporter constructs ( mCherry-S14 Mb and Stella:GFP ) were PGC-LCs induced for 6 days with doxycycline after which the cells were analyzed . Differentiation was carried out in hanging drops as described . The cells were trypsinized , strained through a 30μm cell strainer and analyzed on an LSR Fortessa Analyzer ( BD ) , data were analyzed further using FlowJo . For statistical analysis , Student’s t test was used to compare two normally distributed data sets . The analysis was done in R using unpaired t-test and paired t-test for the same cell population before and after doxycycline treatment . p<0 . 05 was considered to be statistically significant . Dignam nuclear extracts from mESCs were prepared as previously described ( Peng et al . , 2009 ) . For immunoprecipitations , monobodies or antibodies that were used are listed in the antibody section below . Typically , 50–100 pmol of monobody and 50 μl of pre-washed M280 dynabeads were used per immunoprecipitation . For immunoprecipitations performed using antibodies , we used 5 μg of antibody and 75 μl of ProteinG-sepharose ( Sigma ) beads per immunoprecipitation . In double-step IP we first used FLAG M2-beads with peptide elution followed by incubation with HA antibody . If the immunoprecipitation was followed by mass spectrometry peptide identification , then the eluant was run on the one-dimensional SDS-PAGE gel , fixed and excised . Cells were lysed with radioimmunoprecipitation assay buffer ( 50 mM Tris HCl pH 8 , 300 mM NaCl , 1% Triton X-100 , 0 . 5% sodium deoxycholate , 0 . 1% SDS , 1mM ethylenediaminetetraacetic acid [EDTA] ) containing protease inhibitors ( Roche tablet ) and 1 mM DTT . The protein concentration was estimated with Bradford reagent ( Bio-Rad ) and equal or indicated amounts of protein were run on 8% SDS-PAGE gels and transferred to nitrocellulose membranes . Antibodies used in this study are listed in the antibody section . In gel digestion was performed as previously reported ( Shevchenko , et al . , 2007 ) with the addition of Protease Max for increased peptide and protein solubility . The extracted peptides were dried using a speed vac and reconstituted in mobile phase A . The ultra performance liquid chromatography ( UPLC ) was a Waters M-class where mobile phase A was 0 . 2% formic acid , 5% dimethyl sulfoxide ( DMSO ) , 94 . 8% water and mobile phase B was 0 . 2% formic acid , 5% DMSO , 94 . 8% acetonitrile . The UPLC was run at 300 nl/min from 4% mobile phase B to 35% mobile phase B followed by a wash and re-equilibration step . The mass spectrometer was an Orbitrap Fusion mass spectrometer set to acquire in a data dependent fashion to optimize cycle time and fragment ion acquisition . The RAW data was searched using Byonic against the Uniprot mouse database downloaded on 09/29/2015 . The fixed modifications were Cys . propionamide , an the variable Met . oxidation , Asp . deamidation and N-terminal modifications . The data was filtered and presented at a 1% false discovery rate . An expression vector for human PRDM14 ( residues 238–487 ) with an N-terminal biotin-acceptor tag and C-terminal His6 tag based on the p28BIOH-LIC vector ( GenBank accession EF442785 ) was kindly provided by Susanne Gräslund and Cheryl Arrowsmith ( Structural Genomics Consortium ) . The genes for Prdm14 ( residues 184–373 ) and Mtgr1 ( residues 98–206 ) were assembled using synthetic oligonucleotides and cloned in the pHBT vector that adds an N-terminal His6 tag followed biotin-acceptor tag and a TEV cleavage site ( Sha et al . , 2013 ) . The Mtgr1 construct in this vector contained a H200K mutation located outside the NHR1/TAFH domain due to a cloning artifact . For binding assays , all proteins were expressed in BL21 ( DE3 ) cells containing pBirACm plasmid ( Avidity ) in the presence of 50 μM biotin to produce biotinylated proteins . The Prdm14-Mtgr1 fusion protein was designed to have a GSSGSSGS linker separating Prdm14 ( residues 184–373 ) and Mtgr1 ( 98–206 ) . The DNA sequences for these genes have been deposited to the GenBank . All proteins were expressed as His6-tagged proteins as described . Proteins were purified using Ni-Sepharose gravity flow columns ( GE Healthcare ) and the monodispersity of these proteins was assessed by size-exclusion chromatography . For crystallization , the fusion tags were removed using tobacco etch virus ( TEV ) protease cleavage , and the tags were removed using Ni-Sepharose columns . The method of selecting target specific monobodies from phage and yeast display libraries has been previously described ( Koide et al . , 2012a , 2012b ) . Two monobody libraries ( ‘loop’ and ‘side’ ) were used to generate monobodies with diverse binding modes ( Koide et al . , 2012a ) . Each of these libraries contains approximately 10 billion unique monobody clones in which 16–26 residues are diversified using highly tailored amino acid combinations ( Gilbreth and Koide , 2012; Koide et al . , 2012a ) . Four rounds of phage display selection were performed using target concentrations of 100 nM , 100 nM , 75 nM and 50 nM . Streptavidin-coated magnetic beads ( Streptavidin MagneSphere Paramagnetic Particles; Promega , Z5481/2 ) were used for immobilizing the target and captured phages were eluted with 0 . 1M Gly-HCl , pH 2 . 1 . After gene shuffling among the selected clones within the enriched population ( Koide et al . , 2012a ) , the monobody-coding genes were transferred into a yeast display vector . We performed library selection by yeast surface display using magnetic beads in the first round followed by two rounds of FACS-based selection . Binding assay for testing the affinity and specificity of individual monobody clones was performed using yeast surface display as described previously ( Sha et al . , 2013 ) . The general methods for bead-based assays have been described ( Nishikori et al . , 2012 ) . In the assay , streptavidin-coated Dynabeads M280 beads ( Invitrogen ) at 20 μg/ml were incubated with 5 nM biotinylated target protein diluted in BSS/EDTA/DTT buffer ( 50 mM Tris–HCl , 150 mM NaCl , pH 8 , 1 mg/ml bovine serum albumin , 1 mM EDTA , 0 . 1 mM DTT ) for 30 min . The remaining free biotin-binding sites of streptavidin on the M280 beads were blocked with 5 μM free biotin for 30 min . Ten microliters of the target-immobilized beads were transferred to the wells of a 96-well filter plate ( MultiScreen HTS HV , 0 . 45 μm , Millipore ) , drained using a vacuum manifold ( MultiScreen HTS Vacuum Manifold , Millipore ) , and washed with 100 μl of BSS/EDTA/DTT buffer . Next , a biotinylated protein ( biotinylation of the proteins was checked by their ability to bind to streptavidin beads ) to be tested at various concentrations was added to individual wells and incubated for 30 min with gentle shaking . Then the wells of the filter plate were washed twice with 150 μl of the buffer , 20 μl of 10 μg/ml SAV-Dylight650 ( ThermoFisher ) in the buffer was added to the wells , and the plate incubated with shaking for 30 min . The wells were washed again and the beads resuspended in 140 μl buffer and analyzed using a Guava EasyCyte 6/l flow cytometer ( Millipore ) . Purified Prdm14-linker-Mtgr1 and Mb ( S4 ) were mixed in the molar ratio of 1 . 0:1 . 3 and the complex was purified using a Superdex 75 16/600 size exclusion chromatography column ( GE Healthcare ) in 25 mM Tris pH 8 . 0 , 100 mM NaCl , 0 . 2 mM TCEP . The protein complex was then concentrated to a final concentration of 15 mg/ml . Initial crystallization screening of ~ 500 conditions was carried out in 96-well plates using the hanging-drop vapor-diffusion method with a crystallization robot ( Mosquito , TTP Labtech ) . Crystals used for data collection were obtained in 17% PEG3350 and 8% Tascimate pH 5 . 5 , and were cryoprotected in 1:1 mix of Paratone and Paraffin oil and flash-cooled in liquid nitrogen prior to data collection . The Prdm14-linker–Mtgr1 protein was labeled with SeMet as described ( Doublié , 1997 ) , purified as a complex with the monobody and crystallized in a similar manner as the native proteins . X-ray diffraction data were collected at Beamline 19ID of the Advanced Photon Source ( Argonne National Laboratory , Chicago , IL , USA ) ( Table 1 ) . The data were indexed and integrated using HKL3000 ( Minor et al . , 2006 ) . Molecular replacement using PHASER ( McCoy et al . , 2007 ) and MOLREP ( Vagin and Teplyakov , 1997 ) with an hPrdm12 structure ( PDB ID 3EP0 ) and the monobody backbone ( PDB 3UYO ) did not have sufficient scattering power to generate a solution with a signal-to-noise ratio that is high enough to be identified . Thus , we determined the structure through single-wavelength Se anomalous dispersion experiment . A total of eight Se sites were identified and refined using Autosol ( Terwilliger et al . , 2009 ) , resulting in an overall figure of merit of 0 . 45 and Z-score of 43 . 1 . These phases were then used against the SAD data for model building in phenix . autobuild ( Adams et al . , 2010 ) . Iterative model building and refinement were done using the programs COOT ( Emsley and Cowtan , 2004 ) and PHENIX ( Adams et al . , 2010 ) . The structure refined from SAD data was later refined against the higher resolution native data at a 3 . 05 Å resolution . The final structures were analyzed using Procheck and Molprobity ( Davis et al . , 2004 ) . Figures were made using Pymol ( DeLano , 2002 ) . The structure has 100% residues in the allowed regions of the Ramachandran plot with no outliers . The Molprobity score ( 2 . 16 ) is above average for structures refined at comparable resolutions . Uniformly 15N-labeled Prdm14-linker-Mtgr1 and 15N-labeled Prdm14 were prepared by growing bacterial cells in M9 minimal media supplemented with 15N-labeled ammonium sulphate ( 0 . 8 g l-1 , Cambridge Isotope Laboratories ) . The labeled proteins were purified in the same manner as the unlabeled proteins described above . 15N-labeled Prdm14 in complex with unlabeled 14N-Mtgr1 was purified by gel filtration chromatography . NMR data was collected at 30°C on a 600 MHz Bruker AVANCE III Spectrometer . The samples used for data collection contained 50–200 μM protein in 50 mM Tris-Cl buffer pH 8 . 0 containing 150 mM NaCl and 0 . 2 mM TCEP supplemented with 10% D2O . All spectra were processed by the NMRPipe software ( Delaglio , et al . , 1995 ) and analyzed using SPARKY ( Goddard and Kneller ) . Antibodies for Mtgr1 ( Western and IP , ab53363 , lot GR56108-2 , 4; ChIP , ab96161 ) , V5tag ( Western and IP , ab27671 , lot GR186433-4 ) , HA ( ChIP , ab9110 , lot GR146572-8 ) were from Abcam , and Suz12 ( IP , 04–046 ) from Millipore . HA antibody ( Western and IP , H3663 ) , anti-Flag M2 agarose beads ( A2220 ) , M2 Flag antibody ( Western , F1804 ) were from Sigma and dynabeads M280 streptavidin ( 11205D ) were from Life Technologies .	In animals , there are many different types of cells that perform different roles . For example , stem cells divide to produce new cells that may then become other types of cells such as muscle or skin cells . Most stem cells can only produce a limited range of other cell types , except for a subset known as ‘pluripotent’ stem cells that can give rise to cells of any type in the body . A protein called Prdm14 helps to keep stem cells in a pluripotent state . In mouse embryos , Prdm14 binds to and represses particular genes that promote a process by which the stem cells can change into other cell types . If Prdm14 is missing from pluripotent stem cells , these cells become more sensitive to signals that encourage them to become other types of cells , which can lead to the loss of pluripotency . Prdm14 contains a region called the SET domain . In other proteins , this domain can alter how DNA is packaged to help switch particular genes on or off . However , such activity has not been found for the SET domain of Prdm14 , raising questions about how it actually works . Here , Nady , Gupta et al . show that Prdm14 tightly interacts with a protein called Mtgr1 , which belongs to a family of proteins known to be involved in leukemia . The loss of Mtgr1 also leads to the loss of pluripotency in mouse stem cells and disrupts the formation of reproductive stem cells . Furthermore , Mtgr1 binds to the same genes as Prdm14 . Next , Nady , Gupta et al . made synthetic proteins , termed monobodies , that bind to the Prdm14 SET domain . One such monobody enabled the authors to determine the three-dimensional structure of Prdm1 and Mtgr1 , which revealed that the SET domain of Prdm14 has many points of contact with Mtgr1 . Importantly , interaction between the two partners is crucial for these proteins to maintain pluripotency and promote the production of reproductive stem cells . Altogether , these findings identify Mtgr1 as a key binding partner of Prdm14 in pluripotent stem cells and uncover a role for the SET domain in this interaction . A future challenge will be to understand the roles of these proteins in leukemia and other diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "structural", "biology", "and", "molecular", "biophysics" ]	2015	ETO family protein Mtgr1 mediates Prdm14 functions in stem cell maintenance and primordial germ cell formation
Plants react to seasonal change in day length through altering physiology and development . Factors that function to harmonize growth with photoperiod are poorly understood . Here we characterize a new protein that associates with both circadian clock and photoreceptor components , named PHOTOPERIODIC CONTROL OF HYPOCOTYL1 ( PCH1 ) . pch1 seedlings have overly elongated hypocotyls specifically under short days while constitutive expression of PCH1 shortens hypocotyls independent of day length . PCH1 peaks at dusk , binds phytochrome B ( phyB ) in a red light-dependent manner , and co-localizes with phyB into photobodies . PCH1 is necessary and sufficient to promote the biogenesis of large photobodies to maintain an active phyB pool after light exposure , potentiating red-light signaling and prolonging memory of prior illumination . Manipulating PCH1 alters PHYTOCHROME INTERACTING FACTOR 4 levels and regulates light-responsive gene expression . Thus , PCH1 is a new factor that regulates photoperiod-responsive growth by integrating the clock with light perception pathways through modulating daily phyB-signaling . Plants have evolved to coordinate physiology and phenology with seasonal variation in the environment ( Wilczek et al . , 2010 ) . These adaptations to changing day length are called photoperiodic responses , which are regulated by both the circadian clock and specific signaling pathways , including light sensory systems ( Shim and Imaizumi , 2015 ) . In plants , photoperiod regulates myriad processes , including the transition to flowering ( Valverde et al . , 2004 ) , cold acclimation ( Lee and Thomashow , 2012 ) , and growth ( Niwa et al . , 2009; Nomoto et al . , 2012 ) . In Arabidopsis , daily hypocotyl elongation is accelerated in short days compared to long day conditions , and requires both the circadian clock and light signals to properly react to changing photoperiods ( Niwa et al . , 2009; Nozue et al . , 2007 ) . Circadian clocks provide an adaptive advantage by synchronizing internal physiology to the external environment , allowing for an efficient allocation of resources in plants ( Dodd et al . , 2005 ) . More than 20 clock components have been characterized in Arabidopsis , forming a complex network of interlocking transcription-translation feedback loops ( Hsu and Harmer , 2014; Nagel and Kay , 2012; Pokhilko et al . , 2012 ) . Among them , a tripartite protein complex named the Evening Complex ( EC ) regulates circadian rhythms and suppresses hypocotyl growth in the evening ( Nusinow et al . , 2011 ) . Mutations in any of the EC components , EARLY FLOWERING 3 ( ELF3 ) ( Hicks et al . , 2001 ) , EARLY FLOWERING 4 ( ELF4 ) ( Doyle et al . , 2002 ) or LUX ARRHYTHMO ( LUX ) ( Hazen et al . , 2005; Onai and Ishiura , 2005 ) , leads to arrhythmic circadian oscillations , elongated hypocotyls , and early flowering regardless of day length ( Nagel and Kay , 2012 ) . The EC regulates hypocotyl elongation by repressing the expression of two critical bHLH transcription factors PHYTOCHROME INTERACTING FACTOR 4 and 5 ( PIF4 and PIF5 ) ( Nusinow et al . , 2011 ) , which are two key regulators in phytochrome-mediated light signaling pathways ( Huq and Quail , 2002; Khanna et al . , 2004 ) . Furthermore , ELF3 directly binds to the red light photoreceptor phytochrome B ( phyB ) ( Liu et al . , 2001 ) and the E3-ligase CONSTITUTIVE PHOTOMORPHOGENIC 1 ( COP1 ) ( Yu et al . , 2008 ) , connecting the clock to light signaling . Arabidopsis possesses five red/far-red light absorbing phytochromes ( phyA to E ) ( Clack et al . , 1994; Sharrock and Quail , 1989 ) . Phytochromes are converted to the Pfr ( active ) form upon red ( 660 nm ) light treatment , and reverted to the Pr ( inactive ) form either upon far-red ( 730 nm ) light exposure or by incubation in the dark in a process termed dark reversion ( Rockwell et al . , 2006 ) . Signaling through phytochromes regulates germination , shade avoidance , circadian rhythms , photosynthesis , hypocotyl growth and flowering time ( Kami et al . , 2010 ) . During the day , phytochromes play a prominent role sensing environmental light signals to suppress growth: phyB in the Pfr state binds to PIFs ( such as PIF3 , 4 and 5 ) to regulate their post-translational turnover ( Bauer et al . , 2004; Lorrain et al . , 2008; Nozue et al . , 2007 ) . Taken together , daily growth rhythms in seedlings are the result of both post-translational degradation of PIF3 , 4 , and 5 by phytochromes ( Lorrain et al . , 2008; Soy et al . , 2012 ) and transcriptional regulation of PIF4 and PIF5 by the EC ( Nozue et al . , 2007; Nusinow et al . , 2011 ) . Photoconversion of phyB by red light induces its localization to discrete subnuclear domains named photobodies ( Chen and Chory , 2011; Chen et al . , 2003 ) . Light conditions that drive the Pr/Pfr equilibrium towards Pfr will promote formation of large photobodies in vivo ( Chen et al . , 2003 ) , which correlates with the photoinhibition of hypocotyl elongation and the degradation of PIF3 ( Chen et al . , 2003; Van Buskirk et al . , 2014 ) . Since proper degradation of PIFs is critical to regulate growth ( Al-Sady et al . , 2006; Lorrain et al . , 2008 ) , one proposed function of photobodies is to stabilize the phyB Pfr form , which allows active phyB to continue controlling the level of PIFs and suppressing hypocotyl growth in prolonged darkness or in short-days ( Rausenberger et al . , 2010; Van Buskirk et al . , 2014 ) . Current mathematical models of red-light signaling dynamics predict a yet undiscovered factor that directly modulates photobody formation in vivo in response to light ( Klose et al . , 2015 ) . Here we present the characterization of an EC-associated protein called PCH1 ( for PHOTOPERIODIC CONTROL OF HYPOCOTYL 1 ) . Our results define PCH1 as a new clock-regulated phytochrome-binding factor that regulates photoperiodic growth by stabilizing phyB-containing photobodies in the evening , thereby providing a molecular mechanism for prolonging red-light signaling after prior light exposure . A protein encoded by At2g16365 , a gene that was described as required for transcriptional responses to lincomycin-induced chloroplast damage ( Ruckle et al . , 2012 ) , was repeatedly co-purified with the EC by tandem affinity-purification coupled with mass spectrometry ( AP-MS ) analyses ( Huang et al . , 2015 ) . According to the TAIR10 database ( Lamesch et al . , 2011 ) , At2g16365 has four splice variants encoding different protein products , three of which contain an F-box domain ( At2g16365 . 1 , 3 and 4 ) ( Figure 1—figure supplement 1A ) . All peptides from AT2G16365 that co-purified with the EC were mapped to At2g16365 . 2 ( Figure 1—supplement 2 and Huang et al . , 2015 ) , which contains the first two exons and lacks the sequence encoding the F-box domain . Semi-quantitative RT-PCR analysis and RNA-seq reads from a publically available RNA-seq dataset ( Gulledge et al . , 2014 ) indicated that only At2g16365 . 2 is transcribed ( Figure 1—figure supplement 1B and Figure 1—figure supplement 3 ) . Therefore , all presented constructs are based on the dominant isoform At2g16365 . 2 . Furthermore , a T-DNA insertion loss-of-function line in At2g16365 ( SALK_024229 , Ruckle et al . , 2012 ) resulted in a short-day-specific hypocotyl phenotype ( described below ) . Thus , the At2g16365 gene was renamed PHOTOPERIODIC CONTROL OF HYPOCOTYL 1 ( PCH1 ) . 10 . 7554/eLife . 13292 . 003Figure 1 . PCH1 ( At2g16365 . 2 ) encodes a conserved evening-phased protein . ( A ) Time-course gcRMA ( GeneChip Robust Multiarray Averaging ) values of PCH1 expression ( from Diurnal database , http://diurnal . mocklerlab . org/ , Mockler et al . , 2007 ) under short day , 12L:12D and long day conditions ( Light: Dark hours = 8:16 , 12:12 and 16:8 , respectively ) . Grey shading indicates dark period . ( B ) Time-course qPCR analysis of PCH1 expression using cDNA samples ( from ZT 0 to 24 , with 3 hr intervals ) of 4-day-old seedlings grown under short day conditions , normalized to IPP2 and APA1 . Mean ± SD ( n=3 biological reps ) . ( C ) Anti-FLAG immunoblotting detecting PCH1-His6-FLAG3 levels using protein extracts from time-course samples ( from ZT 0 to 24 , with 3 hr intervals ) of 4-day-old , short-day-grown PCH1p::PCH1 and PCH1ox3 plants , which express the tagged PCH1 protein driven by the PCH1 native promoter or the 35S CaMV promoter , respectively . Actin was used for normalization . Rectangles above blots represent light/dark conditions under which samples were flash frozen in liquid N2 , white = light and black = dark . Wild type ( WT ) , pch1 and PCH1ox3 in phyB-9 were controls for immunoblots . ( D ) Normalized gcRMA values of PCH1 orthologs from Arabidopsis thaliana ( At2g16365 ) , Brachypodium distachyon ( Bradi2g46850 ) , Oryza sativa ( Rice , LOC_Os01g49310 ) , and Populus trichocarpa ( Poplar , POPTR_0004s16430 . 1 ) under 12L:12D conditions from Diurnal database , http://diurnal . mocklerlab . org/ , Mockler et al . , 2007 ) . Expression is normalized to min and max value . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 00310 . 7554/eLife . 13292 . 004Figure 1—figure supplement 1 . At2g16365 . 2 is the predominant transcript of PCH1 . ( A ) Schematic structures of four splice variants of At2g16365 . Solid boxes represent exons ( red ) and UTR ( blue ) , while introns are shown as lines . The T-DNA insertion site of SALK_024229 ( pch1 ) , two sets of primers A and B for qPCR , and the F-box domain ( green ) are denoted . ( B ) Semi-quantitative qPCR indicates At2g16365 . 2 is the predominant transcript . Primer set A and B from ( A ) were used to distinguish transcripts of At2g16365 . 2 from those of At2g16365 . 1 , 3 and 4 ( for set A , cDNA amplicon = 109 bp , genomic DNA amplicon = 191 bp; for set B , cDNA amplicon = 163 bp , genomic DNA amplicon = 554 bp ) . Expression of IPP2 was used for normalization . Both cDNA and genomic DNA of WT and pch1 were used for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 00410 . 7554/eLife . 13292 . 005Figure 1—figure supplement 2 . Peptides identified by ELF3/4 AP-MS only mapped to the protein encoded by At2g16365 . 2 . Comparison of the amino acid sequences of protein encoded by At2g16365 . 2 and At2g16365 . 1 ( amino acids encoded by exon 3 and 4 were in gray ) , with peptides that were identified by ELF3 or ELF4 AP-MS ( Huang et al . , 2015 ) highlighted by red color and the F-box motif underscored with green lines . The star symbol means the stop codon . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 00510 . 7554/eLife . 13292 . 006Figure 1—figure supplement 3 . Available RNAseq data suggest only At2g16365 . 2 is expressed . A screen capture from IGB ( Integrated Genome Browser ) analyzing a publically available RNAseq dataset ( Gulledge et al . , 2014 ) . RNAseq reads are represented by solid green blocks and only mapped to the At2g16365 . 2 locus . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 00610 . 7554/eLife . 13292 . 007Figure 1—figure supplement 4 . PCH1 levels peak at dusk under 12L:12D or long day conditions . Time-course protein extracts ( from ZT 0 to 24 , with 3 hr intervals ) from 4-day-old PCH1p::PCH1 and PCH1ox3 lines grown under either 12L:12D ( A ) or long day ( B ) conditions were immunoblotted with anti-FLAG antibodies to detect the PCH1-His6-FLAG3 fusion protein . RPT5 was used as a loading control . The rectangles above represent the light conditions during harvesting: black= lights off , white= lights on . Extracts of WT , pch1 , and PCH1ox3 in phyB-9 were loaded as western controls . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 00710 . 7554/eLife . 13292 . 008Figure 1—figure supplement 5 . Multiple sequence alignments of PCH1 orthologs at the C-terminus . The last 201 amino acids of PCH1 orthologs from Arabidopsis thaliana ( AtPCH1 , At2g16365 . 2 ) , Brachypodium distachyon ( BdPCH1 , Bradi2g46850 ) ( International Brachypodium Initiative , 2010 ) , Oryza sativa ( OsPCH1 , LOC_Os01g49310 ) ( Ouyang et al . , 2007 ) , and Populus trichocarpa ( PtPCH1 , POPTR_0004s16430 . 1 ) ( Tuskan et al . , 2006 ) were compared . Multiple sequence alignments were done using Clustal Omega with default settings ( http://www . ebi . ac . uk/Tools/msa/clustalo/ ) and formatted by BoxShade server ( http://www . ch . embnet . org/software/BOX_form . html ) . Conserved residues were highlighted and labeled with symbols . A conserved region C-terminus ( the last 43 amino acids ) was underscored with blue . A bipartite nuclear localization sequence predicted by cNLS Mapper ( Kosugi et al . , 2009 ) was underscored in red . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 00810 . 7554/eLife . 13292 . 009Figure 2 . PCH1 regulates the photoperiodic response of hypocotyl elongation in the evening . ( A ) Hypocotyl lengths of 4-day-old WT , pch1 , PCH1ox3 , elf3-2 , elf4-2 and phyB-9 seedlings grown under long day , 12L:12D and short day conditions . Mean ± 95% confidence interval ( CI ) ( n=20 ) . ( B ) pch1 grows faster than WT during night . Time-lapse images were taken every hour for each seedling of WT , pch1 , PCH1ox3 and PCH1p::PCH1 grown under short day conditions . Growth rate was calculated as the hypocotyl increase per hour and plotted against time . Solid lines are the regression analyses of data . Mean ± SEM ( n ≥ 14 ) . Grey shading indicates dark period . Also see Figure 2—source data 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 00910 . 7554/eLife . 13292 . 010Figure 2—source data 1 . Raw measurements of hypocotyl lengths for Figure 2A . Hypocotyl lengths ( mm ) of 4-day-old WT , pch1 , PCH1ox3 , elf3-2 , elf4-2 and phyB-9 seedlings grown under long day , 12L:12D and short day conditions , with n = 20 . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 01010 . 7554/eLife . 13292 . 011Figure 2—source data 2 . ANOVA analyses and Bonferroni's multiple comparison tests for Figure 2A . 2-way ANOVA analyses were carried out using GraphPad Prism ( version 6 . 0 , Graphpad . com , La Jolla , California ) , with Bonferroni's multiple comparisons test results . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 01110 . 7554/eLife . 13292 . 012Figure 2—figure supplement 1 . PCH1 levels regulate hypocotyl length under short day conditions . Hypocotyl lengths of short-day-grown , 4-day-old plants of WT , pch1 , two independent overexpression lines ( PCH1ox3 and 4 , in WT ) and two complementation lines ( PCH1p::PCH1-7 and -8 , in pch1 ) were measured . Inset photo of each genotype with scale bar = 5 mm . Mean ± 95% CI ( n=20 ) . ( ***p<0 . 0001 , ns = not significant , each compared to WT ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 012 Examination of microarray data from long day , 12L:12D and short day time courses ( Light: Dark hours = 16:8 , 12:12 and 8:16 , respectively ) from the DIURNAL database ( Michael et al . , 2008b; Mockler et al . , 2007 ) found that expression of PCH1 cycled with a peak of expression occurring in the evening . PCH1 mRNA accumulates after dawn , reaches a maximum at Zeitgeber time 8 ( ZT8 ) and decreases towards the end of night ( Figure 1A ) . This expression pattern was validated by quantitative PCR ( qPCR ) analyses using cDNA samples of short-day grown seedlings ( Figure 1B ) . To test if PCH1 protein levels oscillate , a transgene expressing PCH1-His6-FLAG3 under the control of the PCH1 promoter in pch1 ( SALK_024229 ) ( PCH1p::PCH1 ) was generated and protein abundance was monitored during a short day time course . The PCH1-His6-FLAG3 fusion protein accumulates after dawn , reaches a peak level in the early evening ( ZT9 ) and a trough at subjective dawn ( Figure 1C ) , similar to PCH1 expression ( Figure 1B ) . Under 12L:12D and long day conditions , PCH1-His6-FLAG3 in PCH1p::PCH1 plants continues to peak near the dusk transition ( Figure 1—figure supplement 4 ) . In a PCH1 constitutive expression line ( PCH1ox3 ) , PCH1 protein levels were constant under all photoperiods , unlike PCH1p::PCH1 ( Figure 1C and Figure 1—figure supplement 4 ) . To determine if PCH1 is present in other plant species , PCH1 orthologs were identified . Pairwise alignments of Arabidopsis PCH1 with orthologs from Oryza sativa , Brachypodium distachyon , and Populus trichocarpa indicate percent identity is 19 . 6% , 20 . 81% and 32 . 03% , respectively ( Clustal Omega , http://www . ebi . ac . uk/Tools/msa/clustalo/ ) . The last 43 amino acids of the C-terminus are highly conserved among PCH1 orthologs ( Figure 1—figure supplement 5 ) . In addition , PCH1 orthologs share the evening-phased expression pattern under 12L:12D ( Figure 1D ) , suggesting that PCH1 may have conserved time-of-day-specific functions ( Michael et al . , 2008a ) . The association of PCH1 with the EC and light signaling components suggested that PCH1 may regulate hypocotyl elongation ( Huang et al . , 2015 ) . Therefore , hypocotyl lengths of 4-day-old wild type , phyB-9 ( Reed et al . , 1993 ) , elf4-2 , elf3-2 ( Nusinow et al . , 2011 ) and pch1 loss-of-function mutant ( Ruckle et al . , 2012 ) seedlings were compared under long day , 12L:12D and short day conditions ( Figure 2A ) . Unlike phyB-9 , elf3-2 and elf4-2 , which exhibit longer hypocotyls than wild type under all photoperiods tested , pch1 shows a day-length-specific defect ( Figure 2A ) . Under long days , hypocotyls of pch1 are not longer than wild type ( Figure 2A ) . As the dark period extended to 12 hr , pch1 exhibited a slightly but significantly longer hypocotyl than wild type ( p<0 . 01 ) ( Figure 2A ) . Under short days , pch1 mutants elongated hypocotyls even further ( p<0 . 0001 ) ( Figure 2A ) . Constitutive expression of PCH1 resulted in the opposite phenotype: PCH1ox3 mutants have shorter hypocotyls than wild type under all photoperiods ( p<0 . 0001 ) ( Figure 2A ) . PCH1p::PCH1 rescued the hypocotyl length phenotype of pch1 mutants in two independent lines ( Figure 2—figure supplement 1 ) , showing that the level and/or timing of PCH1 expression is critical for proper regulation of hypocotyl elongation . The evening-phased expression of PCH1 suggested it may function to regulate growth rhythms at a specific time of day . Therefore , hypocotyl growth rates of wild type , pch1 , PCH1ox3 and PCH1p::PCH1 were measured in short days by time-lapse imaging . Wild type plants showed rhythmic hypocotyl growth under short days , with a maximal growth rate at dawn , as described ( Nozue et al . , 2007 ) ( Figure 2B ) . On the third night post-germination ( from ZT56 to ZT72 ) , pch1 seedlings had higher hypocotyl growth rates than wild type during the night , especially during the 3rd to the 5th night ( Figure 2B ) . Supporting the hypothesis that PCH1 is a suppressor of hypocotyl elongation , constitutive expression of PCH1 in PCH1ox3 inhibited hypocotyl elongation throughout the night , while PCH1p::PCH1 restored the growth rate to wild type levels ( Figure 2B ) . Light signaling and the EC are critical for circadian rhythmicity and flowering pathways ( Nagel and Kay , 2012; Shim and Imaizumi , 2015 ) , therefore the role of PCH1 in circadian rhythms and time to flowering was investigated . To determine if PCH1 regulates the circadian oscillator , a CCA1-promoter driven LUCIFERASE ( CCA1::LUC ) was used to monitor endogenous rhythms ( Pruneda-Paz et al . , 2009 ) . The luciferase activity of CCA1::LUC in wild type , pch1 and PCH1ox3 oscillates with a period of ~24 hr ( 23 . 20 ± 0 . 45 , 23 . 13 ± 0 . 37 , and 22 . 97 ± 0 . 32 hr , respectively , mean ± SD , n = 8 ) ( Figure 3A , B and Figure 3—figure supplement 1 ) , showing that PCH1 levels do not affect the circadian period of the reporter . However , the pch1 mutation results in an early flowering phenotype under long days , while PCH1ox3 flowers later than wild type ( Figure 3C ) . Unlike elf4-2 plants , which flower early under both long days and short days ( Doyle et al . , 2002 ) , the flowering time of pch1 or PCH1ox3 is not different from wild type under short days ( Figure 3D ) . Together , the results show that PCH1 is an output rather than a component of the circadian clock and that the pch1 mutant is still sensitive to photoperiod in respect to flowering control . 10 . 7554/eLife . 13292 . 013Figure 3 . Phenotypic characterization of pch1 and PCH1ox3 in circadian and flowering pathways . ( A ) Seedlings of WT , pch1 , or PCH1ox3 carrying the CCA1:LUC luciferase reporter were grown under 12L:12D conditions for five days before transferring to continuous white light . Bioluminescence were plotted against ZT hours . Mean ± SD ( n = 8 ) . Experiments were repeated at least three times . ( B ) Relative amplitude error ( RAE ) versus period of WT , pch1 , and PCH1ox3 rhythms was plotted . RAE = 0 . 5 was used as a cutoff ( dotted line ) , above which a seedling is not considered rhythmic ( n = 8 ) . Experiments were repeated at least three times . ( C ) and ( D ) Flowering assays under either long day ( C ) or short day ( D ) conditions were conducted . Number of rosette leaves from WT , pch1 , PCH1ox3 and PCH1p::PCH1 plants with 1 cm inflorescence stem was counted . Mean ± 95% CI ( n ≥ 20 ) . One-way ANOVA and multiple comparisons were done , with star symbols indicating if it is significantly different from WT ( p=0 . 012 , ***p<0 . 0001 , ns = not significantly different ) . Experiments were repeated twice . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 01310 . 7554/eLife . 13292 . 014Figure 3—figure supplement 1 . Modulating PCH1 levels does not affect the circadian period . Comparison of WT , pch1 , and PCH1ox3 periods using data from Figure 3A and B . Mean ± 95% CI ( n = 8 ) . ns = not significantly different than WT . Experiments were repeated at least three times . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 014 Previous AP-MS studies with the EC components ELF3 and ELF4 identified PCH1 as a co-precipitating protein ( Huang et al . , 2015 ) . PCH1-associated proteins were identified by AP-MS from PCH1ox3 plants expressing PCH1-His6-FLAG3 , which are harvested at dusk ( ZT12 ) . After excluding non-specific binding proteins from negative control GFP-His6-FLAG3 AP-MS and non-specific structural , metabolic , and photosynthetic proteins ( Huang et al . , 2015; Mellacheruvu et al . , 2013 ) , PCH1 AP-MS identified 17 proteins from three biological replicate purifications ( Table 1 , for all co-purified proteins , see Table 1—source data 1 ) . 15 of the 17 proteins that co-precipitated with PCH1 overlap with the ELF3 AP-MS ( Huang et al . , 2015 ) , including the EC components ELF3 , ELF4 , and LUX , all five phytochromes ( A to E ) , TANDEM ZINC KNUCKLE/PLUS3 ( TZP ) , DAYSLEEPER , MUT9-LIKE KINASE 2 ( MLK2 ) , CHLOROPLAST RNA BINDING ( CRB ) , the protease RD21a , and the COP1-SPA1 complex ( Table 1 ) . COP1-SPA1 is part of a complex that mediates the light-dependent turnover of light signaling components ( Saijo et al . , 2003 ) . TZP positively regulates morning-specific plant growth and flowering responses through associating with phyB ( Kaiserli et al . , 2015; Loudet et al . , 2008 ) . DAYSLEEPER is a hAT transposase that is required for proper embryonic development ( Bundock and Hooykaas , 2005 ) . MLK2 is a nuclear kinase that regulates circadian rhythms and osmotic stress responses ( Huang et al . , 2015; Wang et al . , 2015 ) . CRB is a RNA binding protein that regulates circadian rhythms ( Hassidim et al . , 2007 ) . RD21a is a drought-inducible cysteine protease ( Koizumi et al . , 1993 ) . FAR-RED ELONGATED HYPOCOTYL 1 ( FHY1 ) and TOPLESS ( TPL ) were proteins co-purified with PCH1 that were not identified in ELF3 AP-MS . FHY1 interacts with phyA and is required for phyA nuclear import upon light treatment ( Hiltbrunner et al . , 2005 ) . TPL is a Groucho/Tup1-type transcriptional co-repressor that interacts with proteins from circadian and development pathways ( Liu and Karmarkar , 2008; Wang et al . , 2012 ) . In summary , our PCH1 AP-MS results confirm that PCH1 is a component of a reported protein-protein interaction network consisting of the EC , phytochromes and the COP1-SPA1 complex . 10 . 7554/eLife . 13292 . 015Table 1 . Proteins Co-Purified by PCH1 AP-MS in WT and phyB-9 . Proteins co-purified with PCH1 were identified from affinity purification coupled with mass spectrometry ( AP-MS ) analyses using 12L:12D grown , 10-day-old PCH1ox3 plants ( in either WT or phyB-9 mutant backgrounds ) harvested at ZT12 . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 01510 . 7554/eLife . 13292 . 016Table 1—source data 1 . The full list of proteins identified by AP-MS , listing unique peptides and the percent coverage . The full list is generated and exported by Scaffold ( Proteome Software Inc . , Portland , Oregon; v . 4 . 4 . 3 ) showing all co-purified proteins from all replicates of PCH1ox3 AP-MS and the GFP Control . The file contains reports on exclusive unique peptide counts and percent coverage for each co-purified proteins , with their names , accession numbers and molecular weight . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 016AGI numberProtein nameELF3 AP-MSbExclusive unique peptide count/Percent coverageaPCH1ox3 in WTPCH1ox3 in phyB-9rep1rep2rep3rep1rep2At2g16365PCH1cY30/73%41/85%37/79%40/85%32/77%At2g18790phyBY46/69%47/65%41/60%——At5g35840phyCY31/44%23/28%25/29%——At4g16250phyDY22/47%19/34%20/38%22/38%6/11%At4g18130phyEY41/55%40/52%45/60%49/60%31/41%At1g09570phyAY31/46%36/49%35/45%36/48%29/39%At2g37678FHY1N2/22%2/21%4/28%4/28%2/21%At3g42170DAYSLEEPERY5/13%—4/11%3/8%2/6%At1g09340CRBY—d—d4/17%6/24%3/13%At5g43630TZPY9/15%6/12%12/23%——At2g32950COP1Y7/15%8/16%8/18%——At2g46340SPA1Y8/14%5/7%8/12%——At2g25930ELF3Y6/12%11/25%12/26%——At2g40080ELF4Y—d4/60%3/42%——At3g46640LUXY2/6%—d4/15%——At3g03940MLK2Y—d2/6%2/6%——At1g15750TPLN—d3/3%4/5%2/2%—At1g47128RD21aY—d3/8%2/4%—d—dAlso see Table 1—source data 1a All listed proteins match 99% protein threshold , minimum number peptides of 2 and peptide threshold as 95% . Proteins not matching the criteria were marked with "—" . b ELF3 AP-MS ( Huang et al . , 2015 ) was used for comparison . c Percent coverage for PCH1 is calculated using protein encoded by At2g16365 . 2 . d Only one exclusive unique peptide was detected . ELF3 , phyB , and COP1 interact with each other to form a 'triangle core' of the EC-phytochrome-COP1 interactome ( Jang et al . , 2010; Liu et al . , 2001; Yu et al . , 2008 ) , recruiting other proteins into the interaction network ( Huang et al . , 2015 ) . To determine if the association between PCH1 and other co-purified proteins depended on the EC or phyB , PCH1 AP-MS analysis in wild type ( PCH1ox3 ) was compared to those in elf4-2 , elf3-2 or phyB-9 backgrounds . Although PCH1 was originally found co-precipitating with ELF4 and ELF3 , both are dispensable for PCH1 to associate with the light signaling components in the network ( Table 2 ) . In comparison , phyB is critical for recruiting PCH1 to the EC-phytochrome-COP1 interactome . In phyB-9 , PCH1AP-MS did not co-precipitate the EC , the COP1-SPA1 complex , TZP , MLK2 , RD21a , TPL , or phyC ( Table 1 ) . However , the association with DAYSLEEPER , CRB , phyD , phyE , phyA , and FHY1 was retained in phyB-9 . Therefore , our PCH1 AP-MS analysis in phyB-9 suggests that the association of PCH1 with the EC , the COP1-SPA1 complex , MLK2 , and TZP is bridged by phyB , while loss of phyC could be due to a reduction in phyC caused by the phyB mutation ( Clack et al . , 2009 ) . Together , our PCH1 AP-MS analyses in different genetic backgrounds demonstrate that PCH1 is integrated into the EC-phytochrome-COP1 interactome in vivo through the association with phyB . 10 . 7554/eLife . 13292 . 017Table 2 . Proteins Co-Purified by PCH1 AP-MS in elf4-2 and elf3-2 , compared to WT . Proteins co-purified with PCH1 were identified from affinity purification coupled with mass spectrometry ( AP-MS ) analyses using 12L:12D grown , 10-day-old PCH1ox3 plants in either elf4-2 or elf3-2 mutant backgrounds harvested at ZT12 . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 017AGI numberProtein nameELF3 AP-MSbExclusive unique peptide count/Percent coverageaPCH1ox3 in WTcPCH1ox3 in elf4-2PCH1ox3 in elf3-2rep1rep2rep3rep1rep2rep1rep2At2g16365PCH1dY30/73%41/85%37/79%29/77%34/78%42/82%36/82%At2g18790phyBY46/69%47/65%41/60%47/70%31/46%42/63%40/56%At5g35840phyCY31/44%23/28%25/29%30/43%13/16%20/28%15/18%At4g16250phyDY22/47%19/34%20/38%20/42%12/25%20/37%16/30%At4g18130phyEY41/55%40/52%45/60%42/57%37/50%43/55%40/53%At1g09570phyAY31/46%36/49%35/45%32/47%24/34%34/49%27/38%At2g37678FHY1N2/22%2/21%4/28%—e3/21%—e3/21%At3g42170DAYSLEEPERY5/13%—4/11%3/7%5/11%2/5%—eAt1g09340CRBY—e—e4/17%2/9%5/22%3/13%3/13%At5g43630TZPY9/15%6/12%12/23%4/7%—e12/21%2/3%At2g32950COP1Y7/15%8/16%8/18%3/7%—e12/25%7/11%At2g46340SPA1Y8/14%5/7%8/12%5/10%2/4%17/26%7/11%At2g25930ELF3Y6/12%11/25%12/26%4/9%3/6%——At2g40080ELF4Y—e4/60%3/42%————eAt3g46640LUXY2/6%—e4/15%—e———At3g03940MLK2Y—e2/6%2/6%——2/4%—At1g15750TPLN—e3/3%4/5%2/3%3/3%4/5%4/5%At1g47128RD21aY—e3/8%2/4%—e2/8%3/8%—eAlso see Table 1—source data 1a All listed proteins match 99% protein threshold , minimum number peptides of 2 and peptide threshold as 95% . Proteins not matching the criteria were marked with "—" . b ELF3 AP-MS ( Huang et al . , 2015 ) was used for comparison . c PCH1ox3 in WT is as shown in Table 1 , for comparison with PCH1ox3 in elf4-2 and elf3-2 . d percent coverage for PCH1 is calculated using protein encoded by At2g16365 . 2e only one exclusive unique peptide was detected . Next , yeast two-hybrid assays were used to determine if interactions between PCH1 and selected PCH1-associated proteins were direct . Consistent with the AP-MS data , direct interactions between PCH1 and ELF3 , ELF4 , LUX , COP1 or TZP were not observed ( Figure 4A ) . However , PCH1 interacted with the C-terminus of phyB ( Figure 4A ) . PCH1 also interacted with the C-terminal tail of phyD and phyE , but not with either phyA or phyC in yeast ( Figure 4B ) . To validate the PCH1-phyB interaction in planta , PCH1-His6-FLAG3 was transiently co-expressed with a phyB-GFP fusion protein in tobacco ( Nicotiana benthamiana ) leaves . phyB-GFP specifically co-precipitated with PCH1-His6-FLAG3 in an anti-FLAG immunoprecipitation , while PCH1 and GFP alone did not interact ( Figure 4C ) . 10 . 7554/eLife . 13292 . 018Figure 4 . PCH1 directly interacts with phyB in a light-dependent manner . ( A ) and ( B ) yeast two-hybrid between PCH1 ( fused to GAL4 DNA binding domain , DBD ) and preys ( ELF3 , ELF4 , N-/C- termini ( Nt or Ct ) and full length ( FL ) LUX , COP1 , TZP and the Ct of phyA , B , C , D , and E fused to GAL4 activating domain , AD ) . –LW select ( minus Leu and Trp ) for presence of both DBD and AD constructs and–LWH+3AT plates ( minus Leu , Trp and His , with 2 mM 3AT added ) tested interactions . ( C ) Transient tobacco co-immunoprecipitation ( IP ) assay with PCH1-His6-FLAG3 and phyB-GFP or GFP . IPs were done against FLAG followed by westerns using either anti-FLAG , phyB or GFP antibodies . ( D ) The in-vivo PCH1-phyB interaction is light-sensitive . A schematic of the light treatment is above western . PCH1ox3 seedlings entrained in 12L:12D white light ( WL ) were either exposed to WL for 12 hr ( WL ZT12 , lane 1 to 3 ) , subjected to extended dark ( WL to DD ) for 24 or 48 hr ( lane 4 and 5 ) , red light for 12 hr ( WL to Rc ZT12 , lane 6 ) , or an end-of-day far-red pulse for 10 min after 12 hr of WL ( WL EOD-FRp ZT12 , lane 7 ) . WT and PCH1ox3 in phyB-9 plants are western controls . IPs were done against FLAG followed by westerns using either anti-FLAG or phyB antibodies . Anti-RPT5 was used as a loading control . The asterisk at the FLAG-IP / anti-phyB notes an unspecific band that migrates faster than phyB that is present in every lane . ( E ) PCH1 preferentially binds the Pfr form of phyB in in vitro . Recombinant His6-PCH1-His6-FLAG3 or His6-YPet-His6-FLAG3 was incubated with phyB-HA transcribed and translated by rabbit-reticulate lysate . PΦB absent ( apo ) phyB precipitations were incubated in the dark , while red ( Pfr ) or far red light ( Pr ) were incubated with 20 μM PΦB . His-affinity capture was followed by immunoblotting for anti-HA or anti-FLAG . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 01810 . 7554/eLife . 13292 . 019Figure 4—figure supplement 1 . Interaction map of PCH1-associated proteins . An updated interaction map from previous described ( Huang et al . , 2015 ) , which integrate our AP-MS and protein-protein interaction data to illustrate both directly and indirectly interacting proteins that co-precipitate with PCH1 . Solid lines indicate direct interactions determined in this study or previously defined , while dotted lines indicate association . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 019 Light absorption by phytochromes alters their confirmation , subcellular localization , and binding to signaling partners ( Burgie et al . , 2014; Kikis et al . , 2009; Kircher et al . , 2002 ) . We therefore examined the light sensitivity of the PCH1-phyB interaction . PCH1ox3 seedlings were used for phyB co-precipitation reactions after 12L:12D entrainment under white light conditions ( WL ) . Endogenous phyB co-precipitated with PCH1 at the end of the light treatment ( ZT12 ) , confirming the AP-MS results ( Figure 4D , lane 3 ) . Less phyB was co-precipitated with PCH1-His6-FLAG3 under extended dark periods ( 24 or 48 hr in dark , lane 4 and 5 ) , although the levels of phyB in the input were increased in these conditions , indicating that light promotes the PCH1-phyB interaction . The red-light sensitivity of the PCH1-phyB interaction was tested and found that 12 hr of red light treatment on the last day was sufficient to maintain the PCH1-phyB interaction ( Figure 4D , lane 6 ) , suggesting that PCH1 bound to the active Pfr form of phyB . Conversely , a ten-minute pulse of far-red light at the end of day ( EOD-FRp ) that converted phyB to the inactive Pr form reduced the PCH1-phyB interaction ( Figure 4D , lane 7 ) . To test directly if PCH1 preferentially binds the active Pfr form of phyB , a reconstituted light-induced in vitro binding assay was assembled with recombinant PCH1-His6-FLAG3 purified from E . coli as bait . phyB-HA was expressed and translated in rabbit reticulocyte lysate and either the apoprotein , or the phyB holoprotein ( mixed with the chromophore phytochromobilin , PΦB ) were then mixed with PCH1-His6-FLAG3 under dark or red/far-red light , respectively . PCH1 weakly interacts with either the apoprotein or the Pr form of phyB but preferentially binds the active Pfr form of phyB , compared to the YPet ( a YFP variant ) control ( Figure 4E ) . In summary , PCH1 directly interacts with phyB , and the interaction is light- and wavelength-sensitive in vivo and in vitro . Combined with our PCH1 AP-MS analyses in different genetic backgrounds , our protein-protein interaction/association data demonstrate that PCH1 is a new phyB-interacting protein and is integrated into the EC-phytochrome-COP1 interactome in vivo through the association with phyB ( Figure 4—figure supplement 1 ) . A subcellular localization tool ( Kosugi et al . , 2009 ) identified a bipartite nuclear localization signal in PCH1 ( highlighted in Figure 1—figure supplement 5 ) . Transient expression of a PCH1-YPet fusion in tobacco showed that PCH1 was exclusively localized in the nucleus , while the YPet control was localized to both nucleus and cytoplasm ( Figure 5A ) . Furthermore , PCH1-YPet was localized to subnuclear foci ( Figure 5A ) similar to the photobodies that phytochromes form after light exposure ( Kircher et al . , 2002 ) . Indeed , when PCH1-YPet and phyB-CFP were co-expressed , they co-localized into nuclear photobodies ( Figure 5A ) . 10 . 7554/eLife . 13292 . 020Figure 5 . PCH1 is localized in the nucleus to stabilize phyB-containing photobodies . ( A ) PCH1-YPet is nuclear localized when transiently expressed in tobacco and co-localizes with phyB-CFP to photobodies . YPet alone was used as control . Scale bars = 25 µm . ( B ) Representative confocal images showing phyB-GFP-containing photobodies in phyB-9 , pch1 phyB-9 and PCH1ox3 phyB-9 plants at indicated time points during light-to-dark transition . Plants were entrained in short days with 10 µmol·m-2·s-1 of red light for two days before transferring to extended dark ( ZT 56 to 72 ) . The representative images were picked based on the photobody morphology of the majority of the nuclei ( >50% ) . The percentage of nuclei showing the corresponding photobody patterns ( with or without photobodies ) were calculated based on three independent experiments . N represents the total number of nuclei analyzed for each time point . Scale bars equal to 5 μm . No PB = photobodies not detected . ( C ) and ( D ) compare quantitative measurements of large ( >1 μm3 , C ) or small ( <1 μm3 , D ) phyB photobodies in all backgrounds . Mean ± 95% CI ( n ≥ 29 ) . ND = no PB of according size were detected . symbol indicates significantly different ( p<0 . 05 , see text for more details about each p value ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 02010 . 7554/eLife . 13292 . 021Figure 5—figure supplement 1 . Fewer large photobodies were detected in pch1 with higher intensity of red light treatment . ( A ) Representative confocal images showing phyB-containing photobodies in phyB-9 , pch1 phyB-9 and PCH1ox3 phyB-9 plants expressing a phyB-GFP ( PBG ) fusion protein . Plants were entrained by the short day condition supplemented with 40 µmol·m-2·s-1 of red light for two days before extended dark treatment . Scale bars indicate 5 µm . The percentage value in each representative image indicates the mean percentage of all nuclei with the phenotypes shown in the image ( with or without photobodies ) from three independent experiments . n indicates the total number of nuclei to generate the percentage . ( B ) Quantitative measurements of photobodies larger than 1 . 0 μm3 ( Large PB , upper ) or smaller than 1 . 0 μm3 ( Small PB , lower ) were analyzed in all seedlings and indicated by Mean ± 95% CI ( n ≥ 29 ) . * symbol indicates significantly different ( p<0 . 05 , see text for more details about each p value ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 021 Photobodies containing phyB are necessary to suppress hypocotyl elongation in the early evening , and phyB lacking C-terminal tails can neither form photobodies nor properly regulate growth after transfer to dark conditions ( Van Buskirk et al . , 2014 ) . Since PCH1 accumulates towards the early evening to suppress hypocotyl elongation , interacts with the C-terminus of phyB , and localizes with phyB to photobodies , we hypothesized that PCH1 regulates phyB photobody assembly/disassembly . A phyB-GFP fusion protein ( PBG ) was introduced into phyB-9 and crossed into pch1 and PCH1ox3 plants . Photobody formation was examined in short day-entrained ( under 10 µmol·m-2·s-1 red light ) seedlings at and after the transition to dark ( 0 , 4 , 8 and 16 hr in dark or ZT56 , 60 , 64 , and 72 , respectively ) . In wild type ( phyB-9/PBG ) , phyB-GFP formed large ( >1 µm3 ) photobodies ( PB ) after 8 hr of light treatment that gradually dissembled into smaller photobodies ( <1 µm3 ) and a diffuse nuclear GFP signal after 8 hr in dark ( Figure 5B to D ) . PCH1ox3 lines showed more large phyB photobodies at the end of the day and throughout the night compared to wild type ( Figure 5B and C ) . In contrast , pch1 mutants exhibited a significant decrease in large photobodies ( p<0 . 0001 ) and a significant increase in small photobodies ( p<0 . 0001 ) at the dusk transition and during the first four hours of night compared to the wild type ( Figure 5C and D ) . In higher red light ( 40 µmol·m-2·s-1 ) , formation of large phyB photobodies in pch1 was significantly less ( p<0 . 0001 ) at the dusk transition compared to wild type , and significantly more small photobodies was observed throughout the night ( ZT56 , p=0 . 024 and 72 , p=0 . 0125 , ZT60 and 64 , p<0 . 0001 ) ( Figure 5—figure supplement 1 ) . However , PCH1ox3 lines showed significantly more large phyB photobodies ( p≤0 . 0002 ) for all time points ( Figure 5—figure supplement 1 ) . These observations demonstrate that PCH1 levels regulate the fluence-dependent formation and maintenance of large phyB photobodies after illumination . The PCH1-phyB interaction prompted us to test if pch1 results in red-light specific growth defects . The hypocotyls of wild type , pch1 , PCH1ox3 , PCH1p::PCH1 , and phyB-9 seedlings were measured under constant red light of various intensities . Compared with wild type , pch1 seedlings have longer hypocotyls and are hyposensitive to red-light-mediated suppression of hypocotyl elongation . This phenotype was rescued in PCH1p::PCH1 transgenic plants ( Figure 6A ) . Conversely , PCH1ox3 plants showed hypersensitivity to red light under all light fluences ( Figure 6A ) . In either constant far-red or blue light , hypocotyl lengths of pch1 and PCHox3 seedlings resembled those of wild type plants ( Figure 6—figure supplement 1 ) . These data suggest that PCH1 specifically modulates hypocotyl elongation in response to red light . 10 . 7554/eLife . 13292 . 022Figure 6 . pch1 exhibits defects in red light responsive hypocotyl growth and expression of downstream transcription factors . ( A ) Hypocotyl lengths of 4-day-old WT , pch1 , PCH1ox3 , PCH1p::PCH1 and phyB-9 seedlings grown under either dark or constant red light of various intensities ( 25 , 40 and 100 µmol·m-2·s-1 ) . Mean ± 95% CI ( n = 20 ) . Hypocotyl lengths of light-grown seedlings were normalized to dark-grown ( etiolated ) seedlings , and were plotted against light intensities to generate the responsive curve . Etiolated hypocotyl lengths ( mean ± SD ) of WT , pch1 , PCH1ox3 , PCH1p::PCH1 and phyB-9 are 9 . 02 ± 0 . 90 , 8 . 47 ± 0 . 66 , 8 . 04 ± 0 . 70 , 8 . 71 ± 0 . 71 and 7 . 63 ± 0 . 79 , respectively . ( B ) qPCR of HFR1 , ATHB-2 using time-course cDNA samples of short-day grown , 4-day-old WT , pch1 and PCH1ox3 seedlings . Expression was normalized to IPP2 and APA1 . Mean ± SD ( n=3 biological reps ) . Grey shading indicates dark period . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 02210 . 7554/eLife . 13292 . 023Figure 6—figure supplement 1 . pch1 does not affect far-red or blue light mediated hypocotyl elongation . ( A ) Hypocotyl measurements of 4-day-old WT , pch1 , PCH1ox3 and phyA-211 seedlings grown under either dark or constant far-red light conditions ( FRc , 25 µmol·m-2·s-1 ) . Mean ± 95% CI ( n = 20 ) . ( B ) Hypocotyl measurements of 4-day-old WT ( Col ) , pch1 , PCH1ox3 , Wassilewskija ( WS ) and cry1 cry2 ( in WS ) seedlings under either dark or constant blue light conditions ( Bc , 20 µmol·m-2·s-1 ) . Mean ± 95% CI ( n = 20 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 023 To better understand the mechanism underlying PCH1-mediated suppression of hypocotyl growth , the effect of altered PCH1 levels on the expression of transcription factors downstream of phytochrome signaling was measured . LONG HYPOCOTYL IN FAR RED 1 ( HFR1 ) and the homeobox transcription factor ATHB-2 are two transcription factors that are regulated by phytochromes in response to shade and short days , and are positively correlated with hypocotyl elongation ( Kunihiro et al . , 2011; Lorrain et al . , 2008; Steindler et al . , 1999 ) . qPCR analyses were done using time-course cDNA samples of short-day grown wild type , pch1 and PCH1ox3 seedlings . In wild type , transcripts of HFR1 and ATHB-2 are suppressed by daylight and accumulate in the evening , with a peak at dawn ( Figure 6B ) . pch1 mutants up-regulated HFR1 and ATHB-2 during the dark period , in agreement with our growth rate data showing the acceleration of hypocotyl growth in pch1 during night ( Figure 6B ) . Conversely , overexpression of PCH1 suppresses HFR1 and ATHB-2 transcript levels throughout the light/dark cycle ( Figure 6B ) . These data demonstrate that phytochrome photoperception and downstream gene expression is regulated by PCH1 . As PCH1 modulates phyB photobodies ( Figure 5 ) and red light perception ( Figure 6A ) , we sought to determine if altered regulation of the PIFs underlies the gene expression and growth defects observed in pch1 . Since PIF4 directly interacts with phyB and regulates HFR1 and ATHB-2 expression under shade or short day conditions , PIF4 expression and PIF4 levels were analyzed in pch1 mutants ( Lorrain et al . , 2008; Lorrain et al . , 2009; Soy et al . , 2012 ) . PIF4 mRNA levels were upregulated in pch1 compared to wild type in a qPCR assay ( Figure 7A ) . Using a pif4/PIF4p::PIF4-HA line to detect PIF4 protein levels , we observed higher PIF4 levels in the evening in pch1 compared to wild type ( Figure 7B ) , suggesting that PCH1 can modulate PIF4 levels in the early evening . To elucidate if PCH1 regulates hypocotyl elongation also through other PIFs , genetic interactions between pch1 and pifs ( pif3 and pif4 pif5 ) were tested by measuring hypocotyls of seedlings grown in short days ( Figure 7C ) . Single and higher order pif mutants reduced hypocotyl length in the wild type background , as previously reported ( Soy et al . , 2012 ) . Introducing pif mutant alleles into the pch1 mutant background progressively ameliorated the elongated hypocotyl phenotype of the pch1 mutant . Taken together , these results show that altered PIF levels underlie the growth defects seen in pch1 , and that PIF3 , 4 , and 5 are required for the hypocotyl growth defects in pch1 . 10 . 7554/eLife . 13292 . 024Figure 7 . pch1 affects PIF4 levels and PIFs are required for the hypocotyl phenotype in pch1 . ( A ) qPCR of PIF4 using time-course cDNA samples of short-day grown , 4-day-old WT and pch1 seedlings . Expression was normalized to IPP2 and APA1 . Mean ± SD ( n=3 biological reps ) . Grey shading indicates dark period . ( B ) Anti-HA immunoblots for testing PIF4-HA levels in WT ( pif4/PIF4p::PIF4-HA ) and pch1 ( pch1 pif4/PIF4p::PIF4-HA ) genetic backgrounds . Time-course protein extracts ( from ZT 0 to 24 , with 3 hr intervals , plus ZT 8 ) were made from short-day-grown , 4-day-old seedlings . Rectangles above blots represent light/dark conditions under which samples were flash frozen in liquid N2 , white = light and black = dark . Anti-RPT5 used as a loading control . pif4 extracts were used as a negative western control . ( C ) Hypocotyl lengths of 7-day-old , short-day-grown WT , pch1 , pif3 , pif4/5 , pif3/4/5 , pch1 pif3 , pch1 pif4/5 and pch1 pif3/4/5 seedlings were measured . Mean ± 95% CI ( n = 20 ) . Inset shows representative phenotypes with the scale bar = 5 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 024 Here we show that PCH1 is a new phytochrome interacting protein that functions to increase sensitivity to red light and prolongs phyB activity by maintaining photobody formation . PCH1 binding to the C-terminus of phyB likely stabilizes the Pfr conformer ( Chen et al . , 2005 ) , thereby providing a molecular memory of light exposure to prevent inappropriate elongation in response to long nights . Recent models of phyB signaling and photoconversion postulated that specific binding of a yet identified factor to phyB in the Pfr state might prevent dark-reversion and maintain phyB to photobodies to sustain an active pool of phyB in the dark in vivo ( Klose et al . , 2015 ) . We found that loss of PCH1 severely attenuated formation of large photobodies , particularly at low fluence red light ( Figure 5B ) . In both low and high red-light conditions , pch1 mutants had more small photobodies , suggesting that PCH1 regulates either the transition from small to large photobody or the maintenance of large photobodies . The phenotypes of pch1 mutants are distinct from mutations in HEMERA , which is necessary for photobody initiation ( Chen et al . , 2010 ) . Conversely , constitutive overexpression of PCH1 resulted in an increase in the number and prolonged maintenance of large photobodies during the night at both high and low light intensity compared to wild type ( Figure 5 and Figure 5—figure supplement 1 ) . Altering PCH1 levels , however , does not induce a constitutively photomorphogenic phenotype ( Figure 6A and Figure 6—figure supplement 1 ) . We favor a model wherein binding of PCH1 to phyB after light exposure traps phyB in an active conformation and prolongs phyB localization to large photobodies by either slowing dark reversion rates or through maintaining the superstructure of the photobody once formed . Our data demonstrate that PCH1 is a new component that suppresses the photoperiodic response of hypocotyl elongation . Although PCH1 accumulates at dusk , similar to the EC , pch1 mutants are hypersensitive to the extended night , while the EC mutants are insensitive to changing photoperiods , displaying long hypocotyls regardless of day length ( Figure 2A ) . This difference is likely due to the strong transcriptional effects on PIF4 and PIF5 expression when the EC is absent ( Nusinow et al . , 2011 ) . We propose that the short day-specific phenotype of pch1 results from the coincidence of the internal clock-controlled oscillation of PCH1 and PIF4 , and external photoperiodic cues . In short days , PCH1 peaks at dusk , binds photoactivated phyB and prolongs phyB photobody formation to maintain phyB in the Pfr state , which then suppresses PIF4 levels in the early evening , reducing PIF4 activities and hypocotyl growth ( Figure 8 ) . As the daytime increases in long days , the peak of PIF4 ( at ZT8 ) is no longer at dusk but in the middle of the day ( Yamashino et al . , 2014 and Figure 8—figure supplement 1 ) , when light perception by the phytochromes would act as the major suppressor of elongation through post-translation regulation of PIF protein levels , masking the requirement of PCH1 ( Figure 8 ) . PCH1 overexpression lines would constantly suppress hypocotyl elongation due to maintaining phyB photobodies throughout the night ( Figure 5B ) , leading to shortened hypocotyls as observed ( Figure 2A ) . It is likely other PIFs ( e . g . PIF3 and PIF5 ) also contribute to the hypocotyl phenotype of pch1 , as suggested by genetic analysis showing that higher order pif mutations progressively suppresses the long hypocotyl phenotype of pch1 ( Figure 7C ) . 10 . 7554/eLife . 13292 . 025Figure 8 . A model of PCH1-regulated day-length specific growth . A proposed model illustrates the role of PCH1 in controlling the photoperiodic hypocotyl elongation response . In short days , PCH1 peaks at dusk ( ZT 8 ) , maintains phyB photobody formation to suppress PIF4 levels and activities ( downstream gene expression ) to suppress hypocotyl elongation in the early evening . In long days , PIF4 peaks in the middle of the day and is repressed by active phyB . PIF4 protein decreases to basal level prior to dusk ( Figure 8—figure supplement 1 and Yamashino et al . , 2014 ) , therefore no longer requiring the additional suppression mediated by PCH1-regulated phyB photobodies in the evening . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 02510 . 7554/eLife . 13292 . 026Figure 8—figure supplement 1 . Expression of PIF3 , PIF4 and PIF5 under multiple photoperiods . Daily expression ( represented as gcRMA values ) of PIF3 , PIF4 , and PIF5 under short day , 12L:12D ( Col_LDHH ) and long day conditions were generated from microarray data ( Diurnal database , http://diurnal . mocklerlab . org/ , Mockler et al . , 2007 ) . Note high levels of PIFs ( e . g . PIF4s ) expression at ZT8 are at dusk in short days , while shifted gradually to the middle of the day when day length increases ( e . g . long day conditions ) . Grey shadings indicate dark periods . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 026 In summary , we have identified a new factor that binds to active phyB to extend its activity in the dark , and can maintain photomorphogenesis programs even in the long nights of short days . We anticipate that modulating PCH1 levels or its expression pattern could potentially alter light perception and lead to improved growth responses at latitudes where photoperiod changes during the agricultural season or in species whose yield is highly sensitive to photoperiod ( Sadras and Slafer , 2012 ) . All plants used in this study are in the Columbia ( Col ) ecotype of A . thaliana unless noted . pch1 ( SALK_024229 ) , pif3 ( SALK_081927C ) , phyB-9 and phyA-211 were obtained from the ABRC and described previously ( Alonso et al . , 2003; Reed et al . , 1993; Ruckle et al . , 2012; Sung et al . , 2007; Zhong et al . , 2012 ) . elf3-2 , elf4-2 , and pif4 pif5 ( pif4-101 pif5-1 ) lines were described previously ( Nusinow et al . , 2011 ) . WS and cry1 cry2 seeds were kindly provided by Takato Imaizumi ( University of Washington , Seattle ) and are in the Wassilewskija ( WS ) ecotype . pif4/PIF4p::PIF4-HA transgenic plants were kindly provided by Christian Fankhauser ( University of Lausanne Center for Integrative Genomics , Switzerland ) and crossed with pch1 . Homozygous mutant plants were validated by testing luciferase bioluminescence , drug resistance , and by PCR or dCAPS-based genotyping . Seeds were surface sterilized and plated on 1/2X Murashige and Skoog medium supplemented with 0 . 8% agar and 3% sucrose ( w/v ) . Sterilized seeds on plates were then stratified for 2 to 4 days in darkness at 4°C . After stratification , plates were placed horizontally in chambers for 4 days , supplied with white light ( WL , 80 µmol·m-2·s-1 ) and set to 22˚C , under various photoperiodic conditions , including long day , 12L:12D and short day ( Light: Dark= 16: 8 , 12: 12 and 8: 16 hr , respectively ) . For measuring hypocotyl lengths of pch1 and pifs mutants , 7-day-old seedlings grown under short day conditions were compared . For monochromatic wavelength treatments , stratified seedlings were first exposed to white light ( 80 µmol·m-2·s-1 ) for 5 hr to synchronize germination and then were grown under constant red light ( Rc , 40 µmol·m-2·s-1 ) , far-red light ( 24 µmol·m-2·s-1 ) , blue light ( 20 µmol·m-2·s-1 ) conditions ( CLF Plant Climatics , Wertingen , Germany ) or in the dark for 4 days , before hypocotyl measurements were taken . For hypocotyl elongation assays , 4 to 7-day-old seedlings ( as specified in each figure legend ) grown under different photoperiod or light conditions were arrayed , photographed with a ruler for measuring hypocotyl length using the Image J software ( NIH , Bethesda , Maryland ) . For measuring growth rate , a total of 96 time-lapse images were taken every hour for each seedling ( grown under short day conditions ) using an infrared-sensitive camera ( Pi-NoIR , Amazon . com , Seattle , Washington ) with a visible light cut-out filter ( 87 , Lee Filters , Burbank , CA ) and hand assembled 880 nm LED array controlled by a custom script running on a Raspberry Pi ( Amazon . com , Seattle , Washington ) from ZT30 to ZT125 . Hypocotyl lengths were then measured by Image J ( NIH , Bethesda , Maryland ) to calculate growth rates using PRISM software ( version 6 . 0 , Graphpad . com , La Jolla , California ) . For flowering assay , number of rosette leaves from plants with 1cm inflorescence stem was counted . For characterizing clock phenotype , a luciferase-based assay using the CCA1::LUC reporter was monitored as described previously ( Huang et al . , 2015 ) . Statistical analyses ( one-way or two-way ANOVA analysis with Bonferroni's multiple comparisons test ) for all experiments were performed using PRISM software ( Graphpad , La Jolla , California , version 6 . 0 , Graphpad . com ) . pB7HFC vector was used for constitutively expressing C-terminal His6-FLAG3 fusion proteins ( Huang et al . , 2015 ) . To generate the pB7SHHc and pB7YSHHc vectors ( for generating PCH1-YPet fusion protein used in a transient expression assay ) , we first modified the pB7WG2 vector by introducing an AvrII restriction site . The pB7WG2 vector ( Karimi et al . , 2002 ) was used as the template for amplifying two pieces of overlapping DNA fragments with an AvrII site added . These two fragments of AvrIIA ( using primers pDAN0193 and pDAN0202 ) and AvrIIB ( using primers pDAN201 and pDAN0223 ) were diluted , mixed to serve as template and were re-amplified with pDAN0193 and pDAN0223 to generate a longer fragment AvrIIC with the AvrII site in the middle . The pB7WG2 plasmid was then linearized by digestion with EcoRI and XbaI and recombined with AvrIIC fragment using In-Fusion HD cloning ( Clontech , Mountain View , California ) to generate the pB7AVRII vector , which was verified by sequencing and served as the backbone of pB7SHHc and pB7YSHHc vectors . DNA synthesis ( gBlocks Gene Fragments , IDT , Coralville , Iowa ) was used to generate a template sequence of 2xStrepII-HA-His6-TEV-FLAG3-TEV-His6-HA-2xStrepII , which contains 2xStrepII , HA , His6 , Tobacco Etch Virus protease cleavage sites , and FLAG3 epitopes for making all combinations of tags we need to put into the pB7AVRII vector . The sequence of this template is as follows: 5’-GGAAGCTGGAGCCACCCTCAATTTGAAAAGGGAGGAGGATCTGGAGGTGGTTCTGGTGGTGGTTCTTGGTCTCACCCACAATTCGAAAAGGGTTCTTACCCATACGATGTTCCAGATTACGCTCATCACCATCACCATCACGATATTCCAACTACTGCTAGCGAGAATTTGTATTTTCAGGGTGAGCTCGACTACAAAGACCATGACGGTGATTATAAAGATCATGACATCGACTACAAGGATGACGATGACAAGGATATACCTACTACTGCTTCTGAAAATCTGTACTTTCAGGGAGAACTGCACCATCATCATCATCACTACCCTTACGATGTGCCAGACTACGCTGGATCTTGGTCTCATCCACAGTTTGAAAAGGGAGGAGGATCTGGAGGAGGATCTGGAGGAGGATCTTGGAGTCATCCTCAGTTCGAGAAG–3’ . Primer set of pDAN0242 and pDAN0241 was used to amplify 2xSrepII-HA-His6 . The tandem tag was then recombined with pB7AVRII , which was linearized by AvrII digestion , to generate pB7SHHc using In-Fusion HD cloning ( Clontech , Mountain View , California ) . To generate the pB7YSHHc vector , YPet sequence was amplified from pBJ36 containing a YPet-3xHA tag ( pBJ36-YPet-3xHA ) as reported previously ( Krogan et al . , 2012 ) ( a generous gift from Dr . Jeff A . Long ) using primers pDAN0249 and pDAN0250 and recombined with pB7SHHc digested with AvrII using In-Fusion HD cloning ( Clontech , Mountain View , California ) to generate pB7YSHHc . See Table 3 for primer sequences . 10 . 7554/eLife . 13292 . 027Table 3 . Primers used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 13292 . 027Primers used for cloning PCH1 and PCH1 promoter aAmplified FragmentsForward primer ( 5'->3' ) Reverse primer ( 5'->3' ) PCH1-stopCACCATGTCTGAACATGTTATGGTTTTGGCTACCTCAAATCCCTTGCATTCCAPCH1-nonstopCACCATGTCTGAACATGTTATGGTTTTGGCCTCAAATCCCTTGCATTCCAAACPCH1-promoter bAAGCTTAGTTTCCTCATCATTTGCTATTGGCGTAAATCCTCACCGGTCTTPrimers used to generate yeast two-hybrid constructs , all with a stop codon aAmplified fragmentsForward primer ( 5'->3' ) Reverse primer ( 5'->3' ) PCH1CACCATGTCTGAACATGTTATGGTTTTGGCTACCTCAAATCCCTTGCATTCCAELF3CACCATGAAGAGAGGGAAAGATGAGCTAAGGCTTAGAGGAGTCATAGCGTTTELF4CACCATGAAGAGGAACGGCGAGACGATTAAGCTCTAGTTCCGGCAGCACCLUX ( full length ) CACCATGGGAGAGGAAGTACAAATTAATTCTCATTTGCGCTTCCACCTLUX-Nt ( amino acids 1-143 ) CACCATGGGAGAGGAAGTACAAACTATTTAAGTGTTTTCCCAGATAGLUX-Ct ( amino acids 144-324 ) CACCATGCGACCGCGTTTAGTGTGGACATTAATTCTCATTTGCGCTTCCACCTphyA-Ct ( amino acids 606-1123 ) CACCATGGATCTCAAAATTGATGGTATACAACTACTTGTTTGCTGCAGCGAGTTCphyB-Ct ( amino acids 640-1173 ) CACCATGGCGGGGGAACAGGGGATTGATGAGCTAATATGGCATCATCAGCATCATGTCAphyC-Ct ( amino acids 592-1112 ) CACCATGGATAATAGGGTTCAGAAGGTAGATTCAAATCAAGGGAAATTCTGTGAGGATCACphyD-Ct ( amino acids 645-1165 ) CACCATGGTACAGCAAGGGATGCAGTCATGAAGAGGGCATCATCATCAphyE-Ct ( amino acids 583-1113 ) CACCATGAATGGCGTAGCAAGAGATGCCTACTTTATGCTTGAACTACCCTCTGTCOP1CACCATGGAAGAGATTTCGACGGATCACGCAGCGAGTACCAGAACTTTGTZPCACCATGGGAGATGGAGATGAGCAACTAAAAGCCTAACATTTTTCTCTGCTGAPrimers used for qPCRGeneForward primer ( 5'->3' ) Reverse primer ( 5'->3' ) PCH1 set ACCGGCTCCATTTCTTCGTCATCCGGAACAAGAGGTGGTTCTPCH1 set BGAAGTTATTGTTGTCGCCCTGGGAAATCCAAAGCGGTATTIPP2CTCCCTTGGGACGTATGCTGTTGAACCTTCACGTCTCGCAAPA1 ( At1g11910 ) cCTCCAGAAGAGTATGTTCTGAAAGTCCCAAGATCCAGAGAGGTCHFR1TAAATTGGCCATTACCACCGTTTAACCGTGAAGAGACTGAGGAGAAGAATHB-2GAAGCAGAAGCAAGCATTGGCGACGGTTCTCTTCCGTTAGPIF4GTTGTTGACTTTGCTGTCCCGCCCAGATCATCTCCGACCGGTTTPrimers for genotypingMutantfor wild type PCR ( 5'->3' ) for mutant PCR ( 5'->3' ) pch1 ( SALK_024229 ) TGTCAGGTATTTCGGTCCTTG ( LP ) and CACTTGCTTGATGCTCATGAG ( RP ) AAGAACCGGCAAAGATACCAC ( RP ) and ATTTTGCCGATTTCGGAAC ( LBb 1 . 3 ) pif3 ( SALK_081927C ) AGTCTGTTGCTTCTGCTACGC ( LP ) and AAGAACCGGCAAAGATACCAC ( RP ) ACATACAGATCTTTACGGTGG ( RP ) and ATTTTGCCGATTTCGGAAC ( LBb 1 . 3 ) pif4 ( pif4-101 ) dCTCGATTTCCGGTTATGG ( SL42 ) and CAGACGGTTGATCATCTG ( SL43 ) GCATCTGAATTTCATAACCAATC ( PD14 ) and CAGACGGTTGATCATCTG ( SL43 ) pif5 ( pif5-1 ) dTCGCTCACTCGCTTACTTAC ( SL46 ) and TCTCTACGAGCTTGGCTTTG ( SL47 ) TCGCTCACTCGCTTACTTAC ( SL46 ) and GGCAATCAGCTGTTGCCCGTCTCACTGGTG ( JMLB1 ) elf3-2 cTGAGTATTTGTTTCTTCTCGAGC and CATATGGAGGGAAGTAGCCATTACTGGTTATTTATTCTCCGCTCTTTC and TTGTTCCATTAGCTGTTCAACCTAelf4-2 cATGGGTTTGCTCCCACGGATTA and CAGGTTCCGGGAACCAAATTCT , cut with HpyCH4V . WT has 5 cuts while elf4-2 has 4 cuts to give a unique 689 bp band . phyB-9GTGGAAGAAGCTCGACCAGGCTTTG and GTGTCTGCGTTCTCAAAACG , cut with MnlI , phyB-9 gives 167+18 bp bands , WT gives a 185 bp band . Primers for making pB7SHHc and pB7YSHHcPrimer NameSequence ( 5'->3' ) pDAN0193TGCCCGCCTGATGAATGCTCpDAN0202GCGGGATATCACCACCCTAGGCACCACTTTGTACAAGAAAGCTGApDAN201TCAGCTTTCTTGTACAAAGTGGTGCCTAGGGTGGTGATATCCCGCpDAN0223ATTCTCATGTATGATAATTCGAGGpDAN0242TACAAAGTGGTGCCTAGGGGTGGAAGCTGGAGCCACCCTCpDAN0241GCGGGATATCACCACCCTAGTGATGGTGATGGTGATGAGCGpDAN0249GCTTTCTTGTACAAAGTGGTGCCTGCTGCTGCTGCCpDAN0250GGTGGCTCCAGCTTCCACCCCCCTTATAGAGCTCGTTCa CACC ( underscored ) were added to forward primers for cloning into the pENTR/D-TOPO vector . b a Hind III restriction site ( in bold ) was added to the forward primer . c ( Nusinow et al . , 2011 ) . d ( de Lucas et al . , 2008 ) . The pB7HFC vector was described previously ( Huang et al . , 2015 ) . All cDNAs encoding either full-length or fragments of tested genes ( with or without stop codons , as listed in Table 3 ) were first cloned into the pENTR/D-TOPO vector ( Thermo Scientific , Waltham , Massachusetts ) and were verified by sequencing . Transgenes were introduced into various genetic backgrounds by crossing . To generate PCH1 overexpression lines ( Col [35S::PCH1-His6-FLAG3][CCA1::LUC] ) , cDNA of PCH1 ( without the stop codon ) was Gateway cloned ( LR reaction , Invitrogen ) into the pB7HFC vector . The pB7HFC-PCH1 construct was then transformed into Col [CCA1::LUC] plants by the floral dip method ( Clough and Bent , 1998 ) . Two homozygous lines PCH1ox3 and 4 were identified and used in this paper . elf4-2 , elf3-2 and phyB-9 ( all carrying the CCA1::LUC reporter ) were crossed with PCH1ox3 . To generate the pch1 [PCH1pro::PCH1-His6-FLAG3][CCA1::LUC] complementation line ( PCH1p::PCH1-7 and -8 ) , a fragment from ~1 . 5 kb sequence upstream of the transcription start site plus 5’UTR to exon 1 of PCH1 was cloned , using primers that introduced a HindIII restriction enzyme cutting site to its 5’ end ( listed in Table 3 ) . The amplified fragment was then swapped into the pB7HFC-PCH1 construct to replace the 35S promoter by restriction enzyme digestion with HindIII and XhoI and ligation . The pB7HFC-PCH1p::PCH1 construct was then transformed into pch1 [CCA1::LUC] plants . PCH1p::PCH1-7 plants was used in time-course western blottings as well as physiological assays . cDNA of PCH1 without the stop codon was gateway cloned into the pB7YSHHc vector to make the pB7YSHHc-PCH1 construct ( 35S::PCH1-YPet-2xStrepII-HA-His6 ) . Coding sequence of YPet was gateway cloned into the pB7SHHc vector to serve as a control ( 35S::YPet-2xStrepII-HA-His6 , pB7SHHc-YPet ) . The GFP construct has been described previously ( 35S::GFP , pB7GFP ) ( Huang et al . , 2015 ) . The phyB-GFP construct is a generous gift from Dr . A . Nagatani ( Kyoto University , Japan ) that was described previously ( Yamaguchi et al . , 1999 ) and was transformed into phyB-9 plants to generate phyB-9 [35S::phyB-GFP] plants ( PBG ) . PBG plants were then crossed with pch1 phyB-9 and PCH1ox3 phyB-9 to make pch1/PBG and PCH1ox3/PBG lines ( without the CCA1:LUC reporter ) . The pif4/PIF4p::PIF4-HA transgenic line was generated by Séverine Lorrain in Christian Fankhauser’s lab ( University of Lausanne Center for Integrative Genomics , Switzerland ) , which expresses a C-terminal PIF4-3xHA fusion protein driven by the PIF4 native promoter ( ~2 . 1 kb upstream of the start codon ) . Time course RNA samples ( with 3 hr interval ) were made from 4-day-old seedlings of Col , pch1 , and PCH1ox3 ( all carrying the CCA1::LUC reporter ) grown under short day conditions , using the RNeasy Plant Mini Kit ( Qiagen , Hilden , Germany ) . 1 μg of total RNA was reverse transcribed to make cDNA using the iScript cDNA synthesis kit ( Bio-Rad , Carlsbad , CA ) , which was quantified by quantitative real-time PCR ( qPCR ) using a CFX 384 Real-Time System ( C1000 Touch Thermal Cycler , Bio-Rad , Hercules , California ) . PCR was set up as follows: 3 min at 95°C , followed by 40 cycles of 10 s at 95°C , 10 s at 55°C and 20 s at 72°C . A melting curve analysis was conducted right after all PCR cycles are done . Both IPP2 ( At3g02780 ) and APA1 ( At1g11910 ) , expression of which remain stable during the diurnal cycle , were used as the normalization controls ( Hazen et al . , 2005; Michael et al . , 2008a; Nusinow et al . , 2011 ) . PCR efficiencies for each target/reference genes were assessed and qPCR analyses were carried out by applying actual PCR efficiencies to calculate the relative expression of each sample , as described previously ( Hellemans et al . , 2007; Remans et al . , 2014 ) . All qPCR were done using 3 biological replicates . For semi-quantitative qPCR , all cDNA samples of Col [CCA1::LUC] or pch1 [CCA1::LUC] time course ( from ZT0 to ZT24 , with 3 hr intervals ) were pooled and 200 ng of pooled cDNA was used . 30 ng of genomic DNA was used as comparison . PCR conditions are as follows: 5 min at 95°C , followed by 30 cycles of 30 s at 95°C , 30 s at 55°C and 20 s at 72°C for cDNA template or 30 s at 72°C for genomic DNA template ) . See Table 3 for primer sequences . We used the Matchmaker GAL4 Two-Hybrid systems ( Clontech , Mountain View , California ) to analyze protein-protein interactions in yeast . Verified cDNA sequences ( primers listed below ) were cloned into either the pAS2-GW or pACT2-GW vector , which are derived from the pAS2-1 and pACT2 plasmids of Clontech ( Nusinow et al . , 2011 ) , through Gateway LR recombination reactions ( Thermo Scientific , Waltham , Massachusetts ) . Both the DNA binding domain ( DBD ) or activating domain ( AD ) -fused constructs were transformed into Saccharomyces cerevisiae strain Y187 ( MATα ) and the AH109 ( MATa ) , respectively , by the Li-Ac transformation protocol according to the yeast handbook ( Clontech , Mountain View , California ) . Two yeast strains of the same optical density ( OD600 ) were mixed and incubated in low pH YCM media ( 1% yeast extract , 1% bactopeptone , 2% dextrose , pH 4 . 5 ) for 4 . 5 hr at 30°C . Afterwards , cells were transferred to regular YPDA media and incubated overnight at 30°C . Diploid yeast were then grown on CSM –Leu –Trp plates ( Sunrise Science , San Diego , California ) supplemented with extra Adenine ( 30 mg/L final concentration ) for selection of bait and prey vectors and were tested for protein-protein interaction by plate replicating on CSM –Leu –Trp –His media supplemented with extra Adenine and 2 mM 3-Amino-1 , 2 , 4-triazole ( 3AT ) . Pictures were taken after 4-day incubation at 30°C . Empty pAS2-GW and pACT2-GW plasmids were used as negative controls . See Table 3 for primer sequences . Overnight saturated cultures of Agrobacterium tumefaciens strain GV3101 carrying pB7YSHHc-PCH1 , pB7SHHc-YPet , phyB-CFP ( 35S::phyB-CFP ) ( Nito et al . , 2013 ) , pB7HFC-PCH1 , phyB-GFP ( 35S::phyB-GFP , PBG ) and GFP ( 35S::GFP , pB7GFP ) were diluted in 10 mM MgCl2 ( OD600 = 0 . 8 ) and kept at room temperature for 1~2 hr . An Agrobacterium culture of 35S:P19-HA was also diluted into the same concentration and mixed ( at a ratio of 1:1 ) with each culture to suppress gene silencing ( Chapman , 2004 ) . The cultures were then spot-infiltrated into 4 to 5-week-old Nicotiana benthamiana from the abaxial side of leaves . After 48 hr , infected leaves were flash frozen for protein extraction and co-IP experiments or were cut into small square pieces , mounted in water and used for confocal microscopy . For PCH1-YPet and phyB-CFP co-localization assay , confocal microscopy was performed with a Leica TCS SP8 confocal laser scanning microscope and an HC PL APO CS2 63x/1 . 20 WATER objective lens ( Leica Microsystems , Mannheim , Germany ) . Light source is provided by the UV Diode laser ( for CFP ) or the White Light Laser ( WLL , for YPet ) , while all emission fluorescence signals were detected by the HyD detector . CFP fluorescence was monitored by a 460–505 nm band emission and a 405 nm excitation line of the UV Diode laser , with 2% transmission value . YPet fluorescence was sequentially monitored by a 525–600 nm band emission and a 514 nm excitation line of an Ar laser , with 5% transmission value . Line average was set as 16 to reduce noise and frame accumulation was set as 1 . For measuring phyB photobodies in phyB-9 , pch1 phyB-9 and PCH1ox3 phyB-9 plants expressing phyB-GFP ( PBG ) , seedlings were sampled at ZT 56 ( under light ) , 60 ( dark ) , 64 ( dark ) , and 72 ( dark ) for short-day-entrained ( by 10 or 40 μmol·m-2·s-1red light ) seedlings . Fixation was carried out as follow steps: seedlings were first immersed in 2% paraformaldehyde in 1x PBS on ice with 15 min vacuum followed by incubation in 50 mM NH4Cl in 1xPBS for 5 min 3 times , and washed by 1xPBS with 0 . 2% TritonX-100 for 5 min one time and 1xPBS for 5 min 2 times . Fixed seedlings were mounted on Superfrost slides using 1x PBS . Nuclei from hypocotyl were imaged using a Zeiss LSM 510 inverted confocal microscope . GFP signal was detected using a 100× Plan-Apochromat oil-immersion objective , 488-nm excitation from argon laser and 505 to 550 nm bandpass detector setting . The proportion of nuclei with or without photobodies was manually scored . To quantify the number and size of photobodies , confocal images were analyzed by Huygens Essential software . The object analyzer tool was used to threshold the image and to calculate the volume of each photobody in the image . Total number of large photobodies ( >1 . 0 μm3 ) or small photobodies ( < 1 . 0 μm3 ) was presented . For time-course sampling , seedlings were grown on sterilized qualitative filter paper ( Whatman , Maidstone , United Kingdom ) for 4 days , at 22°C under various photoperiods ( long day , 12L:12D and short day ) . 0 . 5 g of PCH1p::PCH1-7 or PCH1ox3 whole seedlings was collected every 3 hr from ZT0 to ZT24 and flash frozen in liquid N2 . For PIF4p::PIF4-HA transgenic plants in pif4 ( WT ) and in pch1pif4 , 4-day-old seedlings grown under short day conditions at 22°C were samples from ZT0 to ZT 24 , with 3 hr interval and with addition of ZT8 . Each time-course sample was put in a 2 mL tube that contained three 3 . 2-mm stainless steel beads ( Biospec Bartlesville , Oklahoma ) . It is noted that samples undergoing dark to light transitions ( e . g . ZT0 and ZT24 ) were collected in the dark before the transition to light , while ZT8 samples were harvested in light . For co-IP experiments testing phyB-PCH1 interaction under different light treatments ( light , dark , red light and end-of-day far red light treatments ) , seedlings were grown under 12L:12D conditions at 22°C , on sterilized qualitative filter paper ( Whatman , Maidstone , United Kingdom ) for four days and sampled at specific ZT timepoints . Frozen plant tissues of either Arabidopsis seedlings or tobacco leaves were homogenized in a reciprocal mixer mill ( Retsch Mixer Mill MM 400 , Newtown , Pennsylvania ) . Homogenized tissue of about 0 . 5 g was gently resuspended in 0 . 5 ml of SII buffer [100 mM sodium phosphate , pH 8 . 0 , 150 mM NaCl , 5 mM EDTA , 5 mM EGTA , 0 . 1% Triton X-100 , 1 mM PMSF , 1x protease inhibitor cocktail ( Roche , Pleasanton , California ) , 1x Phosphatase Inhibitors II & III ( Sigma ) , and 5 µM MG132 ( Peptides International , Louisville , Kentucky ) ] and sonicated twice at 40% power , 1 s on/off cycles for a total of 10 s on ice ( Fisher Scientific model FB505 , with microtip probe , ThermoFisher Scientific , Waltham , Massachusetts ) . For PIF4p::PIF4-HA samples , about 100 μl homogenized tissue powder was mixed with 100 μl denature sample buffer ( 50 mM Tris-HCl , pH 7 . 5 , 150 mM NaCl , 0 . 1% Triton X-100 , 4% SDS ) and denatured in dark by incubation at 95°C for 10 min . Extracts were then clarified by centrifugation twice at 4°C for 10 min at ≥20 , 000 g . For tobacco extracts , a 10% ( w/v ) of polyvinylpolypyrrolidone ( PVPP ) was added to resuspended extracts for 5 min incubation and was discarded after centrifugation . Concentration of total proteins from each sample was measured by using the DC Protein Assay kit ( BIO-RAD ) . 40 ~ 50 µg total proteins were denatured and loaded to a 8% or 10% SDS-PAGE gel , followed by transferred to a nitrocellulose membrane . For western blots , all of the following primary antibodies were diluted into PBS + 0 . 1% Tween + 2% BSA and incubated overnight at 4°C: Anti-GFP-rabbit ( 1:5000 , Abcam , Cambridge , United Kingdom ) , anti-phyB-mouse ( mAB2 , at 1:3000 , a generous gift from Dr . Akira Nagatani at Univeristy of Kyoto ) , and anti-ACTIN-mouse mAB1501 ( 1:2500 , EMD-Millipore , Darmstadt , Germany ) . Anti-HA-HRP ( Roche , Pleasanton , California ) was used as 1:2000 and incubated for 1 hr . Anti-FLAGM2-HRP ( Sigma Aldrich , St Louis , Missouri ) and anti-RPT5-rabbit ( ENZO Life Science , Farmingdale , New York ) was incubated for 1 hr at room temperature and diluted into PBS + 0 . 1% Tween at 1:10 , 000 and 1:5000 , respectively . Anti-Rabbit-HRP and anti-Mouse-HRP secondary antibodies ( Sigma Aldrich , St Louis , Missouri ) were diluted 1:20 , 000 into PBS + 0 . 1% Tween and incubated at room temperature for 1 hr . For in vivo co-IP experiment , 2 mg of protein extract of PCH1ox3 plants ( in 1 ml SII buffer with supplements of inhibitors ) was used . Dynabeads ( ThermoFisher Scientific , Waltham , Massachusetts ) had been conjugated with the Anti-FLAGM2 monoclonal antibody ( Sigma Aldrich , St Louis , Missouri ) ( Nusinow et al . , 2011 ) to precipitate PCH1-His6-FLAG3 and its interacting proteins . 5 μg antibodies conjugated to 30 ul of Dynabeads were used for each FLAG-IP and were incubated with protein extracts on a rotor at 4°C for 1 hr , followed by being washed in SII buffer thrice . IP beads were added with 30 μl 2X SDS sample buffer and incubated at 75°C for 10 min to denature and elute bound proteins . SDS-PAGE and western detections were done as instructed above . It is noted that for co-IPs under different light treatments , all steps were carried out in a cold room supplemented with dim green safety light . For in-vitro co-IP/binding assay , cDNAs of PCH1-His6-Flag3 or YPet-His6-Flag3 was gateway cloned into the pDEST17 vector ( ThermoFisher Scientific , Waltham , Massachusetts ) . The fusion proteins were expressed in BL21 ( DE3 ) pLysS cells ( Promega , Madison , Wisconsin ) ( 1 mM IPTG induction for 3 hr at 30°C ) and purified by Ni-NTA agarose beads ( Qiagen , Hilden , Germany ) following standard procedures . The phyB-HA prey was synthesized using plasmid pCMX-PL2-phyB-HA ( Qiu et al . , 2015 ) and the TNT T7 Quick Coupled Transcription/Translation System ( Promega , Madison , Wisconsin ) as instructed by manual . phyB-HA prey was first resuspended in 500 µl Tris-buffered saline ( TBS ) supplemented with 20 µM phytochromobilin ( PΦB ) and incubated for 45 min at 12°C under constant red ( for Pfr phyB , 50 µmol·m-2·s-1 ) , far red ( for Pr phyB , 25 µmol·m-2·s-1 ) or dark ( without PΦB , for phyB apoprotein ) conditions . 4 . 9 ug purified PCH1-His6-Flag3 protein was then mixed with prey and incubated under the same light treatment for another 45 min at 12°C . 30 ul TALON beads ( incubated for 30 min at 12°C ) were used for immunoprecipitating each sample , followed by being washed with PBS+T buffer thrice . Tandem affinity purifications using PCH1ox3 plants ( in all genetic backgrounds ) were carried out as previously described ( Huang et al . , 2015 ) . In brief , 10-day-old seedlings of PCH1ox3 in Col , elf4-2 , elf3-2 and phyB-9 genetic backgrounds were grown on sterilized qualitative filter paper , under the 12L:12D conditions . 5 g of whole seedlings were harvested at ZT12 and immediately frozen in liquid N2 . Tandem FLAG and His immunoprecipitations were carried out to co-purify proteins associated with PCH1-His6-FLAG3 as described in detail at Bio-protocol ( Huang and Nusinow , 2016 ) . At least two independent biological replications were performed . The proteins were cleaved to peptides with trypsin before analyzed on an LTQ-Orbitrap Velos Pro ( ThermoFisher Scientific , Waltham , MA ) coupled with a U3000 RSLCnano HPLC ( Promega , Madison , Wisconsin ) operated in positive ESI mode using collision induced dissociation ( CID ) to fragment the HPLC separated peptides as previously described ( Huang et al . , 2015 ) . MS data were extracted by Proteome Discoverer ( ThermoFisher Scientific; v . 1 . 4 ) and database searches were done using Mascot ( Matrix Science , London , UK; v . 2 . 5 . 0 ) assuming the digestion enzyme trypsin , two missed cleavages , and using the TAIR10 database ( 20101214 , 35 , 386 entries ) and the cRAP database ( http://www . thegpm . org/cRAP/ ) . Deamidation of asparagine and glutamine , oxidation of methionine and carbamidomethyl of cysteine were specified as variable modifications , while a fragment ion mass tolerance of 0 . 80 Da , a parent ion tolerance of 15 ppm was used in the Mascot search . Scaffold ( Proteome Software Inc . , Portland , Oregon; v . 4 . 4 . 3 ) was used to validate MS/MS based peptide and protein identifications . Peptide identifications were accepted if they could be established at greater than 95 . 0% probability and the Scaffold Local FDR was <1% . Protein identifications were accepted if they could be established at greater than 99 . 0% probability as assigned by the Protein Prophet algorithm ( Keller et al . , 2002; Nesvizhskii et al . , 2003 ) . Proteins that contained similar peptides and could not be differentiated based on MS/MS analysis alone were grouped to satisfy the principles of parsimony and proteins sharing significant peptide evidence were grouped into clusters . Only the proteins identified by PCH1ox3 AP-MS in Col with ≥2 unique peptides were presented in tables , except when proteins with only one peptide were identified in more than one replicate . A full list of all proteins co-purified by PCH1 AP-MS is in Table 1—source data 1 . The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium ( Vizcaino et al . , 2014 ) via the PRIDE partner repository with the dataset identifier PXD003352 and 10 . 6019/PXD003352 .	Most living things possess an internal “circadian” clock that synchronizes many behaviors , such as eating , resting or growing , with the day-night cycle . With the help of proteins that can detect light , known as photoreceptors , the clock also coordinates these behaviors as the number of daylight hours changes during the year . However , it is not known how the clock and photoreceptors are able to work together . The circadian clocks of animals and plants have evolved separately and use different proteins . In plants , a photoreceptor called phytochrome B responds to red light and regulates the ability of plants to grow . Most plants harness sunlight during the day , but grow fastest in the dark just before dawn . In 2015 , researchers identified a new protein in a plant called Arabidopsis that is associated with several plant clock proteins and photoreceptors , including phytochrome B . However , the role of this new protein was not clear . Now , Huang et al . – including many of the researchers from the 2015 work – studied the new protein , named PCH1 , in more detail . The experiments show that PCH1 is a critical link that regulates the daily growth of Arabidopsis plants in response to the number of daylight hours . PCH1 stabilizes the structure of phytochrome B so that it remains active , even in the dark . This prolonged activity acts as a molecular memory of prior exposure to light and helps to prevent plants from growing too much in the winter when there are fewer hours of daylight . Since PCH1 is also found in other species of plants , it may play the same role in regulating growth of major crop plants . The next challenge is to understand how the binding of PCH1 to phytochrome B alters the photoreceptor’s activity . In the future , Huang et al . hope to find out if manipulating the activity of PCH1 can improve the growth of crops in places where there is a large change in day length across the seasons .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology" ]	2016	PCH1 integrates circadian and light-signaling pathways to control photoperiod-responsive growth in Arabidopsis
Gut enzymes can metabolize plant defense compounds and thereby affect the growth and fitness of insect herbivores . Whether these enzymes also influence feeding preference is largely unknown . We studied the metabolization of taraxinic acid β-D-glucopyranosyl ester ( TA-G ) , a sesquiterpene lactone of the common dandelion ( Taraxacum officinale ) that deters its major root herbivore , the common cockchafer larva ( Melolontha melolontha ) . We have demonstrated that TA-G is rapidly deglucosylated and conjugated to glutathione in the insect gut . A broad-spectrum M . melolontha β-glucosidase , Mm_bGlc17 , is sufficient and necessary for TA-G deglucosylation . Using cross-species RNA interference , we have shown that Mm_bGlc17 reduces TA-G toxicity . Furthermore , Mm_bGlc17 is required for the preference of M . melolontha larvae for TA-G-deficient plants . Thus , herbivore metabolism modulates both the toxicity and deterrence of a plant defense compound . Our work illustrates the multifaceted roles of insect digestive enzymes as mediators of plant-herbivore interactions . Plants produce an arsenal of toxic secondary metabolites , many of which protect them against phytophagous insects by acting as toxins , digestibility reducers , repellents , and deterrents ( Mithöfer and Boland , 2012 ) . Insect herbivores commonly metabolize defense metabolites , with important consequences for the toxicity of the compounds ( Heckel , 2014; Pentzold et al . , 2014 ) . Recent studies identified a series of enzymes that metabolize plant defense metabolites and thereby benefit herbivore growth and fitness ( Sun et al . , 2019; Sun et al . , 2020; Poreddy et al . , 2015 ) . However , to date , the behavioral consequences of insect metabolism of plant defense metabolites are little understood , despite the importance of behavioral effects of plant defenses for plant fitness and evolution in nature ( Mithöfer and Boland , 2012; Huber et al . , 2016a; Huber et al . , 2016b; War et al . , 2012 ) . Insect enzymes that were identified to metabolize plant defense compounds belong mainly to a few large enzyme classes including the cytochrome P450 monooxygenases , UDP-glycosyltransferases , and glutathione S-transferases ( Mao et al . , 2007; Heidel-Fischer and Vogel , 2015; Bass et al . , 2013; Maag et al . , 2014; Wouters et al . , 2014 ) . However , members of other enzyme groups can participate in detoxification , some of which are also involved in primary digestive processes for the breakdown of carbohydrates ( β-glucosidases ) , proteins ( proteases ) , and lipids ( lipases ) . For instance , a Manduca sexta β-glucosidase deglycosylates the Nicotiana attenuata diterpene glycoside lyciumoside IV , thus alleviating its toxicity ( Poreddy et al . , 2015 ) . Similarly , the Mexican bean weevil ( Zabrotes subfasciatus ) expresses a protease that degrades α-amylase inhibitors from its host , the common bean ( Phaseolus vulgaris ) ( Ishimoto and Chrispeels , 1996 ) . Finally , several insects degrade antinutritional plant protease inhibitors through intestinal proteases ( Giri et al . , 1998; Zhu-Salzman and Zeng , 2015 ) . Together , these studies suggest that families of typical digestive enzymes should be examined more carefully for possible roles in the detoxification of plant chemicals . Enzymes involved in carbohydrate digestion may play a particular role in processing plant defense glycosides . Such compounds are typically considered protoxins , non-toxic , glycosylated precursors that are brought into contact with compartmentalized plant glycosidases upon tissue damage to yield toxic aglycones ( Wittstock and Gershenzon , 2002 ) . Both plant and insect glycosidases may activate plant defense glycosides ( Pentzold et al . , 2014 ) . The alkaloid glucoside vicine in fava beans , for instance , is hydrolyzed to the toxic aglycone divicine in the gut of bruchid beetles ( Desroches et al . , 1997 ) . Similarly , phenolic glycoside toxins are hydrolyzed rapidly by Papilio glaucus , the eastern tiger swallowtail . P . glaucus subspecies adapted to phenolic glycoside-containing poplars and willow show significantly lower hydrolysis of these metabolites ( Lindroth , 1988 ) . Finally , iridoid glycosides from Plantago species are hydrolyzed and thereby activated by herbivore-derived β-glucosidases , and β-glucosidase activity is negatively correlated with host plant adaptation both within and between species ( Pankoke et al . , 2012; Pankoke and Dobler , 2015 ) . These studies show that herbivore-derived enzymes may cleave plant protoxins and so may be a target of host plant adaptation . However , the genetic basis of protoxin activation by herbivores and the biological consequences of this phenomenon for insect feeding preference and performance are poorly understood . Although the deglycosylation of plant defense metabolites is commonly assumed to be disadvantageous for the herbivore , a recent study in M . sexta showed that deglycosylation of a plant glycoside may decrease rather than increase toxicity ( Poreddy et al . , 2015 ) . Silencing M . sexta β-glucosidase one resulted in developmental defects in larvae feeding on N . attenuata plants producing the diterpene glycoside lyciumoside IV ( Lyc4 ) , but not in larvae feeding on Lyc4-deficient plants , suggesting that deglycosylation detoxifies rather than activates Lyc4 . Although Lyc4 is an atypical defensive glycoside that carries several different sugar moieties and is only partially deglycosylated by M . sexta , these results bring up the possibility that defensive activation by glycoside hydrolysis does not necessarily increase the toxicity of these compounds , but may be a detoxification strategy . Clearly , more research on how glycoside hydrolysis by digestive enzymes impacts herbivores is needed to understand the role of this process in plant-herbivore interactions ( Pentzold et al . , 2014; Poreddy et al . , 2015; Marti et al . , 2013 ) . The herbivore toxins derived from glycoside protoxins have often been investigated for their defensive roles in connection with herbivore growth and development ( Poreddy et al . , 2015; Desroches et al . , 1997; Lindroth , 1988; Pankoke et al . , 2012; Pankoke and Dobler , 2015 ) rather than feeding deterrence , despite the fact that the latter is a well-established mechanism for plant protection in this context ( Pollard , 1992 ) . For example , the maize benzoxazinoid glucoside HDMBOA-Glc reduces food intake by Spodoptera caterpillars as soon as the glucoside moiety is cleaved off by plant β-glucosidases ( Glauser et al . , 2011 ) . Similarly , the deterrent effect of cyanogenic glucosides in Sorghum toward Spodoptera frugiperda is directly dependent on a functional plant β-glucosidase that releases cyanide upon tissue disruption ( Krothapalli et al . , 2013 ) . Furthermore , different glucosinolate breakdown products have been shown to affect oviposition and feeding choices by Pieris rapae and Trichoplusia ni ( de Vos et al . , 2008; Zhang et al . , 2006; Mumm et al . , 2008 ) . However , whether protoxin activation by herbivore-derived enzymes influences herbivore host plant choice remains unknown . All protoxin-activating enzymes that have been characterized so far in insect herbivores are β-glucosidases , which cleave β-D-glucosides and release free glucose ( Pentzold et al . , 2014 ) . The primary role of β-glucosidases in insect digestion is to function in the last steps of cellulose and hemicellulose breakdown by converting cellobiose to glucose ( Zhang et al . , 2012 ) . Most insect β-glucosidases , however , also accept other substrates , including various di- and oligosaccharides , glycoproteins , and glycolipids , which may help herbivores to obtain glucose from various sources and enable the further breakdown of glycosylated proteins and lipids ( Marana et al . , 2000; Azevedo et al . , 2003; Ferreira et al . , 2003; Ferreira et al . , 2001 ) . However , the broad substrate specificity of insect β-glucosidases for plant glucosides with an aryl or alkyl moiety may also result in the activation of defense metabolites , as discussed above ( Terra and Ferreira , 1994 ) . Thus , investigating the substrate specificity and the biochemical function of insect β-glucosidases is important to understand the ecology and evolution of insect-mediated protoxin activation . Known plant protoxins include glucosinolates , salicinoids , and cyanogenic , iridoid , and benzoxazinoid glycosides . Plants produce many other types of glycosides that may also be protoxins , but most of these have not yet been carefully investigated for their toxicity or metabolic stability in herbivores . Among these potential protoxins are the bitter-tasting sesquiterpene lactone glycosides . Sesquiterpene lactones form a large group of over 2000 plant defense compounds found principally in the Asteraceae family , with glycosides especially common in the latex-producing tribe Cichorieae , which enters the human diet through lettuce , endive , and chicory ( Chadwick et al . , 2013 ) . These substances have a long appreciated role in defense against insect herbivores ( Picman , 1986 ) , but it is not clear if glycosylated sesquiterpene lactones should be considered as protoxins that are activated by plant damage . Here , we studied the metabolism of a sesquiterpene lactone glucoside during the interaction between the common dandelion Taraxacum officinale aggregate ( Asteraceae , Chicorieae ) and the larvae of the common cockchafer , Melolontha melolontha ( Coleoptera , Scarabaeidae ) ( Keller et al . , 1986; Hasler , 1986 ) . M . melolontha larvae feed on roots of different plant species including members of Poaceae , Brassicaceae , Salicaceae , and Asteraceae families , which can contain glycosylated defense compounds such as benzoxazinoids , glucosinolates , and salicinoids , as well as sesquiterpene lactone glycosides ( Kondor et al . , 2007; Hauss and Schütte , 1976; Hauss , 1975; Sukovata et al . , 2015 ) . The alkaline gut pH of M . melolontha ( pH = 8 . 0–8 . 5 ) possibly facilitates its polyphagous feeding habit by inhibiting the often acidic activating glucosidases of plant protoxins ( Pentzold et al . , 2014; Egert et al . , 2005 ) . In the third and final instars , M . melolontha prefers to feed on T . officinale , which produces large quantities of latex in its roots ( Hauss and Schütte , 1976; Huber et al . , 2015 ) . The most abundant latex compound , the sesquiterpene lactone glucoside taraxinic acid ( TA ) β-D-glucopyranosyl ester ( TA-G ) , deters M . melolontha feeding and thereby benefits plant fitness ( Huber et al . , 2016a; Huber et al . , 2016b; Huber et al . , 2015 ) . To understand the interaction between TA-G and M . melolontha , we first investigated whether TA-G is deglucosylated during insect feeding and whether plant or insect enzymes are involved . We then identified M . melolontha β-glucosidases that might hydrolyze TA-G through a comparative transcriptomic approach and narrowed down the list of candidate genes through in vitro characterization of heterologously expressed proteins . Finally , we silenced TA-G-hydrolyzing β-glucosidases in M . melolontha through RNA interference ( RNAi ) and determined the effect of these enzymes on TA-G hydrolysis , toxicity , and deterrence in vivo . Taken together , our results reveal that β-glucosidases modify the effects of plant defense metabolites on both herbivore performance and host plant choice , with potentially important consequences for the ecology and evolution of plant-herbivore interactions . To test if TA-G is hydrolyzed during M . melolontha feeding , we analyzed larvae that had ingested defined amounts of TA-G-containing T . officinale latex . The aglycone TA was not detected in the latex itself but was present in substantial amounts in the regurgitant and gut of latex-fed larvae . TA-G on the other hand disappeared as soon as the latex was ingested by the larvae ( Figure 1A ) . TA-glutathione ( TA-GSH ) and TA-cysteine ( TA-Cys ) were also identified in latex-fed larvae based on mass spectral and nuclear magnetic resonance ( NMR ) data , with the Cys sulfhydryl moiety being conjugated to TA at the exocyclic methylene group of the α-methylene-γ-lactone moiety ( Figure 1B–C , Figure 1—figure supplements 1–6 ) . Lower amounts of TA-Cys-Glu and TA-Cys-Gly were also present ( Figure 1—figure supplement 1 ) . No TA-G-GSH or TA-G-Cys conjugates were detected in this experiment . Based on current knowledge of the GSH pathway in insects ( Schramm et al . , 2012 ) , it is likely that TA is first conjugated to GSH and then cleaved sequentially to form the other metabolites , although some conjugation to GSH prior to deglucosylation may also occur ( Figure 1C ) . Quantitative measurements showed that approximately 25% of the ingested TA-G was converted to GSH conjugates and derivatives ( Figure 1D ) , with TA-Cys accounting for 95% of all identified compounds ( Figure 1E ) . TA-Cys mainly accumulated in the anterior midgut ( Figure 1D ) , and this pattern was stable over prolonged exposure of M . melolontha to TA-G ( Figure 1—figure supplement 7 ) . In contrast to the different body parts , the frass only contained a small fraction of TA conjugates and was dominated by trace quantities of intact TA-G ( Figure 1D–E ) . Thus , the deglucosylation and GSH conjugation of TA is a major route for metabolism of this sesquiterpene lactone in M . melolontha . TA-G deglucosylation may be mediated by plant or insect enzymes or a combination of both . TA-G in T . officinale latex incubated at different pH levels at room temperature was readily enzymatically deglucosylated to TA at a pH of 4 . 6 and 5 . 4 , but not at lower or higher pH values ( Figure 2A ) . As the midgut pH of M . melolontha is above 8 ( Figure 2B; Egert et al . , 2005 ) , the deglucosylation of TA-G by plant-derived enzymes is likely inhibited . To test whether TA-G is hydrolyzed by M . melolontha enzymes , various M . melolontha gut sections were dissected and extracted . Strong deglucosylation activity was detected in the proximal parts of the gut , especially in the anterior midgut ( Figure 2B ) . TA-G hydrolysis also occurred when larvae were fed with a diet containing heat-deactivated latex , which no longer hydrolyzes TA-G itself ( Figure 2A and C ) , and the presence of TA-G-hydrolyzing latex proteins in TA-G-containing diets did not result in higher amounts of TA or TA conjugates inside M . melolontha compared to diets with heat-deactivated latex proteins ( Figure 2C ) . Therefore , insect-derived enzymes are sufficient for TA-G deglucosylation in M . melolontha . As the glucose moiety of TA-G is attached through an ester rather than a glycoside linkage , carboxylesterases or glucosidases may deglucosylate TA-G . TA-G deglucosylation by M . melolontha midgut protein extracts was inhibited by the addition of the α- and β-glucosidase inhibitor castanospermine in a dose-dependent manner , but not by the α-glucosidase inhibitor acarbose or the carboxylesterase inhibitor bis ( p-nitrophenyl ) phosphate ( Figure 2—figure supplements 1–2 ) . This suggests that β-glucosidases rather than carboxylesterases catalyze TA-G deglucosylation in M . melolontha . In order to identify TA-G-hydrolyzing β-glucosidases , we separately sequenced 18 mRNA samples isolated from anterior and posterior midguts of larvae that had been feeding on diets coated with crude latex , TA-G-enriched extracts , or water . Putative M . melolontha β-glucosidases were identified based on amino acid similarity to known β-glucosidases from Tenebrio molitor and Chrysomela populi . 19 sequences similar to β-glucosidases had an expression profile matching the observed pattern of high TA-G deglucosylation activity in the anterior midgut . Partial sequences were extended using rapid-amplification of complementary DNA ( cDNA ) ends polymerase chain reaction ( RACE PCR ) , resulting in 12 full-length β-glucosidases sharing between 55 and 79% amino acid similarity ( Figure 3A , Figure 3—figure supplement 1 , Supplementary file 1 ) . The remaining seven transcripts could not be amplified or turned out to be fragments of the other candidate genes . All amplified sequences contained an N-terminal excretion signal and possessed the ITENG and NEP motifs characteristic of glucosidases ( Figure 3—figure supplement 1; Sanz-Aparicio et al . , 1998; Davies and Henrissat , 1995; Barrett et al . , 1995 ) . Expression levels of the candidate genes were 37- to 308-fold higher in the anterior than posterior midgut samples ( padj <10–5 , exact tests , n = 3 ) , thus matching the differences in TA-G deglucosylation rate between these gut compartments ( Figure 3B ) . Average expression of the transcripts did not differ among M . melolontha larvae fed water , TA-G , or latex ( Figure 3B; padj >0 . 50 , exact tests , n = 3 ) . The amplified M . melolontha β-glucosidases were heterologously expressed in an insect cell line and assayed with a variety of plant glycosides , including TA-G , benzoxazinoids , a salicinoid , and a glucosinolate as well as the disaccharide cellobiose . 9 of the 12 β-glucosidases were active with the standard fluorogenic substrate , 4-methylumbelliferyl-β-D-glucopyranoside , and hydrolyzed at least one of the plant metabolites ( Figure 3C , Figure 3—figure supplement 2 ) . For the three remaining enzymes , we did not observe hydrolysis of any substrate . Absence of any enzymatic activity could either be the result of a lack of catalytic activity toward the tested substrates or of low transgene expression and protein secretion by the cell line . All tested substrates were deglucosylated by at least one M . melolontha glucosidase ( Figure 3C ) in agreement with the hydrolysis activity of crude midgut extracts ( Figure 3—figure supplement 3 ) . Five heterologously expressed proteins deglucosylated TA-G ( Figure 3C ) , with the highest TA aglycone formation found for Mm_bGlc17 ( Figure 3—figure supplement 4 ) . Apart from TA-G , Mm_bGlc17 also deglyosylated benzoxazinoids , salicin , and cellobiose . These data suggest that Mm_bGlc17 and up to four other gut-expressed β-glucosidases may play a role in TA-G metabolism in M . melolontha . To test whether M . melolontha β-glucosidases contribute to TA-G deglucosylation , we silenced two β-glucosidases with TA-G deglucosylation activity , Mm_bGlc16 and Mm_bGlc17 , as well as one β-glucosidase without TA-G activity , Mm_bGlc18 , by injecting double-stranded RNA ( dsRNA ) targeting a 500 bp fragment of each gene into the second segment of anesthetized M . melolontha larvae ( Figure 4—figure supplement 1 ) . After 5 days , a stable and specific reduction of the target mRNAs had occurred ( Figure 4—figure supplements 2–3 ) . TA-G deglucosylation was reduced by 75% in gut extracts of larvae that were silenced in Mm_bGlc17 ( Figure 4A ) . Silencing of Mm_bGlc16 and Mm_bGlu18 did not significantly reduce TA-G deglucosylation activity compared to green fluorescent protein ( GFP ) controls ( Figure 4A ) . These results confirm that M . melolontha-derived β-glucosidases hydrolyze TA-G and demonstrate that Mm_Glc17 accounts for most of the TA-G deglucosylation in vivo . To test whether Mm_bGlc17 modulates the impact of TA-G on larval performance , Mm_bGlc17-silenced and GFP-control larvae were allowed to feed on either TA-G-producing wild-type or TA-G-deficient transgenic dandelions . The interaction of Mm_bGlc17 silencing and plant genotype significantly affected larval growth ( Figure 4B; p ( Mm_bGlc17 x TA-G ) = 0 . 009 , two-way analysis of variance ( ANOVA ) ) . On TA-G-containing plants , Mm_bGlc17 silencing reduced larval growth , with GFP-control larvae gaining 4 . 5% body weight and Mm_bGlc17-silenced larvae losing 1 . 4% body weight ( Figure 4B; p = 0 . 009 , Student’s t-test ) . By contrast , on TA-G-deficient plants , Mm_bGlc17 silencing did not affect larval weight gain ( p = 0 . 19 , Student’s t-test ) . GFP-control M . melolontha larvae had higher growth on TA-G-containing than TA-G-lacking plants ( p = 0 . 035 , Student’s t-test; Figure 4 , Figure 4—figure supplement 4 ) , while the reversed pattern was found in tendency for Mm_bGlc17-silenced larvae ( p = 0 . 099 , Student’s t-test; Figure 4—figure supplement 4 ) . The experiment was repeated twice with similar results ( Figure 4—figure supplement 5 ) . As Mm_bGlc17 benefited larval growth in the presence of TA-G , we investigated whether the expression of this gene is induced by TA-G . Mm_bGlc17 gene expression increased by 95% on TA-G-containing compared to TA-G-lacking plants ( Figure 4C; p = 0 . 04 , Kruskal-Wallis rank sum test ) . Taken together , these data show that Mm_bGlc17 expression is induced by TA-G and increases larval performance in the presence of TA-G . As TA-G in T . officinale latex was previously found to deter M . melolontha larvae ( Pollard , 1992 ) , we tested whether TA-G hydrolysis influences the deterrent properties of TA-G . Mm_bGlc17-silenced and GFP-control larvae were allowed to choose between TA-G-producing wild-type and TA-G-deficient transgenic dandelions . GFP-silenced control larvae were deterred by TA-G , with over 60% of the larvae feeding on TA-G-deficient plants and 30-% on the wild-type ( Figure 4D; p ( 3h ) = 0 . 006 , binomial test ) . By contrast , Mm_bGlc17-silenced larvae did not show any preference for TA-G-deficient over TA-G-producing wild-type plants: 44% of the larvae fed on wild-type plants , while 42% fed on TA-G-deficient plants ( Figure 4D; p ( 3h ) = 0 . 86 , binomial test ) . Both patterns were constant over time ( Figure 4—figure supplement 6 ) . Mm_bGlc17 silencing did not significantly affect the total percentage of larvae that made a choice ( 86% Mm_bGlc17 vs 91% GFP ) . These results demonstrate that Mm_bGlc17 expression is required for the deterrent effect of TA-G toward M . melolontha . Herbivore enzymes are well known to modify plant defense metabolites , but only few studies provided clear evidence that these modifications feed back on herbivore performance and fitness . Furthermore , the effects of plant defense metabolizations on herbivore host plant choice are not understood . Here , we show that a herbivore β-glucosidase deglucosylates a plant secondary metabolite , which modifies both its toxic and deterrent properties and thereby determines the interaction between a plant and its major root-feeding natural enemy . Metabolization of plant defense metabolites is considered central for the ability of species to overcome chemical defenses of their host plants ( Heckel , 2014 ) , and recent papers have established direct molecular evidence for this concept ( Sun et al . , 2019; Sun et al . , 2020; Poreddy et al . , 2015 ) . A major metabolization product of TA-G is TA-Cys , with about 25% of the ingested TA-G accumulating in this form . Based on our current knowledge of the GSH pathway in insects ( Schramm et al . , 2012 ) , it is likely that TA-G is deglucosylated prior conjugation to GSH and subsequently sequentially cleaved to TA-Cys by peptidases . The first step of this metabolization pathway , the conjugation of GSH to TA , may occur spontaneously and/or via GSH glucosyl-transferases . When we incubated TA and TA-G with high concentrations of GSH and Cys in vitro , several isomers of the conjugates formed that were not detected inside the M . melolontha gut , suggesting that enzymatic rather than spontaneous conjugation of GSH to TA prevails inside the larva . Interestingly , TA-Cys mostly accumulated in the anterior midgut , and only neglectable amounts of TA-Cys were excreted by the larvae , a pattern that was stable over long-term feeding of M . melolontha on T . officinale . Consequently , larvae must further metabolize TA-Cys to yet unknown products , and either store or excrete these compounds . Future experiments with radioisotope-labeled TA-G may shed light into the ultimate fate of TA inside the M . melolontha larva , and may help to assess whether M . melolontha sequesters TA and uses the compound for its own prupose . As the transformation of defense metabolites by insect enyzmes occurs in the gut , metabolization products are considered unlikely to be tasted via frontal sensory structures of insect herbivores . It is thus commonly assumed that there is no direct impact of this process on herbivore behavior ( Pentzold et al . , 2014; Simon et al . , 2015 ) . By contrast , transformation of defense metabolites by plant enzymes that are activated by tissue disruption is well accepted to have a strong behavioral impact on insect herbivores , which is in line with the rapid and early formation of plant defense catabolites ( Glauser et al . , 2011; Krothapalli et al . , 2013; de Vos et al . , 2008; Zhang et al . , 2006; Mumm et al . , 2008 ) . Here , we have found that the insect β-glucosidase Mm_bGlc17 , which deglucosylates a defensive sesquiterpene lactone ( TA-G ) in the insect gut , is also required to elicit the deterrent effect of this metabolite . Our early work on TA-G showed that , in a community context , the capacity of dandelions to produce the glucosylated sesquiterpene lactone reduces M . melolontha attack and its negative effect on plant growth and fitness ( Huber et al . , 2016b ) , resulting in the selection of high TA-G genotypes under high M . melolontha pressure ( Huber et al . , 2016a ) . As these effects are likely the result of the deterrent , rather than the toxic properties of TA-G , they are likely also directly dependent on the presence of Mm_bGlc17 . Thus , the metabolism of M . melolontha may not only drive the feeding preferences of the herbivore , but also the ecology and evolution of dandelions in their natural habitat . Insect-detoxifying enzymes may thus shape plant defense evolution not only by reducing the toxicity of defense compounds but also by modulating herbivore host plant choice . Many plant defensive metabolites are glycosides , which are typically non-toxic themselves but are deglucosylated upon herbivore damage , forming toxic products . Both plant- and herbivore-derived β-glucosidases can mediate deglycosylation in the insect gut , but their relative contribution is often unclear ( Pentzold et al . , 2014; Desroches et al . , 1997; Lindroth , 1988; Pankoke et al . , 2012 ) . Here we provide several parallel lines of evidence to demonstrate that the deglucosylation of TA-G , a glucosylated secondary metabolite in the latex of T . officinale , depends primarily on β-glucosidases from M . melolontha rather than on plant enzymes . First , T . officinale TA-G hydrolase activity has an acidic pH optimum ( 4 . 8–5 . 4 ) , and the activity is very low at the alkaline pH ( 8 . 0 ) found in the gut of M . melolontha . Second , TA-G is deglucosylated by M . melolontha gut extracts in the absence of plant material . Third , the presence of TA-G-hydrolyzing latex proteins in TA-G-containing diet does not result in higher amounts of TA or TA conjugates inside M . melolontha compared to the diet with heat-deactivated latex proteins . Fourth , M . melolontha expresses several β-glucosidases with TA-G-hydrolyzing activity as demonstrated in in vitro assays . Fifth , silencing the M . melolontha TA-G β-glucosidase Mm_bGlc17 reduces TA-G deglucosylation activity in larval gut extracts and abolishes the avoidance behavior of M . melolontha toward TA-G-containing plants . Together , these results demonstrate that insect rather than plant β-glucosidases hydrolyze ingested TA-G in M . melolontha . A large number of plant glycosides are protoxins that are activated by deglycosylation including glucosinolates , benzoxazinoids , salicinoids , alkaloid glycosides , cyanogenic glycosides , and iridoid glycosides ( Mithöfer and Boland , 2012; Pentzold et al . , 2014; Wittstock and Gershenzon , 2002 ) . But , until now nothing was known about whether sesquiterpene lactone glycosides are also protoxins . Sesquiterpene lactone aglycones are much more potent than their corresponding glycosides in pharmacological studies of cytotoxicity and anti-cancer activity ( Choi et al . , 2002; Seto et al . , 1988 ) . However , the consequences of sesquiterpene lactone deglycosylation for herbivore behavior and performance have not been previously investigated ( Huber et al . , 2015; Sessa et al . , 2000; Graziani et al . , 2015 ) . Our experiments show that deglucosylation of TA-G is associated with an increase rather than a decrease in larval growth on TA-G-producing plants . This suggests that the cleavage of TA-G to TA reduces rather than enhances the toxicity of this sesquiterpene lactone . Several explanations for this phenomenon are possible . First , GSH may be more rapidly conjugated by TA than TA-G , and thus deglucosylation is a step toward detoxification . Second , if the target site of TA-G lies in a hydrophilic compartment ( such as the gut lumen ) , deglucosylation may block its activity . Third , the glucose liberated by TA-G deglucosylation may enhance the nutritional quality of dandelion roots for the larvae . When we compared dandelion roots exposed to different native grassland species in previous studies , we found both positive and negative correlations between root glucose levels and larval growth ( Huang et al . , 2019; Huang et al . , 2018 ) , suggesting a high degree of context dependency . In summary , our results provide evidence that deglycosylation of plant defenses may reduce negative impacts on herbivores . Deglycosylation of a diterpene glycoside of N . attenuata was also found to reduce its toxicity , but in this case , the product still contained two other glycoside moieties and thus differs little from its substrate in terms of polarity compared to the differences between TA and TA-G ( Poreddy et al . , 2015 ) . While Mm_bGlc17 improves larval performance on TA-G-producing plants , the enzyme is also required for M . melolontha larvae to avoid TA-G . We propose two mechanisms that may be responsible for these counterintuitive results . First , the recognition of TA-G through deglucosylation may guide the M . melolontha larva to feeding sites that are most suitable for fast larval growth , independently of the toxicity of TA-G . Exploitation of plant secondary metabolites and sugars to locate nutritious tissue has been reported , for instance , for the specialist root herbivore Diabrotica virgifera virgifera feeding on maize roots ( Robert et al . , 2012; Hu et al . , 2018; Machado et al . , 2021 ) . Melolontha melolontha larvae preferentially feed on side roots of dandelions , which contain lower TA-G and higher soluble protein levels than main roots and also may be more nutritious as they are actively growing ( Huber et al . , 2016b ) . Thus , the larvae may not be avoiding TA-G because of its toxicity , but because avoiding high TA-G levels guides them to nutritious roots , with the avoidance behavior being facilitated by Mm_bGlc17 . A second explanation for the observed patterns may be that herbivore growth by itself gives an incomplete picture regarding the costs of TA-G consumption and metabolism . It has been shown , for instance , that plant secondary metabolites can enhance larval weight gain , but at the same time increase larval mortality , suggesting that growth is not always beneficial ( Veyrat et al . , 2016; Erb , 2018 ) . Furthermore , TA may change the susceptibility of the larvae to parasites and pathogens , as has been shown for the plant volatile indole in maize ( Ye et al . , 2018 ) . In addition , the hydrolysis of TA-G may deplete the level of cysteine inside the larva through conjugation of TA to glutathione , as has been observed in lepidopteran larvae that conjugate food-derived isothiocyanates to GSH ( Jeschke et al . , 2016 ) . The negative consequences of Cys depletion on larval performance may only be observed under stressful conditions , for instance , under nutrient limitations or in the presence of other toxic allelochemicals that require detoxification through GSH conjugation . Thus , it is possible that under natural conditions , Mm_bGlc17-dependent cleavage of TA-G reduces rather than enhances M . melolontha fitness , which may explain Mm_bGlc17-mediated TA-G avoidance . The gene may , nevertheless , be maintained in the insect genome if Mm_bGlc17 is important for the larva to acquire nutrients . In such a scenario , dandelions would exploit a promiscuous β-glucosidase ( Mm_bGlc17 ) whose evolution is constrained by its primary functions in nutrient acquisition . If the evolution of Mm_bGlc17 is constrained by its primary function , such constraints may be alleviated by down-regulating the expression of the gene under harmful conditions , as has been observed for other insect β-glucosidases in the presence of glycosydic protoxins ( Pentzold et al . , 2014 ) . In our study , while the initial RNA-sequencing ( RNAseq ) analysis did not detect differences in Mm_bGlc17 expression in the presence and absence of TA-G , our follow-up quantitative PCR ( qPCR ) analysis on M . melolontha larvae feeding on transgenic plants showed that Mm_bGlc17 is up-regulated in the presence of TA-G . Further experiments are required to determine whether this discrepancy is due to different methods and sample sizes or due to differences in TA-G concentration , the presence of a plant matrix , or the genotypic background of the M . melolontha larvae . Despite these uncertainties , the observed up-regulation of Mm_bGlc17 expression in the presence of TA-G is compatible with a beneficial role of this enzyme for M . melolontha feeding on TA-G-producing dandelions . A more detailed understanding of the role of Mm_bGlc17 and TA-G under natural conditions and over the full 3-year life cycle of M . melolontha would help to shed light on whether the expression of Mm_bGlc17 is indeed beneficial for the larvae . While Mm-bGlc17 is required for the feeding deterrence of TA-G , the underlying physiological mechanisms that lead to TA-G avoidance are unclear . On the one hand , insect feeding preference may be triggered by the presence of the aglycone TA or TA conjugates inside the gut . These metabolites may bind to specific receptors that face the gut lumen or are located inside the gut membrane , and thereby alter herbivore host plant choice . Additionally , reduction of GSH or Cys levels through conjugation to TA may be perceived by the larvae and thereby alter herbivore feeding preference , although this scenario is less likely as GSH or Cys depletion likely requires longer time than the immediately observed feeding responses of naive M . melolontha larvae . On the other hand , TA or its conjugates may be present at low concentrations inside the M . melolontha mouth , for instance , via regurgitation or through direct formation due to the potential presence of Mm_bGlc17 in the saliva , and may thus be detected by gustatory receptors that modulate herbivore host plant choice . We found that forced regurgitants of M . melolontha possess TA-G-hydrolyzing activity; however , it is unclear whether forced regurgitants are informative to infer normal feeding processes of the larvae . While Mm_bGlc17 is highly abundant in the gut , it is unclear whether Mm_bGlc17 is also present in the saliva . Analyzing the expression profile of Mm_bGlc17 across different organs of the M . melolontha larva , as well as performing neurosensory experiments with orally administered TA , would help to shed light onto the exact mechanisms underlying TA feeding deterrence . Interestingly , besides TA-G , Mm_bGlc17 deglycosylates other substrates , including cellobiose and salicinoid and benzoxazinoid defense compounds . The ability of this enzyme to hydrolyze benzoxazinoids seems counterintuitive from the insect’s perspective since benzoxazinoid hydrolysis increases both feeding deterrence as well as toxicity ( Glauser et al . , 2011; de Vos et al . , 2008; Zhang et al . , 2006; Mumm et al . , 2008; Wittstock et al . , 2003 ) , raising the possibility that some plants can co-opt insect enzymes to activate their own defenses . On the other hand , insects are known to have evolved some resistance to plant glycosidic protoxins by inhibiting the activating glycosidases of plants and down-regulating their own activating glycosidases ( Pentzold et al . , 2014; Desroches et al . , 1997; Lindroth , 1988 ) . The fact that Mm_bGlc17 catalyzes the hydrolysis of a range of glucosides plus the glucose ester TA-G is also unusual . There are only a few previous reports of enzymes with this versatility ( Nakano et al . , 1998; Okamoto et al . , 2000 ) . The ability of Mm_bGlc17 to mediate hydrolysis of cellobiose , a disaccharide derived from cellulose , suggests its evolutionary origin as a digestive enzyme that was later recruited for processing plant defenses . The relatively large number of β-glucosidases in many insect herbivores ( Pentzold et al . , 2014; Poreddy et al . , 2015; Beran et al . , 2014 ) and their species-specific phylogenetic clustering ( Beran et al . , 2014 ) indicate that in addition to contributing to the digestion of cell wall carbohydrates—which are mostly shared among plant species—many β-glucosidases also act on a variety of specialized metabolites , such as plant defense compounds . Thus , plant defenses may play an underestimated role in the evolution of β-glucosidases in insect herbivores . Other herbivore digestive enzymes may also interact with plant defenses , leading to changes in herbivore performance and behavior , which likely modulate the ecology and evolution of plants and their consumers . T . officinale plants used for extraction of latex and TA-G were grown in 0 . 7–1 . 2 mm sand and watered with 0 . 01–0 . 05% fertilizer with N-P-K of 15-10-15 ( Ferty 3 , Raselina , Czech Republic ) in a climate chamber operating under the following conditions: 16 hr light/8 hr dark; light supplied by a sodium lamp ( EYE Sunlux Ace NH360FLX , Uxbridge , UK ) ; light intensity at plant height: 58 µmol m2 s–1; temperature: day 22°C; night 20°C; humidity: day 55% , night 65% . Depending on the availability , 3- to 5-month-old wild-type plants of the European A34 , 6 . 56 , or 8 . 13 accession were used unless otherwise indicated ( Verhoeven et al . , 2010 ) . Plants used for the choice experiments were germinated on seedling substrate and transplanted into individual pots filled with potting soil ( five parts landerde , four parts peat , and one part sand ) after 2–3 weeks and grown in a climate chamber operating under the following conditions: 16 hr light/8 hr dark , light supplied by arrays of Radium Bonalux Super NL 39 W/840 white lamps; light intensity at plant height: 250 µmol m2 s-1; temperature: day 22°C; night 18°C; humidity 65% . Plants used for the performance experiments were germinated on seedling substrate , transplanted to individual pots filled with a homogenized mixture of 2/3 seedling substrate ( Klasmann-Deilmann , Switzerland ) and 1/3 landerde ( Ricoter , Switzerland ) and cultivated in a greenhouse operating under the following conditions: 50–70% relative humidity , 16/8 hr light/dark cycle , and 24°C at day and 18°C at night , without extrernal light source . The TA-G-deficient line RNAi-1 and the control line RNAi-15 were used for these experiments ( Huber et al . , 2016b ) . M . melolontha larvae were collected from meadows in Switzerland and Germany . Larvae were reared individually in 200 ml plastic beakers filled with a mix of potting soil and grated carrots in a climate chamber operating under the following conditions: 12 hr day , 12 hr night; temperature: day 13°C , night 11°C; humidity: 70%; lighting: none , except for the RNAi experiment , for which the day and night temperature was 4°C during rearing . All experiments were performed in the dark with larvae in the third larval instar . For heterologous experession , T . ni-derived cells ( High Five cells ) were purchased from Life Technologies ( Carlsbad , CA , USA ) and immediately used for the experiment . The cell lines were tested negatively for mycoplasma infection prior delivery . All statistical analyses were performed in R version 3 . 1 . 1 ( R Development Core Team , 2014 ) . Pairwise comparisons were performed with the Agricolae package ( de Mendiburu , 2014 ) . Results were displayed with gplots , ggplot2 , and RColorBrewer ( Wickham , 2009; Warnes et al . , 2016; Neuwirth , 2014 ) . Differential gene expression was analyzed using DeSeq2 and edgeR ( Robinson et al . , 2010; Love et al . , 2014 ) . Details on the statistical procedure are given in the individual sections . Sample sizes were estimated based on previous experience with the study system . M . melolontha larvae were allocated to treatment groups using restricted randomization to achieve equal sample sizes among groups . In order to test whether TA-G is deglucosylated during digestion in M . melolontha , we screened for TA-G , TA , and other TA-G metabolites in larvae that fed on diets supplemented with either latex or water . 10 M . melolontha larvae were starved for 10 days at room temperature before offering them approximately 0 . 35 cm3 boiled carrot slices that were coated with either main root latex or water . Larvae were allowed to feed for 4 hr inside 180 ml plastic beakers covered with a moist tissue paper , after which the frass and regurgitant were collected in 1 ml methanol . Regurgitant was collected by gentle prodding of the larvae . Left-over food was frozen at –80°C until extraction . The larvae were cooled for 10 min at –20°C and subsequently dissected on ice to remove the anterior midgut , posterior midgut , hindgut , and hemolymph , which were collected in 1 ml methanol . All larval samples were homogenized by vigorously shaking with two to three metal beads for 4 min in a paint shaker ( Fluid Management , Wheeling , IL , USA ) , centrifuged at 4°C for 10 min at 17 , 000 ×g , and the supernatant stored at –20°C until analysis . Left-over food was ground in liquid nitrogen to a fine powder of which 100 mg was extracted with 1 ml methanol by vortexing for 30 s . The samples were subsequently centrifuged at room temperature for 10 min at 17 , 000 ×g and the supernatant was stored at –20°C until analysis . Methanol samples were analyzed on a high-pressure liquid chromatograph ( HPLC 1100 series equipment; Agilent Technologies , Santa Clara , CA , USA ) , coupled to a photodiode array detector ( G1315A DAD; Agilent Technologies ) and an Esquire 6 , 000 ESI-Ion Trap mass spectrometer ( Bruker Daltonics , Bremen , Germany ) . Metabolite separation was accomplished with a Nucleodur Sphinx RP column ( 250 × 4 . 6 mm , 5 µm particle size; Macherey–Nagel , Düren , Germany ) . The mobile phase consisted of 0 . 2% formic acid ( A ) and acetonitrile ( B ) utilizing a flow of 1 ml min–1 with the following gradient: 0 min , 10% B , 15 min: 55% B , 15 . 1 min: 100% B , 16 min: 100% B , followed by column reconditioning ( Huber et al . , 2015 ) . To search for unknown metabolites of TA-G , we visually compared the chromatograms of the anterior midgut of latex- and control-fed larvae and subsequently performed tandem mass spectrometry ( MS2 ) experiments using AutoMS/MS runs on the Esquire 6 , 000 ESI-Ion Trap MS to obtain structure information . Using QuantAnalysis ( Bruker Daltonics ) , TA-G , TA , and the putative TA-GSH conjugates were quantified based on their most abundant ion trace: TA-G: 685 [M+[M-162]] , negative mode , retention time ( RT ) = 12 . 2 min; TA: 263 [M + H] , positive mode , RT = 16 . 8 min; TA-GSH: 570 [M + H] , positive mode , RT = 10 . 1 min; TA-Cys-glycine: 441 [M + H] , positive mode , RT = 9 . 4 min; TA-Cys-glutamate: 513 [M + H] , positive mode , RT = 10 . 4 min , TA-Cys: 384 [M + H] , positive mode , RT = 9 . 8 min . In order to identify the structures of the putative TA conjugates , we allowed 15 M . melolontha larvae to feed for 1 month on T . officinale plants . Larvae were then recovered and dipped for 2 s in liquid nitrogen before dissecting them on ice . The entire midgut was homogenized in 1 ml methanol by shaking the samples for 3 min with three metal beads in a paint shaker . The samples were centrifuged at room temperature for 10 min at 17 , 000 ×g , passed through a 0 . 45 µm cellulose filter , and subsequently purified by high-pressure liquid chromatography ( HPLC ) . NMR analyses were conducted using a 500 MHz Bruker Avance HD spectrometer equipped with a 5 mm TCI cryoprobe . Capillary tubes ( 2 mm ) were used for structure elucidation in MeOH-d4 . The analysis revealed the presence of TA-Cys by comparison with a synthesized standard ( see below , Figure 1—figure supplement 2 and Figure 1—figure supplement 6 ) . Other TA conjugates identified by high-pressure liquid chromatography-mass spectrometry ( HPLC-MS ) were below the detection threshold of NMR . In order to characterize and quantify the TA-G metabolites , we isolated and synthesized TA-G , TA-G-GSH , TA-G-Cys , TA , TA-GSH , and TA-Cys . TA-G was purified from T . officinale latex methanol extracts as described in Huber et al . , 2016b . TA was obtained by incubating 50 mg purified TA-G with 25 mg β-glucosidase from almonds ( Sigma Aldrich ) in 2 . 5 ml H2O at 25°C for 2 days . The sample was centrifuged at room temperature for 5 min at 17 , 000 ×g and supernatant was discarded . The TA-containing pellet was dissolved in 100 µl dimethylsulfoxide ( DMSO ) and diluted in 1 . 9 ml 0 . 01 M TAPS ( [tris ( hydroxymethyl ) methylamino]propanesulfonic acid ) buffer ( pH = 8 . 0 ) . Subsequently , solid-phase extraction was performed with a 500 mg HR-X Chromabond cartridge ( Macherey-Nagel ) . The cartridge was washed and conditioned with two volumes of methanol and H2O , respectively . Separation was accomplished using one volume each of H2O , 30% methanol , and 60% methanol , and two volumes of 100% methanol . TA was eluted in the first 100% methanol fraction , in which no impurities were detected on an Esquire 6 , 000 ESI-Ion Trap-MS . Samples were evaporated under N2 flow at room temperature to almost complete dryness , and 1 ml H2O was added before freeze-drying . To obtain TA-GSH and TA-Cys conjugates , the most abundant TA conjugates in the liquid chromatography-mass spectrometry ( LC-MS ) chromatograms , we dissolved 5 mg isolated TA in 5 µl DMSO in two separate Eppendorf tubes and added 1 . 6 ml 0 . 01 M TAPS buffer ( pH 8 . 0 ) and a 75-fold molar excess of either GSH or Cys to the tubes . Similarly , to obtain TA-G-GSH and TA-G-Cys conjugates , we dissolved 5 mg TA-G in 1 ml 0 . 01 M TAPS ( pH = 8 . 0 ) in two separate Eppendorf tubes and added a 75 molar excess of either GSH or Cys . TA-GSH , TA-G-GSH , and TA-G-Cys samples were incubated for 2 days and TA-Cys for 7 days in the dark at 25°C , after which most of the TA and TA-G had spontaneously conjugated . All samples were stored at –20°C until purification by semi-preparative HPLC . Semi-preparative HPLC was accomplished using an HPLC coupled with ultraviolet ( HPLC-UV ) system coupled to a fraction collector ( Advantec SF-2120 ) using a Nucleodur Sphinx RP column ( 250 × 4 . 6 mm , 5 µm particle size; Macherey-Nagel ) . The mobile phase consisted of 0 . 01% formic acid ( A ) and acetonitrile ( B ) . Flow rate was set to 1 ml min–1 with the following gradient: 0 min: 15% B , 5 min: 30% B , 9 min: 54% B , 9 . 01 min: 100% B , followed by column reconditioning . Compounds were monitored with a UV detector at 245 nm . As the synthesis resulted in the formation of several isomers that differed in retention times , the conjugates with the same retention times as found in M . melolontha larvae were collected . The elution times of the compounds were TA-G-GSH: 6 . 9 min; TA-G-Cys: 6 . 4 min; TA-GSH: 8 . 6 min; TA-Cys: 8 . 3 min . The fractions were concentrated under nitrogen flow at 30°C and subsequently lyophilized . The final yields of the conjugates were TA-G-GSH: 2 . 1 mg; TA-G-Cys: 0 . 38 mg; TA-GSH: 1 . 47 mg; TA-Cys: 0 . 23 mg . Purified fractions were analyzed by NMR spectroscopy for structure verification . Structures with chemical shifts are depicted in Figure 1 - figure supplements 3-6 . Standard curves of the conjugates were prepared using 100 µg of the respective compounds in 100% methanol on an Agilent 1200 HPLC system ( Agilent Technologies , ) coupled to an API 3200 tandem mass spectrometer ( Applied Biosystems , Darmstadt , Germany ) equipped with a turbospray ion source operating in negative ionization mode . Injection volume was 5 μl . Metabolite separation was accomplished on a ZORBAX Eclipse XDB-C18 column ( 50 × 4 . 6 mm , 1 . 8 μm; Agilent Technologies ) . The mobile phase consisted of 0 . 05% formic acid ( A ) and acetonitrile ( B ) using a flow rate of 1 . 1 ml min–1 with the following gradient: 0 min: 5% B , 0 . 5 min: 5% B , 4 min: 55% B , 4 . 1 min: 90% B , 5 min: 90% B , followed by column reconditioning . The column temperature was kept at 20°C . The ion spray voltage was maintained at –4 . 5 keV . The turbo gas temperature was set at 600°C . Nebulizing gas was set at 50 psi , curtain gas at 20 psi , heating gas at 60 psi , and collision gas at 5 psi . Multiple reaction monitoring ( MRM ) in negative mode monitored analyte parent ion → product ion: m/z 423 → 261 ( collision energy ( CE ) –14 V; declustering potential ( DP ) –40 V ) for TA-G; m/z 730 → 143 , ( CE –66 V; DP –80 V ) for TA-G-GSH; m/z 544 → 382 ( CE –26 V; DP –80 V ) for TA-G-Cys; m/z 261 → 217 ( CE –14 V; DP –30 V ) for TA; m/z 568 → 143 ( CE –44 V; DP –50 V ) for TA-GSH; m/z 382 → 120 ( CE –30 V; DP –45 V ) for TA-Cys; m/z 568 → 143 ( CE –44 V; DP –50 V ) for loganic acid . Both Q1 and Q3 quadrupoles were maintained at unit resolution . Analyst 1 . 5 software ( Applied Biosystems ) was used for data acquisition and processing . Weight-based response factors of TA-G , TA , and their conjugates were calculated relative to loganic acid ( Extasynthese , Genay , France ) . The weight-based response factors were as follows: TA-G: 2 . 8; TA-G-GSH: 2 . 5 , TA-G-Cys: 1 . 9; TA: 0 . 3; TA-GSH: 1 . 9; TA-Cys: 1 . 1 . In order to quantify the deglucosylation of TA-G and conjugation to GSH , we performed a Waldbauer assay in which we analyzed the TA-G metabolites in M . melolontha larvae after consumption of a fixed amount of TA-G . Eight larvae were starved for 7 days before offering them 100 mg of an artificial diet ( Huber et al . , 2016b ) supplemented with 100 µg purified TA-G , obtained as described above . Larvae were allowed to feed in the dark for 24 hr in a 180 ml plastic beaker covered with a moist tissue paper , after which the larvae had completely consumed the food . Frass was collected in 500 µl methanol containing 1 µgml–1 loganic acid as an internal standard . Subsequently , larvae were dipped for 2 s in liquid nitrogen and the anterior midgut , posterior midgut , hindgut content and tissue , and hemolymph and fat tissue removed by dissection . For the gut samples , gut content was collected separately from the gut tissue . All samples were homogenized in 500 µl methanol containing 1 µgml–1 loganic acid by vigorously shaking the tubes for 2 min with two to three metal beads in a paint shaker . All samples were centrifuged at room temperature for 10 min at 17 , 000 ×g . Supernatants were analyzed by LC-MS on the API 3200 triple quadrupole mass spectrometer as described above using a 5 µl injection volume . Metabolites were quantified based on loganic acid as an internal standard using the Analyst 1 . 5 software . To assess the distribution of the major TA-G metabolism product , TA-Cys , in M . melolontha exposed for a prolonged time to TA-G , we dissected larvae that were feeding for 1 month on T . officinale plants into anterior and posterior midgut , hindgut , fat tissue , skin , and hemolymph as described above . 10 µl hemolymph was collected inside 100 µl methanol . All other tissue samples were homogenized with 10 µl methanol per mg material by vigorously shaking the tubes for 2 min with two to three metal beads in a paint shaker . All samples were centrifuged at room temperature for 15 min at 17 , 000 ×g . Supernatants were analyzed on the HPLC 1100 series equipment coupled to an Esquire 6 , 000 ESI-Ion Trap mass spectrometer , and the abundance of TA-Cys quantified as described above . In order to test whether TA-G is hydrolyzed by plant enzymes , we analyzed the hydrolysis of TA-G in latex that was extracted in buffers that covered the pH range present in the plant vacuole ( pH 5 ) , plant cytosol ( pH 7 ) , and M . melolontha gut ( pH 8 ) ( Egert et al . , 2005 ) . We cut the main roots of T . officinale plants 0 . 5 cm below the stem-root junction and collected the exuding latex of an entire plant in 1 ml 0 . 05 M MES ( 2- ( N-morpholino ) ethanesulfonic acid ) buffer ( pH 5 . 2 ) , 0 . 05 M TRIS-HCl buffer ( pH 7 . 0 ) , or 0 . 05 M TRIS-HCl ( pH 8 . 0 ) , with three replicates for each buffer . Samples were kept at room temperature for 5 min before stopping the reaction by boiling the samples for 10 min at 98°C , during which TA-G was found to be stable . Samples were centrifuged at room temperature for 10 min at 17 , 000 ×g , and the supernatant was analyzed by an HPLC 1100 series instrument ( Agilent Technologies ) , coupled to a photodiode array detector ( G1315A DAD; Agilent Technologies ) . Metabolite separation was accomplished as described in Huber et al . , 2015 . Peak areas for TA-G and its aglycone TA were integrated at 245 nm . As the absorption spectra of TA-G and TA do not differ , we expressed the deglucosylation activity as the ratio of the peak area of TA/ ( TA + TA G ) . pH-dependent difference in the deglucosylation activity was analyzed using the Kruskal-Wallis rank sum test . To investigate the precise pH optimum of the plant hydrolases , and to test for spontaneous hydrolysis of TA-G at acidic pH , we extracted T . officinale latex in buffers with a pH range of 3–6 . Main root latex was collected as described above , extracted in 2 ml H2O containing 20% glycerol , and 200 µl extract was immediately suspended in equal volumes of a series of 0 . 1 M citrate buffers adjusted to pH 3 . 0 , 3 . 6 , 4 . 2 , 4 . 8 , 5 . 4 , and 6 . 0 . Half of the latex-buffer solution was immediately incubated for 10 min at 95°C to block enzymatic reaction . The remaining samples were kept at room temperature for 15 min to allow enzymatic reaction and subsequently heated for 10 min at 95°C . Samples were centrifuged at room temperature at 17 , 000 ×g and the supernatant was analyzed on HPLC-UV as described above . The peak area of TA-G and TA was integrated at 245 nm , and the deglucosylation activity was expressed as TA/ ( TA + TA-G ) . In order to test for the presence of TA-G-deglucosylating enzymes in M . melolontha , we analyzed the formation of TA in crude extracts of the anterior midgut , posterior midgut , and hindgut . Six M . melolontha larvae were starved for 1 week , after which they were cooled for 10 min at –20°C before dissection . Larvae were dissected into the anterior and posterior midgut and hindgut , with the gut content separated from the gut tissue . Gut samples were weighed and homogenized in 0 . 01 M TAPS buffer ( pH 8 . 0 ) containing 10% glycerol with 10 μl per mg tissue using a plastic pestle . For the deglucosylation assay , 30 µl gut samples that had either been kept on ice or boiled for 10 min at 95°C were incubated with 30 µl latex extract ( prepared as described below ) for 20 min at 25°C , after which the reaction was stopped by heating the samples for 10 min at 95°C . Samples were centrifuged at 17 , 000 ×g at room temperature for 10 min , after which the supernatant was diluted 1:1 in 0 . 01 M TAPS buffer ( pH 8 . 0 ) and stored at –20°C until chemical analysis . Latex extract was obtained by extracting the entire main root latex of six T . officinale plants in 6 ml 0 . 01 M TAPS buffer ( pH = 8 . 0 ) , after which the samples were immediately heated for 10 min at 95°C . The latex samples were centrifuged for 20 min at 17 , 000 ×g and filtered through a 0 . 45 µm cellulose filter . HPLC-UV analysis and quantification of TA-G and TA were carried out as described above . Deglucosylation activity was expressed as the ratio of TA/ ( TA + TA-G ) . Differences between the deglucosylation activity of the gut extract and heat treatment were analyzed with a two-way ANOVA . To test whether M . melolontha enzymes are sufficient to deglucosylate TA-G , we fed larvae with a TA-G-supplemented diet that contained T . officinale latex extracts that had been left intact or heat deactivated . Eight larvae were starved for 2 weeks before offering them approximately 0 . 35 cm3 boiled carrot slices coated with 50 µl of intact or heat-deactivated latex extract . Latex extracts were obtained by cutting the main roots of T . officinale plants 0 . 5 cm below the tiller and collecting the latex of an entire plant in 100 µl of either ice-cooled ( for intact extracts ) or 95°C ( for heat-deactivated extracts ) H2O . M . melolontha larvae were allowed to feed in the dark inside 180 ml beakers covered with soil for 4 hr . Subsequently , regurgitant was collected in 1 ml methanol by gently prodding the larvae . Left-over food was frozen in liquid nitrogen , ground to a fine powder , and 50 mg ground tissue was extracted with 500 µl methanol by vortexing the samples for 30 s . All samples were centrifuged at room temperature for 10 min at 17 , 000 ×g and the supernatant analyzed by LC-MS on an Esquire 6 , 000 ESI-Ion Trap-MS ( Bruker Daltonics ) as described above . TA-G , TA , TA-GSH , and TA-Cys were integrated as described above using QuantAnalysis . Statistical differences in the metabolite abundance between the sample type ( food , regurgitant ) and the presence of active plant enzymes were analyzed with two-way ANOVAs for each metabolite separately . To test whether glucosidases or carboxylesterases mediate the deglucosylation of TA-G , we measured this activity in M . melolontha gut extracts in the presence of either carboxylesterase or glucosidase inhibitors . Bis ( p-nitrophenyl ) phosphate was used as a carboxylesterase inhibitor , whereas castanospermine was deployed as a glucosidase inhibitor that reduces the activity of both α- and β-glucosidases . Six larvae were starved for 12 days before dissection . The anterior midgut content was extracted in 0 . 01 M TAPS buffer ( pH 8 . 0 ) containing 10% glycerol using 10 µl per mg gut material . To obtain TA-G as a substrate for the deglucosylation assay , the entire main root latex of each of the 15 T . officinale plants was collected in 150 µl 0 . 1 M TAPS ( pH 8 . 0 ) and samples were immediately heated for 10 min at 95°C . The samples were centrifuged at room temperature for 10 min at 17 , 000 ×g , and the supernatants were pooled and diluted 1:10 in H2O . The enzymatic assay was performed by incubating 10 µl of the diluted latex TAPS extract with 20 µl gut extract and 30 µl 0 , 0 . 002 , or 0 . 2 mM carboxylesterase or glucosidase inhibitor for 1 hr at room temperature . As a negative control , half volumes of the 0 mM inhibitor samples were immediately incubated at 95°C to stop the enzymatic reaction . Samples were centrifuged at room temperature for 10 min at 17 , 000 ×g and the supernatant was analyzed on an HPLC-UV as described above . TA-G and TA were quantified by integrating the peak area at 245 nm . Deglucosylation activity was expressed as the ratio of TA/ ( TA + TA-G ) . To investigate whether α- or β-glucosidases mediate the hydrolysis of TA-G , we measured deglucosylation activity in M . melolontha midgut extracts in the presence of acarbose , a specific α-glucosidase inhibitor , or castanospermine , which inhibits both α- and β-glucosidases . Three L3 M . melolontha larvae were starved for 5 days , dipped for 2 s in liquid nitrogen , dissected , and the anterior midgut content and tissue extracted in 10 µl 0 . 15 M NaCl per mg material . Samples were homogenized with a plastic pestle and centrifuged at 4°C for 10 min at 17 , 000 ×g . Then , 20 µl of the supernatant was incubated with 20 µl boiled latex TAPS extract ( obtained as described above ) and 0 . 002 , 0 . 2 , or 20 mM acarbose or castanospermine ( added in 40 µl ) for 1 hr at room temperature . The reaction was stopped by heating for 10 min at 95°C . Samples were centrifuged at room temperature for 10 min at 17 , 000 ×g and the supernatant was analyzed on an HPLC-UV as described above . The peak areas of TA-G and TA were integrated at 245 nm . Deglucosylation activity was expressed as the ratio of TA/ ( TA + TA-G ) . In order to identify the putative M . melolontha β-glucosidases , we sequenced 18 anterior and posterior midgut transcriptomes ( three treatments , two gut tissues , three replicates of each ) from larvae feeding on control , TA-G-enriched , or latex-containing diets using Illumina HiSeq 2 , 500 . 15 M . melolontha larvae were starved for 10 days . For 3 consecutive days , larvae were offered 0 . 35 cm3 boiled carrot slices that were coated with either ( i ) 50 µl water ( ‘control’ ) , ( ii ) 50 µl latex water extract that contained heat-deactivated latex of the main root of one T . officinale plant ( ‘TA-G enriched’ ) , or ( iii ) the entire main root latex from one T . officinale plant ( ‘latex enriched’ ) . The latex water extract was obtained by collecting the main root latex of 15 T . officinale plants in a total of 1 . 5 ml 95°C hot water . After 15 min of incubation at 95°C , the sample was centrifuged at room temperature for 10 min at 17 , 000 ×g and the supernatant was stored at –20°C . Food was replaced every day . All larvae consumed at least 95% of the offered food during the entire period of the experiment . On the third day , the larvae were dissected 4 hr after being fed . Larvae were dipped in liquid nitrogen for 2 s and , subsequently , anterior and posterior midguts were removed by dissection . The gut tissue was cleaned from the gut content , immediately frozen in liquid nitrogen , and stored at –80°C until RNA extraction . For RNA extraction , gut tissue was ground to a fine powder using plastic pestles . RNA was extracted from 10 to 20 mg ground tissue using innuPREP RNA Mini Kit ( Analytik Jena , Jena , Germany ) following the manufacturer’s protocol . On-column digestion was performed with the innuPREP DNAse I Digest Kit ( Analytik Jena ) . TrueSeq compatible libraries were prepared and PolyA enrichment performed before sequencing the transcriptomes on an Illumina HiSeq 2 , 500 with 17 Mio reads per library of 100 base pairs , paired-end . Reads were quality trimmed using Sickle with a Phred quality score of >20 and a minimum read length of 80 . De novo transcriptome assembly was performed with the pooled reads of all libraries using Trinity ( version Trinityrnaseq_r20131110 ) , running at default settings . Raw reads were archived in the NCBI Sequence Read Archive ( SRA; BioProject PRJNA728510 ) . Transcript abundance was estimated by mapping the reads of each library to the reference transcriptome using RSEM ( Li and Dewey , 2011 ) with Bowtie ( version 0 . 12 . 9 ) ( Langmead et al . , 2009 ) running at default settings . Differential expression analysis was performed with Wald test in DeSeq2 in which low-expressed genes were excluded . Gene ontology ( GO ) terms were retrieved using Trinotate , and GO enrichment analysis of the up-regulated genes ( Benjamini-Hochberg adjusted <i>p-value < 0 . 05 ) in the anterior midgut of the control and TA-G-enriched samples , as well as the control and latex-enriched samples , was performed using the hypergeometric test implemented in BiNGO using the Benjamini-Hochberg adjusted p-value of <0 . 01 . In order to identify putative M . melolontha β-glucosidases , we performed tBLASTn analysis using the known β-glucosidases from T . molitor ( AF312017 . 1 ) and C . populi ( KP068701 . 1 ) as input sequences ( Ferreira et al . , 2001; Rahfeld et al . , 2015 ) . We retained transcripts with a BitScore larger than 200 , an average FPKM ( fragments per kilobase of transcript per million mapped reads ) value ( all samples ) larger than 2 , and an at least twofold higher average FPKM value in the anterior than posterior midguts of the control samples to match the in vitro deglucosylation activity . Through this analysis , 19 sequences were selected of which 11 appeared to be full-length genes and 8 were gene fragments . In order to verify the gene sequences , RNA was isolated from M . melolontha anterior midgut samples ( three biological replicates ) using the RNeasy Plant Mini Kit ( Qiagen ) , and first-strand cDNA was prepared from 1 . 2 µg of total RNA using SuperScript III reverse transcriptase and oligo d ( T12-18 ) primers ( Invitrogen , Carlsbad , CA , USA ) . RACE PCR ( ‘SMARTer RACE cDNA Amplification Kit’ Clontech , Mountain View , CA , USA ) was used to obtain full-length genes ( see Supplementary file 2 for primer information ) . In the end , 12 full-length open reading frames of putative β-glucosidases could be amplified from M . melolontha cDNA ( see Supplementary file 1 for M . melolontha β-glucosidase nucleotide sequences and Supplementary file 2 for primer information ) , a reduction from the 19 originally selected sequences due to a lack of amplification of some gene fragments , merging of others , and assembly errors in the transcriptome . Signal peptide prediction of the resulting 12 candidate genes was performed with the online software TargetP ( http://www . cbs . dtu . dk/services/TargetP/ ) ( Emanuelsson et al . , 2000 ) . We aligned the amino acid sequences of the 12 candidate sequences , as well as of the known glucosidases of T . molitor ( AF312017 . 1 ) , C . populi ( KP068701 . 1 ) , Brevicoryne brassicae ( AF203780 . 1 ) , and Phyllotreta striolata ( KF377833 . 1 ) ( Beran et al . , 2014; Jones et al . , 2002 ) using the MUSCLE algorithm ( gap open , –2 . 9; gap extend , 0; hydrophobicity multiplier , 1 . 2; clustering method , upgmb ) implemented in MEGA 5 . 05 ( Tamura et al . , 2011 ) , and visualized the alignment in BioEdit version 7 . 0 . 9 . 0 ( Hall , 1999 ) . The alignment was used to compute a phylogeny with a maximum likelihood method ( WAG model; gamma-distributed rates among sites ( five categories ) ; Nearest-Neighbor-Interchange heuristic method; sites with less than 80% coverage were eliminated ) as implemented in MEGA 5 . 05 . A bootstrap resampling strategy with 1000 replicates was applied to calculate tree topology . In order to estimate the expression levels of the putative β-glucosidases , we replaced the previously identified β-glucosidase sequences in the transcriptome with the confirmed full-length genes and estimated transcript abundance by mapping the trimmed short reads of each library to the corrected reference transcriptome as implemented in the Trinity pipeline using RSEM and Bowtie . For differential expression analysis , all contigs that had an average count value of >1 per library were retained . To test whether TA-G or latex affected the expression of the β-glucosidases , differential expression analysis was accomplished by pairwise comparisons of the control and TA-G-enriched anterior midgut samples , and the control and latex-enriched anterior midgut samples , using an exact test in edgeR ( Robinson et al . , 2010 ) . The significance level of 0 . 05 was adjusted for multiple testing using the Benjamini-Hochberg false discovery rate method . To test whether the expression level of β-glucosidases differed between anterior and posterior midgut samples , a pairwise comparison between the control samples of the anterior and posterior midgut was performed as described above . Averaged FPKM values of each treatment and gut section were displayed with a heat map . In order to characterize the isolated M . melolontha β-glucosidase genes , they were heterologously expressed in a line of T . ni-derived cells ( High Five Cells; Life Technologies , Carlsbad , CA , USA ) as described in Rahfeld et al . , 2015 . Briefly , genes were cloned into the pIB/V5-His TOPO vector ( Life Technologies ) . After sequence verification , these vector constructs were individually used with the FuGeneHD-Kit to transfect insect High Five Cells according to the manufacturer’s instructions ( Promega , Madison , WI , USA ) . After 1 day of incubation at 27°C , the cultures were supplied with 60 mgml–1 blasticidin ( Life Technologies ) to initiate the selection of stable cell lines . Afterwards , the insect cells were selected over three passages . The cultivation of the stable cell lines for protein expression was carried out in 75 cm3 cell culture flasks , containing 10 ml Express Five culture medium ( Life Technologies ) , 20 mgml–1 blasticidin , and one x Protease Inhibitor HP Mix ( SERVA Electrophoresis , Heidelberg , Germany ) . After 3 days of growth , the supernatant was collected by centrifugation ( 4000 ×g , 10 min , 4°C ) , concentrated using 10 . 000 Vivaspin 4 ( Sartorius ) , and desalted ( NAP-5; GE Healthcare , Munich , Germany ) into assay buffer ( 100 mM NaPi , pH 8 ) . In order to test the TA-G-hydrolyzing activity and substrate specificity of the M . melolontha glucosidases , the heterologously expressed proteins were assayed with the plant defensive glycosides TA-G , a mixture of maize benzoxazinoids ( BXDs ) , salicin , and 4-methylsulfinylbutyl glucosinolate ( 4-MSOB ) , as well as the disaccharide cellobiose , which were obtained as described below . The standard fluorogenic substrate , 4-methylumbelliferyl-β-D-glucopyranoside ( Glc-MU ) , served as a positive control . Non-transfected insect cells ( WT ) and cells transfected with green fluorescent protein ( GFP ) served as negative controls . For the enzymatic assays , 97 µl concentrated and desalted supernatant of the heterologous expression culture was incubated with 3 µl 10 mM substrate for 24 hr at 25°C , after which the reaction was stopped with an equal volume of methanol . Due to a very rapid deglucosylation of TA-G , incubation time was shortened to 10 s for this compound . After assays , all samples were centrifuged at 11 , 000 ×g for 10 min at room temperature and the supernatant was analyzed with a different method for each substrate as described below . TA-G was purified as described in Huber et al . , 2016b . Deglucosylation activity was measured based on the concentration of the aglycone TA on an HPLC-UV and quantified at 245 nm as described above . BXDs were partially purified from maize seedlings ( cultivar Delprim hybrid ) . Seeds were surface-sterilized and germinated in complete darkness . After 20 days , leaves from approximately 60 seedlings were ground under liquid nitrogen to a fine powder and extracted with 0 . 1% formic acid in 50% methanol with 0 . 25 ml per 100 mg tissue . Methanol was evaporated under nitrogen flow at 40°C . BXDs were enriched using 500 mg HR-X Chromabond solid-phase extraction cartridges ( Macherey-Nagel ) with elution steps ( 5 ml ) using water , 30% ( aq . ) methanol , and 100% methanol . 2 ml water was added to the 100% methanol fraction , which contained the BXDs . Subsequently , methanol was completely evaporated from this fraction under nitrogen flow at 40°C , and after freeze-drying , the freeze-dried material ( ~5 mg ) was dissolved in 1 ml H2O . This enriched BXD solution contained a mixture of different BXD glucosides , with DIMBOA ( 2 , 4-dihydroxy-7-methoxy-1 , 4-benzoxazin-3-one ) -glucoside as the major compound . To test for the deglycosylation of the BXDs , the formation of the aglycone MBOA ( 6-methoxy-benzoxazolin-2-one; a spontaneous degradation product of the DIMBOA aglycone ) was monitored on an Agilent 1200 HPLC system coupled to an API 3200 tandem mass spectrometer ( Applied Biosystems ) equipped with a turbospray ion source operating in negative ionization mode . Injection volume was 5 μl using a flow rate of 1 mlmin–1 . Metabolite separation was accomplished with a ZORBAX Eclipse XDB-C18 column ( 50 × 4 . 6 mm , 1 . 8 μm; Agilent Technologies ) using the following gradient of 0 . 05% formic acid ( A ) and methanol ( B ) : 0 min: 20% B , 9 min: 25% B , 10 min: 50% B , 12 min: 100% B , followed by column reconditioning . The column temperature was kept at 20°C . MRM was used to monitor analyte parent ion → product ion: m/z 164 → 149 ( CE –20 V; DP –24 V ) for MBOA . Analyst 1 . 5 software ( Applied Biosystems ) was used for data acquisition and processing . Salicin ( Alfa Aeser ) was purchased and its deglucosylation was quantified based on the formation of the deglucosylation product salicyl alcohol , which was analyzed on an HPLC-UV using the same procedure as described for TA-G . The peak of salicyl alcohol ( elution time = 9 . 3 min ) was integrated at 275 nm . 4-MSOB was isolated from 50 g of broccoli seeds ( Brokkoli Calabraise; ISP GmbH , Quedlingburg , Germany ) , which were homogenized in 0 . 3 l of 80% aqueous methanol and centrifuged at 2500 ×g for 10 min , and the supernatant separated on a DEAE-Sephadex A25 column ( 1 g ) . After the supernatant was loaded , the column was washed three times with 5 ml formic acid + isopropanol + water ( 3 : 2 : 5 by volume ) and four times with 5 ml water . Intact glucosinolates were eluted from the DEAE Sephadex with 25 ml of 0 . 5 M K2SO4 ( containing 3% isopropanol ) dropped into 25 ml of ethanol ( Thies , 1988 ) . The collected solution was centrifuged to spin down the K2SO4 and the supernatant was dried under vacuum . The residue was resuspended in 3 ml of water , and 4-MSOB was isolated by HPLC as described in Schramm et al . , 2012 . Purification was performed on an Agilent 1 , 100 series HPLC system using a Supelcosil LC-18-DB Semi-Prep column ( 250 × 10 mm , 5 µm; Supelco , Bellefonte , PA , USA ) with a gradient of 0 . 1% ( v/v ) aqueous trifluoroacetic acid ( solvent A ) and acetonitrile ( solvent B ) . Separation was accomplished at a flow rate of 4 ml min–1 at 25°C as follows: 0–3% B ( 6 min ) , 3–100% B ( 0 . 1 min ) , a 2 . 9-min hold at 100% B , 100–0% B ( 0 . 1 min ) , and a 3 . 9-min hold at 0% B , and the fraction containing 4-MSOB was collected with a fraction collector . The fraction was dried under vacuum and resuspended in 10 ml methanol to which 40 ml ethanol was added to precipitate the glucosinolate as the potassium salt . The flask was evaporated under vacuum to remove the solvents , and the residue was recovered as a powder . The identity and purity of the isolated 4-MSOB were checked by LC-MS ( Bruker Esquire 6000; Bruker Daltonics , Bremen ) and 1 H NMR ( 500 MHz model; Bruker BioSpin GmbH , Karlsruhe , Germany ) . The deglucosylation of 4-MSOB was quantified based on the formation of 4-MSOB isothiocyanate with an API 3200 LC-MS as described above operating in positive ionization mode . Injection volume was 5 μl using a flow rate of 1 . 1 mlmin–1 . Metabolite separation was accomplished with ZORBAX Eclipse XDB-C18 column ( 50 × 4 . 6 mm , 1 . 8 μm; Agilent Technologies ) using the following gradient of 0 . 05% formic acid ( A ) and acetonitrile ( B ) : 0 min: 3% B , 0 . 5 min: 15% B , 2 . 5 min: 85% B , 2 . 6 min: 100% B , 3 . 5 min: 100% B , followed by column reconditioning . The column temperature was kept at 20°C . MRM was used to monitor analyte parent ion → product ion: m/z 178 → 114 ( CE –13 V; DP –26 V ) . Analyst 1 . 5 software ( Applied Biosystems ) was used for data acquisition and processing . Cellobiose ( Fluka ) was purchased and its deglucosylation was quantified based on the decrease of substrate on an API 3200 LC-MS as described above operating in negative ionization mode . Injection volume was 5 μl , using a flow rate of 1 mlmin–1 . Metabolite separation was accomplished with an apHera NH2 column ( 15 cm x 4 . 6 mm x 3 μm ) using the following gradient of H2O ( A ) and acetonitrile ( B ) : 0 min: 20% A , 0 . 5 min: 20% A , 13 min: 45% A , 14 min: 20% A , followed by column reconditioning . The column temperature was kept at 20°C . MRM was used to monitor analyte parent ion → product ion: m/z 341 → 161 ( CE –10 V; DP –25 V ) . Analyst 1 . 5 software was used for data acquisition and processing . Glc-MU ( Sigma Aldrich ) , a fluorogenic substrate , served as a rapid positive control for the presence of β-glucosidases . Hydrolysis of Glc-MU was scored visually by the presence of fluorescence in samples excited with UV light at 360 nm using a gel imaging system ( Syngene ) . Activity of the heterologously expressed β-glucosidases was categorized into presence and absence based on the formation of the respective aglycones of TA-G , BXDs , salicin , and 4-MSOB , and the decrease of the substrate for cellobiose . For the secondary metabolites , activity was accepted if the aglycone concentration was threefold higher than the mean aglycone concentration of the controls ( GFP , WT; except only WT for TA-G ) . For cellobiose , activity was scored as positive if the cellobiose concentration after the assay was lower than 30% of the cellobiose concentration of the controls ( GFP , WT ) . The enzymatic assays were performed three times ( except TA-G only twice ) with freshly harvested recombinant proteins within 2 weeks , which gave similar results ( Figure 3—figure supplement 2 ) . The averaged categorization results are displayed in Figure 3C . In order to test whether M . melolontha gut proteins deglucosylate BXDs , 4-MSOB , and salicin , we tested glucohydrolase activity of crude extracts of the anterior midgut in vitro . 10 M . melolontha larvae were starved for 24 h , after which the larvae were dipped for 2 s in liquid nitrogen , and , subsequently , anterior midgut tissue and gut content were removed by dissection . The samples were extracted with 10 µl ice-cold 0 . 1 M TAPS ( pH 8 . 0 ) per mg material as described above . All samples were centrifuged at 17 , 000 ×g for 5 min at 4°C and the supernatant stored at –20°C until the enzymatic assay . Deglucosylation activity was measured by incubating 20 µl gut extract that had been either kept on ice or boiled for 10 min at 95°C with a 6 mM mixture of BXDs , salicin , or 4-MSOB ( substrates were obtained as described above added in a 20 µl volume ) in 0 . 01 M TAPS ( pH 8 . 0 ) for 1 hr at room temperature , after which the reaction was stopped by the addition of an equal volume of methanol . All samples were centrifuged at 3220 ×g for 5 min at room temperature and the supernatant stored at –20°C until analysis . For BXDs , salicin , and 4-MSOB , the formation of the aglycone was quantified using HPLC-MS and HPLC-UV as described above . Deglycosylation activity was standardized by dividing the peak area of the aglycone of each sample by the maximal peak area of all samples ( ‘relative aglycone formation’ ) . Differences in the relative aglycone formation between boiled and non-boiled gut samples , as well as between anterior midgut content and tissue samples , were analyzed with two-way ANOVAs . In order to establish RNAi in M . melolontha , we injected different doses of dsRNA targeting tubulin and GFP ( negative control ) into the larvae . As a template for dsRNA synthesis , we chose an approximately 500 bp fragment of each gene ( see Supplementary file 3 for fragment nucleotide sequence ) . The fragments were amplified using the Q5 High-Fidelity DNA Polymerase ( New England Biolabs , Ispwich , MA , USA ) according to the manufacturer’s procedure and the specific primer combinations Mm-tubulin-fwd and Mm-tubulin-rev for tubulin , as well as GFP-RNAi_fwd and GFP-RNAi_rev for GFP ( Supplementary file 2 ) . Isolated and purified M . melolontha cDNA served as a template for tubulin . Plasmid pGJ 2648 , which encodes for the emerald variant for GFP and was kindly supplied by Dr . Christian Schulze-Gronover , served as a template for GFP . Amplified fragments were separated by agarose gel electrophoresis and purified using GeneJET Gel Extraction Kit ( Thermo Fisher Scientific , Waltham , MA , USA ) according to the manufacturer’s procedure . An A-tail was added using DreamTaq DNA Polymerase ( Thermo Fisher Scientific ) , and the A-tailed fragments were then cloned into T7 promoter sequence containing pCR2 . 1-TOPO plasmids ( Life Technologies ) according to the manufacturer’s instructions . Plasmids with the insert in both orientations with regard to the T7 promoter were identified by sequencing . dsRNA was synthesized using the MEGAscript RNAi Kit ( Thermo Fisher Scientific ) according to the manufacturer’s procedure . The above-described tubulin and GFP plasmid templates were linearized downstream of the insert using the restriction enzyme BamHI ( New England Biolabs ) . Sense and antisense single-stranded ( ss ) RNAs were synthesized in separate reactions . The complementary RNA molecules were then annealed and purified using MEGAscript RNAi Kit according to the manufacturer’s instructions ( Thermo Fisher Scientific ) . In order to investigate the required dsRNA concentration and duration of the silencing , we injected 2 . 5 and 0 . 25 µg dsRNA of tubulin or GFP per g of larva into M . melolontha . The larvae were anesthetized under CO2 . Subsequently , larvae were punctured with a sterile syringe ( Ø 0 . 30 × 12 mm ) between the second and the third segment , and approximately 50 µl tubulin or GFP dsRNA ( 100 ngµl–1 for 2 . 5 µg per g larva and 10 ngµl–1 for 0 . 25 µg per g larva ) was injected into the hemolymph of the second segment of nine M . melolontha larvae per concentration . Every second day , the larvae were weighed . 5 days after injection , the larvae received fresh carrots to feed on . 2 , 5 , and 10 days after injection , three larvae per concentration were frozen in liquid nitrogen . The entire larvae were ground to a fine powder using mortar and pestle under liquid nitrogen and stored at –80°C until RNA extraction . Total RNA was isolated using the GeneJET Plant RNA Purification Kit following the manufacturer’s instructions . On-column RNA digestion was performed with RNase-free DNase ( Qiagen , Netherlands ) . cDNA synthesis was performed using SuperScript II Reverse Transcriptase ( Thermo Fisher ) and oligo ( dT21 ) ( Microsynth , Switzerland ) according to the manufacturer’s instructions . Consequently , the qPCR was performed with the KAPA SYBR FAST qPCR Kit Optimized for LightCycler 480 ( Kapa Biosystems , Wilmington , MA , USA ) in a Nunc 96-well plate ( Thermo Fisher Scientific ) on a LightCycler 96 ( Roche Diagnostics , Switzerland ) with one technical replicate per sample . Tubulin gene expression was quantified relative to actin using the qPCR primers qPCR_Mm_Tubulin_fwd and qPCR_Mm_Tubulin_rev for tubulin , as well as qPCR_Mm_actin_fwd and qPCR_Mm_actin_rev for actin ( Supplementary file 2 ) . Differences in the relative expression of tubulin to actin and between tubulin- and GFP dsRNA-treated larvae were analyzed with the Student’s t-test . In order to test whether Mm_bGlc17 accounts for the TA-G deglucosylation in vivo , we silenced Mm_bGlc16 , Mm_bGlc17 , and Mm_bGlc18 in M . melolontha using RNAi and analyzed TA-G deglucosylation activity in vitro using anterior midgut extracts . M . melolontha in which a dsGFP fragment was injected served as a control . GFP dsRNA was synthesized as described above . To obtain dsRNA for the glucosidase genes , we chose approximately 500 bp fragments of Mm_bGlc16 , Mm_bGlc17 , and Mm_bGlc18 cDNA as templates for dsRNA synthesis that showed maximal sequence divergence with other M . melolontha β-glucosidases as well as among each other ( see Supplementary file 3 for fragment nucleotide sequence ) . The fragments were amplified using the Q5 High-Fidelity DNA Polymerase ( New England Biolabs ) according to the manufacturer’s procedure and specific primer combinations of which one primer was fused to the T7 promoter sequence . The plasmids obtained from the heterologous expression were used as PCR templates ( see above ) . For each β-glucosidase , we performed two PCRs to yield two dsRNA templates that are identical except for a single T7 promoter sequence at opposite ends . For Mm_bGlc16 fragment amplification , the primer combinations Mm_bGlc_16_fwd_T7 and Mm_bGlc_16_rev , as well as Mm_bGlc_16_fwd and Mm_bGlc_16_rev_T7 , were used . For the amplification of Mm_bGlc17 and Mm_bGlc18 fragments , the respective primers were deployed . Amplified fragments were separated by agarose gel electrophoresis and purified using GeneJET Gel Extraction Kit ( Thermo Fisher Scientific ) according to the manufacturer’s procedure . An A-tail was added using DreamTaq DNA Polymerase ( Thermo Fisher Scientific ) and the A-tailed fragments were then cloned into pIB/V5-His-TOPO plasmids . dsRNA was synthesized and linearized as described above using the restriction enzymes Xhol , for the glucosidase genes , and BamHI , for GFP ( New England Biolabs ) . The dsRNA was synthesized using the MEGAscript RNAi Kit ( Thermo Fisher Scientific ) according to the manufacturer’s procedure . The above-described M . melolontha β-glucosidase and GFP plasmid templates were linearized downstream of the insert using restriction enzymes XhoI and BamHI ( New England Biolabs ) , respectively , and annealed and purified as described above . To silence M . melolontha glucosidases in vivo , we injected dsRNA of the respective glucosidases or GFP as a control into M . melolontha larvae as described above using 50 µl of a 10 ngµl–1 Mm_bGlc16 , Mm_bGlc17 , Mm_bGlc18 , or GFP dsRNA . In addition , we performed a co-silencing of Mm_bGlc16 and Mm_bGlc17 ( Mm_bGlc16&17 ) , for which 25 µl of 10 ngµl–1 Mm_bGlc16 and Mm_bGlc17 was injected . Larvae were kept at room temperature for 7 days , after which the larvae were dissected as described above . The anterior midgut content was extracted with 10 µl 0 . 01 M TAPS ( pH 8 . 0 ) per mg material and centrifuged at 17 , 000 ×g for 10 min at 4°C . For the enzymatic reaction , 10 µl supernatant that was either kept at 4°C or had been boiled for 1 hr at 98°C was incubated with 40 μl 0 . 01 M TAPS ( pH 8 . 0 ) and 50 μl 2 mM latex water extract . After 3 hr , the reaction was stopped by adding equal volumes of methanol . The samples were centrifuged at 17 , 000 ×g for 10 min at room temperature and the supernatant analyzed on a Waters ACQUITY UPLC series equipment coupled to an ACQUITY photodiode array and an ACQUITY QDa mass detector . Metabolite separation was accomplished using an ACQUITY UPLC column with 1 . 7 μm BEH C18 particles ( 2 . 1 × 100 mm ) . The mobile phase consisted of 0 . 05% formic acid ( A ) and acetonitrile ( B ) utilizing a flow rate of 0 . 4 mlmin–1 with the following gradient: 0 min: 5% B , 1 . 5 min: 20% B , 2 . 5 min: 40% B , 3 min: 95% B , 5 min: 95% B , followed by column reconditioning . The peak areas of TA and TA-G were integrated at 245 nm using Waters MassLynx49 . Deglucosylation activity was expressed as the ratio of TA/ ( TA + TA-G ) . In addition , to account for the spontaneous deglucosylation of TA-G , the deglucosylation activity was normalized by subtracting the average TA/ ( TA + TA-G ) of the boiled samples from each non-boiled sample ( ‘normalized deglucosylation activity’ ) . Difference in the normalized and non-normalized deglucosylation activities between the RNAi-silenced larvae was analyzed with one-way ANOVAs , and significant differences between the groups were determined using Tukey’s Honest Significance test . To test for the silencing efficiency and specificity of the Mm_bGlc17 dsRNA injection , we injected M . melolontha with 0 . 25 µg of Mm_bGlc17 or GFP dsRNA per g larva as described above . Non-injected larvae were set as controls . After injections , larvae were kept at room temperature for 2 days , after which the larvae were dissected and the individual midguts were isolated . Then , total RNA of the midgut was extracted using RNeasy Lipid Tissue Mini Kit ( QIAGEN ) , coupled with on-column DNA digestion following the manufacturer’s instructions . One microgram of each total RNA sample was reverse transcribed with SuperScript III Reverse Transcriptase ( Invitrogen ) . The quantitative reverse transcription PCR ( RT-qPCR ) assay ( N = 7–8 ) was performed on the LightCycler 96 Instrument ( Roche ) using the KAPA SYBR FAST qPCR Master Mix ( Kapa Biosystems ) . The actin gene was used as an internal standard to normalize cDNA concentrations . The relative gene expressions of Mm_bGlc16 , Mm_bGlc17 , and Mm_bGlc18 to actin were calculated with 2−∆∆Ct method . Primers ( qPCR_Mm _bGlc_16_fwd , qPCR_Mm _bGlc_16_rev , qPCR_Mm _bGlc_17_fwd , qPCR_Mm _bGlc_17_rev , qPCR_Mm _bGlc_18_fwd , qPCR_Mm _bGlc_18_rev , qPCR_Mm _actin-fwd , and qPCR_Mm _actin-rev ) are listed in Supplementary file 2 . In order to test whether Mm_bGlc17 activity affects the performance of M . melolontha larvae in the presence and absence of TA-G , we assessed the growth of Mm_bGlc17- and control ( GFP ) -silenced larvae on TA-G-deficient and control T . officinale plants . T . officinale seeds were germinated on seedling substrate . After 15 days , plants were transplanted into 1 l rectangular pots ( 18 × 12 × 5 cm , length × width × height ) filled with a homogenized mixture of 2/3 seedling substrate ( Klasmann-Deilmann , Switzerland ) and 1/3 landerde ( Ricoter , Switzerland ) . Each pot consisted of four plants in two parallel rows of two plants , which were arranged along the short edges of the pots . Rows were spaced 9 cm apart and had a distance of 4 . 5 cm from the short edges , and plants within each row were grown 4 cm apart from each other . After 50 days of growth , half of the pots ( N = 15 per genotype ) were randomly selected to examine the performance of Mm_bGlc17-silenced larva , and the second half of the pots ( N = 15 per genotype ) were used for GFP-control larva . dsRNA was synthesized as described above . Larvae were treated with 0 . 25 µg of Mm_bGlc17 or GFP dsRNA per g larva as previously described . One pre-weighed larva was added into a hole ( 4 cm depth , 1 cm diameter ) in the center of the pots and covered with moist soil . After 3 weeks of infestation , larvae were recovered from the pots , reweighed , and the midgut was extracted for subsequent RNA extraction following the above-mentioned protocol . To reduce the possible effects of environmental heterogeneity within the greenhouse , the position and direction of the pots were randomly re-arranged weekly . Total RNA of the midgut was extracted using RNeasy Lipid Tissue Mini Kit ( QIAGEN ) , coupled with on-column DNA digestion following the manufacturer’s instructions . One microgram of each total RNA sample was reverse transcribed with SuperScript III Reverse Transcriptase ( Invitrogen ) . The RT-qPCR assay was performed on the LightCycler 96 Instrument ( Roche ) using the KAPA SYBR FAST qPCR Master Mix ( Kapa Biosystems ) . The actin gene was used as an internal standard to normalize cDNA concentrations . The relative gene expressions to actin were calculated with 2−∆∆Ct method . Differences in M . melolontha weight gain between larval and plant RNAi treatments were analyzed with a two-way ANOVA . Differences in larval weight gain between Mm_bGlc17-silenced and GFP-control larvae were analyzed with Student’s t-tests for larvae grown on wild-type and TA-G-deficient plants separately . Differences in larval weight gain on TA-G-containing and TA-G-lacking T . officinale plants were analyzed with Student’s t-tests for the Mm_bGlc17-silenced and GFP-control larvae separately . A two-way ANOVA was applied to analyze differences in relative Mm_bGlc17 expression between larval and plant RNAi treatments . Relative Mm_bGlc17 expression was thereto log-transformed to improve model assumptions . Differences in relative Mm_bGlc17 expression between larvae growing on TA-G-containing and TA-G-lacking plants were analyzed with Kruskal-Wallis rank sum tests based on untransformed data for Mm_bGlc17-silenced and GFP-control larvae separately . To repeat the above-described experiment , T . officinale seeds of TA-G-deficient and control plants were cultivated in the greenhouse as previously described , with some slight modifications . Seedlings were germinated on seedling substrate and transplanted into individual pots ( 11 × 11 x 11 cm ) after 21 days of growth ( N = 40 per line ) . After 70 days of growth , larvae were treated with 0 . 25 µg of Mm_bGlc17 or GFP dsRNA per g larva as described above . 4 days later , for each T . officinale line , half of the plants were infested with one pre-weighed Mm_bGlc17-silenced larva and the other half was infested with one pre-weighed GFP-control larva . After 3 weeks of infestation , larvae were carefully recaptured from the pots , weighed , and added into the pots again . 5 weeks later , larvae were recaptured again and weighed . Differences in M . melolontha weight gain between larval and plant RNAi treatments were analyzed with two-way ANOVAs for three time periods ( 3 weeks , 3–8 weeks , and 8 weeks after the start of the experiment ) separately . Differences in larval weight gain between Mm_bGlc17-silenced and GFP-control larvae in these three time periods were analyzed with Student’s t-tests for wild-type and TA-G-deficient monocultures separately . In order to test whether M . melolontha glucosidase activity affects the deterrence of TA-G , we assessed the choice of Mm_bGlc17- and control ( GFP ) -silenced larvae between TA-G-deficient and control T . officinale plants . M . melolontha larvae were injected with 0 . 025 µg*g-1 Mm_bGlc17 or GFP dsRNA as described above . 1 week after dsRNA injection , the larvae were starved for 3 days and placed individually into the center of 250 ml plastic beakers filled with vermiculite . 5-week-old TA-G-deficient and control T . officinale seedlings were embedded into the vermiculite-filled beaker at opposite edges , with 37 replicated beakers for each of the Mm_bGlc17 and GFP treatments . The feeding site was scored visually 3 hr after the start of the experiment by inspecting the beakers from outside . Differences in the choice between TA-G-deficient and control T . officinale plants were analyzed with binomial tests for the Mm_bGlc17- and GFP-silenced larvae separately .	Plants produce certain substances to fend off attackers like plant-feeding insects . To stop these compounds from damaging their own cells , plants often attach sugar molecules to them . When an insect tries to eat the plant , the plant removes the stabilizing sugar , ‘activating’ the compounds and making them toxic or foul-tasting . Curiously , some insects remove the sugar themselves , but it is unclear what consequences this has , especially for insect behavior . Dandelions , Taraxacum officinale , make high concentrations of a sugar-containing defense compound in their roots called taraxinic acid β-D-glucopyranosyl ester , or TA-G for short . TA-G deters the larvae of the Maybug – a pest also known as the common cockchafer or the doodlebug – from eating dandelion roots . When Maybug larvae do eat TA-G , it is found in their systems without its sugar . However , it is unclear whether it is the plant or the larva that removes the sugar . A second open question is how the sugar removal process affects the behavior of the Maybug larvae . Using chemical analysis and genetic manipulation , Huber et al . investigated what happens when Maybug larvae eat TA-G . This revealed that the acidity levels in the larvae’s digestive system deactivate the proteins from the dandelion that would normally remove the sugar from TA-G . However , rather than leaving the compound intact , larvae remove the sugar from TA-G themselves . They do this using a digestive enzyme , known as a beta-glucosidase , that cuts through sugar . Removing the sugar from TA-G made the compound less toxic , allowing the larvae to grow bigger , but it also increased TA-G’s deterrent effects , making the larvae less likely to eat the roots . Any organism that eats plants , including humans , must deal with chemicals like TA-G in their food . Once inside the body , enzymes can change these chemicals , altering their effects . This happens with many medicines , too . In the future , it might be possible to design compounds that activate only in certain species , or under certain conditions . Further studies in different systems may aid the development of new methods of pest control , or new drug treatments .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology" ]	2021	A beta-glucosidase of an insect herbivore determines both toxicity and deterrence of a dandelion defense metabolite
ApoE , ApoE receptors and APP cooperate in the pathogenesis of Alzheimer’s disease . Intriguingly , the ApoE receptor LRP4 and APP are also required for normal formation and function of the neuromuscular junction ( NMJ ) . In this study , we show that APP interacts with LRP4 , an obligate co-receptor for muscle-specific tyrosine kinase ( MuSK ) . Agrin , a ligand for LRP4 , also binds to APP and co-operatively enhances the interaction of APP with LRP4 . In cultured myotubes , APP synergistically increases agrin-induced acetylcholine receptor ( AChR ) clustering . Deletion of the transmembrane domain of LRP4 ( LRP4 ECD ) results in growth retardation of the NMJ , and these defects are markedly enhanced in APP−/−;LRP4ECD/ECD mice . Double mutant NMJs are significantly reduced in size and number , resulting in perinatal lethality . Our findings reveal novel roles for APP in regulating neuromuscular synapse formation through hetero-oligomeric interaction with LRP4 and agrin and thereby provide new insights into the molecular mechanisms that govern NMJ formation and maintenance . Apolipoprotein E ( ApoE ) ε4 genotype is the strongest and most prevalent risk factor for late-onset Alzheimer’s disease ( AD ) ( Corder et al . , 1993; Schmechel et al . , 1993 ) . ApoE binds to and modulates the function of ApoE receptors , a family of LDL receptor-related proteins ( LRPs ) and postsynaptic signal transducers that regulate glutamatergic neurotransmission in the CNS ( Weeber et al . , 2002; Beffert et al . , 2005; Herz and Chen , 2006 ) and the formation of the neuromuscular synapse in the periphery ( Weatherbee et al . , 2006; Zhang et al . , 2008; Kim et al . , 2008b ) . Mutations in the amyloid precursor protein ( APP ) cause early-onset AD ( Goate et al . , 1991 ) . APP and its pathogenic cleavage product , β-amyloid , physically and functionally interact with ApoE receptors on multiple levels ( Kounnas et al . , 1995 ) , by regulating the trafficking and processing of APP ( Ulery et al . , 2000; Pietrzik et al . , 2002; Andersen et al . , 2005; Hoe et al . , 2005; Pietrzik and Jaeger , 2008; Marzolo and Bu , 2009 ) , mediating amyloid clearance ( Andersen et al . , 2005; Deane et al . , 2008 ) and by preventing amyloid-induced synaptic suppression at the synapse ( Durakoglugil et al . , 2009 ) . Intriguingly , APP is also expressed at the neuromuscular junction ( Akaaboune et al . , 2000 ) and APP family members are required for normal formation and function of the neuromuscular junction ( Torroja et al . , 1999; Wang et al . , 2005 , 2009 ) , although the underlying mechanisms remain unclear . APP family members have also been shown to contribute to synaptic function in the CNS ( Weyer et al . , 2011 ) . Although APP occupies such a central role in the pathogenesis of AD , its physiological functions and how they relate to the disease process on the molecular level remains poorly understood . ApoE ε4 genotype strongly predisposes to an earlier age of AD onset , increasing the relative risk in individuals >65 years of age by approximately 10-fold ( Corder et al . , 1993; Schmechel et al . , 1993 ) . APP and ApoE interact with LRPs , thus suggesting a role of ApoE receptors in AD pathogenesis ( Herz and Beffert , 2000 ) . Besides serving as cargo receptors that can mediate the endocytosis of ApoE containing lipoprotein particles , ApoE receptors have also been shown to function as signal transducers that regulate essential signaling pathways during development and in the adult organism , where they control glutamate receptor function , synaptic plasticity , memory and learning ( reviewed in Herz and Chen , 2006 ) . Examples include brain development ( LRP8 , VLDLR ) , vascular development and maintenance ( LRP1 ) , kidney and neuromuscular junction formation ( LRP4 ) , and others ( reviewed in Dieckmann et al . , 2010 ) . The NMJ is a specialized peripheral cholinergic synapse ( Hall and Sanes , 1993; Sanes and Lichtman , 2001; Wu et al . , 2010 ) . Impaired cholinergic neurotransmission has been implicated in AD and this forms the conceptual basis for the therapeutic use of cholinesterase inhibitors . Both APP and LRP4 are required for normal NMJ formation , and the LRP4 ligand agrin is abundantly expressed in the CNS ( Burgess et al . , 2000; O’Connor et al . , 1994; Stone and Nikolics , 1995 ) . Furthermore , MuSK localizes to excitatory synapses ( Ksiazek et al . , 2007 ) and has been shown to mediate cholinergic responses , synaptic plasticity and memory formation ( Garcia-Osta et al . , 2006 ) , while agrin was found to regulate synaptogenesis ( McCroskery et al . , 2009 ) and prevent synapse loss in the cortex ( Ksiazek et al . , 2007 ) . These intriguing similarities between MuSK , agrin and ApoE receptors in synaptic maintenance and function prompted us to explore the NMJ as a model system on which potentially novel functional interactions between APP and ApoE receptors could be investigated . We found that LRP4 and APP physically bind to each other and that APP can also bind agrin . Ligation of LRP4 by immobilized APP can activate MuSK in the absence of neural agrin . Simultaneous and stoichiometric interaction of the three proteins favors the formation of a hetero-oligomeric complex , which may serve to focus the formation of the NMJ at the surface of the myotube . A common mechanism by which ApoE receptors directly transduce or indirectly modulate extracellular signals involves the interaction of adaptor proteins with their intracellular domain ( ICD ) ( Trommsdorff et al . , 1998 , 1999; Gotthardt et al . , 2000 ) . We therefore generated three mutant Lrp4 alleles in mice to test which domains of LRP4 are required for NMJ development . Consistent with the previous report that mice with point mutations in Lrp4 die at birth ( Weatherbee et al . , 2006 ) , mice carrying a novel Lrp4 null allele , which we generated by deleting exon 1 of murine Lrp4 ( Figure 1A ) , also die perinatally from a complete failure to form NMJs ( Figure 2A ) . By contrast , mice carrying an Lrp4 allele encoding a truncated receptor consisting of only the extracellular domain ( ECD ) , but lacking the transmembrane segment and intracellular domain ( ICD ) , are viable ( Johnson et al . , 2005 ) , indicating that at least partially functional NMJs must form and that thus neither membrane anchoring of the LRP4 ECD nor its ICD is absolutely required for the formation of the NMJ ( Dietrich et al . , 2010; Gomez and Burden , 2011 ) . To further test this hypothesis , we first set out to determine to what extent membrane anchoring of the LRP4 receptor is required for NMJ formation . We examined NMJs of Lrp4ECD/ECD mice , in which we introduced a stop codon immediately before the transmembrane segment . This results in the expression of LRP4 with the extracellular domain , but without the transmembrane domain or the intracellular domain ( Figure 1B ) ( Johnson et al . , 2005 ) . 10 . 7554/eLife . 00220 . 003Figure 1 . Targeting vectors . Diagrammatic representation of gene replacement strategies used to generate Lrp4−/− ( null allele , A ) , Lrp4ECD/ECD ( hypomorphic allele without membrane anchor , B ) , and Lrp4ΔICD/ΔICD ( C ) mice . Exons encoding the LRP4 ECD are depicted in blue and indicated by brackets . The exon encoding the transmembrane segment is shown in red , cytoplasmic domain encoding exons in green . To generate the Lrp4 null allele , the transcription start site and exon 1 were replaced with a neo cassette ( A ) . To generate the Lrp4 ΔICD allele , a cDNA cassette encoding the transmembrane segment of LRP4 ( TM , red ) followed by a Myc epitope and bovine growth hormone 3′UTR was introduced into the targeting vector described by Johnson et al . ( 2005 ) . This results in the normal expression of the LRP4 ECD and transmembrane segment , but complete replacement of the ICD with a Myc tag ( C ) . The generation of the LRP4 ECD allele has been described in Johnson et al . ( 2005 ) ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00220 . 00310 . 7554/eLife . 00220 . 004Figure 2 . NMJ development in Lrp4 mutant mice . ( A ) Embryonic diaphragm muscles ( E16 . 5 ) from wild-type ( top row ) , Lrp4–/– ( middle row ) and Lrp4ECD/ECD mice ( bottom row ) were double-labeled with anti-syntaxin antibodies ( green ) for decorating innervating fibers and with α-bungarotoxin ( red ) to detect AChRs . In both wild-type and Lrp4ECD/ECD muscles , AChRs are clustered along the central region juxtaposed to nerve terminals . By contrast , AChR clusters are completely absent and NMJs fail to form in Lrp4–/– muscles . ( B ) Triangularis sterni muscles at postnatal day 12 were stained for nerves in green ( anti-NF150 and anti-Syt2 antibodies ) and AChRs in red ( α-bungarotoxin ) . In both wild-type ( left column ) and Lrp4ΔICD/ΔICD mice ( middle column ) , neuromuscular synapses were distributed within a narrow band near the nerve trunk . By contrast , neuromuscular synapses in Lrp4ECD/ECD mice ( right column ) were distributed across a broader region of the muscle , as prolonged nerve branches extended from the nerve trunk . Furthermore , AChR clusters are markedly smaller in the Lrp4ECD/ECD muscle , compared with the wild-type or Lrp4ΔICD/ΔICD muscle ( bottom row: high-power views of AChR clusters ) . Scale bars , A: 300 µm; B: 200 µm ( top three rows ) and 50 µm ( bottom row ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00220 . 004 One of the hallmarks of NMJ formation is that presynaptic nerve terminals progressively accumulate synaptic vesicles in juxtaposition with AChR clusters that accumulate on the postsynaptic membrane ( Sanes and Lichtman , 1999; Wu et al . , 2010 ) . We therefore double-labeled embryonic diaphragm muscles ( E16 . 5 ) with anti-syntaxin antibodies and α-bungarotoxin to detect presynaptic nerve terminals and postsynaptic AChRs , respectively . In contrast to the complete absence of pre- and postsynaptic differentiation in Lrp4 null mice ( Figure 2A , middle row ) , neuromuscular synapses were present in the E16 . 5 Lrp4ECD/ECD embryos . Presynaptic terminals were properly juxtaposed to postsynaptic AChR clusters ( Figure 2A , bottom row ) , which aligned along the central region of the muscle , although the density of the AChR appeared reduced . These results indicate an initial formation of the NMJs in E16 . 5 Lrp4ECD/ECD embryos . Next , we tested whether membrane anchoring of LRP4 further enhances the function of the ECD or whether the ICD is required to achieve full functionality . To do this , we took advantage of a mouse line carrying a novel Lrp4 knockin allele in which the ICD had been replaced with a Myc-tag ( Lrp4ΔICD/ΔICD , Figure 1C ) . Triangularis sterni muscle from these mice was isolated at P12 and NMJs were stained with anti-syntaxin antibodies and α-bungarotoxin . NMJs developed normally in the Lrp4ΔICD/ΔICD mice ( Figure 2B , middle column ) , with synapse size indistinguishable from wild type ( left column ) . These results are also consistent with the lack of an obvious neuromuscular phenotype in a strain of cattle affected by Mulefoot disease , where a naturally occurring splice site mutation in Lrp4 results in truncation of the protein and a nearly complete loss of the cytoplasmic domain ( Johnson et al . , 2006 ) . By contrast , the impaired development of NMJs in Lrp4ECD/ECD mice continues during the postnatal period . AChR clusters were distributed more diffusely in P12 , the triangularis sterni muscle of Lrp4ECD/ECD mice , compared with wild-type or Lrp4ΔICD/ΔICD mice ( Figure 2B ) . In adult muscles , the NMJ from Lrp4ECD/ECD mice remained noticeably smaller than that from age-matched wild-type mice . As shown in Figure 3A , in the triangularis sterni muscle from 3-month old Lrp4ECD/ECD mice , the pre-synaptic nerve terminal and post-synaptic AChR clusters remained markedly smaller compared with wild type . By contrast , the NMJ in Lrp4ΔICD/ΔICD mice developed to similar size as wild type ( Figure 3B ) . Thus , although the NMJ was established in Lrp4ECD/ECD embryos , it remained considerably smaller ( compared with age-matched wild-type control ) at postnatal and adult stages , indicating that LRP4 membrane anchoring is required for the growth and maturation of the NMJ . Intriguingly , the ultrastructure of the nerve terminal and post-synaptic membrane appeared largely normal—with abundant presynaptic vesicles and elaborate junctional folds ( Figure 3C ) . We examined 55 nerve terminal profiles from 18 Lrp4ECD/ECD NMJs and 74 profiles from 23 control NMJs and found no statistical difference between control and Lrp4ECD/ECD . 10 . 7554/eLife . 00220 . 005Figure 3 . Reduced NMJ size in Lrp4ECD/ECD mice . ( A ) Triangularis sterni ( TS ) muscles from wild type ( WT ) and 3-month-old Lrp4ECD/ECD mice double-labeled with anti-syntaxin antibodies ( green , nerve ) and α-bungarotoxin ( red , AChRs ) . The NMJ in the Lrp4ECD/ECD muscle is markedly smaller than in the wild-type muscle . ( B ) Size comparison of neuromuscular synapse areas in TS muscles in 3-month-old wild-type , Lrp4ECD/ECD and Lrp4ΔICD/ΔICD mice . ( C ) Electron micrographs of NMJs from lumbrical muscles in 2-month old Lrp4WT/ECD and littermate Lrp4ECD/ECD mice . Nerve terminals containing numerous synaptic vesicles and mitochondria are embedded in the muscle surface and junctional folds are apparent in both genotypes . Scale bars , 1 μm . 55 nerve terminal profiles from 18 Lrp4ECD/ECD NMJs and 74 profiles from 23 control NMJs were analyzed . Representative images are shown . Arrow indicates junctional fold . SV: synaptic vesicle; mito: mitochondria . DOI: http://dx . doi . org/10 . 7554/eLife . 00220 . 005 Furthermore , despite a significant reduction in sizes , only a minor fraction ( 2% ) of synapses was innervated by more than one axon in Lrp4ECD/ECD NMJs at P12 , as was the case in the wild type . Moreover , the γ-ε switch also occurred normally in the Lrp4ECD/ECD mutant . By P16 γ-AChR subunit expression was no longer detectable on the protein or mRNA level in wild-type , Lrp4wt/ECD or Lrp4ECD/ECD muscle ( data not shown ) . Therefore , deleting the transmembrane and intracellular domains of LRP4 had no impact on the process of synapse elimination and AChR subunit switch ( from the embryonic γ to the adult ε form ) at the NMJ . These findings indicate that the LRP4 intracellular domain harbors no essential elements for relaying the molecular signals that govern NMJ formation , but that membrane anchoring of LRP4 is required for maintaining the level of signal strength necessary for postnatal growth and maturation of the neuromuscular synapses . On the other hand , secretion of the extracellular domain of LRP4 into the pericellular space is apparently sufficient to initiate clustering of AChRs and thus manifestation of immature NMJs that can then be maintained at the prevailing reduced signal strength . This is consistent with the findings in conditional Musk knockout mice ( Hesser et al . , 2006 ) , which showed NMJ disassembly in postnatal muscle upon conditional MuSK inactivation , indicating the requirement for continuous MuSK activity in postnatal muscle . This low level signaling may in part be mediated by the interaction of LRP4 ECD with MuSK ( Zhang et al . , 2008; Kim et al . , 2008b ) . However , several ApoE receptors have also been reported to interact directly or indirectly with APP ( Kounnas et al . , 1995; Ulery et al . , 2000; Pietrzik et al . , 2002; Andersen et al . , 2005; Hoe et al . , 2005; Pietrzik and Jaeger , 2008; Marzolo and Bu , 2009 ) , and APP itself has been shown to participate in NMJ development ( Wang et al . , 2005 ) . Taken together , these findings suggested that LRP4 might act synergistically with APP to regulate NMJ development and maintenance . To test this hypothesis , we bred Lrp4ECD/ECD mice with App−/− mice to generate Lrp4ECD/ECD and App double mutant mice . We found that postnatal survival of compound mutant mice was markedly reduced when three wild-type Lrp4 and App alleles were deleted ( e . g . , Lrp4+/ECDApp–/– or Lrp4ECD/ECDApp+/– ) and that the survival of double mutant mice ( Lrp4ECD/ECDApp–/– ) was significantly reduced ( p<0 . 01 ) ( Table 1 ) . Because of this accelerated loss of the compound double mutant animals before weaning , we analyzed the NMJ in E18 . 5 embryos rather than in adults , as normal Mendelian distribution was observed at E18 . 5 in these mice ( data not shown ) . As has been reported earlier ( Wang et al . , 2005 ) , normal size of AChR clusters was observed at the NMJ of App–/– embryos ( Figure 4A , B , d–f , compared to wild type in subpanels a–c , Figure 4C ) . In Lrp4ECD/ECD embryos , AChR cluster size was similar to wild-type muscles , but AChR density was decreased ( Figure 4C ) . However , in Lrp4ECD/ECDApp–/– double mutant embryos , both the density and the size of AChR clusters were significantly reduced compared to wild-type , App–/– or Lrp4ECD/ECD single mutant mice ( Figure 4C ) . Furthermore , nerve terminal sprouting was markedly increased in the double mutant mice ( Figure 4A , B , j–l ) . The significant reduction in the synaptic area of individual double mutant NMJs compared to wild type and Lrp4ECD/ECD is also apparent in the three-dimensional rendition of individual synapses ( Figure 4A , m–o ) . Similar findings were obtained in Lrp4ECD/ECD mice lacking the amyloid precursor protein family member APLP2 ( Figure 5 ) , which cooperates with APP in the development of the NMJ ( Wang et al . , 2005 ) . 10 . 7554/eLife . 00220 . 006Table 1 . Postnatal survival of Lrp4ECD/ECD;App−/− double mutant miceDOI: http://dx . doi . org/10 . 7554/eLife . 00220 . 006GenotypeExpectedObservedChi-square testLrp4+/ECD;App+/−6374n . s . Lrp4+/ECD;App−/−4445n . s . Lrp4ECD/ECD;App+/−4949n . s . Lrp4ECD/ECD;App−/−3815Note: 41% ( 20/49 ) of Lrp4ECD/ECD;App+/− died within five months of age; 13% ( 6/45 ) of Lrp4+/ECD;App−/− mice died within five months of age . n . s . , not significant . Significant at p<0 . 01 . Survival was scored at weaning ( P25 ) . The ‘Expected’ column states the number of offspring of the indicated genotype according to the rules of Mendelian inheritance . The ‘Observed’ column states the number of offspring found to have the indicated genotype . Chi-square analysis of these data indicates whether the observed number significantly deviates from the expected number or not . 10 . 7554/eLife . 00220 . 007Figure 4 . Impairment of pre-and post-synaptic development in Lrp4ECD/ECD;App–/– double mutant mice . ( A ) ( a–l ) Wholemount staining of TS muscles ( E18 . 5 ) double-labeled with anti-neurofilament antibodies and anti-Syt2 antibodies ( nerve , a , d , g , j ) and α-bungarotoxin ( AChR , b , e , h , k ) . Merged images are shown in panels c , f , i and l . The inset in each panel shows a high-power view of the image . AChR clusters were markedly reduced both in number and size in Lrp4ECD/ECD;App−/− mice ( j–l ) , compared to wild type ( a–c ) , App−/− ( d–f ) or Lrp4ECD/ECD ( g–i ) . Furthermore , numerous terminal sprouts ( arrowheads in the inset of j and l ) were seen in Lrp4ECD/ECD;App−/− mutant mice , whereas the nerve terminals in the wild type ( arrowhead in a ) juxtaposed with postsynaptic AChR clusters . m–o: 3D reconstruction of confocal images of the NMJs in wholemount diaphragm muscles in wild type ( m ) , Lrp4ECD/ECD ( n ) and Lrp4ECD/ECD;App−/− ( o ) illustrate the reduced size of the NMJ in ( o ) . Scale bars: a–l: 100 µm; inset: 20 µm . ( B ) Wholemount staining of diaphragm muscles ( E18 . 5 ) double-labeled with anti-neurofilament antibodies and anti-Syt2 antibodies ( nerve , a , d , g , j ) and α-bungarotoxin ( AChR , b , e , h , k ) . Low-power views of the left dorsal region of the diaphragm muscles . Scale bar , 400 µm . The number of AChR clusters was notably reduced in Lrp4ECD/ECD mice , compared to wild-type and App−/− mice . However both the size and number of AChR clusters were markedly reduced in Lrp4ECD/ECD;App−/− mice , compared to wild-type , App−/− and Lrp4ECD/ECD mice . ( C ) Quantitative analysis of AChR clusters size and numbers . The number of AChR clusters at the ventral regions of right hemi-diaphragm muscles ( E18 . 5 ) , normalized by muscle area ( upper panel ) . Average size of AChR clusters ( lower panel ) . The numbers of AChR clusters analyzed are: 144 wild type , 132 App−/− , 131 Lrp4ECD/ECD and 96 Lrp4ECD/ECD;App−/− ( N = 3 mice for each genotype ) . Data are shown as average ± S . E . M . Pairwise multiple comparisons were carried out using Tukey’s test and the statistical differences determined by one-way analysis of variance ( ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00220 . 00710 . 7554/eLife . 00220 . 008Figure 5 . Impairment of pre-and post-synaptic development in Lrp4ECD/ECD;Aplp2–/– double mutant mice . Wholemount staining of triangularis sterni muscles ( E18 . 5 ) double-labeled with anti-neurofilament antibodies and anti-Syt2 antibodies ( nerve , a , d , g , j ) and α-bungarotoxin ( AChR , b , e , h , k ) . Merged images are shown in panels c , f , i and l . The inset in each panel shows a high-power view of the image . The nerve terminals and AChR clusters were markedly reduced ( both in number and size ) in Lrp4ECD/ECD;Aplp2−/− mice ( j–l ) , compared to wild type ( a–c ) , Aplp2−/− ( d–f ) or Lrp4ECD/ECD ( g–i ) . Arrowhead in the inset of l indicates a nerve terminal sprout . Scale bars: a–l: 100 µm; inset: 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00220 . 008 These results indicate that Lrp4 and the App family members , App and Aplp2 , interact genetically and functionally in the formation of NMJs and that loss of App or Aplp2 greatly enhances the synaptic defect in Lrp4ECD/ECD mice . To test , whether APP and LRP4 interact directly , analogous to the interactions of APP with LRP1 ( Kounnas et al . , 1995 ) and Apoer2 ( Hoe et al . , 2005 ) , we performed a binding analysis using recombinant APP and LRP4 fusion proteins . The LRP4 ligand-binding domain ( LBD ) was fused to maltose-binding protein ( MBP ) and the APP ECD was fused to the constant Fc region of human IgG . Figure 6A shows that specific , ectodomain-dependent interaction of both fusion proteins does indeed occur . To further corroborate these findings , we performed a cellular aggregation analysis in which we transiently co-transfected 293 cells with GFP and full length APP , or RFP and full length LRP4 plasmids , respectively . Equal numbers of red , green or both types of fluorescent cells were mixed , and the number of aggregates was scored ( Figure 6B ) . Significantly more and larger aggregates formed in the presence of APP and LRP4 than in control experiments in which cells were only transfected with GFP , RFP , APP+GFP , and LRP4+RFP , respectively ( Figure 6C , note approximately equal numbers of APP and LRP4 expressing cells in the aggregates in Figure 6B ) . Although homophilic interaction of APP has been reported ( Soba et al . , 2005 ) , no increase in the number of large clusters was detected in wells containing only APP or LRP4 expressing cells , indicating a higher affinity of APP for LRP4 than for itself . 10 . 7554/eLife . 00220 . 009Figure 6 . APP interacts with LRP4 . ( A ) APP-Fc-bead conjugates or blank beads were incubated with 100 ng purified MBP or MBP-LRP4LBD for 4 hr at 4°C prior to immunoprecipitation with anti-MBP followed by immunoblotting with anti-hFc antibody ( lane 3–7 ) . Lanes 1 and 2 show MBP ( control protein ) and MBP-LRP4LBD , respectively . ( B ) GFP- and RFP-expressing HEK293T or APP/GFP- and LRP4/RFP-expressing HEK293T cells were mixed in suspension and incubated for 30 min at 37°C before capturing cell aggregates under a fluorescent microscope . ( C ) The number of aggregates bigger than 3000 µm2 from twelve randomly selected images was determined . A significantly ( p<0 . 0001 ) increased number of aggregates were present only when APP and LRP4 expressing cells were mixed . ( D ) APP and LRP4 protein levels in the transfected HEK293T cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00220 . 009 The direct interaction between the LRP4 LBD and APP ectodomain suggested that pre- and/or post-synaptically expressed APP itself might be capable of engaging LRP4 , thereby inducing MuSK phosphorylation and thus triggering AChR clustering on its own or cooperatively with the neural form of agrin ( henceforth referred to as agrin unless stated otherwise ) . To test for this possibility , we exposed isolated myotubes to recombinant dimeric APP-Fc or RAP-Fc , an ER chaperone that binds the extracellular domains of ApoE receptors , that is LDL receptor family members ( Fisher et al . , 2006 ) including LRP4 ( Ohazama et al . , 2008 ) , with high affinity and thereby prevents the binding of most cognate ligands to this class of receptors ( Herz et al . , 1991 ) . To facilitate MuSK ligation , the Fc fusion proteins were immobilized on protein A beads and beads were incubated with isolated myotubes . Figure 7A shows that APP-Fc beads , but not RAP-Fc beads , significantly increased AChR clustering in the absence of exogenous neuronal agrin . When the same experiment was performed in the presence of a low concentration of agrin ( 0 . 1 ng/ml ) , again only APP-Fc , but not RAP-Fc , significantly increased AChR clustering . These results indicate that APP and agrin function synergistically through LRP4 to stimulate AChR clustering and NMJ formation . Interestingly , soluble APP-Fc not bound to beads was also able to promote AChR clustering , suggesting that focal clustering of APP on opposing beads is not required , but that the interaction of APP , either in a dimeric ( APP-Fc ) or monomeric ( APP-FLAG ) form ( Figure 7B ) , with the myotube surface is sufficient to induce initial clustering . The soluble LRP4-ligand binding domain ( MBP-LRP4 ) was also capable of inducing clustering , alone or together with the APP-ectodomain ( Figure 7B ) . Moreover , the failure of RAP-Fc to induce AChR clustering suggests that LRP4 dimerization alone is not sufficient , but that possibly the formation of an extracellular scaffold ( Bromann et al . , 2004 ) through interactions with other surface proteins , which can be mediated by APP and LRP4 ECD , is required . 10 . 7554/eLife . 00220 . 010Figure 7 . APP activates MuSK and promotes AChR clustering . ( A ) C2C12 myotubes were incubated with the indicated beads preloaded with APP-Fc or RAP-Fc , respectively , in the absence or presence of 0 . 1 ng/ml agrin for 24 hr prior to labeling AChRs . The number of average AChR clusters per 200 µm myotube was counted from 50 randomly captured images and normalized to control levels . ( B ) Monomeric and divalent recombinant APP fusion proteins and MBP-LRP4 , alone or in combination with APP fusion proteins , are equally effective in inducing AChR clustering . C2C12 myotubes were incubated with 1 µg/ml of soluble monovalent APP ( APP-Flag ) , divalent APP ( APP-Fc ) , or MBP-LRP4 , alone or in combination with APP fusion proteins , for 24 hr prior to labeling AChRs . ( C ) Myotubes were incubated for 30 min at 37°C with the indicated relative molar concentrations of proteins prior to immunoblotting for total MuSK and tyrosine phosphorylated MuSK ( pTyr-MuSK ) . 0 . 1 nM Agrin = 9 ng/ml , 0 . 1 nM APP-Fc = 13 ng/ml , 10 nM RAP-Fc = 720 ng/ml . ( D ) Relative amounts of pTyr-MuSK were quantified from three independent experiments . At all APP concentrations MuSK phosphorylation was significantly increased for agrin alone ( 0 . 1 nM APP , p=0 . 0003; 1 nM APP , p=0 . 0002; 10 nM APP , p=0 . 0031 ) . ( E ) APP-induced AChR clustering requires LRP4 and MuSK . Primary myotubes cultured from wild-type ( WT ) , Lrp4−/− , and Musk−/− mouse embryos ( E18 . 5 ) were incubated in the absence ( control ) or presence of APP-Fc for 24 hr prior to labeling AChRs . Statistical analysis by Student’s t-test . Scale bars , 30 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00220 . 010 These findings were further corroborated by determining the effect of APP-Fc on MuSK phosphorylation in the absence or presence of different concentrations of agrin ( Figure 7C ) . At low concentrations ( 0 . 1 nM ) agrin alone increased MuSK phosphorylation by 5 . 4-fold over baseline . RAP-Fc was used as a control and had no effect , whereas APP-Fc alone also dose-dependently increased MuSK phosphorylation , although not as effectively as agrin ( 2 . 4 , 3 . 2 and 4 . 3-fold over baseline at 1× , 10× or 100× molar concentrations , respectively , compared to 0 . 1 nM agrin alone ) . However , at low ( 0 . 1 nM ) to intermediate ( 1 nM ) concentrations APP-Fc significantly increased MuSK phosphorylation by agrin further ( Figure 7C , D ) . At higher concentration ( 10 nM ) , APP-Fc became less effective , suggesting that APP and agrin cooperatively enhance MuSK phosphorylation within a narrow physiological concentration range . As was expected from the analysis of mice genetically deficient for LRP4 or MuSK , respectively , recombinant APP-Fc failed to induce any detectable AChR clustering in LRP4 or MuSK deficient myotubes ( Figure 7E ) , indicating that APP can enhance but not bypass the essential activities of LRP4 and MuSK in NMJ formation . These observations suggested that APP itself might interact directly with agrin and that this interaction might serve to cooperatively enhance a hetero-oligomeric interaction between APP , LRP4 and agrin to maximize MuSK phosphorylation and thereby focus AChR clustering to the narrow patch of the membrane destined to harbor the emerging NMJ . To test this hypothesis , we determined whether APP , LRP4 and MuSK form a stable complex in muscle in vivo . Wild-type embryonic muscle proteins were precipitated with antibodies against LRP4 , as well as Apoer2 and LRP1 , two other LDL receptor family members that are expressed in muscle and are known to interact with APP . Remarkably , only LRP4 co-precipitated with APP and with MuSK , indicating that these three proteins are already present in muscle in a preformed stable complex ( Figure 8A ) . 10 . 7554/eLife . 00220 . 011Figure 8 . Interactions between agrin , APP and LRP4 . ( A ) APP , LRP4 and MuSK form a stable complex in muscle in vivo . Proteins were extracted from wild-type embryo ( E14 . 5 ) muscles and incubated with polyclonal anti-LRP4 , anti-Apoer2 , or anti-LRP1 antibodies overnight at 4°C followed by adsorption to Protein A Dynabeads . Supernatants were efficiently immunodepleted of LRP4 , Apoer2 and LRP1 with the respective antibodies . APP and MuSK coprecipitate with LRP4 , but not with the other LRP members Apoer2 and LRP1 , indicating the presence of a physiological complex consisting of LRP4 , APP and MuSK in muscle in vivo . ( B ) Control ( non-transfected ) , LRP4-Fc , APP-Fc , or RAP-Fc containing HEK293T supernatants were adsorbed with Protein A sepharose beads . Conjugated beads were incubated with 1 µg neural agrin in a final volume of 1 ml for 4 hr at 4°C , followed by immunoblotting for bound Fc proteins and agrin . ( C ) APP-Fc-bead conjugates or blank beads were incubated with purified 100 ng of MBP-LRP4LBD and 0–1000 ng of neural agrin for 4 hr at 4°C . The precipitate was analyzed by immunoblotting with anti-MBP , anti-agrin , and anti-Fc antibodies to detect cooperative protein interactions among LRP4 , APP , and agrin . ( D ) Control ( lanes 1–3 ) , agrin Z0 ( lanes 4–6 ) , agrin Z8 ( lanes 7–9 ) , or agrin Z19 ( lanes 10–12 ) containing culture supernatants were incubated overnight at 4°C in the absence ( lanes 1 , 4 , 7 , 10 ) or presence of RAP-Fc ( lanes 2 , 5 , 8 , 11 ) or APP-Fc ( lanes 3 , 6 , 9 , 12 ) prior to adsorption to Protein A coupled Dynabeads and magnetic isolation . Bound agrin isoforms and Fc proteins were determined by immunoblotting with anti-Myc and anti-Fc antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 00220 . 011 Next , we performed a qualitative interaction assay to determine whether agrin can interact independently with the APP ectodomain . To this end , LRP4-Fc , APP-Fc and RAP-Fc fusion proteins were prepared and incubated with a fixed concentration of agrin ( 100 ng/ml ) . LRP4-Fc efficiently bound agrin at much lower concentrations than was required to achieve comparable binding to APP-Fc ( Figure 8B ) , while RAP-Fc did not interact with agrin at all . To test whether the ability of APP to bind agrin , albeit at much lower affinity than LPR4 , translates into a cooperative enhancement of the interaction of APP with LRP4 , we incubated a fixed concentration of APP-Fc ( 100 ng/ml ) with a fixed amount of LRP4-MBP ( 100 ng/ml ) in the absence or presence of increasing amounts of agrin . As shown in Figure 8C , agrin dose-dependently enhanced the interaction of LRP4 with APP at 10 ng/ml ( 1 . 3-fold ) up to an optimal concentration ratio of 30 ng/ml ( threefold higher binding than in the absence of agrin ) . Higher concentrations of agrin potently inhibited the interaction of LRP4 with APP to 60% of baseline at 100 ng/ml and 10% at 300 ng/ml , respectively . At 1000 ng/ml , which equates to an approximately 15-fold molar excess of agrin , LRP4 binding to APP was completely abolished . This finding suggests that LRP4 and APP may bind agrin through different epitopes . Independent binding of different agrin molecules to LRP4 and APP may thus induce steric inhibition , which prevents the formation of the favored hetero-oligomeric complex consisting of LRP4 , APP , agrin and , on the surface of the myotube , MuSK . Agrin occurs in multiple splice forms ( McMahan et al . , 1992; Ferns et al . , 1992 , 1993; Burgess et al . , 2000 ) of which the Z0 form is expressed by non-neural tissues such as muscle , while Z8 and Z19 are neural-specific . To test whether APP selectively interacts with specific agrin isoforms , we performed pull-down analysis with APP-Fc . Figure 8D shows that all three agrin isoforms ( Z0 , Z8 and Z19 ) interact with APP , although Z0 binding to APP appears less robust than Z8 or Z19 does . To determine whether muscle prepatterning is affected in App–/–;Aplp2–/– double mutant mice , we analyzed E14 . 5 muscles . Consistent with previous report that AChRs clusters are prepatterned in E14 . 5 muscle ( Lin et al . , 2001 , Chen et al . , 2011 ) , AChR clusters were aligned along the central region of E14 . 5 wild-type muscle , and the majority of AChR clusters were near , but not directly apposed by presynaptic nerve terminals ( Figure 9 ) . Similarly , prepatterned AChRs were distributed along the central region of App–/–;Aplp2–/– muscle , although individual AChR clusters appear less robust compared with those in the wild-type , App–/– or Aplp2–/– muscles . These results demonstrate that pre-patterned AChR clustering is maintained in the absence of APP and APLP2 . 10 . 7554/eLife . 00220 . 012Figure 9 . AChR prepatterning in the absence of APP and APLP2 . Wholemounts of embryonic diaphragm muscles ( E14 . 5 ) from wild type ( a–c ) , App–/– ( d–f ) , Aplp2–/– ( g–i ) and App–/–;Aplp2–/– ( j–l ) embryos were double-labeled with α-bungarotoxin ( red ) and anti-syntaxin antibodies ( green ) . In wild-type and all mutant embryos , AChR clusters were confined to the central regions of diaphragm muscles indicating AChR prepatterning is independent of APP and APLP2 . However , individual AChR clusters appear less robust in App–/–;Aplp2–/– muscle compared with those in the wild-type , App–/– or Aplp2–/– muscles . Scale bar: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00220 . 012 We next asked if removing membrane anchoring of LRP4 ( as in Lrp4ECD/ECD mice ) or deleting App and Aplp2 ( as in App/Aplp2 double mutant mice ) might affect LRP4 protein distribution in muscles . We generated anti-LRP4 antibodies against the extracellular domain of LRP4 and performed immunofluorescence staining on embryonic muscles . As expected , no LRP4 labeling was detected in Lrp4–/– muscle ( Figure 10A , k; Figure 10C , b ) . By contrast , specific LRP4 staining was clearly visible at the NMJ in E18 . 5 WT muscle ( Figure 10A , h ) , although the staining was markedly weaker at earlier developmental stages ( E14 . 5 [Figure 10A , b] and E16 . 5 [Figure 10A , e] ) . Similarly , APP was also localized at the NMJ in E18 . 5 WT muscle ( Figure 10B ) . In Lrp4ECD/ECD muscle ( Figure 10C , e ) , we detected LRP4 staining at the NMJ , but at reduced levels compared with WT muscle ( Figure 10D ) . Strikingly , LRP4 staining was also reduced in App–/–;Aplp2–/– double mutant muscle ( Figure 10C , h ) , compared to WT muscle ( Figure 10D ) . In adult muscle ( hindlimb , 10-week-old ) , higher LRP4 protein ( Figure 10E ) and mRNA ( Figure 10F ) expression were detected in Lrp4ECD/ECD compared to WT muscle . Taken together , these data indicate a decreased LRP4 level at the NMJ in Lrp4ECD/ECD and App–/–;Aplp2–/– embryonic muscles . 10 . 7554/eLife . 00220 . 013Figure 10 . LRP4 localization in Lrp4ECD/ECD and App–/–;Aplp2–/– mutant mice . ( A ) Cross sections of hindlimb muscle from E14 . 5 ( a–c ) , E16 . 5 ( d–f ) and E18 . 5 ( g–i ) wild type ( WT ) and E18 . 5 Lrp4–/– ( j–l ) mice were double-labeled with anti-LRP4 antibodies ( green ) and Texas-Red conjugated α-bungarotoxin ( red ) , which marks the site of the NMJ ( arrowhead in a , d , g ) . LRP4 staining appears at low level in E14 . 5 WT and E16 . 5 WT muscle ( compared with Lrp4–/– muscle ) , and become highly concentrated at the NMJ in E18 . 5 WT muscle ( arrowhead in h ) . Scale bar in A: 20 µm . ( B ) Localization of APP at the NMJ . Wholemount diaphragm muscle from E18 . 5 WT mice was double-labeled with anti-APP antibodies ( b ) and Texas-Red conjugated α-bungarotoxin ( a ) . APP ( arrowhead in b ) is localized at the NMJ ( arrowhead in a ) , as shown in the merged image in c . Scale bar in B: 10 µm . ( C ) Cross sections of hindlimb muscles from Lrp4–/– ( a–c ) , Lrp4ECD/ECD ( d–f ) and App–/–;Aplp2–/– ( g–i ) mice ( E18 . 5 ) were double-labeled with anti-LRP4 antibodies ( green ) and α-bungarotoxin ( red ) . LRP4 was detected at the NMJ in Lrp4ECD/ECD ( arrowhead in e ) and App–/–;Aplp2–/– ( arrowhead in h ) , but not in Lrp4–/– muscle ( b ) . LRP4 expression appeared diffused in Lrp4ECD/ECD and in App–/–;Aplp2–/– muscles compared with age-matched WT muscle ( see A ) . Scale bar in C: 30 μm . ( D ) Quantification of relative fluorescence intensity for LRP4 immunostaining on hindlimb myofibers from WT , Lrp4ECD/ECD and App–/–;Aplp2–/– mice ( E18 . 5 ) . Gray value of LRP4 staining at the NMJ ( defined as signal ) and within the sarcoplasm ( defined as background ) of the same myofiber was separately measured using ImageJ . The ratio of signal to background was then calculated for each individual myofiber . The bar graph shows significant ( p<0 . 05 ) decreases in LRP4 staining in Lrp4ECD/ECD ( 1 . 38 ± 0 . 05 , n = 23 myofibers ) and App−/−;Aplp2−/− ( 1 . 33 ± 0 . 05 , n = 22 myofibers ) , compared with WT muscle ( 1 . 78 ± 0 . 18 , n = 19 myofibers ) . ( E ) Anti-LRP4 and beta-actin ( loading control ) immunoblot shows increased level of LRP4 protein expression in Lrp4ECD/ECD and Lrp4WT/ECD muscle , compared with WT muscle ( from the hindlimb muscles of 10-week-old mice ) . ( F ) Lrp4 mRNA expression in hindlimb muscles ( 10-week-old mice ) was determined by quantitative PCR . Levels of Lrp4 mRNA were normalized to cyclophilin mRNA levels . Results are shown as average ± SD of triplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 00220 . 013 Our results have revealed previously unrecognized interactions at the NMJ between LRP4 and APP on the one hand and APP and agrin on the other . Genetic epistasis experiments show that LRP4 and APP functionally interact to regulate NMJ development and maintenance . APP may accomplish this through several independent mechanisms: first , APP interacts directly with LRP4 and thereby is capable of activating MuSK ( at a low level and independent of agrin ) by ligating and clustering LRP4 . Second , APP can also directly bind to agrin , albeit at lower affinity than LRP4 . However , agrin cooperatively increases the stoichiometric interaction of APP with LRP4 , which may serve to focus MuSK activation and thereby prevent aberrant AChR clustering . Third , APP and its homologues , which are expressed pre- and post-synaptically ( Wang et al . , 2005 ) , bind homomerically and heteromerically to each other ( Soba et al . , 2005 ) , thereby generating or strengthening attachment sites for nerve terminals ( Torroja et al . , 1999; Akaaboune et al . , 2000; Soba et al . , 2005; Wang et al . , 2005 ) . By interacting simultaneously with LRP4 and agrin , APP and potentially APP-like proteins such as APLP2 , which also participates in NMJ formation ( Wang et al . , 2005 ) , would further serve to recruit all components necessary for AChR clustering into an area of maximal density , thereby ensuring that AChR clustering is restricted in vivo to the site of nerve contact . A rendition of this model is shown in Figure 11A . 10 . 7554/eLife . 00220 . 014Figure 11 . Hypothetical model of the interactions of APP/APLP2 , LRP4 , agrin and MuSK during NMJ formation . ( A ) Our genetic , biochemical and functional results suggest that APP interacts with LRP4 and agrin to regulate NMJ formation . APP–LRP4 interaction in the absence of agrin may be able to activate MuSK signaling ( Figure 7C ) at low levels to promote AChR clustering on the muscle fiber membrane , raising the possibility that nerve and muscle-derived APP and APLP2 homo- or heterodimerization may be able to cooperatively promote AChR clustering in the absence of agrin , which provides a potential mechanism for the earlier observation that initial AChR clustering is independent of agrin ( Lin et al . , 2001; Yang et al . , 2001 ) . Neural agrin ( red ) enhances the APP–LRP4 interaction and strongly activates MuSK ( Figures 7C and 8C ) , which is consistent with the role of agrin in the stabilization and maintenance of NMJs . Cooperative binding of APP and agrin to LRP4 ( Figure 8C ) would further strengthen LRP4–APP interaction on muscles , promoting synaptic differentiation and postnatal maintenance of NMJs ( solid line with arrow in muscle fiber cytoplasm ) . By contrast , muscle agrin ( gray ) , which cannot activate MuSK , would compete with decreasing concentrations of diffusible neural agrin at increasing distance from the NMJ and thus keep AChR clustering outside the nerve contact site suppressed . ( B ) Loss of the LRP4 membrane anchor results in a secreted ectodomain that remains partially functional through bivalent interactions with MuSK ( Zhang et al . , 2008; Kim et al . , 2008b ) and APP , resulting in reduced signaling ( dashed line with arrow ) and thus a hypomorphic NMJ developmental defect compatible with embryonic NMJ formation , but impaired postnatal maturation and maintenance of the NMJ . DOI: http://dx . doi . org/10 . 7554/eLife . 00220 . 014 We have shown that LRP4 ECD is sufficient to initiate NMJ formation at embryonic stages but not sufficient for postnatal maturation , which requires a progressive increase of MuSK activation . The secreted LRP4 ectodomain that is encoded by our LRP4 ECD allele is presumably retained at the cell surface by simultaneous interactions with APP ( possibly also APLPs ) , MuSK and agrin , but at levels that are insufficient to sustain the signal strength required for postnatal expansion of the synapse ( Figure 11B ) . Also supporting this model is the finding that initial AChR clustering is independent of neuronal agrin ( Lin et al . , 2001; Yang et al . , 2001 ) and that rudimentary NMJs form in agrin-deficient mice ( Gautam et al . , 1996 ) , which suggested the existence of a second nerve derived organizing signal . By contrast , NMJs completely fail to form in Musk ( DeChiara et al . , 1996 ) and Lrp4 ( Weatherbee et al . , 2006 ) knockout mice , indicating an absolute requirement for MuSK and LRP4 . Our finding that APP-Fc , but not RAP-Fc , can interact with LRP4 and induce MuSK phosphorylation at low level independent of agrin is consistent with both observations . Interestingly , however , and in contrast to our results with LRP4 , the intracellular domain of APP appears to be required for normal NMJ development ( Li et al . , 2010 ) . This suggests a role of these highly conserved C-terminal sequences in either signaling with APP serving as a co-receptor in the MuSK signaling complex , or a role in regulating the trafficking of all or part of the components of this complex . It is conceivable that promotion of AChR clustering by homo- or heteromeric interaction of APP with itself or APLP2 , and through inclusion of LRP4 and MuSK , requires an additional protein that interacts with the APP cytoplasmic domain , and that this requirement resembles the need for the interaction of rapsyn with the AChR and MuSK ( Bromann et al . , 2004 ) . In their study , Bromann and colleagues further showed that immobilized , but not soluble , agrin is able to aggregate MuSK and promote AChR clustering independent of MuSK’s kinase activity , suggesting that MuSK scaffolding plays an important role in AChR clustering . Our findings show that APP is independently capable of promoting AChR clustering , and that it further enhances clustering in the presence of agrin . Moreover , agrin potentiates the interaction of APP with LRP4 . Thus , it is conceivable that APP cooperates with agrin and LRP4 to establish such an extracellular scaffold to promote MuSK aggregation and AChR clustering . AChR clustering can also be mediated by a variety of agrin independent mechanisms , including treatment with neuraminidase and VVA lectin ( Martin and Sanes , 1995; Grow et al . , 1999a , b ) and by defucosylation of muscle agrin , which unmask its ability to induce MuSK phosphorylation ( Kim et al . , 2008a ) . In addition , we have also obtained further evidence showing functional interaction of LRP4 , APP and agrin during NMJ formation . We have analyzed mutant mice that lack the App family member Aplp2 on the background of homozygous LRP4 ECD ( Lrp4ECD/ECD;Aplp2–/– ) . Like App , Aplp2 is also required for normal NMJ formation , and mice lacking both App family members exhibit aberrant apposition of presynaptic marker proteins with postsynaptic acetylcholine receptors and excessive nerve terminal sprouting and a reduced number of synaptic vesicles at presynaptic terminals ( Wang et al . , 2005 ) . This phenotype is consistent with the role of App family members in mediating trans-synaptic adhesion functions ( Wang et al . , 2009 ) through homo- and hetero-oligomeric interactions ( Soba et al . , 2005 ) . Loss of both APP and APLP2 , but not of either protein alone , might critically reduce this trans-synaptic interaction ( Figure 11 ) , resulting in aberrant apposition of nerve terminals and AChR clusters . We did not observe such an aberrant apposition in our Lrp4ECD/ECD;App−/− model or in Lrp4ECD/ECD;Aplp2−/− mice . However , as in Lrp4ECD/ECD;App−/− , functional NMJ formation , as evidenced by greatly impaired postnatal survival ( Table 2 ) and AChR clustering ( Figure 5 ) , is also significantly compromised in the Lrp4ECD/ECD;Aplp2–/– double mutants . This would suggest that while APP and APLP2 are important for establishing trans-synaptic contact , LRP4 rather serves , through a hetero-oligomeric interaction with APP , MuSK and likely APLP2 , to integrate agrin through cooperative binding into a spatially restricted signaling complex that focuses the molecular instructions required for AChR clustering at the surface of the muscle fiber . 10 . 7554/eLife . 00220 . 015Table 2 . Survival analysis of Lrp4ECD/ECD;Aplp2−/− double mutant pupsDOI: http://dx . doi . org/10 . 7554/eLife . 00220 . 015GenotypeExpectedObservedChi-square testLrp4+/ECD;Aplp2+/−6882n . s . Lrp4+/ECD;Aplp2−/−6975n . s . Lrp4ECD/ECD;Aplp2+/−5851n . s . Lrp4ECD/ECD;Aplp2−/−5417n . s . , not significant . Significant at p<0 . 01 . Pups were genotyped at weaning ( ∼25 days of age ) to determine postnatal survival rates . This oligomeric interaction of the LRP4 ECD may explain the residual function of the nonmembrane-associated soluble receptor in the LRP4 ECD mutants . While membrane anchoring firmly restricts LRP4 to the two-dimensional surface of the muscle fiber and thereby ensures its continued presence in the MuSK/agrin/APP/APLP2 signaling complex , the loss of the transmembrane segment can be partially compensated for by the combined interactions of APP , APLP2 , MuSK and agrin with the LRP4 ECD . The reduced signal strength that results in impaired NMJ size and number can be explained by the inevitable reduction of LRP4 ectodomain at the NMJ and relative dilution of secreted ECD throughout the muscle surface and the interstitial space ( Figure 10 ) . This in turn is partially compensated for by the increased expression of LRP4 ECD in the heterozygous and homozygous mutants ( Figure 10E , F ) . We have further found that the non-neural Z0 as well as the neural Z8 and Z19 splice forms can bind to APP ( Figure 8D ) . Interaction of Z0 ( muscle ) agrin , which cannot efficiently promote AChR clustering at physiological levels ( Kim et al . , 2008a ) , may have nevertheless physiological importance in this context by negatively regulating the recruitment of the MuSK activating , neural agrin isoforms outside of nerve contact sites , where their concentrations would be reduced . In doing so , muscle agrin would serve to potently suppress the formation of aberrant signaling complexes by blocking LRP4 and APP family proteins with the inactive form . This inhibition would be overcome by the higher concentrations of neural agrin at the nerve terminals . On another note , we have previously shown that LRP4 can interact with several modulators of the Wnt pathways , that is Wise , Dkk , and Sost ( Dietrich et al . , 2010; Karner et al . , 2010 ) . These modulators bind to the homologous β-propeller domain in LRP5 . It is intriguing to speculate that these interactions can also contribute to the modulation of NMJ development or function . In summary , the findings we have reported here reveal new insights into the molecular mechanisms that govern the formation of the neuromuscular synapse and that are consistent with the recent findings from the Burden and Mei laboratories ( Wu et al . , 2012; Yumoto et al . , 2012; Zong et al . , 2012 ) . Furthermore , they raise interesting new questions about the interplay of agrin , MuSK , APP and LRP4 in the CNS where all these proteins are also expressed . There they might , in principle , work together in an analogous or similar manner to regulate synaptogenesis and promote neuronal survival . LRP4 belongs to an evolutionarily conserved class of ApoE receptors with essential functions in the regulation of neurotransmission at glutamatergic synapses ( Herz and Chen , 2006 ) . We have recently shown that ApoE in an AD-associated , isoform-specific manner impairs ApoE receptor recycling at central synapses ( Chen et al . , 2010b ) and that this impairs the ability of ApoE receptor-dependent signals to prevent β-amyloid induced synaptic suppression ( Durakoglugil et al . , 2009 ) . The findings we have reported here , together with those by Wang et al . , who have proposed that perturbed APP synaptic adhesion activity may contribute to synaptic dysfunction and AD pathogenesis ( Wang et al . , 2009 ) , reaffirm the NMJ as a versatile and useful experimental model system on which the cell biology of ApoE , APP and ApoE receptor trafficking and its pathobiology at peripheral and central synapses can be investigated . The generation of App−/− ( Zheng et al . , 1995 ) and Lrp4ECD/ECD ( Johnson et al . , 2005 ) mice has been reported . Lrp4ΔICD/ΔICD mice were generated using the same targeting construct , except that sequences containing the LRP4 transmembrane segment followed by a Myc-epitope were introduced in place of the stop codon ( Johnson et al . , 2006 ) . The Lrp4ΔICD/ΔICD knockin mice were generated using a replacement vector based on the construct described for the LRP4 ECD mutant . The same long arm of homology , but lacking the introduced premature stop codon , was used . A cDNA insert expressing the transmembrane segment followed by a Myc epitope was cloned using an upstream Bst1107I site in the long arm and a BsrGI site in the bovine growth hormone 3′UTR . The oligonucleotide KI5 ( 5′-GTATACTGCTGATTTTGTTGGTGATCGCGGCTTTG-3′ ) was used as the 5′ primer and MEJ380 ( 5′-GTGTGTTGTACATCAGCTATTCAGATCCTCTTCTGAGATGAGTTTTTGTTCCTTGGATTTCCTGTGTCTGTATAGCATCAAAG-3′ ) was used as the 3′ primer to amplify the cDNA insert for the TM-Myc Epitope cassette . Lrp4 null mice were generated as described in Figure 1 by replacing the first exon with a neomycin resistance cassette . The long arm of homology upstream of the first exon of Lrp4 was generated by PCR amplification using the primers MEJ23 ( 5′-GCGGCCGCCAGGTCATGAAGTGAGTGCTGAGCCACTGGG-3′ ) and MEJ24 ( 5′-CCACCACCGCCTCATGGTGCTGCGGCCGCC-3′ ) . The short arm of homology downstream of the first exon of Lrp4 was generated by PCR amplification using the primers MEJ33 ( 5′-CTCGAGGAGCGGTCTGCAGATCCTGGCGATTCACGG-3′ ) and MEJ35 ( 5′-CTCGAGGGTTACAGACTCTGCAACTGCTCTACCTCATTG-3′ ) . The long arm and short arm of homology were cloned into pJB1 using the NotI and XhoI restriction sites , respectively . Animals were maintained on a mixed 129SvEv Bradley;C57BL/6J background by heterozygous intercrossing . Wild-type ( WT ) control mice were obtained from the same crosses . Animals were maintained on 12-hr light/12-hr dark cycles and fed a standard rodent chow diet ( Diet 7001; Harlan Teklad , Madison , WI ) and water ad libitum . No sexual dimorphism of phenotype was observed . All procedures were performed in accordance with the protocols approved by the Institutional Animal Care and Use Committee of the University of Texas Southwestern Medical Center at Dallas . A cDNA encoding the LRP4 ligand binding domain ( LBD , aa 27–349 , accession number CAM24075 ) was inserted downstream of the maltose-binding protein ( MBP ) coding sequence into pMAL-p4x ( NEB ) . pMAL-p4x and pMAL-LRP4LBD-p4x constructs were expressed in Escherichia coli ( BL21 ) to produce properly folded MBP and MBP-LRP4LBD fusion protein in the periplasm . Periplasmic extracts were subjected to amylose column ( NEB , Ipswich , MA ) to obtain affinity-purified MBP and MBP-LRP4LBD . Fc fusion proteins were generated by fusing the extracellular domains of mouse LRP4 ( residues 1–1650 , LRP4-Fc ) , APP695 ( residues 1–596 , APP-Fc ) , or full length RAP ( RAP-Fc ) to the constant region of human IgG ( Fc ) . Secreted fusion proteins were produced by transfecting HEK293A cells with pCDNA3 . 1-LRP4-Fc , pCDNA3 . 1-APP-Fc , or pCDNA3 . 1-RAP-Fc constructs using FuGENE 6 ( Roche , Indianapolis , IN ) . Fc fusion proteins secreted into media were collected and purified on Protein A-Sepharose columns ( Sigma , St . Louis , MO ) . APP-FLAG was generated by fusing 3xFLAG in place of Fc to the carboxyl-terminus of APP-695 ( residues 1–596 ) in pCDNA3 . 1 . To investigate binding between LRP4 and APP , 100 ng of purified APP-Fc was incubated with Protein A-agarose beads ( Sigma ) in phosphate-buffered saline ( PBS ) containing 0 . 1% bovine serum albumin ( BSA ) for 4 hr at 4°C to generate Fc-agarose conjugates . The conjugates were then incubated with 100 ng of purified MBP or MBP-LRP4 LBD in PBS containing 0 . 1% BSA for 4 hr at 4°C . Protein coprecipitated with APP-Fc was detected by Western blotting with anti-MBP antibody ( NEB ) , and the membrane was reprobed with anti-human IgG ( Fc specific ) antibody ( Sigma ) to determine the levels of APP-Fc . To investigate binding between agrin and LRP4 or APP , 1 . 5 ml of LRP4-Fc , APP-Fc , or RAP-Fc conditioned medium was incubated with Protein A-agarose beads for 4 hr at 4°C to generate Fc-agarose conjugates . The conjugates were then incubated with 1 µg of neuronal agrin ( R&D Systems ) in Dulbecco’s modified Eagle’s medium ( DMEM ) containing 0 . 1% BSA for 4 hr at 4°C . Agrin coprecipitated with Fc fusion protein was detected by Western blotting with anti-agrin antibody ( R&D Systems ) . The amount of Fc fusion protein present in the reaction was determined by reprobing the membrane with an anti-Fc antibody . To investigate the role of agrin in LRP4–APP interaction , 100 ng of purified APP-Fc were conjugated to Protein A-agarose beads in PBS containing 0 . 1% BSA for 4 hr at 4°C . The conjugates were then incubated with 100 ng of purified MBP-LRP4LBD and 0–1000 ng of neuronal agrin in PBS containing 0 . 1% BSA for 4 hr at 4°C . The precipitates were then subjected to anti-MBP , anti-Agrin , or anti-Fc immunoblotting . To investigate the differential interactions of Agrin isoforms with APP , we used constructs encoding the C-terminal 110 kDa fragments of agrin isoforms Z0 , Z8 and Z19 containing N-terminal Flag tag and C-terminal Myc and 6xHis tags , respectively ( Burgess et al . , 2000; Bogdanik and Burgess , 2011 ) . These plasmids were generously provided by Dr Robert Burgess ( Jackson Laboratory , Bar Harbor , ME ) . Agrin fragments were produced as secreted proteins in HEK293A cells grown in DMEM containing 0 . 1% BSA . 100 ng of purified APP-Fc or RAP-Fc was incubated with 300 µl of agrin containing supernatants overnight at 4°C followed by adsorption to Protein A Dynabeads ( Invitrogen , Grand Island , NY ) for 10 min at room temperature . After magnetic isolation , bound proteins and supernatants were analyzed by Western blotting using an anti-Myc antibody ( 9E10 ) . Equal binding of the respective Fc proteins to the Dynabeads was detected by Western blotting using an anti-human IgG ( Fc specific ) antibody ( Sigma , St . Louis , MO ) . Hind limb muscle proteins in WT embryo ( E14 . 5 ) were extracted in lysis buffer ( PBS supplemented with 5 mM EDTA , 5 mM EGTA , 1% digitonin , and completed protease inhibitor tablet [Roche] ) . After centrifugation at 18 , 000×g for 5 min , the supernatant was incubated with 10 μl of an affinity purified anti-LRP4 , anti-ApoER2 , or anti-LRP1 antibody overnight at 4°C . Each antibody and bound proteins were absorbed to Protein A Dynabeads ( Invitrogen ) for 10 min at room temperature . After magnetic isolation , bound proteins and supernatants were analyzed by immunoblotting . HEK293T cells were co-transfected for 48 hr with a 3:1 ratio of pDsRed:pCDNA3 . 1-LRP4 full-length constructs or of pEGFP:pCDNA3 . 1-APP full-length constructs to generate LRP4-expressing red-fluorescent and APP-expressing green-fluorescent cells , respectively . HEK293T cells transfected with pDsRed or pEGFP vector were used as control cells: red- or green-fluorescent cells . Cells were incubated with 0 . 05% Trypsin/0 . 53 mM EDTA in calcium- and magnesium-free Hank’s balanced salt solution ( HBSS ) for 5 min followed by trituration , washed twice with HBSS and resuspended in aggregation buffer ( 0 . 1g glucose , 0 . 1 g BSA , 0 . 26 g HEPES , 13 . 75 mg CaCl2 , and 10 mg DNase I in 100 ml HBSS ) . LRP4 and APP expression was determined from single cell suspensions by immunoblotting . LRP4-expressing red-fluorescent and APP-expressing green-fluorescent cells or red- and green-fluorescent cells were mixed and mutated at 37°C for 30 min in 16-mm wells precoated with 1% BSA . Aggregation was stopped by fixation with 2 . 5% glutaraldehyde . Cell suspensions were drop-plated and covered with coverslips . Fluorescent cell images were captured with a Zeiss 10x/0 . 30 NA dry objective on a Zeiss Axioplan 2 microscope . Aggregates of >3000 µm2 in random fields were scored . Morphological analysis of the NMJ was carried out in both diaphragm and triangularis sterni muscles using procedures described previously ( Liu et al . , 2008 ) . Briefly , diaphragm ( E14 . 5 , E16 . 5 , E18 . 5 or P12 ) or triangularis sterni ( E18 . 5 , P12 or 3-month-old ) muscles were dissected out , fixed with 2% paraformaldehyde in 0 . 1 M phosphate buffer ( pH 7 . 3 ) overnight at 4°C , washed thoroughly with PBS , and incubated with 0 . 1 M glycine in PBS for 30 min . Samples were incubated with Texas Red-conjugated α-bungarotoxin ( 2 nM , Molecular Probes ) in antibody dilution buffer ( 0 . 01 M phosphate buffer , 500 mM NaCl , 3% BSA and 0 . 01% thimerosal ) for 30 min at room temperature to label postsynaptic AChR , washed with PBS and incubated with rabbit polyclonal anti-neurofilament ( 1:1500; Chemicon , Billerica , MA ) and rabbit polyclonal anti-synaptotagmin 2 ( 1:1000 , a generous gift from Dr Thomas Südhof , Stanford University School of Medicine , Palo Alto , CA ) antibodies diluted in antibody dilution buffer overnight at 4°C to stain presynaptic motor axons . For sections , samples were incubated in antibodies against LRP4 ( rabbit polyclonal , 1:100 ) or APP ( rabbit polyclonal , 1:500 ) overnight at room temperature . After washing three times with 0 . 5% Triton X-100 in PBS , samples were incubated with FITC-conjugated anti-rabbit IgG overnight at 4°C , washed in PBS , and mounted in Vectashield mounting medium . Fluorescent images were captured using a Zeiss LSM 510 confocal microscope . Quantitative measurements of AChR cluster number and size were made using NIH ImageJ software . For the analysis of AChR cluster number , Texas Red-labeled AChR clusters on the same right ventral region of each diaphragm were counted . For the analysis of AChR size , images acquired at high magnification were used . To determine LRP4 distribution , hindlimb muscles from E14 . 5 , E16 . 5 and E18 . 5 embryos were transversely sectioned ( 20 μm ) . Sections were incubated with 2 nM of Texas Red-conjugated α-bungarotoxin for 30 min at room temperature to label postsynaptic AChR . Then , sections were incubated with rabbit polyclonal antibody against LRP4 extracellular domain ( 1:100 ) overnight at 4°C . After washing with 0 . 1% Triton X-100 in PBS , the samples were incubated with fluorescein isothiocyanate-conjugated anti-rabbit IgG for 2 hr at room temperature , washed with PBS and mounted in Vectashield mounting medium . Images were acquired under an upright fluorescence microscope ( Olympus BX51 ) using a Hamamatsu ORCA-285 camera , and fluorescence intensity ( mean gray value ) was measured using NIH ImageJ . Images were captured using a Zeiss LSM 510 META confocal microscope with a Zeiss 40x/1 . 3 NA oil immersion objective . Excitation wavelengths used for the red and green channels were 543 nm ( HeNe1 Laser ) and 488 nm ( Argon Laser ) , respectively . Z-stacks were acquired at an optical slice interval of 0 . 6 μm . The Reuse feature from the Zeiss LSM acquisition software allowed capturing of all images at the same laser power , detector gain and amplifier offset for all specimens . 3D reconstruction of the Z-stack images was done using Imaris software ( Bitplane Inc , MN ) . Initial image analysis included applying the background subtraction algorithm in Imaris . The intensity for the green and red channels was adjusted in Blend mode during volume rendering and an Iso Surface model was created for the background-subtracted images . Ultrastructural analysis of the NMJ was carried out as previously described ( Liu et al . , 2009; Chen et al . , 2010a ) . Deeply anesthetized animals were fixed via cardiac perfusion with a mixture of 1% glutaraldehyde and 4% paraformaldehyde in 0 . 1 M phosphate buffer , pH 7 . 4 . Lumbrical muscles were dissected and post-fixed in the same solution overnight at 4°C . Tissues were then rinsed with the 0 . 1 M phosphate buffer , trimmed to small pieces and post-fixed with 1% osmium tetroxide for 3 hr on ice . Tissues were then dehydrated in a graded series of ethanol , infiltrated and polymerized in Epon 812 ( Polysciences , Warrington , PA ) . Epon blocks were cut at 1 μm and stained with toluidine blue for light microscopic observation . Ultrathin ( 70 nm ) sections were mounted on Formvar-coated grids and stained with uranyl acetate and lead citrate . Electron micrographs were acquired using a Tecnai ( Netherlands ) electron microscope operated at 120 kV . Mouse C2C12 cells maintained as undifferentiated myoblasts in DMEM supplemented with 20% fetal bovine serum and 0 . 5% chick embryo extract were seeded on 8-well plastic chamber slides ( Nalge Nunc International , Rochester , NY ) and grown to ∼90% confluence . Myoblasts were fused into myotubes in differentiation medium ( DMEM supplemented with 2% horse serum ) for 3 days . Myotubes were incubated with 300 μl differentiation medium containing 3 μl Protein A-agarose beads ( 50% slurry in PBS containing 0 . 1% BSA ) conjugated without or with saturating amounts of APP-Fc or RAP-Fc for 24 hr in the absence or presence of 0 . 1 ng/ml agrin . At the end of the incubation , myotubes were fixed with 1% paraformaldehyde in 0 . 1 M phosphate buffer for 30 min , washed with PBS and incubated with 2 nM of Texas-Red conjugated α-bungarotoxin ( Molecular Probes , Grand Island , NY ) for 30 min to label AChRs . Myotubes were then washed with PBS and mounted in Vectashield mounting medium . For each sample , 50 fluorescent images were randomly captured using a Hamamatsu ORCA-285 camera under 400× magnification . A cluster with a length >2 μm was defined as a patch of AChRs . All the myotubes that had AChR clusters in the fields were analyzed blindly , and the size and number of AChR clusters/200 µm myotube length were measured using NIH ImageJ software and normalized to control levels . Muscles were quickly removed from the limbs of WT , Lrp4−/− , and Musk−/− embryos ( E18 . 5 ) in Hanks’ Balanced Salt Solution ( Invitrogen ) . Muscles were minced into small pieces before enzymatic digestion for 45 min at 37°C in HBSS supplemented with 2 mg/ml of Type II collagenase ( Worthington ) and 0 . 4 mg/ml of DNAse l ( Worthington , Lakewood , NJ ) . Dissociated muscle cells were collected by centrifugation at 2000×g for 5 min , resuspended in culture medium ( DMEM supplemented with 10% horse serum , 5% fetal calf serum and 1% chick embryo extracts ) , and filtered through a Cell Strainer with diameter of 100 µm ( BD Falcon , San Jose , CA ) . Cells were then seeded on a 100-mm culture dish and incubated in a 37°C , 5% CO2 incubator for 2 hr in order to attach and remove fibroblasts . Suspended myoblasts were collected by centrifugation at 2000×g for 5 min and seeded on 8-well plastic chamber slides ( Fisher , Pittsburgh , PA ) . After growing cells in the culture medium until approximately 90% confluence , the culture medium was replaced with differentiation medium ( DMEM supplemented with 2% horse serum ) , and this medium was changed every day for 3 days to fuse myoblasts into myotubes for the AChR clustering assay as described . Differentiated myotubes were incubated with indicated concentrations of agrin , APP-Fc , or RAP-Fc for 30 min at 37°C . The myotubes were rinsed once on ice with PBS containing 1 mM sodium orthovanadate and 50 mM sodium fluoride and extracted in lysis buffer ( 30 mM triethanolamine , 1% NP-40 , 50 mM NaCl , 5 mM EDTA , 5 mM EGTA , 50 mM sodium fluoride , 2 mM sodium orthovanadate , 1 mM sodium tetrathionate , 1 mM N-ethylmaleimide , 10 µM Pepstatin , 0 . 5 mg/ml Pefabloc , and complete protease inhibitor tablet [Roche] ) . Lysates were centrifuged at 18 , 000×g for 5 min and the supernatant was used as total myotube extracts . To immunoprecipitate MuSK , the extracts were incubated overnight at 4°C with anti-MuSK antibody ( Abcam , Cambridge , MA ) followed by precipitation of the antibody with Protein A/G PLUS-Agarose ( Santa Cruz , Dallas , TX ) . Precipitated MuSK was separated by SDS-PAGE and the phosphorylation ( activation ) status of MuSK was determined by Western blotting using anti-phosphotyrosine antibody 4 G10 Platinum ( Millipore , Billerica , MA ) . Molarities are stated for monomers . Data are presented as mean ± standard error of the mean ( SEM ) . Statistical analyses of differences between or among control , mutant and treatment groups were carried out using Student’s t test or one-way analysis of variance ( ANOVA ) . Post hoc test was carried out by using Tukey analyses when a significant F-value was obtained in ANOVA . A p value of <0 . 05 was considered to be significant .	One of the hallmarks of Alzheimer’s disease is the formation of plaques in the brain by a protein called β-amyloid . This protein is generated by the cleavage of a precursor protein , and mutations in the gene that encodes amyloid precursor protein greatly increase the risk of developing a familial , early-onset form of Alzheimer’s disease in middle age . Individuals with a particular variant of a lipoprotein called ApoE ( ApoE4 ) are also more likely to develop Alzheimer’s disease at a younger age than the rest of the population . Due to its prevalence—approximately 20% of the world’s population are carriers of at least one allele—ApoE4 is the single-most important risk factor for the late-onset form of Alzheimer’s disease . Amyloid precursor protein and the receptors for ApoE—in particular one called LRP4—are also essential for the development of the specialized synapse that forms between motor neurons and muscles . However , the mechanisms by which they , individually or together , contribute to the formation of these neuromuscular junctions are incompletely understood . Now , Choi et al . have shown that amyloid precursor protein and LRP4 interact at the developing neuromuscular junction . A protein called agrin , which is produced by motor neurons and which must bind to LRP4 to induce neuromuscular junction formation , also binds directly to amyloid precursor protein . This latter interaction leads to the formation of a complex between LRP4 and amyloid precursor protein that robustly promotes the formation of the neuromuscular junction . Mutations that remove the part of LRP4 that anchors it to the cell membrane weaken this complex and thus reduce the development of neuromuscular junctions in mice , especially if the animals also lack amyloid precursor protein . These three proteins thus seem to influence the development and maintenance of neuromuscular junctions by regulating the activity of a fourth protein , called MuSK , which is present on the surface of muscle cells . Activation of MuSK by agrin bound to LRP4 promotes the clustering of acetylcholine receptors in the membrane , which is a crucial step in the formation of the neuromuscular junction . Intriguingly , Choi et al . have now shown that amyloid precursor protein can , by interacting directly with LRP4 , also activate MuSK even in the absence of agrin , albeit only to a small extent . The work of Choi et al . suggests that the complex formed between agrin , amyloid precursor protein and LRP4 helps to focus the activation of MuSK , and thus the clustering of acetylcholine receptors , to the site of the developing neuromuscular junction . Since all four proteins are also found in the central nervous system , similar processes might well be at work during the development and maintenance of synapses in the brain . Further studies of these interactions , both at the neuromuscular junction and in the brain , should shed new light on both normal synapse formation and the synaptic dysfunction that is seen in Alzheimer’s disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2013	APP interacts with LRP4 and agrin to coordinate the development of the neuromuscular junction in mice
Olfactory receptor usage is highly heterogeneous , with some receptor types being orders of magnitude more abundant than others . We propose an explanation for this striking fact: the receptor distribution is tuned to maximally represent information about the olfactory environment in a regime of efficient coding that is sensitive to the global context of correlated sensor responses . This model predicts that in mammals , where olfactory sensory neurons are replaced regularly , receptor abundances should continuously adapt to odor statistics . Experimentally , increased exposure to odorants leads variously , but reproducibly , to increased , decreased , or unchanged abundances of different activated receptors . We demonstrate that this diversity of effects is required for efficient coding when sensors are broadly correlated , and provide an algorithm for predicting which olfactory receptors should increase or decrease in abundance following specific environmental changes . Finally , we give simple dynamical rules for neural birth and death processes that might underlie this adaptation . In vertebrates , axons from olfactory neurons converge in the olfactory bulb on compact structures called glomeruli , where they form synapses with dendrites of downstream neurons ( Hildebrand and Shepherd , 1997 ) ; see Figure 1a . To good approximation , each glomerulus receives axons from only one type of OSN , and all OSNs expressing the same receptor type converge onto a small number of glomeruli , on average about two in mice to about 16 in humans ( Maresh et al . , 2008 ) . Similar architectures can be found in insects ( Vosshall et al . , 2000 ) . The anatomy shows that in insects and vertebrates , olfactory information passed to the brain can be summarized by activity in the glomeruli . We treat this activity in a firing-rate approximation , which allows us to use available receptor affinity data ( Hallem and Carlson , 2006; Saito et al . , 2009 ) . This approximation neglects individual spike times , which can contain important information for odor discrimination in mammals and insects ( Resulaj and Rinberg , 2015; DasGupta and Waddell , 2008; Wehr and Laurent , 1996; Huston et al . , 2015 ) . Given data relating spike timing and odor exposure for different odorants and receptors , we could use the time from respiratory onset to the first elicited spike in each receptor as an indicator of activity in our model . Alternatively , we could use both the timing and the firing rate information together . Such data is not yet available for large panels of odors and receptors , and so we leave the inclusion of timing effects for future work . A challenge specific to the study of the olfactory system as compared to other senses is the limited knowledge we have of the space of odors . It is difficult to identify common features shared by odorants that activate a given receptor type ( Rossiter , 1996; Malnic et al . , 1999 ) , while attempts at defining a notion of distance in olfactory space have had only partial success ( Snitz et al . , 2013 ) , as have attempts to find reduced-dimensionality representations of odor space ( Zarzo and Stanton , 2006; Koulakov et al . , 2011 ) . In this work , we simply model the olfactory environment as a vector 𝐜={c1 , … , cN} of concentrations , where ci is the concentration of odorant i in the environment ( Figure 1a ) . We note , however , that the formalism we describe here is equally applicable for other parameterizations of odor space: the components ci of the environment vector 𝐜 could , for instance , indicate concentrations of entire classes of molecules clustered based on common chemical traits , or they might be abstract coordinates in a low-dimensional representation of olfactory space . Once a parameterization for the odor environment is chosen , we model the statistics of natural scenes by the joint probability distribution P⁢ ( c1 , … , cN ) . We are neglecting temporal correlations in olfactory cues because we are focusing on odor identity rather than olfactory search where timing of cues will be especially important . This simplifies our model , and also reduces the number of olfactory scene parameters needed as inputs . Similar static approximations of natural images have been employed powerfully along with the efficient coding hypothesis to explain diverse aspects of early vision ( e . g . , in Laughlin , 1981; Atick and Redlich , 1990; Olshausen and Field , 1996; van Hateren and van der Schaaf , 1998; Ratliff et al . , 2010; Hermundstad et al . , 2014 ) . To construct a tractable model of the relation between natural odor statistics and olfactory receptor distributions , we describe the olfactory environment as a multivariate Gaussian with mean 𝐜0 and covariance matrix Γ , ( 1 ) environment P ( c ) ∼𝒩 ( c0 , Γ ) . This can be thought of as a maximum-entropy approximation of the true distribution of odorant concentrations , constrained by the environmental means and covariances . This simple environmental model misses some sparse structure that is typical in olfactory scenes ( Yu et al . , 2015; Krishnamurthy et al . , 2017 ) . Nevertheless , approximating natural distributions with Gaussians is common in the efficient-coding literature , and often captures enough detail to be predictive ( van Hateren , 1992a; van Hateren , 1992b; Van Hateren , 1993; Hermundstad et al . , 2014 ) . This may be because early sensory systems in animals are able to adapt more effectively to low-order statistics which are easily represented by neurons in their mean activity and pairwise correlations . The number N of odorants that we use to represent an environment need not be as large as the total number of possible volatile molecules . We can instead focus on only those odorants that are likely to be encountered at meaningful concentrations by the organism that we study , leading to a much smaller value for N . In practice , however , we are limited by the available receptor affinity data . Our quantitative analyses are generally based on data measured using panels of 110 odorants in fly ( Hallem and Carlson , 2006 ) and 63 in mammals ( Saito et al . , 2009 ) . We next build a model for how the activity at the glomeruli depends on the olfactory environment . We work in an approximation in which the responses depend linearly on the concentration values: ( 2 ) ra=Ka∑iSaici+ηaKa , where ra is the response of the glomerulus indexed by a , Sa⁢i is the expected response of a single sensory neuron expressing receptor type a to a unit concentration of odorant i , and Ka is the number of neurons of type a . The second term describes noise , with ηa , the noise for a single OSN , modeled as a Gaussian with mean 0 and standard deviation σa , ηa∼𝒩 ( 0 , σa2 ) . The approximation we are using can be seen as linearizing the responses of olfactory sensory neurons around an operating point . This has been shown to accurately capture the response of olfactory receptors to odor mixtures in certain concentration ranges ( Singh et al . , 2018 ) . While odor concentrations in natural scenes span many orders of magnitude and are unlikely to always stay within the linear regime , the effect of the nonlinearities on the information maximization procedure that we implement below is less strong ( see Appendix 3 for a comparison between our linear approximation and a nonlinear , competitive binding model in a toy example ) . One advantage of employing the linear approximation is that it requires a minimal set of parameters ( the sensing matrix coefficients Sai ) , while nonlinear models in general require additional information ( such as a Hill coefficient and a maximum activation for each receptor-odorant pair for a competitive binding model; see Appendix 3 ) . We can predict the optimal distribution of receptor types given the sensing matrix S and the statistics of odors by maximizing the mutual information in Equation ( 4 ) while keeping the total number of neurons Ktot=∑aKa constant . We tested the effect of changing the variance of a single odorant , and found that the effect on the optimal receptor abundances depends on the context of the background olfactory environment . Increased exposure to a particular ligand can lead to increased abundance of a given receptor type in one context , but to decreased abundance in another ( Figure 3 ) . In fact , patterns of this kind have been reported in recent experiments ( Santoro and Dulac , 2012; Zhao et al . , 2013; Cadiou et al . , 2014; Ibarra-Soria et al . , 2017 ) . To understand this context-dependence better , we analyzed the predictions of our model in various signal and noise scenarios . One factor that does not affect the optimal receptor distribution in our model is the average concentration vector 𝐜0 . This is because it corresponds to odors that are always present and therefore offer no new information about the environment . This is consistent with experiment ( Ibarra-Soria et al . , 2017 ) , where it was observed that chronic odor exposure does not affect receptor abundances in the epithelium . In the rest of the paper , we thus restrict our attention to the covariance matrix of odorant concentrations , Γ . The problem of maximizing the amount of information that OSN responses convey about the odor environment simplifies considerably if these responses are weakly correlated . In this case , standard efficient coding theory says that receptors whose activities fluctuate more extensively in response to the olfactory environment provide more information to brain , while receptors that are active at a constant rate or are very noisy provide less information . In this circumstance , neurons expressing receptors with large signal-to-noise ratio ( SNR , i . e . signal variance as compared to noise variance ) should increase in proportion relative to neurons with low signal-to-noise ratio ( see Appendix 2 for a derivation ) . In terms of our model , the signal variance of glomerular responses is given by diagonal elements of the overlap matrix Q ( Equation 5 ) , while the noise variance is σa2; so we expect Ka , the number of OSNs of type a , to increase with Qa⁢a/σa2 . Responses are less correlated if receptors are narrowly tuned , and we find indeed that if each receptor type responds to only a small number of odorants , the abundances of OSNs of each type correlate well with their variability in the environment ( narrow-tuning side of Figure 2d ) . This is also consistent with the results at high SNR: we saw above that in that case Ka≈C-σa2⁢ ( Q-1 ) a⁢a , and when response correlations are weak , Q is approximately diagonal , and thus ( Q-1 ) a⁢a≈1/Qa⁢a . The biological setting is better described in terms of widely tuned sensing matrices ( Hallem and Carlson , 2006 ) , and an intermediate SNR level in which noise is important , but does not dominate the responses of most receptors . We therefore generated sensing matrices with varying tuning width by changing the number of odorants that elicit strong activity in each receptor ( as detailed in Appendix 1 ) . We found that as receptors begin responding to a greater diversity of odorants , the correlation structure of their activity becomes important in determining the optimal receptor distribution; it is no longer sufficient to just examine the signal to noise ratios of each receptor type separately as a conventional theory suggests ( wide-tuning side of Figure 2d ) . In other words , the optimal abundance of a receptor type depends not just on its activity level , but also on the context of the correlated activity levels of all the other receptor types . These correlations are determined by the covariance structures of the environment and of the sensing matrix . In fact , across the range of tuning widths the optimal receptor abundances Ka are correlated with the inverse of the overlap matrix , A=Q-1 ( Figure 2e ) . For narrow tuning widths , the overlap matrix Q is approximately diagonal ( because correlations between receptors are weak ) and so Q-1 is simply the matrix of the inverse diagonal elements of Q . Thus , in this limit , the correlation with Q-1 simply follows from the correlation with Q that we discussed above . As the tuning width increases keeping the total number of OSNs Ktot constant , the responses from each receptor grow stronger , increasing the SNR , even as the off-diagonal elements of the overlap matrix Q become significant . In the limit of high SNR , the analytical formula Ka≈C-σa2⁢Qa⁢a-1 ( Equation 8 ) ensures that the OSN numbers Ka are still correlated with the diagonal elements of Q-1 , despite the presence of large off-diagonal components . Because of the matrix inversion in Q-1 , the optimal abundance for each receptor type is affected in this case by the full covariance structure of all the responses and not just by the variance Qa⁢a of the receptor itself . Mathematically , this is because the diagonal elements of Q-1 are functions of all the variances and covariances in the overlap matrix Q . This dependence of each abundance on the full covariance translates to a complex context-dependence whereby changing the same ligand in different background environments can lead to very different adapted distributions of receptors . In Appendix 6 we show that the correlation with the inverse overlap matrix has an intuitive interpretation: receptors which either do not fluctuate much or whose values can be guessed based on the responses of other receptors should have low abundances . To investigate how the structure of the optimal receptor repertoire varies with the olfactory environment , we first constructed a background in which the concentrations of 110 odorants were distributed according to a Gaussian with a randomly chosen covariance matrix ( e . g . , Figure 4a; see Appendix 4 for details ) . From this base , we generated two different environments by adding a large variance to 10 odorants in environment 1 , and to 10 different odorants in environment 2 ( Figure 4b ) . We then considered the optimal distribution in these environments for a repertoire of 24 receptor types with odor affinities inferred from ( Hallem and Carlson , 2006 ) . We found that when the number of olfactory sensory neurons Ktot is large , and thus the signal-to-noise ratio is high , the change in odor statistics has little effect on the distribution of receptors ( Figure 4c ) . This is because at high SNR , all the receptors are expressed nearly uniformly as discussed above , and this is true in any environment . When the number of neurons is smaller ( or , equivalently , the signal-to-noise ratio is in a low or intermediate regime ) , the change in environment has a significant effect on the receptor distribution , with some receptor types becoming more abundant , others becoming less abundant , and yet others not changing much between the environments ( see Figure 4d ) . This mimics the kinds of complex effects seen in experiments in mammals ( Schwob et al . , 1992; Santoro and Dulac , 2012; Zhao et al . , 2013; Dias and Ressler , 2014; Cadiou et al . , 2014; Ibarra-Soria et al . , 2017 ) . In the comparison above , the two environment covariance matrices differed by a large amount for a small number of odors . We next compared environments with two different randomly generated covariance matrices , each generated in the same way as the background environment in Figure 4a . The resulting covariance matrices ( Figure 5a , top ) are very different in detail ( the correlation coefficient between their entries is close to zero; distribution of changes in Figure 5b , red line ) , although they look similar by eye . Despite the large change in the detailed structure of the olfactory environment , the corresponding change in optimal receptor distribution is typically small , with a small fraction of receptor types experiencing large changes in abundance ( red curve in Figure 5c ) . The average abundance of each receptor in these simulations was about 1000 , and about 90% of all the abundance change values \|ΔKi\| were below 20% of this , which is the range shown on the plot in Figure 5c . Larger changes also occurred , but very rarely: about 0 . 1% of the abundance changes were over 800 . If we instead engineer two environments that are almost non-overlapping so that each odorant is either common in environment 1 , or in environment 2 , but not in both ( Figure 5a , bottom; see Appendix 4 for how this was done ) , the changes in optimal receptor abundances between environments shift away from mid-range values towards higher values ( blue curve in Figure 5c ) . For instance , 40% of abundance changes lie in the range \|ΔK\|>50 in the non-overlapping case , while the proportion is 28% in the generic case . It seems intuitive that animals that experience very different kinds of odors should have more striking differences in their receptor repertoires than those that merely experience the same odors with different frequencies . Intriguingly , however , our simulations suggest that the situation may be reversed at the very low end: the fraction of receptors for which the predicted abundance change is below 0 . 1 , \|ΔK\|<0 . 1 , is about 2% in the generic case but over 9% for non-overlapping environment pairs . Thus , changing between non-overlapping environments emphasizes the more extreme changes in receptor abundances , either the ones that are close to zero or the ones that are large . In contrast , a generic change in the environment leads to a more uniform distribution of abundance changes . Put differently , the particular way in which the environment changes , and not only the magnitude of the change , can affect the receptor distribution in unexpected ways . The magnitude of the effect of environmental changes on the optimal olfactory receptor distribution is partly controlled by the tuning of the olfactory receptors ( Figure 5d ) . If receptors are narrowly tuned , with each type responding to a small number of odorants , changes in the environment tend to have more drastic effects on the receptor distribution than when the receptors are broadly tuned ( Figure 5d ) , an effect that could be experimentally tested . Our study opens the exciting possibility of a causal test of the hypothesis of efficient coding in sensory systems , where a perturbation in the odor environment could lead to predictable adaptations of the olfactory receptor distribution during the lifetime of an individual . This does not happen in insects , but it can happen in mammals , since their receptor neurons regularly undergo apoptosis and are replaced . A recent study demonstrated reproducible changes in olfactory receptor distributions of the sort that we predict in mice ( Ibarra-Soria et al . , 2017 ) . These authors raised two groups of mice in similar conditions , exposing one group to a mixture of four odorants ( acetophenone , eugenol , heptanal , and R-carvone ) either continuously or intermittently ( by adding the mixture to their water supply ) . Continuous exposure to the odorants had no effect on the receptor distribution , in agreement with the predictions of our model . In contrast , intermittent exposure did lead to systematic changes ( Figure 6a ) . We used our model to run an experiment similar to that of Ibarra-Soria et al . ( 2017 ) in silico ( Figure 6b ) . Using a sensing matrix based on odor response curves for mouse and human receptors ( data for 59 receptors from Saito et al . ( 2009 ) ) , we calculated the predicted change in OSN abundances between two different environments with random covariance matrices constructed as described above . We ran the simulations 24 times , modifying the odor environments each time by adding a small amount of Gaussian random noise to the square roots of these covariance matrices to model small perturbations ( details in Appendix 4; range bars in Figure 6b ) . The results show that the abundances of already numerous receptors do not change much , while there is more change for less numerous receptors . The frequencies of rare receptors can change dramatically , but are also more sensitive to perturbations of the environment ( large range bars in Figure 6b ) . These results qualitatively match experiment ( Figure 6a ) , where we see the same pattern of the largest reproducible changes occurring for receptors with intermediate abundances . The experimental data is based on receptor abundance measured by RNAseq which is a proxy for counting OSN numbers ( Ibarra-Soria et al . , 2017 ) . In our model , the distinction between receptor numbers and OSN numbers is immaterial because a change in the number of receptors expressed per neuron has the same effect as a change in neuron numbers . In general , additional experiments are needed to measure both the number of receptors per neuron and the number of neurons for each receptor type . We have explored the structure of olfactory receptor distributions that code odors efficiently , that is are adapted to maximize the amount of information that the brain gets about odors . The full solution to the optimization problem , Equation ( 7 ) , depends in a complicated nonlinear way on the receptor affinities S and covariance of odorant concentrations Γ . The distribution of olfactory receptors in the mammalian epithelium , however , must arise dynamically from the pattern of apoptosis and neurogenesis ( Calof et al . , 1996 ) . At a qualitative level , in the efficient coding paradigm that we propose , the receptor distribution is related to the statistics of natural odors , so that the life cycle of neurons would have to depend dynamically on olfactory experience . Such modulation of OSN lifetime by exposure to odors has been observed experimentally ( Santoro and Dulac , 2012; Zhao et al . , 2013 ) and could , for example , be mediated by feedback from the bulb ( Schwob et al . , 1992 ) . To obtain a dynamical model , we started with a gradient ascent algorithm for changing receptor numbers , and modified it slightly to impose the constraints that OSN numbers are non-negative , Ka≥0 , and their sum Ktot=∑aKa is bounded ( details in Appendix 5 ) . This gives ( 9 ) dKadt=α{Ka−λKa2−σa2 ( R−1 ) aaKa2} , where α is a learning rate , σa2 is the noise variance for receptor type a , and R is the covariance matrix of glomerular responses , ( 10 ) Ra⁢b=⟨ra⁢rb⟩-⟨ra⟩⁢⟨rb⟩ , with the angle brackets denoting ensemble averaging over both odors and receptor noise . In the absence of the experience-related term ( R-1 ) a⁢a , the dynamics from Equation ( 9 ) would be simply logistic growth: the population of OSNs of type a would initially grow at a rate α , but would saturate when Ka=1/λ because of the population-dependent death rate λ⁢Ka . In other words , the quantity M/λ sets the asymptotic value for the total population of sensory neurons , Ktot→M/λ , with M being the number of receptor types . Because of the last term in Equation ( 9 ) , the death rate in our model is influenced by olfactory experience in a receptor-dependent way . In contrast , the birth rate is not experience-dependent and is the same for all OSN types . Indeed , in experiments , the odor environment is seen to have little effect on receptor choice , but does modulate the rate of apoptosis in the olfactory epithelium ( Santoro and Dulac , 2012 ) . Our results suggest that , if olfactory sensory neuron lifetimes are appropriately anti-correlated with the inverse response covariance matrix , then the receptor distribution in the epithelium can converge to achieve optimal information transfer to the brain . The elements of the response covariance matrix Ra⁢b could be estimated by temporal averaging of co-occurring glomerular activations via lateral connections between glomeruli ( Mori et al . , 1999 ) . Performing the inverse necessary for our model is more intricate . The computations could perhaps be done by circuits in the bulb and then fed back to the epithelium through known mechanisms ( Schwob et al . , 1992 ) , Within our model , Figure 8a shows an example of receptor numbers converging to the optimum from random initial values . The sensing matrix used here is based on mammalian data ( Saito et al . , 2009 ) and we set the total OSN number to Ktot=2000 . The environment covariance matrix is generated using the random procedure described earlier ( details in Appendix 4 ) . We see that some receptor types take longer than others to converge ( the time axis is logarithmic , which helps visualize the whole range of convergence behaviors ) . Roughly speaking , convergence is slower when the final OSN abundance is small , which is related to the fact that the rate of change d⁢Ka/d⁢t in Equation ( 9 ) vanishes in the limit Ka→0 . For the same reason , OSN populations that start at a very low level also take a long time to converge . In Figure 8b , we show convergence to the same final state , but this time starting from a distribution that is not random but was optimized for a different environment . The initial and final environments are the same as the two environments used in the previous section to compare the simulations to experimental findings ( Figure 6b ) . Interestingly , many receptor types actually take longer to converge in this case compared to the random starting point , perhaps because there are local optima in the landscape of receptor distributions . Given such local minima , stochastic fluctuations will allow the dynamics to reach the global optimum more easily . In realistic situations , there are many sources of such variability , for example , sampling noise due to the fact that the response covariance matrix R must be estimated through stochastic odor encounters and noisy receptor readings . In fact , in Figure 8b , we added a small amount of noise ( corresponding to ±0 . 05⁢Ktot/M ) to the initial distribution of receptors to improve convergence rates . We built a model for the distribution of receptor types in the olfactory epithelium that is based on efficient coding , and assumes that the abundances of different receptor types are adapted to the statistics of natural odors in a way that maximizes the amount of information conveyed to the brain by glomerular responses . This model predicts a non-uniform distribution of receptor types in the olfactory epithelium , as well as reproducible changes in the receptor distribution after perturbations to the odor environment . In contrast to other applications of efficient coding , our model operates in a regime in which there are significant correlations between sensors because the adaptation of OSN abundances occurs upstream of the brain circuitry that can decorrelate olfactory responses . In this regime , OSN abundances depend on the full correlation structure of the inputs , leading to predictions that are context-dependent in the sense that whether the abundance of a specific receptor type goes up or down due to a shift in the environment depends on the global context of the responses of all the other receptors . All these striking phenomena have been observed in recent experiments and had not been explained prior to this study . In our framework , the sensitivity of the receptor distribution to changes in odor statistics is affected by the tuning of the olfactory receptors , with narrowly tuned receptors being more readily affected by such changes than broadly tuned ones . The model also predicts that environments that differ in the identity of the odors that are present will lead to greater deviations in the optimal receptor distribution than environments that differ only in the statistics with which these odors are encountered . Likewise , the model broadly predicts a monotonic relationship between the number of receptor types found in the epithelium and the total number of olfactory sensory neurons , all else being equal . A detailed test of our model requires more comprehensive measurements of olfactory environments than are currently available . Our hope is that studies such as ours will spur interest in measuring the natural statistics of odors , opening the door for a variety of theoretical advances in olfaction , similar to what was done for vision and audition . Such measurements could for instance be performed by using mass spectrometry to measure the chemical composition of typical odor scenes . Given such data , and a library of receptor affinities , our GitHub ( RRID:SCR_002630 ) online repository provides an easy-to-use script that uses our model to predict OSN abundances . For mammals , controlled changes in environments similar to those in Ibarra-Soria et al . ( 2017 ) could provide an even more stringent test for our framework . To our knowledge , this is the first time that efficient coding ideas have been used to explain the pattern of usage of receptors in the olfactory epithelium . Our work can be extended in several ways . OSN responses can manifest complex , nonlinear responses to odor mixtures . Accurate models for how neurons in the olfactory epithelium respond to complex mixtures of odorants are just starting to be developed ( e . g . Singh et al . , 2018 ) , and these can in principle be incorporated in an information-maximization procedure similar to ours . More realistic descriptions of natural odor environments can also be added , as they amount to changing the environmental distribution P⁢ ( 𝐜 ) . For example , the distribution of odorants could be modeled using a Gaussian mixture , rather than the normal distribution used in this paper to enable analytic calculations . Each Gaussian in the mixture would model a different odor object in the environment , more closely approximating the sparse nature of olfactory scenes discussed in , for example , Krishnamurthy et al . ( 2017 ) . Of course , the goal of the olfactory system is not simply to encode odors in a way that is optimal for decoding the concentrations of volatile molecules in the environment , but rather to provide an encoding that is most useful for guiding future behavior . This means that the value of different odors might be an important component shaping the neural circuits of the olfactory system . In applications of efficient coding to vision and audition , maximizing mutual information , as we did , has proved effective even in the absence of a treatment of value ( Laughlin , 1981; Atick and Redlich , 1990; van Hateren , 1992a; Olshausen and Field , 1996; Simoncelli and Olshausen , 2001; Fairhall et al . , 2001; Lewicki , 2002; Ratliff et al . , 2010; Garrigan et al . , 2010; Tkacik et al . , 2010; Hermundstad et al . , 2014; Palmer et al . , 2015; Salisbury and Palmer , 2016 ) . However , in general , understanding the role of value in shaping neural circuits is an important experimental and theoretical problem . To extend our model in this direction , we would replace the mutual information between odorant concentrations and glomerular responses by a different function that takes into account value assignments ( see , e . g . Rivoire and Leibler , 2011 ) . It could be argued , though , that such specialization to the most behaviorally relevant stimuli might be unnecessary or even counterproductive close to the sensory periphery . Indeed , a highly specialized olfactory system might be better at reacting to known stimuli , but would be vulnerable to adversarial attacks in which other organisms take advantage of blind spots in coverage . Because of this , and because precise information regarding how different animals assign value to different odors is scarce , we leave these considerations for future work . One exciting possibility suggested by our model is a way to perform a first causal test of the efficient coding hypothesis for sensory coding . Given sufficiently detailed information regarding receptor affinities and natural odor statistics , experiments could be designed that perturb the environment in specified ways , and then measure the change in olfactory receptor distributions . Comparing the results to the changes predicted by our theory would provide a strong test of efficient coding by early sensory systems in the brain . The code ( written in Matlab , RRID:SCR_001622 ) and data that we used to generate all the results and figures in the paper is available on GitHub ( RRID:SCR_002630 ) , at https://github . com/ttesileanu/OlfactoryReceptorDistribution ( Teşileanu , 2019; copy archived at https://github . com/elifesciences-publications/OlfactoryReceptorDistribution ) .	A mouse’s nose contains over 10 million receptor neurons divided into about 1 , 000 different types , which detect airborne chemicals – called odorants – that make up smells . Each odorant activates many different receptor types . And each receptor type responds to many different odorants . To identify a smell , the brain must therefore consider the overall pattern of activation across all receptor types . Individual receptor neurons in the mammalian nose live for about 30 days , before new cells replace them . The entire population of odorant receptor neurons turns over every few weeks , even in adults . Studies have shown that some types of these receptor neurons are used more often than others , depending on the species , and are therefore much more abundant . Moreover , the usage patterns of different receptor types can also change when individual animals are exposed to different smells . Teşileanu et al . set out to develop a computer model that can explain these observations . The results revealed that the nose adjusts its odorant receptor neurons to provide the brain with as much information as possible about typical smells in the environment . Because each smell consists of multiple odorants , each odorant is more likely to occur alongside certain others . For example , the odorants that make up the scent of a flower are more likely to occur together than alongside the odorants in diesel . The nose takes advantage of these relationships by adjusting the abundance of the receptor types in line with them . Teşileanu et al . show that exposure to odorants leads to reproducible increases or decreases in different receptor types , depending on what would provide the brain with most information . The number of odorant receptor neurons in the human nose decreases with time . The current findings could help scientists understand how these changes affect our sense of smell as we age . This will require collaboration between experimental and theoretical scientists to measure the odors typical of our environments , and work out how our odorant receptor neurons detect them .	[ "Abstract", "Introduction", "Discussion", "Materials", "and", "methods" ]	[ "physics", "of", "living", "systems" ]	2019	Adaptation of olfactory receptor abundances for efficient coding
An outstanding challenge has been to understand the mechanism whereby proteins associate . We report here the results of exhaustively sampling the conformational space in protein–protein association using a physics-based energy function . The agreement between experimental intermolecular paramagnetic relaxation enhancement ( PRE ) data and the PRE profiles calculated from the docked structures shows that the method captures both specific and non-specific encounter complexes . To explore the energy landscape in the vicinity of the native structure , the nonlinear manifold describing the relative orientation of two solid bodies is projected onto a Euclidean space in which the shape of low energy regions is studied by principal component analysis . Results show that the energy surface is canyon-like , with a smooth funnel within a two dimensional subspace capturing over 75% of the total motion . Thus , proteins tend to associate along preferred pathways , similar to sliding of a protein along DNA in the process of protein-DNA recognition . Interactions between proteins play a central role in various aspects of the structural and functional organization of the cell . To recognize its partner , a protein must align its binding interface , usually a small fraction of the total surface , with a similarly small binding interface on the other protein ( Berg and von Hippel , 1985; Ubbink , 2009 ) . Since all interactions are of relatively short range , the process must start with a diffusive search governed by Brownian motion , which brings the proteins to a ‘macrocollision’ to yield a transition state also known as the encounter complex ( Berg and von Hippel , 1985; von Hippel and Berg , 1989 ) . The encounter complex can be thought of as an ensemble of conformations in which the two molecules can rotationally diffuse along each other , or participate in a series of ‘microcollisions’ that properly align the reactive groups . The second step of association consists of conformational rearrangements to the native complex . While it has been generally recognized that association proceeds through a transition state , little was known of the encounter complex structures and configurations as their populations are low , their lifetimes are short , and they are difficult to trap , rendering them essentially invisible to conventional structural and biophysical methods ( Iwahara and Clore , 2006 ) . Novel experimental and improved computational methods , developed during the last decade , have the potential to provide information leading to better understanding of the nature of encounter complexes . On the experimental side , the major progress is due to the application of NMR paramagnetic relaxation enhancement ( PRE ) , a technique that is exquisitely sensitive to the presence of lowly populated states in the fast exchange regime ( Clore , 2008; Clore and Iwahara , 2009; Fawzi et al . , 2010 ) . The detection of such intermediates requires introducing paramagnetic labels , one at a time , at a few sites on one of the interacting proteins , and measuring the transverse paramagnetic relaxation enhancement ( PRE ) rates , Γ2 , of the backbone amide protons ( 1HN ) of the partner protein ( Tang et al . , 2006 ) . In a fast exchanging system , the observed value of Γ2 is the weighted average of the values for the various states present in solution ( Iwahara et al . , 2004; Iwahara and Clore , 2006 ) . Because Γ2 is dependent on the inverse sixth power of the distance ( <r−6> ) between the unpaired electron on the paramagnetic center and the observed proton , and because the Γ2 rates at short distances are very large owing to the large magnetic moment of the unpaired electrons , low-population intermediates can be detected . In particular , the observed intermolecular 1HN−Γ2 rates and those back-calculated from the structure of the native complex generally differ for a number of residues , revealing regions that participate in transitional interactions ( Tang et al . , 2006 ) . Paramagnetic relaxation enhancement ( PRE ) techniques can provide distributions of distances between a paramagnetic ion and protons , indicating the presence and relative ratio of conformational sub-ensembles ( Tang et al . , 2006; Suh et al . , 2007; Clore and Iwahara , 2009; Fawzi et al . , 2010 ) , but determining the detailed structure of encounter complexes requires computational approaches . A semi-quantitative depiction of the minor species can be obtained by using restrained rigid-body simulated annealing refinement to minimize the difference between observed and calculated 1HN−Γ2 rates ( Tang et al . , 2006; Kim et al . , 2008 ) . However , with the development of docking methods it is possible to globally sample the entire conformational space of two interacting proteins , generating all low energy states . Although in some cases binding is inherently coupled with folding ( Shoemaker et al . , 2000; Zheng et al . , 2012 ) , a large class of protein complexes can be adequately described by a model that assumes essentially rigid association , possibly followed by refinement that allows for local changes in side chains and interacting loops ( Smith and Sternberg , 2002; Kozakov et al . , 2006; Vajda and Kozakov , 2009 ) . As demonstrated by the results of CAPRI ( Critical Assessment of Prediction of Interactions ) communitywide protein docking experiment , for such cases modern computational docking methods , including automated servers , are capable of generating docked conformations that agree well with the X-ray structure of the complex ( Lensink and Wodak , 2013 ) . In particular , our program PIPER , based on the Fast Fourier transform ( FFT ) correlation approach , globally and systematically samples the conformational space of two interacting proteins on a dense grid using a physics based energy function ( Kozakov et al . , 2006 ) . The program is implemented in the heavily used server ClusPro ( Comeau et al . , 2007 ) , which yields good results when docking X-ray structures of two proteins with at most moderate backbone conformational changes upon binding ( Kozakov et al . , 2010 ) . Based on the results of CAPRI , Cluspro has been the best protein–protein docking server for the last 5 years ( Kozakov et al . , 2013; Lensink and Wodak , 2013 ) . It is well known that , in addition to near-native structures , docking generally yields a large number of models that are similar to near-native ones in terms of energy , but may substantially differ in terms of geometry ( Vajda and Kozakov , 2009 ) . Since such ‘false positive’ models do not predict the final bound complex , they are usually regarded as artifacts . However , using molecular mechanics energy functions without ‘built-in’ information on the native state it is reasonable to assume that the alternative low energy models represent encounter complexes that , in view of their favorable interactions , may occur along association pathways . Accordingly , we show here that using the large ensemble of low energy structures generated by docking provides better approximation of experimental PRE profiles than the one calculated only from the coordinates of the final complex . Since in docking calculations we start from unbound protein structures and systematically sample the entire conformational space , based on this result we can easily generate ensembles of encounter complexes for any pair of associating proteins . Once it is established that the energy function used for sampling the conformational space enables us to accurately predict both the native state and the ensemble of encounter complexes , and thus the energy function is valid beyond selecting the native structure of the complex , we proceed to characterizing the energy landscape in the 6D translational/rotational space near the native state ( McCammon , 1998; Camacho et al . , 1999; Zhang et al . , 1999; Tovchigrechko and Vakser , 2001 ) . We focus on the main binding funnel in a neighborhood of the native state , which is the most important region of the conformational space , containing over 90% of complex structures observed in the PRE experiments ( Iwahara and Clore , 2006; Tang et al . , 2006; Clore and Iwahara , 2009; Fawzi et al . , 2010 ) . Since rigid body association occurs in a low dimensional space , the shape of the binding energy landscape can be studied in detail , in contrast to protein folding , which occurs in a very high dimensional space ( Dill and Chan , 1997 ) . In spite of the low dimensionality , the analysis is far from simple in this highly curved space due to the interdependence of the coordinates ( Park , 1995; Park and Ravani , 1997; Shen et al . , 2008; Mirzaei et al . , 2012 ) . However , one can transform the rotational space into a product of axis-angle representations using complex exponentials ( Park and Ravani , 1997 ) . These so-called exponential parameters create a local one-to-one mapping between the nonlinear manifold of potential conformations and an Euclidean space ( Mirzaei et al . , 2012 ) , and thus the shape of low energy regions in the conformational space can be studied by classical principal component analysis ( PCA ) in the Euclidean space . Using PCA we will be able to determine whether any subspace can accommodate a large fraction of the structures , and whether there are energy barriers that restrict the distribution of encounter complexes in the vicinity of the native state . The most important result of our analysis is that the region of the space in a neighborhood of the native state invariably includes high energy barriers preventing the ligand from moving into a one- or two-dimensional restrictive subspace . Orthogonal to the restrictive subspace is a permissive subspace , in which the energy is relatively flat . Based on these results one can visualize the energy landscape as resembling a canyon-like terrain where the low energy areas ( at the bottom of the canyon ) lie in a lower dimensional subspace . Thus , within the range of physical interactions , two proteins sample only relatively small fractions of the conformational space , and converge toward the native state along preferred pathways . This result represents information that , in principle , could have been obtained by running molecular dynamics or Brownian dynamics simulations . However , a sufficiently dense sampling of conformational space by molecular dynamics is computationally very demanding , even when restricting considerations to a neighborhood of the native state ( Wang and Wade , 2003 ) , whereas Brownian dynamics simulations usually rely on highly simplified protein models ( Camacho et al . , 2000; Spaar and Helms , 2005 ) . In contrast , we use detailed all-atom models and map the energy surface using two different physics-based energy functions . As will be shown , these differences do not significantly affect the results of the principal component analysis , suggesting that the reduction of dimensionality is an inherent property of the free energy landscape in protein–protein association . We focused on the modeling of the association between the N-terminal domain of Enzyme I ( EIN ) and the histidine-containing phosphocarrier protein ( HPr ) ( Figure 1A ) , because the complex has been studied in a series of PRE titration experiments ( Fawzi et al . , 2010 ) . Specific association between the two proteins occurs in the first step of the bacterial phosphotransfer system , resulting in phosphoryl transfer between EIN and HPr upon proper alignment of active site histidines of the two sides of the interface ( Garrett et al . , 1999 ) . The binding has an equilibrium dissociation constant of 4 . 3 μM ( Suh et al . , 2008 ) . For the computational study of encounter complexes we have placed the center of EIN at the origin of a coordinate system , and systematically sampled the entire rotational/translational space of HPr . Unbound structures were used both for the receptor , EIN ( chain A from PDB entry 1ZYM ) and for the ligand , HPr ( chain P from PDB entry 2JEL ) . Sampling was performed using the docking program PIPER , which performs exhaustive evaluation of a physics-based energy function in discretized 6D space of mutual orientations of two proteins using the Fast Fourier transform ( FFT ) correlation approach ( Kozakov et al . , 2006 ) . We sample 70 , 000 rotations , which approximately correspond to sampling at every 5° in the space of Euler angles . In the translational space , the sampling is defined by the 1 . 0 Å grid cell size . PIPER is used with a ‘smooth’ energy function that includes terms describing attractive and repulsive van der Waals interactions , electrostatic interactions calculated by a simplified generalized Born-type expression , and a desolvation terms , the latter represented by a pairwise interaction potential ( Chuang et al . , 2008 ) . We call the energy function ‘smooth’ because the repulsive contributions to the van der Waals interaction are selected to allow for a certain amount of overlaps . 10 . 7554/eLife . 01370 . 003Figure 1 . Docking results for the EIN–HPr complex . Unbound structures were used both for the receptor , EIN ( chain A from PDB entry 1ZYM ) and for the ligand , HPr ( chain P from PDB entry 2JEL ) . Encounter complexes were generated using Fast Fourier transform ( FFT ) based sampling . ( A ) Cartoon of the specific complex formed by EIN and HPr , shown in grey and yellow , respectively . The locations of the paramagnetic tags E5C-EDTA-Mn+ and E32C-EDTA-Mn2+ on HPr are encircled and are shown in red and blue , respectively . ( B ) Centers of HPr structures in the encounter complex ensemble . Colors indicate classification as follows ( 8 ) : blue , Class I ( i . e . , overlapping with the specific complex ) ; magenta , patch 1 of Class II ( i . e . , non-overlapping ) positions; red , patch 2 of Class II positions; and pink , additional Class II position outside the main patches . ( C ) Ligand IRMSD vs PIPER energy score . ( D ) Two representative HPr poses , colored light blue and dark blue , from Class I . ( E ) Two representative HPr poses ( in different shades of magenta ) from Patch 1 of Class II . ( F ) View of the EIN–HPr complex and the centers of HPr poses after rotating 180° around the vertical axis ( the bound HPr is now on the left side , almost completely hidden by EIN ) . ( G ) Representative HPr poses ( in different shades of red ) from Patch 2 of Class II , shown in the rotated view . DOI: http://dx . doi . org/10 . 7554/eLife . 01370 . 00310 . 7554/eLife . 01370 . 004Figure 1—figure supplement 1 . Rotamers of the paramagnetic labels E5C-EDTA-Mn2+ and E32C-EDTA-Mn2+ on HPr . The C-EDTA moiety is shown as sticks , with carbon atoms colored cyan . The rest of the HPr structure is shown as yellow cartoon . The Mn2+ ions are shown as magenta spheres . ( A ) Three rotamers of HPr-E5C-EDTA-Mn2+ . ( B ) Three rotamers of HPr-E32C-EDTA-Mn2+ . DOI: http://dx . doi . org/10 . 7554/eLife . 01370 . 004 Since the generated structures will be used for calculating PRE profiles to compare them to experimental PRE data ( Fawzi et al . , 2010 ) , two sets of docking calculations were performed using HPr structures that included the paramagnetic label EDTA-Mn2+ , placed either at E5C , which is distal to the EIN/HPr interface in the native state , or at E32C , which is close to the edge of the interface ( Figure 1A ) . Each C-EDTA-Mn2+ label has three potential rotameric states , and hence Mn2+ can occupy three different positions ( Figure 1—figure supplement 1 ) . A separate docking was performed for each rotameric state of each EDTA-Mn2+ probe . We retained the 10 , 000 lowest energy structures from each docking simulation , thus a total of 30 , 000 low energy structures for each of the two probes . These structures were then used for the analysis of encounter complexes and for the calculation of intermolecular PRE rates . Figure 1B shows the center of each low energy HPr structure , generated by the docking , as a small sphere , and indicates that these structures form three major clusters . Figure 1C shows the Interface Root Mean Square Deviation ( IRMSD ) from the native complex vs the PIPER energy score of the docked structures . For the calculation of IRMD we first select the interface residues of HPr that are within 10 Å of any EIN atom in the native complex . For each docked structure we than superimpose EIN onto EIN in the X-ray structure of the complex , and calculate the RMSD between the Cα atoms of the HPr interface residues in docked and native structures . The structures in the largest cluster ( shown in blue in Figure 1B ) overlap with the native state , with the lowest energy conformations being within 5 Å IRMSD from the native ( Figure 1C ) . The structures in this cluster , termed Class I , are the results of rigid body rotations and small translations around the native binding mode . Two representative Class I structures are shown in Figure 1D . Some Class I structures have less than 1 Å IRMSD , but the cluster extends as far as 15 Å IRMSD from the native . The two other clusters , termed Class II patch 1 and Class II patch 2 , consist of structures that can coexist with the native complex . We note that while the three clusters clearly separate in the 3D representation shown in Figure 1B , they substantially overlap when projected into one dimension as a function of their IRMSD values . Nevertheless , Figure 1C shows at least three distinguishable energy minima . Class II patch 1 ( magenta in Figure 1B ) centers around a local energy minimum at around 17 Å IRMSD ( Figure 1C ) . Figure 1E shows two representative conformations for this patch . The third large cluster , Class II patch 2 ( red in Figure 1B ) , is located on the opposite side of the Class I cluster , and is better seen after rotating the complex by 180° around its vertical axis ( Figure 1F ) . The local energy minimum in this cluster is located at about 30 Å IRMSD from the native state . Figure 1G shows two representative conformations for the Class II patch 2 . In addition to the complexes that belong to Class I and the two patches of Class II , there are a number of smaller patches , shown in pink in Figure 1B . To detect encounter complexes in the EIN/HPr system by PRE titration experiments , HPr was labeled with a paramagnetic EDTA-Mn2+ moiety conjugated via a disulfide bond to surface cysteine mutations at specific sites ( Fawzi et al . , 2010 ) . We consider the mutations E5C and E32C that are both located outside the specific interaction surface with EIN ( Figure 1A ) and thus the labels do not interfere with the formation of the native complex . Intermolecular 1HN−Γ2 rates for the backbone amide protons of U-[2H , 15N]-labeled EIN were measured in the presence of 150 mM NaCl to eliminate potential spurious nonspecific interactions not relevant at physiological ionic strength . PRE measurements were carried out at six different concentrations of the paramagnetically labeled HPr ( ranging from 60 to 450 μM ) , corresponding to HPr:EIN molar ratios of 0 . 2–1 . 5 . At each point in the titration , the intermolecular PREs were summed over their respective residues and normalized to the highest value of each titration curve . The data points in Figure 2 show the normalized intermolecular PRE values and their standard errors observed in these titration experiments ( Fawzi et al . , 2010 ) . 10 . 7554/eLife . 01370 . 005Figure 2 . Normalized intermolecular PRE profiles for the EIN–HPr complex . PRE measurements were carried out at 300 μM EIN , 300 μM HPr , and 150 mM NaCl ( Fawzi et al . , 2010 ) . Theoretical intermolecular PREs , calculated only from the coordinates of the specific EIN/HPr complex , are shown as black lines . Calculated PRE values , based on all generated encounter complexes , are shown as blue lines , and reveal substantial contributions by the non-specific structures . The experimental PRE rates ( Γ2 ) are displayed as filled-in magenta circles . Points representing Γ2 values that were too large ( >60 s−1 ) to be determined accurately are placed at the saturation level Γ2/Γ2max = 1 . Interface residues are indicated by red ticks on the x-axis . ( A ) Results for EIN/HPr-E5C-EDTA-Mn2+ complexes . ( B ) Results for EIN/HPr-E32C-EDTA-Mn2+ complexes . DOI: http://dx . doi . org/10 . 7554/eLife . 01370 . 00510 . 7554/eLife . 01370 . 006Figure 2—figure supplement 1 . Controls emphasizing the need for accurate energy function in docking: theoretical PRE profiles for the EIN/HPr complex , based on complexes generated by using only the van der Waals energy ( blue line ) . Theoretical intermolecular PREs , calculated from the coordinates of the specific EIN/HPr complex , are also shown as reference ( black line ) . The experimental PRE rates ( Γ2 ) are displayed as filled-in magenta circles . ( A ) Results for EIN/HPr-E5C-EDTA-Mn2+ compexes . ( B ) Results for EIN/HPr-E32C-EDTA-Mn2+ complexes . DOI: http://dx . doi . org/10 . 7554/eLife . 01370 . 00610 . 7554/eLife . 01370 . 007Figure 2—figure supplement 2 . Normalized intermolecular PRE profiles and encounter compexes for the IIAMannitol/HPr interactions . ( A ) Native structure of the complex formed by IIAMannitol ( grey ) and HPr ( yellow ) . The location of the paramagnetic tag , HPr-E5C-EDTA-Mn2+ , is colored red and is indicated by a circle . The PDB ID of the complex is 1 J6T . ( B ) Centers of HPr structures , shown as blue spheres , in the encounter complex ensemble generated by the PIPER docking program . IIAMannitol , shown as grey solid , is considered the receptor . The native binding pose of HPr is shown as yellow cartoon . ( C ) Theoretical intermolecular PRE profiles calculated from the coordinates of the native structure only ( black line ) , and based on all encounter complexes generated by the docking ( blue line ) . The experimental PRE rates ( Γ2 ) are displayed as filled-in magenta circles ( Tang et al . , 2006 ) . Points representing Γ2 values that were too large ( >60 s−1 ) to be determined accurately are placed at the saturation level Γ2/Γ2max = 1 . The interface residues of IIAMannitol are indicated by red ticks on the x-axis . DOI: http://dx . doi . org/10 . 7554/eLife . 01370 . 00710 . 7554/eLife . 01370 . 008Figure 2—figure supplement 3 . Normalized intermolecular PRE profiles and encounter compexes for the HPr/HPr interactions . ( A ) Normalized intermolecular PRE profiles for the HPr/HPr complex . The theoretical intermolecular PREs profile , calculated from low energy encounter complexes , is shown as a green line . The experimental PRE rates ( Γ2 ) are displayed as filled-in magenta circles ( Tang et al . , 2008 ) . Points representing Γ2 values that were too large ( >60 s−1 ) to be determined accurately are placed at the saturation level Γ2/Γ2max = 1 . Interface residues are indicated by red ticks on the x-axis . ( B ) Ensemble of low energy conformations of HPr/HPr interactions ( Tang et al . , 2008 ) . One of the two HPr molecules , considered he receptor , is shown as grey cartoon . The centers of the other HPr positions generated by the docking are shown as small blue spheres . These structures were used for back-calculating the theoretical PRE profile ( green curve ) shown in A . DOI: http://dx . doi . org/10 . 7554/eLife . 01370 . 00810 . 7554/eLife . 01370 . 009Figure 2—figure supplement 4 . Encounter complexes in the Cytochrome c–Cytochrome c peroxidase interactions as reported on the basis of PRE experiments ( Bashir et al . , 2010 ) , shown as pink spheres , and the ones generated by the PIPER docking program , shown as blue spheres . DOI: http://dx . doi . org/10 . 7554/eLife . 01370 . 009 Given the coordinates of a complex , one can back-calculate theoretical PRE profiles ( ‘Materials and methods’ ) . As shown in Figure 2 , the theoretical profile calculated from the coordinates in the X-ray structure of the native complex ( black curve ) substantially deviates from the experimental values for a number of residues . For HPr-E5C the largest differences occur at positions 59–97 and 160–167 of EIN , with smaller deviations at 23–37 and 183–189 ( Figure 2A ) . For HPr-E32C the large differences are at positions 22–51 , 56–74 , 96–106 , and 160–167 of EIN , with smaller differences at 184–189 ( Figure 2B ) . These deviations show that the observed PRE rates cannot be explained well on the basis of the native binding mode of HPr alone , and provide at least qualitative evidence for the existence of lowly populated encounter states in rapid exchange with the final native complex ( Tang et al . , 2008 ) . Our hypothesis is that the non-native structures generated by the docking represent encounter complexes , and hence accounting for these structures predicts the experimental PRE values better than the profile calculated only from the coordinates of the native complex . Therefore we calculate the theoretical PRE profiles based on all 30 , 000 low energy structures obtained by the docking for each position of the paramagnetic label ( blue lines in Figure 2 ) . The details of the calculation are given in the ‘Materials and methods’ . We emphasize that these curves are based only on the docked structures , and the native binding mode in the X-ray structure of the complex is not used . The results show that the hypothesis is certainly true for EIN/HPr-E5C , because the full encounter ensemble provides much better approximation than the native complex , particularly for residues 59–97 and 160–167 ( Figure 2A ) . The correlation coefficient between the experimental PRE rates and the ones based on the encounter complexes is 0 . 705 . In contrast , the correlation coefficients between the experimental PRE rates and the ones back-calculated from the native structure ( black line in Figure 2A ) is only 0 . 47 . We note that the agreement improves even for the interface residues in the 67–85 region of EIN ( indicated by red ticks in Figure 2 ) . The explanation is that the E5C-EDTA-Mn2+ label is on the far side of HPr from the interface , and in some encounter complexes this label is much closer to interface residues than in the native complex . For HPr with the E32C-EDTA-Mn2+ label , accounting for the encounter complexes generated by docking ( blue curve in Figure 2B ) improves the correlation coefficient more moderately , from 0 . 709 to 0 . 77 . This is due to the fact that E32C is close to the edge of the interface in the native complex , already providing a strong PRE signal for residues 67–85 , resulting in a high correlation coefficient with the experimental PRE values . As will be discussed , since the PRE data are sensitive to small conformational changes and thus are inherently noisy , it is difficult to further improve an already high correlation coefficient . However , even for HPr-E32C , the PRE profile back-calculated from the low energy models still yields better prediction than considering only the native binding mode . Since it is well known that the PRE profiles heavily depend on the location of the paramagnetic tag relative to the native interface ( Fawzi et al . , 2010 ) , this result does not contradict to our hypothesis that accounting for all structures generated by the docking improves the prediction of PRE rates . To provide a control and to demonstrate that the use of an accurate energy function is very important for generating a meaningful ensemble of encounter complexes we have also performed docking calculations using a scoring function without long-range energy terms , that is , considering only the attractive and repulsive components of the van der Waals energy . This simplified energy function yields docked structures that have good shape complementarity , but have no favorable electrostatic or chemical interactions . The 30 , 000 structures with the lowest van der Waals energy from this ‘shape-complementarity only’ docking were then used for back-calculating theoretical PRE profiles . The results of these calculations , shown in Figure 2—figure supplement 1 , make it absolutely clear that the back-calculated PRE profiles based on the ensemble of structures generated without a proper energy function do not show any resemblance to the observed PRE data . In fact , both correlation coefficients between theoretical and experimental PRE rates are negative , −0 . 36 and −0 . 58 , respectively , for the probes at positions E5C and E32C . Considering encounter complexes generated by PIPER using its physics based energy function we also obtained good agreement with experimental PRE data for other pairs of proteins . The first is the IIAMannitol/HPr complex ( Cornilescu et al . , 2002; Tang et al . , 2006 ) . Figure 2—figure supplement 2 , B show the native structure of the complex and the ensemble of docked structures generated by PIPER . The paramagnetic label is placed at E5C of HPr , colored red and indicated by a small circle in Figure 2—figure supplement 2A . It is important that , similarly to the EIN/HPr system , the E5C-EDTA-Mn2+ label is on the far side of HPr from the interface in the native structure of the complex . Figure 2—figure supplement 2C shows the experimental PRE data , the theoretical PRE profile based on the native complex ( black line ) , and the theoretical profile obtained by considering the 30 , 000 low energy structures generated by the docking ( blue line ) . The correlation coefficient between the experimental PRE rates and the ones back-calculated from the native structure is 0 . 58 , whereas using the docked structures for the PRE calculation increases the correlation coefficient to 0 . 78 , demonstrating substantially improved prediction . Although some of the improvements occur at the interface residues , it is clearly helpful that the HPr-E5C-EDTA-Mn2+ label is far from the interface . The distance between the label and a number of IIAMannitol residues is substantially reduced in some of the encounter complexes , which makes the presence of minor species more pronounced . We also show observed PRE data and theoretical profiles calculated from the ensemble of structures generated by docking for the complexes HPr/HPr ( Tang et al . , 2008; Figure 2—figure supplement 3 ) , and cytochrome c/cytochrome c peroxidase ( Bashir et al . , 2010 ) ( Figure 2—figure supplement 4 ) , demonstrating good qualitative agreement in both cases . Having established that the sampling algorithm and energy function are accurate enough for predicting ensembles of encounter complexes , we proceeded to the characterization of the energy landscape in a neighborhood of the native complex conformation . As in the previous section , we focused on the rigid-body motions of the ligand protein in the space fixed on the receptor protein , although local structural adjustments of the proteins were allowed for more accurate energy calculation . Geometrically the 6D translational/rotational space is the so-called Special Euclidean Group SE ( 3 ) , which is the semidirect product of R3 of the translations and SO ( 3 ) of the rotations ( Park , 1995; Park and Ravani , 1997; Shen et al . , 2008; Mirzaei et al . , 2012 ) . Restricting considerations to encounter complexes in which the surfaces of the two proteins touch each other removes the distance of the two proteins as a variable , and the space can be parameterized in terms of 5 angular coordinates . Two angles are needed to define the direction from the center of the receptor to the center of the ligand interface , and the other three angles specify the rotation of the ligand . Although the resulting space is nonlinear and thus the 5 angular coordinates are interdependent , a 5D Euclidean space can be mapped onto this nonlinear space using exponential maps ( Shen et al . , 2008 ) , and hence we will be able to use analysis tools such as PCA , developed for application in Euclidean space . Further details justifying the need for the use of exponential maps will be given in ‘Materials and methods’ . Once an appropriate coordinate system was defined , we selected and densely sampled a region in the neighborhood of the native state to obtain information on the shape of the binding funnel ( Camacho et al . , 1999; Selzer and Schreiber , 2001; Wang and Wade , 2003; Miyashita et al . , 2004 ) . Since the apparent properties of the landscape depend both on the energy evaluation model and the method of sampling , to assess the generality of the results we used both PIPER ( Kozakov et al . , 2006 ) and the very different docking program RosettaDock ( Gray et al . , 2003 ) , which is based on Monte Carlo minimization and rebuilds side chain conformations during the search . From each sampling calculation , performed either by PIPER or by RosettaDock , we selected the conformations below a certain energy threshold to delineate the floor of the energy funnel . Encounter complexes were generated from unbound protein structures ( Chen et al . , 2003 ) for a diverse set of 42 interacting protein pairs ( Table 1 ) . For each of these complexes , selected from the protein docking benchmark ( Chen et al . , 2003 ) , both PIPER and RosettaDock found an energy funnel near the native state . Since this is generally not the case for complexes involving multiple subunits or large conformational changes upon binding , such complexes in the benchmark set were not considered . Structures were retained within 10 Å IRMSD from the native state . After sampling , the exponential coordinates were normalized to ensure that the variances in the sample set are the same along each coordinate axis , and the shape of the energy landscape over the selected region was studied by applying principal component analysis ( PCA ) to 5% of conformations with the lowest energy values . As will be emphasized in ‘Materials and methods’ , the use of exponential maps , resulting in independent coordinates , is crucial for the success of our study , as only in this case can PCA separate the essential hyperspaces that bound the low energy ensemble . 10 . 7554/eLife . 01370 . 010Table 1 . Eigenvalues ( in % ) obtained by PCA , and the angle between restrictive subspacesDOI: http://dx . doi . org/10 . 7554/eLife . 01370 . 010PDB IDPIPERRosettaDockDiscrepancy ( degrees ) λ1λ2λ3λ4λ5λ1λ2λ3λ4λ51AVX59 . 432 . 86 . 21 . 20 . 367 . 415 . 513 . 43 . 40 . 351B6C72 . 119 . 16 . 91 . 30 . 584 . 210 . 23 . 31 . 90 . 441 E6E59 . 118 . 111 . 310 . 01 . 557 . 616 . 310 . 98 . 86 . 4291EAW44 . 331 . 722 . 21 . 00 . 957 . 933 . 04 . 64 . 00 . 4251 E6J78 . 713 . 57 . 10 . 30 . 347 . 631 . 518 . 71 . 21 . 0161GLA58 . 926 . 99 . 33 . 71 . 241 . 533 . 115 . 18 . 02 . 321IQD74 . 713 . 57 . 83 . 60 . 458 . 026 . 712 . 71 . 90 . 7131K7447 . 828 . 019 . 03 . 61 . 561 . 022 . 010 . 55 . 21 . 2191MAH60 . 322 . 411 . 74 . 41 . 252 . 822 . 013 . 57 . 74 . 0141N8O56 . 123 . 413 . 36 . 40 . 966 . 922 . 310 . 30 . 30 . 2201PPE56 . 426 . 414 . 91 . 70 . 647 . 144 . 47 . 90 . 40 . 131PXV68 . 317 . 09 . 64 . 30 . 832 . 127 . 223 . 614 . 52 . 781 R0R55 . 026 . 815 . 32 . 70 . 269 . 320 . 26 . 73 . 00 . 9132SNI49 . 331 . 617 . 01 . 50 . 679 . 615 . 53 . 51 . 20 . 2161KXQ47 . 730 . 016 . 84 . 11 . 366 . 330 . 04 . 40 . 20 . 1297CEI44 . 728 . 520 . 94 . 61 . 347 . 727 . 918 . 93 . 61 . 9192SIC58 . 623 . 49 . 47 . 21 . 484 . 28 . 84 . 12 . 40 . 531AY756 . 920 . 315 . 05 . 42 . 442 . 132 . 512 . 69 . 23 . 7271OPH72 . 615 . 59 . 22 . 20 . 584 . 29 . 35 . 90 . 40 . 2211UDI64 . 618 . 612 . 62 . 41 . 851 . 227 . 114 . 26 . 41 . 1331BUH44 . 827 . 717 . 69 . 20 . 640 . 332 . 916 . 68 . 21 . 9211FSK45 . 028 . 022 . 13 . 51 . 442 . 629 . 719 . 95 . 81 . 9211JPS57 . 125 . 712 . 44 . 00 . 856 . 328 . 813 . 70 . 70 . 6301DQJ51 . 431 . 315 . 01 . 40 . 946 . 519 . 817 . 412 . 34 . 0172B4255 . 627 . 712 . 83 . 40 . 545 . 423 . 115 . 711 . 54 . 4242FD665 . 118 . 19 . 94 . 62 . 236 . 423 . 921 . 013 . 75 . 0202HQS80 . 111 . 37 . 21 . 00 . 454 . 136 . 27 . 81 . 60 . 492I2570 . 318 . 59 . 90 . 80 . 556 . 015 . 312 . 810 . 55 . 3122MTA45 . 826 . 120 . 55 . 22 . 545 . 532 . 212 . 27 . 42 . 7301MLC59 . 331 . 56 . 81 . 21 . 142 . 730 . 117 . 37 . 02 . 8172HRK57 . 731 . 09 . 70 . 90 . 761 . 216 . 410 . 58 . 63 . 3301AHW74 . 416 . 37 . 71 . 10 . 546 . 731 . 517 . 23 . 11 . 5241Z5Y66 . 118 . 08 . 95 . 41 . 557 . 132 . 39 . 70 . 60 . 4292HLE54 . 028 . 113 . 03 . 41 . 467 . 012 . 912 . 15 . 92 . 122NZ869 . 414 . 18 . 45 . 22 . 944 . 828 . 315 . 17 . 34 . 5341BVN61 . 420 . 314 . 33 . 60 . 437 . 228 . 517 . 810 . 06 . 5321CGI66 . 915 . 410 . 95 . 41 . 957 . 927 . 910 . 73 . 00 . 5481GPW53 . 820 . 914 . 26 . 44 . 746 . 526 . 517 . 95 . 04 . 0272JEL72 . 416 . 38 . 12 . 20 . 947 . 831 . 612 . 96 . 80 . 8311NCA76 . 718 . 83 . 01 . 20 . 455 . 924 . 315 . 63 . 21 . 0272UUY69 . 415 . 412 . 71 . 60 . 973 . 616 . 29 . 01 . 00 . 2111KAC48 . 934 . 613 . 32 . 01 . 253 . 819 . 515 . 66 . 84 . 230 The eigenvalues obtained by PCA are normalized to add to 100% . Each eigenvalue λ i can be then interpreted as the percentage of the total variance that is accounted for by the variance along the corresponding eigenvector vi . The smallest eigenvalue , λ5 , is less than 5% for almost all complexes ( Table 1 ) . In many cases both λ5 and λ4 are small ( their sum is less than 10% ) , indicating that the eigenvectors v4 and v5 span a ‘restrictive’ subspace where the low energy structures barely deviate from the native complex . In contrast , λ1 and λ2 typically sum up to more than 75% of variance . Thus , it is expected that in the ‘permissive’ subspace spanned by v1 and v2 the low energy structures may substantially differ from the native conformation . As an example , Figure 3A shows IRMSD and energy distributions along the five eigenvectors , calculated from the low energy structures generated by PIPER , for the complex between the retinoid X-receptor α ( RXRα ) and the peroxisome proliferator-activated receptor γ ( PPARγ ) , considered here as the receptor and the ligand , respectively . The PDB entry of the complex is 1K74 , but we docked the unbound ( separately crystallized ) RXRα and PPARγ structures rather than the components from the complex . Figure 3B shows the distributions of the same quantities , but based on the low energy structures generated by RosettaDock . The largest eigenvalue , λ1 , is close to 50% for both energy functions . The corresponding movements along v1 are rotations of helix H12 of PPARγ around a hydrophobic patch , formed by the side chains of F432 , A433 , and L436 , which binds to a large hydrophobic pocket of RXRα and remains almost at the same position in all low energy encounter complexes ( Figure 3C , Figure 3—figure supplement 1; Video 1 ) . As helix H12 rotates , the entire PPARγ moves with it until a loop formed by PPARγ residues 394 to 403 reaches a favorable position on the surface of RXRα . We note that hydrophobic patch on helix H12 and the residues connecting it to the rest of the protein ( residues 413–433 ) are known to be essential for forming the heterodimer ( Chan and Wells , 2009 ) . Along the eigenvector v2 that correspond to the second largest eigenvalue λ2 we can observe how the amino end of helix H12 with the hydrophobic patch on it moves into its binding pocket ( Video 2 ) . Based on the eigenvalues λ1 and λ2 ( Table 1 ) , over 75% of all movement of PPARγ approaching RXRα occurs in the subspace spanned by the eigenvectors v1 and v2 . Thus , this subspace can be regarded as the essential consensus of a very large number of association trajectories . In contrast to the permissive subspace , changes are very small along v5 ( Figure 3D ) . Since the higher energy structures ( not included in the data considered for PCA ) can be substantially further from the native state than the ones with low energy , we conclude that the valley based on energy is much narrower than the valley based on geometry . 10 . 7554/eLife . 01370 . 011Figure 3 . Shape of the energy landscape along the five PCA eigenvectors for the complex of PPAR-γ and RXR-α ( PDB code 1K74 ) . ( A ) Distributions of IRMSD ( green ) and energy ( cyan ) values based on structures generated by PIPER as functions of the ‘balanced’ coordinates shown on the x-axis . Dark blue diamonds indicate low energy data points used for the PCA . The IRMSD ( y-axis in the left column ) is given in Å . The energy values ( on the y-axis in the right column ) are given by the PIPER scoring function . ( B ) Same as Figure 3A , but based on structures generated by RosettaDock . The energy values ( on the y-axis in the right column ) are given by the RosettaDock scoring function . ( C ) Encounter complexes along the most permissive direction v1 . The ensemble includes mostly translations from the native state . ( D ) Encounter complexes along the most restrictive direction v5 . DOI: http://dx . doi . org/10 . 7554/eLife . 01370 . 01110 . 7554/eLife . 01370 . 012Figure 3—figure supplement 1 . Helix H12 of PPARγ with residues of the hydrophobic patch indicated . The receptor , RXRα , is shown in surface representation . Color code: oxygen red , nitrogen blue , and carbon white . DOI: http://dx . doi . org/10 . 7554/eLife . 01370 . 01210 . 7554/eLife . 01370 . 013Video 1 . Movement of PPARγ , shown as green cartoon , along the most permissive eigenvector v1 . The receptor , RXRα , is shown as grey surface . DOI: http://dx . doi . org/10 . 7554/eLife . 01370 . 01310 . 7554/eLife . 01370 . 014Video 2 . Movement of PPARγ , shown as green cartoon , along the second most permissive eigenvector v2 . The receptor , RXRα , is shown as grey surface . DOI: http://dx . doi . org/10 . 7554/eLife . 01370 . 014 As a second example , we show IRMSD and energy distributions and PCA results for an enzyme–inhibitor complex , subtilisin Carlsberg and its protein inhibitor , OMTKY3 ( Figure 4 ) . The PDB entry of the complex is 1R0R . For this pair of proteins , the low energy encounter complexes along the eigenvectors v4 and v5 show even narrower distributions than in the previous example , both for PIPER and RosettaDock ( Figure 4A , B ) . Since the essentially planar inhibitor loop ( residues 13 to 19 of OMTKY3 ) is locked into the crevice at the enzyme’s active site , we expected that the motion along the most permissive direction would be the rigid body rotation of the entire inhibitor , possibly with slight readjustments of the loop . However , we have found that the motion along v1 is the move of the loop , and particularly the primary specificity residue L18 , deeper into the binding pocket of the enzyme ( Figure 4C , Figure 4—figure supplement 1; Video 3 ) . The rotation along the loop is also present , but along the eigenvector v2 rather than v1 ( Video 4 ) . Based on the eigenvectors λ1 and λ2 ( Table 1 ) , 81 . 8% of the movements of OMTKY3 upon binding occurs in the subspace spanned by eigenvectors v1 and v2 for this complex . In contrast , the motion along the most restrictive direction v5 is a very small translation along the bottom of the active site ( Figure 4D ) . It is important to note that , in principle , small eigenvalues identified by PCA could have also occur by chance due to undersampling a subspace . We performed simple Monte Carlo analyses to exclude this possibility ( ‘Materials and methods’ ) . 10 . 7554/eLife . 01370 . 015Figure 4 . Shape of the energy landscape along the five PCA eigenvectors for the complex of subtilisin Carlsberg and its protein inhibitor , OMTKY3 . All notations are as in Figure 3 . ( A ) Distributions of interface IRMSD and energy values based on the structures generated by PIPER . ( B ) Same as Figure 4A , but based on the RosettaDock dataset . ( C ) Encounter complexes along the most permissive direction v1 . The ensemble consists of small rotations that leave the inhibitory loop position largely invariant . ( D ) Encounter complexes along the most restrictive direction v5 . DOI: http://dx . doi . org/10 . 7554/eLife . 01370 . 01510 . 7554/eLife . 01370 . 016Figure 4—figure supplement 1 . Movement of the OMTKY3 inhibitory loop into the active site of subtilisin Carlsberg . Two snapshot of the motion are shown ( in green and cyan ) for residues 16 to 19 ( CTLE ) , with L18 indicating the primary specificity residue . DOI: http://dx . doi . org/10 . 7554/eLife . 01370 . 01610 . 7554/eLife . 01370 . 017Video 3 . Movement of the protein inhibitor , OMTKY3 , shown as green cartoon , along the most permissive eigenvector v1 . The receptor , subtilisin Carlsberg , is shown as grey surface . DOI: http://dx . doi . org/10 . 7554/eLife . 01370 . 01710 . 7554/eLife . 01370 . 018Video 4 . Movement of the protein inhibitor , OMTKY3 , shown as green cartoon , along the second most permissive eigenvector v2 . The receptor , subtilisin Carlsberg , is shown as grey surface . DOI: http://dx . doi . org/10 . 7554/eLife . 01370 . 018 As shown by the eigenvalues in Table 1 and by Figures 3 and 4 , the energy funnels derived from the PCA of the energy landscapes generated by PIPER and RosettaDock slightly differ . This is not surprising , because we specifically selected two docking programs that are very different both in terms of their sampling algorithms and scoring functions ( ‘Materials and methods’ ) . PIPER performs systematic rigid body sampling on a dense grid using a ‘smooth’ potential that allows for some overlaps ( Kozakov et al . , 2006 ) . In contrast , RosettaDock samples the region of interest using a Monte Carlo minimization algorithm , which biases the search toward low energy regions , and thus the sampling is less exhaustive than the systematic sampling by PIPER . The method periodically rebuilds the complete set of interface side chains , followed by the optimization of the rigid body displacement . The energy is locally minimized in every iteration cycle of a Monte Carlo search algorithm ( Gray et al . , 2003 ) , and since the clashes are continuously removed , RosettaDock can use an energy function that is more sensitive to small changes in the coordinates than the energy function used in PIPER . Accordingly , Panels A and B of Figures 3 and 4 display somewhat different shapes of the energy distributions around the native state placed at the center of the coordinate system . Using rigid structures without local minimization , the minimum of the PIPER energy may be somewhat shifted from the native state , resulting in a more skewed energy landscape than the one obtained with RosettaDock , which generally places the energy minimum closer to the native structure and hence has a more symmetric energy landscape . In view of the differences between the two energy functions we consider it important that the PCA analyses of low energy structures generated by either PIPER or RosettaDock yield similar distributions of the eigenvalues for all 42 complexes . For each complex , both programs result in two small eigenvalues λ5 and λ4 . Although these eigenvalues are generally somewhat smaller for PIPER , because the rigid body approximation leads to a steeper increase in energy as we move away from the minimum along the most restrictive direction , both programs clearly indicate the existence of a restrictive subspace . In addition , Monte Carlo studies , described in the ‘Materials and methods’ , show that the restrictive subspaces predicted PIPER and RosettaDock are fairly similar . The similarity of these subspaces obtained by two very different energy functions for all 42 complexes indicates that the reduction of dimensionality is an inherent property of protein–protein association . Both programs predict that , on the other end of the spectrum , the two largest eigenvalues , λ1 and λ2 , together exceed 75% of the total variance for most complexes . Thus , in a neighborhood of the native state the encounter complexes are essentially restricted to a two dimensional permissive subspace in the rotational/translational space , and this conclusion is independent of the docking program used . Assuming moderate conformational changes and using grid approximation , the FFT based global and systematic sampling of the configurational space of two interacting proteins using a physics based energy function converts the docking problem into an exactly solvable problem of statistical mechanics ( Kozakov et al . , 2013 ) . According to the CAPRI community-wide protein–protein docking experiment , this type of approximation gives good results for a large fraction of complexes ( Lensink and Wodak , 2013 ) . However , it has been well known that , for most protein pairs , such global search yields low energy structures in several regions of the conformational space , some of which are far from the structure of the native complex . Physics-based energy functions are expected to be globally valid for modeling interactions between proteins , including the non-native states . Thus , one can assume the energy values that are low relative to the average energy but still exceed the energy at the global minimum may lead to the formation of relatively short-lived encounter complexes along the association pathways . As shown in this paper , the agreement between experimental PRE data and theoretical PRE profiles calculated from the ensemble of structures generated by docking confirms this hypothesis , and thus structures of encounter complexes can be obtained simply as byproducts of docking without any further computational expense . While this result is not unexpected , in view of the limited structural information available on encounter complexes it is potentially significant . To detect intermediate structures in the association of proteins EIN and HPr , paramagnetic labels were introduced at two sites on HPr , one at a time , and the transverse paramagnetic relaxation enhancement ( PRE ) rates , Γ2 , of the backbone amide protons ( 1HN ) of EIN were measured . Since the population of the intermediate structures is generally much lower than the population of the native complex , it is important to discuss why PRE can detect the presence of encounter complexes . The major factor is that the magnitude of the PREs is proportional to < r−6> , where r is the distance between the nucleus of interest and the paramagnetic center , and <> denotes averaging over the ensemble of structures . Due to the large magnetic moment of an unpaired electron , the effect is detectable for sizeable separations ( up to ∼34 Å for Mn2+ ) . A hypothetical example can be used to explain why the method can detect states with very low populations . We consider an ensemble that includes a major species A with the population , pA , of 99% , and with a paramagnetic center to proton distance of 30 Å , and a minor species B with the population , pB , of 1% , and with a paramagnetic center to proton distance of 8 Å . We calculate Γ2 for this proton in a two-site exchange system between A and B , where Γ2 is defined as the difference in the transverse relaxation rates of the paramagnetic and diamagnetic states ( Iwahara and Clore , 2006 ) . For a ∼30-kDa complex , for species A the 1H-Γ2 arising from Mn2+ is ∼2 s−1 ( Γ2 , A ) , and for species B it is ∼5 . 6 × 103 s−1 ( Γ2 , B ) . Considering B as a short-lived encounter complex and A as the native state , and assuming that the system is in the fast exchange regime , the apparent PRE rate , Γ2 , is the population weighted average of the Γ2 rates of the two species , that is , Γ2 = pA Γ2 , A + pB Γ2 , B ( Iwahara and Clore , 2006 ) . Based on this expression Γ2 is ∼30-fold larger than Γ2 , A , thereby permitting one to both infer the presence of , and obtain some structural information on , the minor species , because the PRE is a highly distance-dependent quantity . Thus , according to this simple explanation , the PREs can clearly capture the footprint of minor species that exchange rapidly with the native complex , in spite of their much lower concentration . In a realistic protein–protein association the PRE rate , Γ2 , is the population weighted average of the Γ2 rates over the native state and the entire ensemble of encounter complexes . The strong distance dependence of Γ2 implies that the observed values are sensitive even to small conformational changes that may occur , for example , due to changes in the rotameric state of the EDTA-Mn2+ paramagnetic probe . Thus , as shown in Figure 2 , the PRE data , while sensitive to the presence of minor species , are also fairly noisy ( ‘Materials and methods’ ) . In spite of their substantial variance , the data are informative , since PREs generally also occur at residues that are far from the paramagnetic label in the native complex but are getting closer to it in some members of the encounter ensemble , clearly indicating the presence of non-native transition states . As the examples studied in this paper show , the minor species can be better detected if the label is far from the interface . In fact , a label placed close to interface generates a strong PRE signal , and thus the PRE profile back-calculated from the native structure already correlates well with the data . However , accounting for the encounter complex ensemble most likely improves the prediction even in such cases , but the improvement is generally smaller than the one with the paramagnetic label placed far from the interface . The reduction of dimensionality in molecular association was originally proposed to explain high binding rates ( von Hippel and Berg , 1989 ) , particularly the ability of proteins to locate their target sites along DNA ( Riggs et al . , 1970 ) . Dimensionality reduction is caused by interaction forces that are non-specific and thus do not lead to binding at a specific site , but keep the macromolecules in proximity for a prolonged time , allowing an extensive search of the surface along certain directions while restraining the search along others ( Ubbink , 2009 ) . This is clearly the case for DNA , whose negative charge attracts positively charged proteins without providing a specific interaction site ( Iwahara et al . , 2006; Gorman and Greene , 2008 ) . It is well known that long-range electrostatic interactions can also increase the rates of association of two proteins with net opposite charges or with strong charge dipoles , as the search for the reactive patches is facilitated by dipolar pre-orientation of the proteins upon their approach ( Schreiber et al . , 2009 ) . Such charge interactions prolong the lifetime of the transition state and increase the fraction of productive complexes , and thus can reduce dimensionality . However , it is frequently assumed that , due to specific charge–charge interactions and their irregular surface , proteins do not have ensembles of orientations having similar energies and thus allowing for search along the surface . We have shown here that this is definitely not the case because the energy landscape of interacting proteins , at least within the 10 Å IRMSD neighborhood of the native state , always includes a permissive subspace along which the conformation of the complex can substantially change without crossing significant energy barriers . Thus , there is no reason to assume that the interactions are nonspecific in protein-DNA association but are specific when two proteins associate . In fact , for all 42 protein pairs , some of which have strong electrostatic interactions , the energy landscape is smooth funnel in a two dimensional permissive subspace . In all cases this subspace captures at least 75% of the total motion as the two molecules approach the native state . For each of the 42 complexes we also detect a high energy subspace , which reduces the dimensionality of the space available to encounter complexes along the association pathways . Thus , there is much less difference between protein-DNA and protein–protein association than it was previously believed . Finally we note that the reduced dimensionality of the search space can potentially simplify docking calculations , and thus the results of PCA provide several opportunities for improving the efficiency of docking methods . First , it is well known that any type of optimization is more efficient along the principal components , as large steps can be taken along the permissive directions . This is particularly the case for second-order methods such as the Newton–Raphson optimization that uses a local quadratic approximation of the energy function to find the next minimum in each iteration ( Fletcher , 1981 ) . It is also important that such methods require the inversion of the Hessian matrix of the energy function , and in the case of reduced dimensionality the matrix is nearly singular , leading to numerical difficulties and loss of accuracy . Once such directions are known , the problem can be avoided by regularization methods ( Fletcher , 1981 ) . In fact , after developing and testing a medium-range optimization method SDU , which employs quadratic semi-definite underestimation in the 5D angular space with the exponential parameters also used in this work ( Shen et al . , 2008 ) , we understood that the approach can be made more efficient by accounting for the reduced dimensionality of the search space and adding regularization based on PCA . Another potential use is optimally selecting perturbation vectors using the relative magnitudes of the eigenvalues in biased Monte Carlo methods ( Lee et al . , 1996 ) . In order to fully explore the conformational space in protein–protein association we perform exhaustive evaluation of an energy function in the discretized space of mutual orientations of the two proteins using the docking program PIPER , which is based on the Fast Fourier transform ( FFT ) correlation approach ( Kozakov et al . , 2006 ) . The center of mass of the first protein , defined here as the receptor , is fixed at the origin of the coordinate system , whereas the second protein ( usually the smaller of the two ) , defined as the ligand , is rotated and translated . The translational space is represented as a grid of 1 . 0 Å displacements of the ligand center of mass , and the rotational space is sampled using 70 , 000 rotations based on a deterministic layered Sukharev grid sequence , which quasi-uniformly covers the space . The energy expression used for the FFT based sampling includes simplified van der Waals energy Evdw with attractive ( Eattr ) and repulsive ( Erep ) contributions , the electrostatic interaction energy Eelec , and a statistical pairwise potential Epair , representing other solvation effects ( Chuang et al . , 2008 ) :E=Evdw+w2Eelec+w3Epair The individual energy terms are calculated by the Evdw=Eattr+w1Erep , Eelec=∑i∑j[qiqj/{r2+D2 exp ( −r2/4D2 ) }1/2] , and Epair = Σi Σj εij , where r is the distance between atoms i and j , D is an atom-type independent approximation of the generalized Born radius , and εij is a pairwise interaction potential between atoms i and j . All energy expressions are defined on the grid . The coefficients w1 = 4 , w2 = 600 , w3 = 5 , weight the different contributions to the scoring function , and are based on calorimetric considerations . In order to evaluate the energy function E by FFT , it must be written as a sum of correlation functions . The first two terms , Evdw and Eelec , satisfy this condition , whereas Epair is written as a sum of a few correlation functions , using an eigenvalue-eigenvector decomposition ( Kozakov et al . , 2006 ) . For each rotation , this expression can be efficiently calculated using P forward and one inverse Fast Fourier transforms . The calculations are performed for each of the 70 , 000 rotations , and one or several lowest energy translations for each rotation are retained . The results are clustered with a 10 Å IRMSD radius around the native coordinate . Unbound structures were used both for the receptor , EIN ( chain A from PDB ( Berman et al . , 2000 ) entry 1ZYM ) and for the ligand , HPr ( chain P from PDB entry 2JEL ) . Encounter complexes were generated using the global systematic Fast Fourier Transform based docking program PIPER ( Kozakov et al . , 2006 ) . The docking was performed with each of the three conformers of EDTA-Mn2+ group , both at positions E5C and E32C of HPr ( Fawzi et al . , 2010 ) . For each conformer , the 10 , 000 lowest energy complex structures were retained for the calculation of PRE rates . To calculate the transverse PRE rates ( Γ2 ) from the ensemble of encounter complexes generated by the FFT based sampling we use the Nst = 30 , 000 ( 10 , 000 for each of the three conformers of EDTA-Mn2+ ) lowest energy structures . The observed PRE values are the PRE rates averaged over a population ( Iwahara et al . , 2004; Tang et al . , 2006 ) , and hence<Γ2 ( i ) >= ( ∑jΓji ) /Nstwhere Γji is the PRE rate for residue i of the jth structure in the ensemble . Each individual Γji value is proportional to the inverse sixth power of distance rij between the backbone amide proton ( directly bonded to 15N ) of the ith residue and the paramagnetic ion Mn2+ . To account for magnetic trap flexibility , we use the three state EDTA-Mn2+ Solomon-Bloembergen approximation ( Iwahara et al . , 2004; Tang et al . , 2006 ) . With these assumptions the PRE rates are given by Γji = C × Σk ( rikj ) −6 , where C = 1 . 2 × 1010 Å6/s and rikj is the distance between residue i and the kth state of EDTA-Mn2+ in the jth structure in the low energy ensemble of docked configurations . The Γji values are limited to 90s−1 , that is , Γji = 90 if Γji ≥ 90 . Introducing this threshold is based on the observation that both theoretical and experimental Γji values become very uncertain over this threshold because small distance variations strongly affect the result ( Kim et al . , 2008; Fawzi et al . , 2010 ) . The free energy landscape near the native complex was explored using both PIPER , based on the Fast Fourier transform ( FFT ) correlation approach ( Kozakov et al . , 2006 ) and RosettaDock , a docking program based on the Monte Carlo minimization ( MCM ) algorithm ( Gray et al . , 2003 ) . In each MCM cycle , RosettaDock perturbs the position of the ligand by random translations and rotations , followed by adjusting the distance between the ligand and receptor to create a contact . Next , a fast MCM at low resolution optimizes the complex orientation with respect to features that do not depend on the explicit conformations of the side chains . Finally , the side chains are added , and an all-atom optimization locates the local minimum energy conformation . The complete set of interface side chains is repacked every eight cycles , followed by the optimization the rigid body displacement . After each move , side chain packing , and minimization , an energy score is calculated . The new position is kept or rejected according to the standard Metropolis acceptance criterion ( Gray et al . , 2003 ) . RosettaDock uses a detailed energy function which includes van der Waals interactions with a linear term serving as the repulsive part , a solvation term based on a pairwise Gaussian solvent exclusion model , hydrogen bonding energies using an orientation-dependent empirical function , a rotamer probability term , residue–residue atom pair interactions for charged residues , and a simple electrostatic term across the protein–protein interface . The origin of the reference frame is placed at the center of the receptor , and the z-axis is directed toward the center of the interface of the native conformation . The translation of the ligand is described by the vector from the center of the receptor to the center of the ligand interface ( as opposed to the center of the ligand ) , and the rotation of the ligand is also defined around the center of its interface . This choice is made to decouple , as much as possible , the effects of translation on the locations of the interface atoms from those of the rotation . A translation vector y ∈ R3 can be represented by a triplet ( r , θ , φ ) , where r = \|\|y\|\| is the distance , θ is the azimuth angle between the projection of y on the xy-plane and the x-axis ( longitude , 0 ≤ θ < 2π ) , and φ is the zenith angle between the z-axis and the vector y ( colatitude , 0 ≤ φ < π ) . As will be discussed , we consider r separately , and focus on selecting appropriate parameterizations for ( θ , φ ) compounded with the rotational space SO ( 3 ) . Parameterizing rotations is problematic because rotations are non-Euclidean in nature ( i . e . , travelling infinitely far in any direction will bring you back to your starting point an infinite number of times ) . Any attempt to parameterize the entire set of three degrees-of-freedom ( DOF ) rotations by an open subset of Euclidean space ( as do Euler angles ) will suffer from gimbal lock , the loss of rotational degrees of freedom , due to singularities in the parameter space ( Grassia , 1998 ) . Parameterizations that are themselves defined over non-Euclidean spaces ( such as the set of unit quaternions embedded in R4 ) may remain singularity-free , and thus avoid gimbal lock . Employing such parameterizations is complicated , however , since the numerical tools such as optimization and PCA assume Euclidean parameterizations; therefore we must either develop new tools whose domains are non-Euclidean , or complicate our systems by imposing explicit constraints , for example , to assure that the quaternions stay on the unit sphere ( Grassia , 1998 ) . In this paper we use exponential maps ( Park , 1995; Park and Ravani , 1997; Shen et al . , 2008 ) to project an Euclidean space onto the nonlinear rotational space . As an example , the simplest exponential map defines a local one-to-one correspondence between the unit circle , which is a nonlinear space ( called the circle group ) , centered at 0 in the complex plane , and the tangent space at 1 , which can be identified with the imaginary line in the complex plane . The exponential map for the circle group is given by it → eit , where i = √ ( −1 ) , it specifies a point on the tangent line , and the exponential function projects it into the corresponding point of the circle ( Figure 5A ) . Extending the map to three dimensions , the exponential map projecting the tangent plain of a 3D sphere onto the surface of the sphere is shown in Figure 5B . 10 . 7554/eLife . 01370 . 019Figure 5 . Examples of simple exponential maps . ( A ) Parameterization of the unit circle using an exponential map . The function eit is a local one-to-one mapping of the tangent line around p = 1 onto the unit circle . ( B ) Prameterization of the 3D unit sphere using exponential parameters . DOI: http://dx . doi . org/10 . 7554/eLife . 01370 . 019 For ( θ , φ ) the exponential coordinates are ( σ1 , σ2 ) = ( -φ sinθ , φ cosθ ) . For SO ( 3 ) , the exponential coordinates are ω = ( ω1 , ω2 , ω3 ) ∈ R3 , where ω1 , ω2 , and ω3 are elements of a skew-symmetric matrix ( Park , 1995; Shen et al . , 2008; Mirzaei et al . , 2012 ) . The vector ( σ1 , σ2 , ω1 , ω2 , ω3 ) defines the relative orientation of the two rigid proteins . Once this relative orientation is given , the binding distance r along the translation vector y ( which connects the centers of the receptor with the center of the ligand’s interface ) is determined by the assumptions that the two proteins are in contact but do not overlap ( Shen et al . , 2008 ) . Complexes generated by the docking methods used in this work always have this property . In particular , given a complex structure generated by the FFT sampling approach PIPER ( Kozakov et al . , 2006 ) , we can uniquely determine the corresponding ( σ1 , σ2 , ω1 , ω2 , ω3 ) coordinates and the value of r . Since RosettaDock ( Gray et al . , 2003 ) generally changes the conformations of the component proteins , to isolate the rigid body motion we have to fit the initial protein structures to the complex and determine the closest rigid body transformation , projecting the higher dimensional motion into the rotational-translational space . Since small eigenvalues identified by PCA might also occur by chance due to undersampling a subspace , we performed a simple Monte Carlo analysis to show that this is not the case . PCA was based on at least 100 low energy structures for each of the 42 complexes , and thence we generated 100 random vectors in the 5D space , applied PCA to derive the eigenvalues , and performed this experiment 1000 times . Results confirmed that the probability of λ5 < 10% is less than 0 . 01 . Thus , in view of the large number of sample points we used , it is very unlikely that the small eigenvalues shown in Table 1 occur due to undersampling . The PCA analyses of the 5D exponential coordinates of the low energy complex conformations show clear distinctions between permissive and restrictive directions in all 42 cases . The rigid FFT-based method ( Kozakov et al . , 2006 ) and the Monte Carlo approach in RosettaDock ( Gray et al . , 2003 ) yield similar results , in spite of the fact that two methods implement fundamentally different sampling and scoring schemes . To perform a rigorous comparison of the results from the two methods we introduce the notations λ1P , . . . , λ5P and λ1R , . . . , λ5R for the eigenvalues based on the low energy structures generated by , respectively , PIPER and RosettaDock . Both sets of eigenvalues are ordered in descending magnitude , and v1P , . . . , v5P and v1R , . . . , v5R denote the corresponding eigenvectors . Since our main goal is to show that the restrictive subspace is largely independent of the method used , as a measure of discrepancy we will determine the angle between the restrictive subspace spanned by eigenvectors v4P and v5P based on PIPER , and the subspace spanned by v4R and v5R based on RosettaDock . As shown in the last column of Table 1 , this angle is less than or equals to 30° for all but 5 of the complexes . Accepting that the restrictive subspaces match if they differ by less than 30° seems to be a somewhat relaxed condition . However , it is easy to show by Monte Carlo simulations that the probability of such agreement by chance for 37 of the 42 structures is negligibly small . We applied PCA to two sets of 100 random vectors in 5D , calculated the covariance matrix and its eigenvalues , for each set we selected the subspace spanned by eigenvectors corresponding to the two smallest eigenvalues , and finally determined the angle between these two ‘restrictive’ subspaces . By repeating this calculation 1000 times we could show that the probability of obtaining an angle below 30° is p=0 . 131 . Although this is not a very small number , the probability that this occurs for 37 of the 42 complexes is less than 10−22 . Thus , the results overwhelmingly support the claim that the restrictive directions found by two very different methods are similar , and thus the reduction in dimensionality is an inherent property of protein–protein association . Due to the orthogonality of eigenvectors , the same similarity between PIPER and RosettaDock results also applies to the permissive subspaces spanned by the first 3 eigenvectors . Additionally we provide Figures and Videos analogous to Figures 3 , 4 and Videos 1–4 for all 42 complexes studied in this paper ( Kozakov et al . , 2014 ) .	Proteins rarely act alone . Instead , they tend to bind to other proteins to form structures known as complexes . When two proteins come together to form a complex , they twist and turn through a series of intermediate states before they form the actual complex . These intermediate states are difficult to study because they don’t last for very long , which means that our knowledge of how complexes are formed remains incomplete . One promising approach for studying the formation of complexes is called paramagnetic relaxation enhancement . In this technique certain areas in one of the proteins are labelled with magnetic particles , which produce signals when the two proteins are close to each other . Repeating the measurement several times with the magnetic particles in different positions provides information about the overall structure of the complex . Computational modelling can then be used to work out the fine details of the structure , including the shapes of the intermediate structures made by the proteins as they interact . A computer method called docking can be used to predict the most favourable positions that the proteins can take , relative to one another , in a complex . This involves calculating the energy contained in the system , with the correct structure having the lowest energy . Docking methods also predict protein models with slightly higher energies , but with structures that are radically different . Modellers usually ignore these structures , but comparing the docking results to paramagnetic relaxation enhancement data , Kozakov et al . found that these structures actually represent the intermediate states . Analysing the structure of the intermediate states revealed that the movement of the two proteins relative to one another is severely restricted as they form the final complex . Kozakov et al . found that proteins associate along preferred pathways , similar to the way a protein slides along DNA in the process of protein-DNA recognition . Knowing that the movement of the proteins is restricted in this way will enable researchers to improve the efficiency of docking calculations .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2014	Encounter complexes and dimensionality reduction in protein–protein association
Delineating the basic cellular components of cortical inhibitory circuits remains a fundamental issue in order to understand their specific contributions to microcircuit function . It is still unclear how current classifications of cortical interneuron subtypes relate to biological processes such as their developmental specification . Here we identified the developmental trajectory of neurogliaform cells ( NGCs ) , the main effectors of a powerful inhibitory motif recruited by long-range connections . Using in vivo genetic lineage-tracing in mice , we report that NGCs originate from a specific pool of 5-HT3AR-expressing Hmx3+ cells located in the preoptic area ( POA ) . Hmx3-derived 5-HT3AR+ cortical interneurons ( INs ) expressed the transcription factors PROX1 , NR2F2 , the marker reelin but not VIP and exhibited the molecular , morphological and electrophysiological profile of NGCs . Overall , these results indicate that NGCs are a distinct class of INs with a unique developmental trajectory and open the possibility to study their specific functional contribution to cortical inhibitory microcircuit motifs . Cortical microcircuit function relies on the coordinated activity of a variety of GABAergic interneuron subtypes , which play critical roles in controlling the firing rate of glutamatergic pyramidal neurons , synchronizing network rhythms and regulating behavioral states ( Cardin et al . , 2009; Fu et al . , 2014; Kepecs and Fishell , 2014; Pfeffer et al . , 2013; Pi et al . , 2013; Pinto and Dan , 2015; Sohal et al . , 2009; Zhang et al . , 2014 ) . Different subtypes of cortical interneurons ( INs ) emerge during development and their specification arises through the complex interaction of cell-intrinsic mechanisms and cell-extrinsic cues ( Bartolini et al . , 2013; Fishell and Rudy , 2011; Huang , 2014; Kessaris et al . , 2014 ) . Cortical INs are generated in a variety of subpallial regions and the combinatorial expression of transcription factors ( TFs ) in these domains is believed to play a critical role in their fate specification ( Kessaris et al . , 2014; Anastasiades and Butt , 2011; Flames et al . , 2007; Wonders and Anderson , 2006 ) . The largest fraction ( about 60–70% ) of cortical INs is generated from NKX2 . 1-expressing progenitors located in the medial ganglionic eminence ( MGE ) ( Butt et al . , 2008; Xu et al . , 2008 ) and their specification is under the control of the TFs LHX6 ( Du et al . , 2008; Liodis et al . , 2007 ) and SOX6 ( Azim et al . , 2009; Batista-Brito et al . , 2009 ) . MGE-derived INs develop into fast-spiking parvalbumin ( PV ) + basket and chandelier cells , as well as into Martinotti and multipolar somatostatin ( SST ) + INs ( Butt et al . , 2008; Xu et al . , 2008; Du et al . , 2008; Butt et al . , 2005; Fogarty et al . , 2007; Taniguchi et al . , 2013 ) . The second largest fraction of cortical INs arises from the caudal ganglionic eminence ( CGE ) ( Miyoshi et al . , 2010; Nery et al . , 2002 ) and expresses TFs such as PROX1 , SP8 and NR2F2 ( Cai et al . , 2013; Ma et al . , 2012; Miyoshi et al . , 2015; Rubin and Kessaris , 2013 ) . CGE-derived INs also express the ionotropic serotonin receptor 3A ( 5-HT3AR ) and give rise to a large diversity of INs , including reelin+ cells , vasointestinal peptide ( VIP ) +/calretinin+ bipolar cells and VIP+/cholecystokinin+ basket cells ( Miyoshi et al . , 2010; Armstrong and Soltesz , 2012; Prönneke et al . , 2015; Lee et al . , 2010; Murthy et al . , 2014; Vucurovic et al . , 2010 ) . Finally , lineage-tracing experiments using Hmx3 ( Nkx5 . 1 ) -Cre ( Gelman et al . , 2009 ) and Dbx1-Cre driver lines ( Gelman et al . , 2011 ) have shown that a small fraction ( about 10% ) of cortical INs originate from the preoptic area ( POA ) ( Gelman et al . , 2009; Gelman et al . , 2011 ) . Among cortical INs , neurogliaform cells ( NGCs ) display unique characteristics . They represent the main source of ‘slow’ cortical inhibition by acting on metabotropic GABAB receptors ( Tamás et al . , 2003 ) , and are thought to be key effectors of a powerful inhibitory circuit recruited by long-range connections such as interhemispheric and thalamic projections ( Craig and McBain , 2014; Palmer et al . , 2012; De Marco García et al . , 2015 ) . Whether the current description of NGCs captures an IN subtype related to a distinct developmental specification process is unclear . Here we used in vivo genetic lineage-tracing to follow the developmental origin and trajectory of NGCs . We found that they originate from a distinct pool of 5-HT3AR-expressing Hmx3+ cells located in the rostral POA region , ventrally to the anterior commissure . In the embryonic POA , Htr3a-GFP+ INs in the Hmx3+ domain expressed CGE-enriched TFs such as PROX1 and NR2F2 , but only rarely , if not , MGE-related TFs such as NKX2 . 1 or LHX6 . In the cortex , Hmx3-derived Htr3a-GFP+ INs expressed markers of CGE-derived INs such as NPY and/or reelin , as well as CGE-enriched TFs such as SP8 , NR2F2 and PROX1 , but neither MGE-specific markers such as PV or SST nor TFs such as LHX6 or SOX6 . Finally , single-molecule in situ hybridization and electrophysiological recordings followed by post hoc reconstructions indicated that Hmx3-derived Htr3a-GFP+ cells exhibited the molecular , electrophysiological and morphological profile of NGCs . Taken together , these results demonstrate that cortical NGCs have a precise developmental trajectory that is linked to the expression of the transcription factor ( TF ) Hmx3 in a discrete embryonic subpallial region . To determine whether the POA could contribute to Htr3a-GFP INs , we crossed Htr3a-GFP; R26R-tdTOMfl/fl mice with Hmx3-Cre mice , a reporter line previously shown to fate map a population of cortical INs derived from cells located in the POA ( Gelman et al . , 2009 ) . Examination of brains from Hmx3-Cre::Htr3a-GFP; R26R-tdTOMfl/fl matings at embryonic age 14 . 5 ( E14 . 5 ) revealed a large fraction of Hmx3; tdTOM+ cells co-labelled with Htr3a-GFP ( 85 . 2 ± 0 . 9%; 1675/1986 cells in the overlap zone ) in a restricted region of the POA , located ventrally to the anterior commissure ( Figure 1A , C , Figure 1—source data 1 ) . Individual co-labelled Hmx3; tdTOM+/Htr3a-GFP+ cells displaying migratory profiles were observed at more caudal levels entering the CGE ( Figure 1B ) , suggesting that a fraction of Hmx3; tdTOM+/Htr3a-GFP+ cells migrate from the POA into the CGE . In situ hybridization indicated that the vast majority ( 95 . 8 ± 0 . 9%; 115/120 cells ) of Hmx3; tdTOM+/Htr3a-GFP+ cells located in the POA expressed the endogenous Htr3a mRNA , in contrast to Hmx3; tdTOM+cells negative for Htr3a-GFP ( 0 . 5 ± 0 . 5%; 1/158 cells ) ( Figure 1D , Figure 1—source data 1 ) . In addition , a large fraction of Hmx3; tdTOM+/Htr3a-GFP+ cells in the POA expressed the TFs PROX1 ( 54 . 0 ± 2 . 2%; 249/463 cells ) and NR2F2 ( 65 . 9 ± 1 . 3%; 795/1212 cells ) , which have previously been shown to be enriched in CGE-derived INs ( Cai et al . , 2013; Ma et al . , 2012; Miyoshi et al . , 2015; Rubin and Kessaris , 2013 ) , but more rarely the TF NKX2 . 1 ( 15 . 3 ± 1%; 175/1146 cells ) ( Figure 1E , F , Figure 1—source data 1 ) . Collectively , these results indicate that a fraction of Hmx3+ cells located in the POA express the 5-HT3AR and a pattern of TFs related to CGE-derived INs . To determine whether Hmx3; tdTOM+/Htr3a-GFP+ cells in the POA eventually give rise to a specific subpopulation of cortical INs , we examined brains at various postnatal ages . From P5 to P21 , Hmx3; tdTOM+/Htr3a-GFP+ INs were found distributed preferentially in superficial cortical layers and in a variety of other brain regions including in the hippocampus ( Figure 2A , Figure 2—figure supplement 1 ) . Hmx3; tdTOM+/Htr3a-GFP+ cells were rarely observed at postnatal ages in the subpallial brain regions corresponding to the embryonic POA ( i . e . , the preoptic nuclei ) ( Figure 2—figure supplement 2 ) . In situ hybridization for Htr3a mRNA indicated that Hmx3; tdTOM+/Htr3a-GFP+ cells expressed the Htr3a transcript , similarly to Htr3a-GFP+ cells negative for Hmx3; tdTOM ( Figure 2C ) . About half ( 51 . 9 ± 2 . 1%; 863/1653 cells ) of Hmx3-derived cells in the cortex were co-labelled with Htr3a-GFP+ and virtually all Hmx3; tdTOM+/Htr3a-GFP+ ( 96 . 1 ± 0 . 5%; 357/372 cells ) were positive for the neuronal marker NeuN ( Figure 2D , E , Figure 2—source data 1 ) . In contrast , the fraction of Hmx3; tdTOM+ cells negative for Htr3a-GFP mostly did not express NeuN ( 3 . 4 ± 1 . 3%; 28/758 cells ) ( Figure 2D , E , Figure 2—source data 1 ) , and remained relatively constant across postnatal ages ( Figure 2—figure supplement 3A , Figure 2—figure Supplement 3—source data 1 ) . These cells displayed the morphology of glial cells and expressed the astrocytic markers GFAP and S100β as well as the oligodendrocytic marker SOX10 ( Figure 2—figure supplement 3B–D ) . Overall , these findings indicate that the cortical Hmx3-derived lineage observed in the POA differentiate into INs that are Htr3a-GFP+ , glial cells that are Htr3a-GFP negative and , for a small fraction , to NeuN+ neurons negative for Htr3a-GFP . A second distinct region in the POA expressing Dbx1 was previously reported to give rise to subsets of cortical INs ( Gelman et al . , 2011 ) . To determine whether a fraction of Htr3a-GFP+ INs also originate from Dbx1-expressing cells , we examined Dbx1-Cre::Htr3a-GFP; R26R-tdTOMfl/fl brains at postnatal periods . While the overall contribution of Hmx3-derived cells to the Htr3a-GFP IN population in the cortex increased with postnatal maturation from P5 ( 6 . 8 ± 0 . 2%; 77/1118 cells ) to P21 ( 16 . 0 ± 0 . 3%; 863/5405 cells ) ( Figure 3A , C ) , only minimal fractions ( 1 . 44 ± 0 . 2%; 81/5741 cells at P5; 0 . 8 ± 0 . 2%; 20/2551 cells at P21 ) of Htr3a-GFP+ INs were fate-mapped with Dbx1; tdTOM ( Figure 3B , C , Figure 3—source data 1 ) . Moreover , Dbx1; tdTOM+ cells were preferentially found in deep cortical layers and expressed the MGE-enriched TF SOX6 ( 30 . 4 ± 2 . 2%; 82/266 cells ) , while PROX1 was found only in a very small fraction of Dbx1; tdTOM+ cells expressing also the Htr3a-GFP ( 2 . 2 ± 0 . 7%; 6/266 cells ) ( Figure 3D , Figure 3—source data 1 ) . In addition , Dbx1; tdTOM+ INs expressed Lhx6 mRNA ( Figure 3E ) , and the MGE-related markers SST and PV ( Figure 3—figure supplement 1 ) , and only very rarely the Htr3a mRNA ( Figure 3F ) . Overall , these results indicate that Hmx3-derived 5-HT3AR+ cortical INs largely originate from Hmx3-expressing cells but not from the Dbx1+ domain , which gives rise to INs expressing MGE-related markers . We next examined whether Hmx3; tdTOM+/Htr3a-GFP+ cells expressed distinct patterns of TFs involved in cortical IN subtype specification . At P21 , we found that , similarly to Htr3a-GFP+ INs , a large fraction ( 65 . 8 ± 3 . 4%; 202/308 cells ) of Hmx3; tdTOM+/Htr3a-GFP+ INs expressed the CGE-enriched but not the MGE-related TFs . Indeed , a large fraction of them ( 65 . 8 ± 3 . 4%; 202/308 cells ) expressed PROX1 but not SOX6 ( Figure 4A , B , Figure 4—source data 1 ) , as well as NR2F2 ( 32 . 7 ± 5 . 9%; 71/218 cells ) , SP8 ( 9 . 8 ± 2 . 6%; 22/218 cells ) , and both SP8 and NR2F2 ( 8 . 8 ± 2 . 0%; 18/218 cells ) ( Figure 4C , D , Figure 4—source data 1 ) . The fraction of Hmx3; tdTOM+/Htr3a-GFP+ expressing at least one of these two latter TFs was smaller and biased toward NR2F2 expression , when compared to Htr3a-GFP+ INs ( Figure 4D , Figure 4—figure supplement 1 , Figure 4—source data 1 ) . These findings indicate that Hmx3; tdTOM+/Htr3a-GFP+ cortical INs express a repertoire of TFs related to CGE but not to MGE-derived INs . We next examined whether Hmx3; tdTOM+/Htr3a-GFP+ INs expressed classical CGE markers such as reelin , NPY and VIP ( Lee et al . , 2010; Murthy et al . , 2014; Vucurovic et al . , 2010 ) . Quantification across layers revealed that a large fraction of Hmx3; tdTOM+/Htr3a-GFP+ INs expressed reelin or NPY . This was particularly striking in layer 1 ( L1 ) for reelin ( Figure 5A , C ) and in L2–6 for NPY ( Figure 5B , E ) , respectively ( Figure 5—source data 1 ) . Overall , Hmx3; tdTOM+/Htr3a-GFP+ INs accounted for approximately a third of all reelin+/Htr3a-GFP+ INs ( 34 . 5 ± 2 . 3%; 267/797 cells ) and of all NPY+/Htr3a-GFP+ INs ( 27 . 7 ± 2 . 3%; 149/571 cells ) ( Figure 5D , F , Figure 5—source data 1 ) . Given that INs expressing reelin have been shown to co-express NPY ( Lee et al . , 2010 ) , we assessed reelin and NPY co-expression in Hmx3; tdTOM+/Htr3a-GFP+ cells . At P21 , only a small fraction ( 8 . 0 ± 0 . 9%; 17/232 cells ) of these cells expressed NPY without reelin , thus indicating that reelin labels the largest fraction ( 66 . 1 ± 8 . 6%; 267/398 cells ) of Hmx3-derived Htr3a-GFP+ INs ( Figure 5G , H , Figure 5—source data 1 ) . In contrast , Hmx3; tdTOM+/Htr3a-GFP+ INs did not co-label nor with the CGE-specific marker VIP ( Figure 5I ) neither with the MGE-enriched markers SST and PV ( Figure 5—figure supplement 1 ) . These results indicate that Hmx3-derived Htr3a-GFP+ INs mainly belong to the reelin but not to the VIP subtypes and account for an important fraction of all reelin+/Htr3a-GFP+ INs . Two distinct profiles of reelin-expressing INs have been identified in L1 of the neocortex , namely neurogliaform ( NGCs ) and single bouquet-like cells ( SBCs ) ( Cadwell et al . , 2016; Jiang et al . , 2015 ) . At a molecular level , NGCs are strongly enriched in the carbonic anhydrase 4 ( Car4 ) transcript in contrast to SBCs ( Cadwell et al . , 2016 ) . Using single-molecule fluorescent in situ hybridization experiments ( Wang et al . , 2012 ) , we found that Hmx3; tdTOM+/Htr3a-GFP+ INs in L1 exhibited significantly higher levels of Car4 transcripts in contrast to Htr3a-GFP+ INs ( Figure 6A , Figure 6—source data 1 ) , thus indicating that Hmx3-derived Htr3a-GFP+ INs share the molecular profile of NGC . To further verify whether their morphological and electrophysiological features could fit with NGCs , we performed whole-cell recordings ( Figure 6B ) and reconstructions ( Figure 6C , D; Figure 6—figure supplement 1 ) of Hmx3-derived versus non Hmx3-derived Htr3a-GFP+ INs in L1 of the barrel cortex . There , Hmx3; tdTOM+/Htr3a-GFP+ INs displayed the characteristic morphology of elongated NGCs with dense axonal ramifications mostly restricted to L1 ( Figure 6C , Figure 6—figure supplement 1 ) , whereas Htr3a-GFP+ INs negative for tdTOM had the morphology of SBCs with less developed axonal processes that extended deeper in cortical layers ( Figure 6D ) . With regard to their first action potential ( AP ) at rheobase , NGCs are reported to display a type 1 profile with only an after-hyperpolarization potential ( AHP ) , whereas SBCs show a type 2 profile consisting of an AHP followed by an after-depolarization potential ( ADP ) ( Cadwell et al . , 2016; Jiang et al . , 2015 ) . Strikingly , all ( 29 out of 29 ) recorded Hmx3; tdTOM+/Htr3a-GFP+ INs were of type 1 , thus confirming their NGC identity . Moreover , the vast majority of them ( 23 out of 29 ) were of type 1A with a deep and wide AHP and only a few ( 6 out of 29 ) were of type 1B with a shallow and narrow AHP ( Figure 6E , left , Figure 6—source data 2 ) . Htr3a-GFP+ INs negative for Hmx3; tdTOM had more variable profiles , but the majority of them ( 24 out of 28 ) were displaying a type two profile with an average ADP amplitude of 2 . 30 ± 0 . 51 mV , suggesting that they were SBCs . Most of them ( 20 out of 28 ) were of type 2B with a small ADP below the spike threshold , a few others ( 4 out of 28 ) were of type 2A with a big ADP above the spike threshold ( Figure 6E , right , Figure 6—source data 2 ) and another few of them ( 4 out of 28 ) had not measurable ADP ( not shown ) . Hmx3; tdTOM+/Htr3a-GFP+ INs showed also a higher tendency to late-spiking when compared to Htr3a-GFP+ INs ( Figure 6F , Figure 6—source data 2 ) . Hmx3; tdTOM+/Htr3a-GFP+ INs had bigger AP delay average , but not significantly different from Htr3a-GFP+ INs negative for Hmx3; tdTOM ( Figure 6—figure supplement 2C , Figure 6—source data 2 ) . However , the variability of individual cell values was higher for Hmx3; tdTOM+/Htr3a-GFP+ INs , indicating that these cells tend to be more late-spiking , a characteristic of NGCs ( Cadwell et al . , 2016; Jiang et al . , 2015 ) . Alignement of the first APs at rheobase revealed other putative differences between the two groups ( Figure 6G , Figure 6—source data 2 ) . After quantification , Hmx3; tdTOM+/Htr3a-GFP+ INs significantly differed from Htr3a-GFP+ INs negative for Hmx3; tdTOM in the first AP amplitude ( Peak ) , AHP amplitude ( AHP ) , membrane resistance ( Rm ) ( Figure 6H , Figure 6—source data 2 ) and threshold potential ( Vthr ) ( Figure 6—figure supplement 2C ) . We next aimed to determine whether Hmx3 and non Hmx3-derived Htr3a-GFP+ INs classes in L1 could be predicted from single-cell electrophysiological properties . Using an automatic cell type classifier based on combined electrophysiological measures , we were able to predict the Hmx3-derived class with 80 . 7% accuracy , with highest weights found on ADP , Peak and AHP but not Vthr ( Figure 6I , J; Figure 6—figure supplement 2A , B , Figure 6—source data 2 ) . Finally , we analysed Hmx3; tdTOM+/Htr3a-GFP+ INs in other cortical layers to determine whether they displayed the same NGC characteristics . Similarly to L1 cells , Car4 expression in L2-6 was significantly higher in Hmx3; tdTOM+/Htr3a-GFP+ INs as compared to Htr3a-GFP+ INs negative for tdTOM ( Figure 6—figure supplement 3A , B , Figure 6—source data 1 ) . Morphological recovery of Hmx3; tdTOM+/Htr3a-GFP+ INs located in L2–6 revealed that all cells ( 7 out of 7 ) had also the characteristic morphology of NGCs ( Figure 6—figure supplement 4 ) . Furthermore , characteristic properties of NGC like the tendency to late spiking , the depth of AHP , and the level of Vthr were significantly more pronounced in these cells compared to L1 cells ( Figure 6—figure supplement 3C , E ) . Overall , these data indicate that Htr3a-GFP+ INs displaying the molecular , morphological and electrophysiological properties of NGC INs originate from Hmx3-expressing cells in the embryonic POA ( Figure 7 , orange ) , whereas SBCs in layer 1 , as well as VIP +INs , are more likely to originate from the CGE ( Figure 7 , green ) . Here , we find that a fraction ( about 15% ) of Htr3a-GFP+ cortical INs originate from Hmx3+ but not Dbx1+ cells in the POA . The overall fraction of Hmx3; tdTOM+/Htr3a-GFP+ INs to the total Htr3a-GFP+ IN population in the cortex almost doubled from P9 to P21 , a period during which neural migration is largely achieved . Given that about 40% of developing cortical INs undergo apoptosis during early postnatal life ( Southwell et al . , 2012 ) higher levels of programmed cell death in Htr3a-GFP+ INs negative for tdTOM could thus account for the relative postnatal increase in the cortical Hmx3; tdTOM+/Htr3a-GFP+ cell population . Overall , our data support the general view that the differential expression of TFs in progenitor cells originating from distinct subpallial germinal zones controls the specification of cortical IN subtypes ( Huang , 2014; Kessaris et al . , 2014; Anastasiades and Butt , 2011; Flames et al . , 2007; Gelman et al . , 2009 ) . A striking example in the field relates to chandelier INs , which have been shown to derive from Nkx2 . 1+ cells produced specifically at late embryonic time-points in a restricted region of the MGE ( Taniguchi et al . , 2013 ) . Three major germinal zones contribute to the generation of cortical INs , including the MGE , the CGE and the POA ( Kessaris et al . , 2014 ) . The majority of cortical IN subtypes ( about 60–70% ) originates from Nkx2 . 1+ progenitors in the MGE and includes fast-spiking PV+ basket INs , chandelier cells and SST+ Martinotti cells . In addition to NKX2 . 1 , sequential expression of the TFs such as LHX6 ( Anastasiades and Butt , 2011; Du et al . , 2008; Liodis et al . , 2007 ) and SOX6 ( Azim et al . , 2009; Batista-Brito et al . , 2009 ) controls the specification and migration of MGE-derived IN subtypes . Here we find that Htr3a-GFP+ cortical INs originating from Hmx3+ cells in the POA do not express MGE-enriched TFs such as LHX6 or SOX6 . In the embryonic POA , we observe that only a small fraction of Hmx3; tdTOM+/Htr3a-GFP+ cells expresses the TF NKX2 . 1 , which has been shown to be strongly expressed in the ventricular zone of the POA ( Flames et al . , 2007 ) . This could be due to either down-regulation of NKX2 . 1 in postmitotic Hmx3; tdTOM+/Htr3a-GFP+ cells as previously observed in migrating MGE-derived INs ( Nóbrega-Pereira et al . , 2008 ) or to the fact that the majority of Hmx3; tdTOM+/Htr3a-GFP+ cells do not originate from NKX2 . 1 progenitors . In line with this second possibility , recent genetic fate-mapping experiments using a Nkx2 . 1-ires-Flpo knock-in mouse line did not appear to label INs in L1 ( He et al . , 2016 ) . Overall , further work needs to be done to clarify the precise origin of mitotic cells giving rise to the pool of Hmx3; tdTOM+/Htr3a-GFP+ cells observed in the embryonic POA . In contrast to the absence of co-localization with MGE-enriched TFs , we find that Hmx3; tdTOM+/Htr3a-GFP+ INs express TFs such as PROX1 and NR2F2 in the embryonic POA and in the postnatal cortex . PROX1 and NR2F2 have been shown to be expressed in CGE cells and these TFs are maintained in subsets of cortical INs as they mature in the developing cortex ( Cai et al . , 2013; Ma et al . , 2012; Rubin and Kessaris , 2013; Murthy et al . , 2014; Kanatani et al . , 2008 ) . Our results thus indicate that the specification of Hmx3-derived and CGE-derived Htr3a-GFP+ INs shares common transcriptional controls and that the expression of Hmx3 in a fraction of Htr3a-GFP+ defines the distinct subtype of NGCs . To gain insights on the requirement of Hmx3 in the specification of Hmx3-derived Htr3a-GFP+ NGCs , cell-type specific genetic deletion strategies are needed . Finally , the molecular pathways specifically controlled by Hmx3 in NGCs remain to be identified . MGE-derived INs express the neurochemical markers PV or SST and are preferentially distributed in lower cortical layers , whereas CGE-derived INs specifically express the 5-HT3AR , but not PV or SST , and populate more superficial cortical layers ( Fishell and Rudy , 2011; Huang , 2014; Rudy et al . , 2011 ) . Using in situ hybridization , we confirmed that Hmx3+ lineage give rise to superficial cortical Htr3a-GFP+ INs expressing the Htr3a transcript . Reelin , VIP and NPY have been used as neurochemical markers to identify different subtypes of Htr3a-GFP+ cortical INs ( Lee et al . , 2010; Murthy et al . , 2014; Vucurovic et al . , 2010 ) . Expressions of reelin and VIP are mutually exclusive in Htr3a-GFP+ INs , whereas a fraction of them is found to co-express reelin and NPY ( Lee et al . , 2010 ) . Using these markers , we find that Hmx3; tdTOM+/Htr3a-GFP+ INs express reelin and/or NPY , but not VIP , PV or SST . This is in line with previous results showing that Hmx3+ INs express NPY and not VIP , PV or SST ( Gelman et al . , 2009 ) . Finally , we find that cortical INs from the Dbx1+ domain express the MGE-enriched markers PV or SST and only rarely co-label with Htr3a-GFP+ INs . In addition , Dbx1-derived cortical INs express the MGE-related TFs SOX6 and LHX6 but not the CGE-enriched TF PROX1 . Taken together , our findings thus indicate that Hmx3+ but not Dbx1+ cells give rise to a subpopulation of cortical Htr3a-GFP+ INs , which share molecular similarities with CGE but not MGE-derived INs . However , given that both Hmx3+ and Hmx3- Htr3a-GFP+ INs express reelin and/or NPY , these classical neurochemical markers are not sufficient to segregate Hmx3- and non-Hmx3- derived Htr3a-GFP+ IN subtypes . Electrophysiological recordings obtained from Htr3a-GFP+ INs revealed the existence of many different subtypes of INs ( Lee et al . , 2010 ) . Recently , electrophysiological and morphological characterization of L1 INs combined to single-cell transcriptomics delineated two main types of INs , namely NGCs and SBCs ( Cadwell et al . , 2016 ) . Our findings support this observation and indicate that Hmx3-derived Htr3a-GFP+ INs exhibit the morphological and electrophysiological signature of NGCs and strongly express Car4 , a transcript present at high level in NGCs , but not in SBCs . In contrast , Htr3a-GFP+ INs in L1 that do not derive from Hmx3+ cells , have low levels of Car4 and display the electrophysiological profile of SBCs . These results indicate that Htr3a-GFP+ cortical INs in L1 can be subdivided in two major groups characterized by distinct intrinsic properties and that these subgroups are determined by their sites of origin and the differential expression of the TF Hmx3 . Finally , we show that all Hmx3; tdTOM+/Htr3a-GFP+ INs analysed in deeper cortical layers also display molecular , morphological and electrophysiological profiles of NGCs , indicating that the Hmx3-Cre line labels NGCs across neocortical layers . In vivo studies of the canonical cortical microcircuit have mainly relied on the use of the mutually exclusive SST- , PV- and VIP-Cre driver lines ( Cardin et al . , 2009; Fu et al . , 2014; Kepecs and Fishell , 2014; Pfeffer et al . , 2013; Pi et al . , 2013; Pinto and Dan , 2015; Sohal et al . , 2009; Zhang et al . , 2014 ) but they do not give access to NGCs . These cells are the main source of ‘slow’ GABAB-receptor mediated inhibition in the neocortex ( Tamás et al . , 2003 ) and are thought to constitute the core cellular component of a canonical inhibitory circuit in L1 ( Craig and McBain , 2014 ) . NGCs acts through GABAB-receptors to inhibit the activity of projection neurons and halt ongoing network activity through dendritic calcium channels ( Craig and McBain , 2014 ) . Long-range interhemispheric inhibition has been shown to be mediated through a GABAB-receptor dependent mechanism and it has been proposed that this process requires the recruitment of L1 cortical INs , possibly of the neurogliaform-type ( Craig and McBain , 2014; Palmer et al . , 2012 ) . However , given the diversity of L1 cortical INs ( Cadwell et al . , 2016; Jiang et al . , 2013 ) and the lack of molecular tools to specifically target NGCs in vivo , it has so far not been possible to manipulate and interrogate exclusively NGCs in cortical networks . Our findings redefine the Hmx3-Cre mice as a valuable tool to specifically investigate the functional contribution of NGCs in the cortical microcircuit motif . Animal experiments were approved by the local Geneva animal care committee ( GE113/16 ) and conducted according to international and Swiss guidelines . Mice were housed in the conventional area of the animal facility of the Geneva Medical Center . Water and food were provided ad libitum and both temperature ( 22 ± 2°C ) and dark/light cycles ( 12 hr each ) were controlled . Timed-pregnant mice were obtained by overnight mating and the following morning was counted as embryonic day ( E ) E0 . 5 . Tg ( Htr3a-EGFP ) DH30Gsat mice expressing the enhanced GFP under the control of the Htr3a regulatory sequences ( Htr3a-GFP ) were provided by the GENSAT Consortium and maintained on a C57Bl/6 background ( Murthy et al . , 2014 ) . B6 . Cg-Gt ( ROSA ) 26Sortm14 ( CAG-tdTomato ) Hze/J loxP flanked reporter mice ( R26R-tdTOMfl/fl ) were obtained from Jackson Laboratory . Htr3a-GFP mice were crossed with R26R-tdTOMfl/fl mice to obtain Htr3a-GFP; R26R-tdTOMfl/fl mice . Tg ( Hmx3-icre ) 1Kess ( Hmx3-Cre ) mice were obtained from Oscar Marin and previously described ( Gelman et al . , 2009 ) . Dbx1tm2 ( cre ) Apie ( Dbx1-Cre ) mice were obtained from Alessandra Pierani and previously described ( Gelman et al . , 2011 ) . Details of the genotyping procedure are given in the Key Resources Table . Pregnant females were euthanized by lethal intraperitoneal ( i . p . ) injection of pentobarbital ( 50 mg/kg ) , embryos were collected by caesarian cut and brains dissected and fixed overnight ( O . N . ) in cold 4% paraformaldehyde dissolved in 0 . 1M phosphate buffer ( PFA ) pH 7 . 4 . For postnatal brains , animals were deeply anesthetized by i . p . injection of pentobarbital and transcardially perfused with 0 . 9% saline/liquemin followed by cold 4% PFA . Brains were cut on a Vibratome ( Leica VT1000S ) at 60 µm for immunohistochemistry ( IHC ) or at 80–100 µm for free-floating in situ hybridization ( ISH ) . Sections were kept in a cryoprotective solution at −20°C or processed directly for IHC or ISH as described ( Murthy et al . , 2014 ) . The following primary antibodies were used: rabbit anti-GFAP ( 1:2000 , Abcam ) , goat anti-GFP ( 1:2000 , Abcam ) , rabbit anti-GFP ( 1:500 , Millipore ) , mouse anti-NeuN ( 1:500 , Millipore ) , rabbit anti-NKX2 . 1 ( 1:100 , Santa Cruz Biotechnology ) , rabbit anti-NPY ( 1:500 , Abcam ) , rabbit anti-NR2F2 ( 1:500 , Abcam ) , goat anti-PROX1 ( 1:250 , R and D System ) , mouse anti-Parvalbumin ( PV ) ( 1:2000 , Swant ) , mouse anti-Reelin ( 1:500 , Abcam ) , rabbit anti-S100β ( 1:2000 , Abcam ) , rat anti-Somatostatin ( SST ) ( 1:500 , Millipore ) , rabbit anti-SOX6 ( 1:500 , Abcam ) , goat anti-SOX10 ( 1:100 , Santa Cruz Biotechnology ) , goat anti-SP8 ( 1:50 , Santa Cruz Biotechnology ) , goat anti-tdTomato ( tdTOM ) ( 1:500 , Sicgen ) , rabbit anti-VIP ( 1:500 , Abcam ) . Secondary goat or donkey Alexa-405 , –488 , −568 and −647 antibodies ( Abcam , Invitrogen ) raised against the appropriate species were used at a dilution of 1:500 and sections were counterstained with Hoechst 33258 ( 1:10000 ) when no Alexa-405 staining was done . A list of the antibodies is given in the Key Resources Table . Sections were hybridized with the respective DIG-labeled RNA probes as described previously ( Murthy et al . , 2014 ) . The Htr3a plasmid probe was linearized with HindIII-HF for antisense RNA probe synthesis by T7 polymerase ( kind gift from Dr . B . Emerit ) . The Lhx6 plasmid probe ( Liodis et al . , 2007 ) was linearized with Not1 for antisense RNA probe synthesis by T3 polymerase ( kind gift from Dr . M . Denaxa ) . The unbound probe was washed and slices incubated with alkaline phosphatase-conjugated anti-DIG antibody ( 1:2000 , Roche ) O . N . at 4°C . Fast Red ( Kem-En-Tech ) was used as an alkaline phosphatase fluorescent substrate to reveal the hybridized probe . We took advantage of the removal of both GFP and tdTOM endogenous fluorescence due to protocol treatments and revealed them by IHC using green and far-red emitting secondary antibodies , respectively . For illustration purposes , the bound probe ( red ) and the tdTOM ( far red ) are shown in blue and red , respectively . For RNAscope experiments , P30 brains were rapidly extracted and fresh frozen . After dehydration and protease treatment , coronal 12 µm-thick brain sections were processed using the RNAscope Multiplex Fluorescent Reagent Kit ( Advanced Cell Diagnostics ) according to the manufacturer’s protocol . Probes targeting mRNAs of the GFP and tdTOM transgenes and of the endogenous Car4 gene were designed by Advanced Cell Diagnostics . Images were acquired using confocal microscopes ( Nikon A1R or Axio Imager . Z2 Basis LSM 800 ) equipped with oil-immersion 40x , 60x and 63x objectives ( CFI Plan Fluor 40x/1 . 3 and CFI Plan Apo VC H 60x/1 . 4 , Nikon or Plan-APO ( UV ) VIS-IR 40x/1 . 4 and Plan-Apochromat f/ELYRA 63x/1 . 4 , LSM ) . For widefield illustrations ( Figure 2—figure supplement 1 ) , images were taken with Axioscan . Z1 slidescanner ( Zeiss ) , equipped with Plan-Apochromat 10x/0 . 45 objective ( Zeiss ) . Images were lightly treated ( gamma , brightness and and despeckle filter only ) for visual purpose with Photoshop CC and manual counts were achieved with Fiji . Data are presented as brain averages calculated from at least three slices at different rostro-caudal levels per brain ( except for P5 Dbx1 brain 3 ) . A detailed description of the counts , cells and brains in the different experiments is given in Supplementary file 1 . 300 µm-thick coronal brain slices were prepared from 3 to 4 weeks old Hmx3; tdTOM+/Htr3a-GFP+ mice with a vibratome ( Leica VT 1000S ) . In the recording chamber , slices were continuously superfused with ACSF ( 32°C ) containing ( in mM ) : NaCl ( 119 ) , KCl ( 2 . 5 ) , CaCl2 ( 2 . 5 ) , MgSO4 ( 1 . 3 ) , NaH2PO4 ( 1 . 0 ) , NaHCO3 ( 26 . 2 ) , and glucose ( 22 ) , and equilibrated with 95% O2/5% CO2 , pH 7 . 4 . Whole-cell recordings were obtained from visually identified Hmx3; tdTOM+/Htr3a-GFP+ in cortical layers 1–6 and Hmx3; tdTOM-/Htr3a-GFP+ INs in L1 , using an upright microscope ( Zeiss Axioskop FS ) equipped with differential interference contrast and standard epifluorescence . Borosilicate glass patch pipettes had a resistance of 5–6 MΩ when filled with an internal solution containing ( in mM ) : K gluconate ( 135 ) , KCl ( 4 ) , HEPES ( 10 ) , Phosphocreatine ( 10 ) , Mg-ATP ( 4 ) , Na-GTP ( 0 . 3 ) , and biocytin ( 8 . 1 ) . Current clamp recordings were performed at rest and firing properties were studied by delivering consecutive current pulses , 500 ms duration each , ranging from −20 to +360 pA with a 5 pA increment , every 3 s . Data were acquired using a Multiclamp 700B Amplifier ( Molecular Devices ) , and digitized at 10 kHz ( National Instruments ) , using MATLAB ( MathWorks ) -based Ephus software ( Ephus; The Janelia Farm Research Center ) . Offline analysis was performed using Clampfit ( Version 10 . 1 . 0 . 10 , Molecular Devices ) . Cells were accepted for analysis only if their series resistance was below 30 MΩ and did not change more than 20% during recordings . Following patch-clamp recordings , slices were incubated in ACSF for 1–2 hr at room temperature , then fixed overnight with 4% PFA / 2% Glutaraldehyde in 0 . 1 M phosphate buffer . Biocytin-filled recorded cells were revealed with IHC , using streptavidin-Alexa 647 conjugate ( 1:500 , Thermo Fisher Scientific ) , and confirmed being in L1 and expressing Hmx3; tdTOM and/or Htr3a-GFP . For detailed morphology , slices were quenched for endogenous peroxidase activity in methanol/0 . 5% H202 , blocked in 0 . 05 M Tris buffer pH 7 . 4/0 . 6% NaCl/0 . 3% Triton X-100/10% normal horse serum ( NHS ) and incubated ( O . N . , 4°C ) with avidin-biotin complex ( Vectastain Elite ABC-HRP Kit ) in 0 . 1M Tris buffer pH 7 . 7 . 3 , 3’-diaminobenzidine ( DAB ) revelation was performed following the manufacturer’s protocol ( Vectastain DAB Kit; SK-4100 ) . Slices were finally dehydrated in graded series of ethanol/xylene and mounted in Eukitt ( Sigma ) . Morphological reconstructions of biocytin-filled cells were performed with Neurolucida software ( v . 11 . 02 . 1 , MBF Bioscience , Microbrightfield ) , linked to a microscope ( Nikon eclipse 80i ) equipped with an oil-immersion 100x objective ( Plan Apo VC/1 . 4 , Nikon ) . Brightfield images of the reconstructed cells were acquired with the same microscope and a 10x objective ( Plan Apo/0 . 45 , Nikon ) . Traces were extracted with Neurolucida Explorer ( v . 11 . 02 . 1 , MBF Bioscience , Microbrightfield ) . 14 Hmx3; tdTOM+/Htr3a-GFP+ cells ( 7 in L1 , 5 cells in L2/3 and 2 cells in L5 ) and 3 Hmx3; tdTOM-/Htr3a-GFP+ INs ( 3 in L1 ) from four brains were recovered for morphology . For L1 cells , border artefacts in morphological tracings due to tissue compression were corrected . Traces from 14 Hmx3; tdTOM+/Htr3a-GFP+ and 3 Hmx3; tdTOM-/Htr3a-GFP+ INs from four brains were analysed blindly . The membrane resistance ( Rm ) , the membrane resting potential and five properties of the first action potential ( AP ) at rheobase - i ) threshold potential ( Vth ) ; ii ) AP amplitude ( peak ) ; iii ) AP latency from current step onset ( delay ) ; iv ) after-hyperpolarization potential amplitude ( AHP ) and when present; v ) after-depolarization potential amplitude ( ADP ) - were measured for all recorded cells . Both electrophysiological features and morphological tracings were analyzed blindly and data were attributed back to their corresponding cell . Values for each recorded cell are provided in Figure 6—source data 2 and Figure 6—figure Supplement 3—source data 1 . Animals were used regardless of their sex and statistical analysis was done with R programming language and GraphPad Prism . No statistics were used to determine optimal group sample size; however , sample sizes were similar to those used in previous publications from our group and others . Normality of the samples was assessed with D’Agostino-Pearson test and when distribution was not normal , non-parametric tests were applied . Using bmrm ( v3 . 3 ) package for L1-regularized logistic regression model , data were standardized , and a L1-regularized logistic regression model was trained to distinguish between Htr3a-GFP+ INs that were Hmx3-derived and those which were not . This model assigned a linear weight that reflects the power of each feature in the model logistic regression and computed a probability that a given cell is Hmx3-derived in such a way that the misclassification error on the training data was minimized ( Figure 6I–J ) . Classification performance of the L1-regularized logistic regression algorithm was assessed by leave-one-out-cross-validation ( LOOCV ) . It consists in training a model on all but one cell , feeding the model with this isolated cell to predict its origin and finally assessing if the prediction is correct . Looping it over all cells , yields a prediction value for each cell , which is used to estimate the generalization error of the classifier . Finally , in order to determine if the prediction made by the logistic regression model improved over the signal contained into each feature taken individually , receiver operating characteristic ( ROC ) curves were drawn to visualize the sensitivity/specificity ratio for each feature and for the leave-one-out predictions . Areas under the curves ( AUC ) were analyzed to determine the strongest signals ( Figure 6—figure supplement 2 , Figure 6—source data 2 ) .	Our brain contains over a 100 billion nerve cells or neurons , and each of them is thought to connect to over 1 , 000 other neurons . Together , these cells form a complex network to convey information from our surroundings or transmit messages to designated destinations . This circuitry forms the basis of our unique cognitive abilities . In the cerebral cortex – the largest region of the brain – two main types of neurons can be found: projection neurons , which transfer information to other regions in the brain , and interneurons , which connect locally to different neurons and harmonize this information by inhibiting specific messages . The over 20 different types of known interneurons come in different shapes and properties and are thought to play a key role in powerful computations such as learning and memory . Since interneurons are hard to track , it is still unclear when and how they start to form and mature as the brain of an embryo develops . For example , one type of interneurons called the neurogliaform cells , have a very distinct shape and properties . But , until now , the origin of this cell type had been unknown . To find out how neurogliaform cells develop , Niquille , Limoni , Markopoulos et al . used a specific gene called Hmx3 to track these cells over time . With this strategy , the shapes and properties of the cells could be analyzed . The results showed that neurogliaform cells originate from a region outside of the cerebral cortex called the preoptic area , and later travel over long distances to reach their final location . The cells reach the cortex a few days after their birth and take several weeks to mature . These results suggest that the traits of a specific type of neuron is determined very early in life . By labeling this unique subset of interneurons , researchers will now be able to identify the specific molecular mechanisms that help the neurogliaform cells to develop . Furthermore , it will provide a new strategy to fully understand what role these cells play in processing information and guiding behavior .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Neurogliaform cortical interneurons derive from cells in the preoptic area
Neuregulin 1 ( NRG1 ) and the γ-secretase subunit APH1B have been previously implicated as genetic risk factors for schizophrenia and schizophrenia relevant deficits have been observed in rodent models with loss of function mutations in either gene . Here we show that the Aph1b-γ-secretase is selectively involved in Nrg1 intracellular signalling . We found that Aph1b-deficient mice display a decrease in excitatory synaptic markers . Electrophysiological recordings show that Aph1b is required for excitatory synaptic transmission and plasticity . Furthermore , gain and loss of function and genetic rescue experiments indicate that Nrg1 intracellular signalling promotes dendritic spine formation downstream of Aph1b-γ-secretase in vitro and in vivo . In conclusion , our study sheds light on the physiological role of Aph1b-γ-secretase in brain and provides a new mechanistic perspective on the relevance of NRG1 processing in schizophrenia . Schizophrenia ( SZ ) is a severe disorder that affects neuronal circuits involved in social behaviour and cognitive processes ( Harrison and Weinberger , 2005; Lewis and Sweet , 2009 ) . Increasing evidence suggests that different risk genes interact synergistically to contribute to SZ , mainly by affecting excitatory and/or inhibitory circuitries in the cortex ( Harrison and Weinberger , 2005; Lisman et al . , 2008; Lewis and Sweet , 2009; Glausier and Lewis , 2013 ) . In particular , polymorphisms in Neuregulin 1 ( NRG1 ) have been consistently linked to schizophrenia in different populations ( Stefansson et al . , 2002; Mei and Xiong , 2008 ) . The NRG1 gene encodes more than 30 isoforms that differ in structure , expression pattern , processing and signalling modes which complicates the study of the NRG1 family ( Mei and Xiong , 2008 ) . Most Ig-Nrg1 isoforms apparently function as diffusible paracrine signals . Conversely , the cysteine-rich domain- ( CRD- ) containing Nrg1 isoform ( also known as Type III Nrg1 ) is membrane bound and , in addition to canonical forward signalling via ErbB4 , can also signal backward via its intracellular domain ( Nrg1-ICD ) ( Bao et al . , 2003; Mei and Xiong , 2008; Chen et al . , 2010a; Pedrique and Fazzari , 2010 ) . Converging studies demonstrate that Nrg1/ErbB4 forward signalling controls the establishment of cortical inhibitory circuits and is implicated in the control of neuronal synchronisation ( Chen et al . , 2010b; Fazzari et al . , 2010; Wen et al . , 2010; Rico and Marin , 2011; Cahill et al . , 2012 ) . However , the physiological role of CRD-Nrg1 intracellular signalling , and thus the function of the membrane bound and intracellular domain of Nrg1 remains unclear . In analogy to Notch signalling ( De Strooper et al . , 1999 ) , the intracellular part of Nrg1 , Nrg1-ICD , is released by intramembrane processing . It is known that γ-secretase activity is responsible for this cleavage ( Bao et al . , 2003; Dejaegere et al . , 2008; Chen et al . , 2010a; Pedrique and Fazzari , 2010; Marballi et al . , 2012 ) , but it remains unclear which specific γ-secretase is involved . γ-secretases are a family of intramembrane proteases composed of four different subunits: presenilin ( PSEN ) , anterior pharynx homologue 1 ( APH1 ) , nicastrin ( NCT ) , and presenilin enhancer 2 ( PEN2 ) ( De Strooper , 2003 ) . In the human genome two presenilin ( PSEN1 and PSEN2 ) and two APH1 ( APH1A and APH1B ) are present; thus , at least four different γ-secretase complexes can be generated . One of the major challenges in the γ-secretase field is to understand whether these different γ-secretase complexes have different biological or pathological functions . This question is particularly relevant for understanding the mechanisms that contribute to the molecular pathogenesis of SZ since indirect evidence indicates that NRG1 intracellular signalling might be involved in the risk for this disease . In this regard , a Val-to-Leu mutation in the NRG1 transmembrane domain increases the risk for development of SZ ( Walss-Bass et al . , 2006 ) , impairs intramembrane γ-secretase cleavage of Nrg1 ( Dejaegere et al . , 2008 ) and abnormal NRG1 processing was found in schizophrenic patients ( Chong et al . , 2008; Mei and Xiong , 2008; Marballi et al . , 2012 ) . Moreover , putative loss of function variants of APH1B , a crucial component of the γ-secretase complex , were found to aggregate with NRG1 risk alleles in schizophrenia patients ( Hatzimanolis et al . , 2013 ) and Aph1b-loss of function mutations in rodents display behavioural phenotypes that are relevant for schizophrenia ( Coolen et al . , 2005 , 2006; Dejaegere et al . , 2008 ) . Rodents have duplicated the Aph1b gene during evolution into highly homologous Aph1b and Aph1C . In order to model human APH1B loss of function , we previously generated double mutant mice for Aph1b and Aph1C ( Serneels et al . , 2005 ) , hereafter referred to as Aph1bcfl/fl or Aph1bc−/− upon Cre-dependent deletion . We also conditionally targeted the Aph1a locus , referred to as Aph1afl/fl . We have found that Aph1a-γ-secretase complexes are necessary to activate Notch signalling and genetic deletion of Aph1a leads to a Notch related embryonic lethality ( Serneels et al . , 2005 ) . Conversely , deletion of the Aph1bc-γ-secretase complex does not affect Notch signalling but hampers Nrg1 processing and alters sensory motor gating , working memory and sensitivity to psychotropic drugs , thereby mimicking Nrg1 deficiency and various phenotypes related to schizophrenia ( Coolen et al . , 2005 , 2006; Dejaegere et al . , 2008 ) . However , the selective role of Aph1bc-γ-secretase complexes and the Aph1bc-γ-secretase-dependent processing of Nrg1 in brain wiring and function remained unstudied . In the current study , we address the question of whether the Aph1bc subunit provides specificity to Nrg1 processing and whether this subunit would indeed be involved in the postulated intracellular signalling of Nrg1 . We show here that the Aph1bc-γ-secretase complex controls excitatory circuitry via Nrg1 intracellular signalling . Mice mutant for Aph1bc display altered expression of excitatory synaptic markers , impaired synaptic transmission and decreased long term potentiation . Furthermore , single cell deletion of Aph1bc in vivo impaired dendritic spine formation which could be rescued by the expression of the Nrg1-ICD . Taken together , these data indicate that Nrg1 intracellular signalling downstream of Aph1bc-γ-secretase complexes promotes in a cell autonomous fashion the formation of excitatory connections in cortical neurons . Hence , our study provides a cellular and molecular mechanistic explanation for the cognitive deficits observed in Aph1bc-γ-secretase deficient mice ( Dejaegere et al . , 2008 ) . More importantly , it provides unique insight into the importance of Nrg1 intracellular signalling in the establishment of functional synapses and the potential aetiological role of misprocessing of NRG1 in the pathogenesis of schizophrenia . We reasoned that the behavioural deficits observed in Aph1bc−/− mice ( Dejaegere et al . , 2008 ) could be due to abnormal development of the brain . To perform the morphometric analysis of control and Aph1bc−/− cortices , we immunolabelled control and mutant brains for Cux1 , a marker for layers II/III and IV , and for the panneuronal marker NeuN ( Figure 1A ) . We found that Aph1bc deletion did not alter the size of cortical layers or the relative distribution of neurons in different layers ( Figure 1B–C ) . Hence , the observed behavioural abnormalities could not be attributed to a gross morphological alteration of the brain structure . 10 . 7554/eLife . 02196 . 003Figure 1 . Normal cortical layer formation and altered expression of synaptic markers in Aph1bc−/− deficient mice . ( A ) Representative pictures of neuronal cortices from Control and Aph1bc−/− null mice at P30 immunostained for the upper layers marker Cux1 and for the pan neuronal marker NeuN . Nuclei were stained with DAPI . Scale bars , 100 µm . ( B ) Quantification of cortical layers size at Bregma −1 . 4 mm . Ctrl: n = 4; KO: n = 3; Histogram show average ± SD , two way ANOVA . ( C ) Neuronal distribution , as measured by relative NeuN fluorescence intensity along bins ordered from top to bottom , was unchanged in Aph1bc−/− mutant brains at P30 . n = 6 sections from three mice; Graph show means ± SD , two way ANOVA . ( D and E ) Western blot analysis of synaptic markers VGluT1 , Synaptophysin and PSD95 in prefrontal cortex homogenates show decreased expression of these proteins in Aph1bc−/− . n = 9 replicates out of n = 3 mice per group; the histogram shows signal intensity normalized for tubulin signal , means ± SEM , p<0 . 05 , p<0 . 01 . VGluT1: Ctrl = 100 ± 3% , Aph1bc−/− = 85 ± 6%; Syn: Ctrl = 100 ± 4% , Aph1bc−/− = 80 ± 4%; PSD95: Ctrl = 100 ± 4% , Aph1bc−/− = 78 ± 7% . ( F ) Representative confocal pictures of layer II/III of prefrontal cortices from control and Aph1bc−/− mice at P30 immunolabelled for the excitatory presynaptic marker VGlut1 and for the excitatory postsynaptic marker Homer1 . Scale bar 10 µm . ( G and H ) Cumulative probability of VGluT1 and Homer1 puncta intensities in control and Aph1bc−/− mice . VGlut1 , n >338 puncta; Homer1 , n >1160 puncta; three animals per genotype each , Komolgorov–Smirnov test , p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 02196 . 00310 . 7554/eLife . 02196 . 004Figure 1—figure supplement 1 . Aph1bc deletion does not affect the expression of inhibitory synaptic markers . ( A ) Representative pictures of layer II/III of prefrontal cortices from control and Aph1bc−/− mice at P30 immunolabelled for inhibitory markers VGAT and PV . Scale bar 10 µm . ( B and C ) Cumulative probability plots of VGAT and PV puncta intensities in control and Aph1bc−/− null mice . PV , n >1352 puncta; VGAT , n >804 puncta , three animals per genotype each , Komolgorov–Smirnov test , p>0 . 05 for both VGAT and PV . DOI: http://dx . doi . org/10 . 7554/eLife . 02196 . 004 Schizophrenia is characterized by dysfunction in the prefrontal cortex ( Arnsten et al . , 2012 ) , where Aph1bc is highly expressed ( Dejaegere et al . , 2008 ) . In particular , excitatory circuitry is impaired in prefrontal cortex of schizophrenic patients ( Lewis and Sweet , 2009 ) . Thus , we scrutinized the expression of different pre- and post-synaptic markers in control and Aph1bc−/− prefrontal cortices to test if neuronal connectivity was properly established . Western blot data indicate a small but significant decrease in the expression of the presynaptic markers VGluT1 and Synaptophysin and of the postsynaptic protein PSD95 ( Figure 1D , E ) . In addition , we carried out confocal quantitative analysis of the expression of VGluT1 and the postsynaptic marker Homer1 . We found that the staining intensity of VGluT1 and Homer1 positive puncta shifted toward lower staining intensities ( Figure 1F–H ) . On the other hand , Aph1bc loss of function did not affect the intensity of positive puncta for VGAT , that labels all inhibitory terminals , and for PV , a specific marker for fast spiking interneurons ( Figure 1—figure supplement 1A–C ) . In sum , these data suggested that the glutamatergic circuitry is impaired in Aph1bc−/− mice . The observed synaptic phenotype prompted us to further investigate synaptic function in Aph1bc−/− mice . We showed that Aph1bc is expressed in CA1 and CA3 layers of the hippocampus ( Serneels et al . , 2009 ) . As it was previously reported that deletion of Presenilins in hippocampal pyramidal neurons impairs synaptic plasticity ( Saura et al . , 2004; Zhang et al . , 2009 ) , we decided to analyse the Schaffer collateral pathway of the hippocampus , in Aph1bc−/− mice . The Input-Output ( I/O ) curves , that show the field excitatory postsynaptic potentials ( fEPSP ) in response to stimuli of increasing strength , indicate that baseline synaptic transmission is impaired in Aph1bc+/− heterozygous and Aph1bc−/− homozygous mutant mice as compared to controls ( Figure 2A ) . Conversely , paired-pulse facilitation ( PPF ) , a presynaptic form of short term plasticity which reflects release probability , was undistinguishable in control and mutant mice ( Figure 2B ) . Next , we studied the relevance of Aph1bc in long-term synaptic plasticity . Induction of long term potentiation ( LTP ) by three trains of theta burst stimulation is impaired in both heterozygous and homozygous Aph1bc deficient mice compared to controls ( Figure 2C ) . Even though reduced basal transmission might in principle interfere with LTP analysis , these results suggest that synaptic plasticity is affected by Aph1bc deletion . Furthermore , we recorded miniature excitatory currents ( mEPSCs ) which showed normal amplitude but slightly increased inter-event intervals ( IEIs ) in Aph1bc−/− as compared to control mice ( Figure 2D–F ) . Since PPF analysis indicates that Aph1bc deletion does not alter release probability , the increased IEIs in mEPSCs suggest a decrease in the number of release sites that is consistent with the decreased expression of excitatory synaptic markers assessed by immunofluorescence . Altogether , Aph1bc is required for synaptic transmission and long-term synaptic plasticity . 10 . 7554/eLife . 02196 . 005Figure 2 . Aph1bc deletion impairs synaptic transmission and plasticity . ( A ) Input-Output curves recorded in the Schaffer collaterals of the hippocampus show that basic synaptic transmission is impaired in Aph1bc+/− heterozygous and homozygous Aph1bc−/− mutant mice as compared to control littermates . Graph shows means ± SEM , RM-ANOVA for the three groups: F ( 2 , 18 ) = 4 . 163 , p<0 . 05; Ctrl: n = 6; Aph1bc+/−: n = 7; Aph1bc−/−: n = 6 . ( B ) Paired pulse facilitation ( PPF ) , a presynaptic form of short term synaptic plasticity , is not significantly affected by genetic Aph1bc deletion . RM-ANOVA , p>0 . 05 . ( C ) Long term potentiation elicited by three bursts of theta stimulations ( black arrows ) is reduced in heterozygous and homozygous Aph1bc mutant mice in comparison to control mice . The insets show representative traces from mutant and control mice . Means ± SEM , RM-ANOVA for the three groups . F ( 2 , 18 ) = 9 . 74 , p=0 . 0014 . Ctrl n = 6; Aph1bc+/− n = 7; Aph1bc−/− n = 6 . ( D ) Representative traces from mEPSC recordings in slices from control , Aph1bc+/− and Aph1bc−/− mice plotted in ( E ) and ( F ) . ( E ) Cumulative plot of inter-event intervals of mEPSCs in control , heterozygous and homozygous Aph1bc deficient mice . Kruskal–Wallis test followed by Dunn's multiple comparison test , p<0 . 01 . Median , Ctrl 1651 ms , BC+/− 1835 ms , BC−/− 1947 ms . Mean , Ctrl 2465 ms ± 56 , BC+/− 2550 ms ± 58 , BC−/− 2622 ms ± 56 . n >1679 each out of 37 control , 36 Aph1bc+/− and 35 Aph1bc−/− neurons . ( F ) Cumulative probability of mEPSCs amplitude in control and mutant Aph1bc mice . Komolgorov–Smirnov test , p>0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 02196 . 005 A reduction in the density of dendritic spines , that receive excitatory inputs , is an hallmark of schizophrenia ( Lewis and Sweet , 2009 ) . Hence , we investigated the effects of Aph1bc deficiency on the establishment of dendritic spines . We generated hippocampal cultures from Aph1bcfl/fl conditional mutant mice . We transfected Aph1bcfl/fl neurons at DIV8 with GFP as a control or with GFP-ires-Cre to obtain single-cell deletion of Aph1bc , and we quantified dendritic spine density at DIV15 . This study showed that Aph1bc loss of function impaired the formation of dendritic spines ( Figure 3A–C ) . We then hypothesized that a deficit in Nrg1 intracellular signalling underpinned the Aph1bc loss of function phenotype . This model predicts that restoring Nrg1 intracellular signalling would rescue the observed dendritic spine deficit in Aph1bc−/− neurons . Therefore , we co-transfected GFP-ires-Cre with Nrg1-ICD or with CRD-Nrg1-FL ( Figure 3B ) . The expression of Nrg1-ICD indeed rescued the impairment of spine formation in Aph1bc−/− neurons in a cell autonomous way ( Figure 3A–C ) . Moreover , also CRD-Nrg1-FL expression rescued the Aph1bc dependent phenotype ( Figure 3A–C ) . 10 . 7554/eLife . 02196 . 006Figure 3 . Spine formation is impaired by single cell Aph1bc-γ-secretase loss of function and is rescued by Nrg1 intracellular signalling . ( A ) Representative pictures at DIV15 of cultured hippocampal from Aph1bcfl/fl conditional mutant mice transfected at DIV8 with GFP as control , GFP-ires-Cre to obtain single cell Aph1bc−/− neurons , GFP-ires-Cre and CRD-Nrg1-FL or Nrg1-ICD to restore Nrg1 intracellular signalling in Aph1bc−/− neurons . Scale bars in left column: 50 µm , right column: 5 µm . ( B ) The schemata show the structure of CRD-Nrg1 full length ( CRD-Nrg1-FL ) , of Nrg1 intracellular domain ( Nrg1-ICD ) . CRD , Cysteine Rich Domain , EGF epidermal growth factor-like domain; TM , transmembrane domain; NLS , nuclear localization signal . ( C ) Quantification of spine density . Selective single cell genetic deletion of Aph1bc-γ-secretase decreased spine density . Co-expression of CRD-Nrg1-FL and of Nrg1-ICD in Aph1bc−/− neurons rescued the impairment in spine formation . Means ± SEM , one-way ANOVA . **p<0 . 001 . Ctrl: n = 36; Aph1bc−/−: n = 36; Aph1bc−/−+CRD-Nrg1-FL: n = 38; Aph1bc−/−+Nrg1-ICD: n = 37 . ( D ) Representative traces from mEPSC recordings shown in ( E ) and ( F ) . ( E ) Cumulative probability of inter-event intervals of mEPSCs recorded in non-transfected and GFP positive control neurons , in Aph1bc−/− deficient neurons and in Aph1bc−/− neurons transfected with Nrg1-ICD . nt = non transfected . The inset graph shows means ± SEM . Kruskal–Wallis test followed by Dunn's multiple comparison test , ns p>0 . 05; p<0 . 05 . nt: n = 457 out of 9 neurons; GFP: n = 359 out of 6 neurons; Aph1bc−/−: n = 107 out of 7 neurons; Aph1bc−/−+Nrg1-ICD: n = 220 out of 8 neurons . ( F ) Cumulative probability plot of mEPSCs amplitude recorded in non-transfected or GFP positive control neurons , in Aph1bc−/− deficient neurons and in Aph1bc−/− neurons transfected with Nrg1-ICD . nt = non transfected . The inset graph shows means ± SEM , Kruskal–Wallis test followed by Dunn's multiple comparison test , ns , p>0 . 05; p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 02196 . 006 To further asses the cell autonomous relevance of Aph1bc-γ-secretase/Neuregulin 1 intracellular signalling in excitatory synaptic function , we recorded mEPSCs in neuronal cultures . As additional control we also measured mEPSCs in non-transfected ( nt ) neurons which were undistinguishable from GFP expressing neurons . The IEIs and the amplitude of mEPSCs were impaired in Aph1bc−/− relative to control neurons ( Figure 3D–F ) . This phenotype was substantially , although not completely , rescued by the expression of Nrg1-ICD in Aph1bc−/− neurons ( Figure 3D–F ) . These results are coherent with the morphological analysis of dendritic spine density and further support a role for Aph1bc-γ-secretase/Nrg1 intracellular signalling in excitatory connections . Pyramidal neurons express both Aph1a- and Aph1bc-γ-secretase complexes ( Serneels et al . , 2009 ) . We reasoned that the rescue of spine formation observed upon exogenous expression of CRD-Nrg1-FL might indicate that Aph1a-γ-secretase , which is also expressed by pyramidal neurons ( Serneels et al . , 2009 ) , could compensate for the loss of Aph1bc-dependent Nrg1 processing under these experimental conditions of overexpression of the substrate . Hence , we investigated the involvement of Aph1a in dendritic spine development in two additional experimental paradigms . First , we analysed hippocampal cultures from Aph1afl/fl conditional mutant mice . Aph1a deletion by GFP-ires-Cre expression did not alter spine formation as compared to control neurons expressing GFP , suggesting that Aph1a has a redundant function in spine formation in vitro when Aph1bc is present ( Figure 4A , B ) . We then established primary neuronal cultures from Aph1abcfl/fl triple conditional mutant mice to obtain complete genetic abrogation of γ-secretase activity ( Serneels et al . , 2009 ) . Deletion of all of the Aph1 genes by GFP-ires-Cre expression at DIV8 decreased spine density at DIV15 similarly to Aph1bc deletion . Moreover , expression of CRD-Nrg1-FL did not rescue decreased spine density in Aph1abc-/- triple mutant neurons , indicating that CRD-Nrg1-FL cleavage by the γ-secretase is necessary to rescue spine formation ( Figure 4C , D ) . Taken together with the conditional Aph1bc loss of function experiments , these data provide a proof of concept that specific γ-secretase complexes are differentially involved in spine formation . In addition , they indicate that γ-secretase is required to trigger Nrg1 intracellular signalling in this biological process . 10 . 7554/eLife . 02196 . 007Figure 4 . Selective function of different γ-secretase complexes in spine formation . ( A ) Cultured hippocampal neurons from Aph1afl/fl conditional mutant mice were transfected at DIV8 with GFP as control or with GFP-ires-Cre to delete Aph1a and fixed at DIV15 . ( B ) Single cell deletion of Aph1a indicates that Aph1a-γ-secretase activity is not necessary for spine formation in these experimental conditions . Means ± SEM , t test . p>0 . 05 . Ctrl: n = 25; Aph1a−/−: n = 19 . ( C ) Hippocampal neurons from Aph1abcfl/fl triple conditional mutant mice were transfected with GFP as control , with GFP-ires-Cre to completely abrogate γ-secretase activity in single neurons or co-transfected with GFP-ires-Cre and CRD-Nrg1-FL . ( D ) Complete γ-secretase loss of function by Aph1abc−/− triple deletion impaired spine formation . This phenotype could not be rescued by CRD-Nrg1-FL indicating that γ-secretase dependent Nrg1 intracellular signalling is necessary to restore spine formation . Means ± SEM , one-way ANOVA . p<0 . 001 . Ctrl: n = 13; Aph1abc−/−: n = 22; Aph1bc−/− + CRD-Nrg1-FL: n = 19 . Scale bars in A , C , left column: 50 µm , right column: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02196 . 007 We further established the role of Nrg1 in spine formation by performing additional gain-of-function experiments . We transfected wild type hippocampal neurons in vitro with GFP as control or with constructs co-expressing GFP and Nrg1 at DIV8 , and we analysed transfected neurons at DIV15 . We found that expression of CRD-Nrg1-FL increased spine density as compared to control neurons expressing GFP only ( Figure 5A , B ) . To selectively test the role of Nrg1 intracellular signalling , we co-transfected neurons with a construct encoding Nrg1-ICD . Exogenous expression of Nrg1-ICD increased the density of dendritic spines as compared to controls , indicating that the activation of Nrg1 intracellular signalling promotes spine formation ( Figure 5A , B ) . Nrg1-ICD contains a nuclear localization signal ( NLS ) , which is required for the translocation of Nrg1-ICD to the nucleus where it regulates gene expression ( Bao et al . , 2003 ) . Here , we found that the expression of the Nrg1-ICD lacking the NLS ( Nrg1-ΔNLS-ICD ) does not alter the formation of dendritic spines ( Figure 5A , B ) . Collectively , these data indicate that cell autonomous Nrg1 intracellular signalling cell autonomously enhances dendritic spine formation and the localization of the Nrg1-ICD to the nucleus is required for this function . 10 . 7554/eLife . 02196 . 008Figure 5 . Nrg1 intracellular signalling cell-autonomously promotes spine formation in vitro . ( A ) Representative pictures of cultured hippocampal neurons transfected at DIV8 , at the beginning of synaptogenesis , with either GFP alone as control or GFP and the CRD-Nrg1-Fl , GFP and the Nrg1-ICD and GFP and Nrg1-ΔNLS-ICD . Neurons were fixed and analysed at DIV15 . ( B ) Quantification of spine density in Nrg1 transfected neurons . Single cell exogenous expression of Nrg1-Fl and of Nrg1-ICD enhanced spine formation . Conversely , Nrg1-ΔNLS expression did not increase spine density indicating that nuclear localization signal of Nrg1 is required for this function . Means ± SEM , one-way ANOVA . p<0 . 001 . Ctrl , n = 19; CRD-Nrg1-FL , n = 15; Nrg1-ICD , n = 16; Nrg1-ΔNLS-ICD , n = 21 . Scale bars in B , left column: 50 µm , right column: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02196 . 008 The observation that Aph1bc was required for spine formation in neuronal cultures in vitro led us to investigate the neuronal morphology and spine density in the brain of Aph1bc−/− deficient mice in vivo . To this aim , we compared Golgi stained neurons from Aph1bc−/− mutant and Aph1bc+/+ control mice at P30 . Since synaptic transmission is impaired in hippocampal Schaffer collaterals of Aph1bc−/− mice , we scrutinized spine density in dendrites of CA1 hippocampal neurons that receive input from CA3 axons . Morphological observation of Aph1bc−/− neurons did not reveal overt abnormalities ( Figure 6A ) . In addition , Sholl analysis did not show a significant difference in dendritic arborisation between Aph1bc+/+ control and Aph1bc−/− mice ( Figure 6B ) . However , we found that dendritic spine density was significantly decreased in the apical dendrites of Aph1bc−/− neurons as compared to controls ( Figure 6C , D ) . 10 . 7554/eLife . 02196 . 009Figure 6 . Aph1bc deletion cell autonomously disrupts spine formation which is rescued by Nrg1-ICD expression in vivo . ( A ) Representative drawings of Golgi stained CA1 hippocampal neurons from control and Aph1bc−/− null brains at P30 . so , stratum oriens; sp , stratum pyramidale; sr , stratum radiatum; slm , stratum lacunosum moleculare . ( B ) Sholl analysis of dendritic arbour of neurons from Aph1bc−/−mice did not reveal overt defects in neuronal morphology as compared to control in neither basal nor apical dendrites . Basal , means ± SEM , two-way ANOVA . p>0 . 05 . Ctrl: n = 40; Aph1bc−/−: n = 32 . Apical; Apical , means ± SEM , two-way ANOVA . p>0 . 05 . Ctrl: n = 22; Aph1bc−/−: n = 22 . ( C ) Representative images of apical dendrites of CA1 hippocampal neurons that receive input from Schaffer collaterals from control and Aph1bc−/− mice . ( D ) Histogram shows that spine density is decreased in apical dendrites of Aph1bc−/− deficient neurons . Means ± SEM , t test . p<0 . 001 . Ctrl , n = 31; Aph1bc−/− , n = 46 . ( E ) Schema summarizing the experimental paradigm for cell autonomous Aph1bc loss of function and rescue by Nrg1-ICD via in utero electroporation ( IUE ) at E14 . 5 . ( F ) Basal dendrites of layer II/III cortical pyramidal neurons from Aph1bcfl/fl mutant mice electroporated at E14 . 5 with either GFP as control , GFP-ires-Cre alone to perform single cell Aph1bc−/− deletion or with GFP-ires-Cre and Nrg1-ICD to rescue spine formation and fixed at P30 . ( G ) Quantification of spine density . Spine formation was impaired by single cell deletion of Aph1bc and it was rescued by expression of Nrg1-ICD construct in Aph1bc−/− neurons . Means ± SEM . One-way ANOVA . **p<0 . 001 . Ctrl , n = 41; Aph1bc−/− , n = 78; Aph1bc−/−+Nrg1-ICD , n = 47 . Scale bar in A 50 µm , in C and F 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02196 . 009 Our in vitro experiments suggest that the role of Aph1bc was cell autonomous and that expression of the Nrg1-ICD could rescue Aph1bc loss of function . Thus , to investigate the cell autonomous role of Aph1bc under physiological conditions , we analysed the effect of Aph1bc deletion in single cells developing on a wild type background . For that we performed in utero electroporation ( IUE ) of GFP or GFP-ires-Cre in Aph1bcfl/fl conditional mice . We found that single cell conditional deletion of Aph1bc impaired spine formation as compared to GFP electroporated neurons ( Figure 6E–G ) . Moreover , co-electroporation of Nrg1-ICD together with GFP-ires-Cre could rescue decreased spine density ( Figure 6E–G ) . Altogether , these data indicated that Aph1bc-γ-secretase activity cell autonomously controls dendritic spine formation in vivo at least in part via the activation of Nrg1 intracellular signalling . Here we scrutinized the involvement of Aph1bc in neuronal synaptic function and investigated the hypothesis that Nrg1 intracellular signalling controls synaptogenesis downstream of Aph1bc-γ-secretase . Aph1bc−/− deficient brains did not show overt signs of neuronal degeneration or an alteration in neuronal layering or dendritic branching . Nonetheless , we found that the expression of excitatory synaptic markers , synaptic transmission , mEPSCs and long-term plasticity were impaired in Aph1bc−/− deficient mice . Taking advantage of conditional selective γ-secretase mutant mice , we demonstrated in neuronal cultures that Aph1bc is cell autonomously required for dendritic spine formation and that Nrg1-ICD and CRD-Nrg1-FL expression could rescue the Aph1bc loss of function phenotype . The relevance of Aph1bc-γ-secretase/Nrg1 intracellular signalling in excitatory synaptic function was further supported by the impairment in mEPSCs in Aph1bc−/− neuronal cultures which was significantly alleviated by Nrg1-ICD expression . Notably , exogenous expression of CRD-Nrg1-FL could not rescue the spine formation deficit in Aph1abc−/− neurons , which are completely devoid of γ-secretase activity ( Serneels et al . , 2005 ) . Therefore , the γ-secretase-mediated cleavage of Nrg1 is necessary for Nrg1 rescue of spine formation . Besides , even though Aph1a-γ-secretase might redundantly contribute to Nrg1 processing , collectively our data indicate that Aph1bc-γ-secretase is the major regulator of Nrg1 intracellular signalling in this biological process . Finally , single cell Aph1bc deletion in vivo by in utero electroporation impaired spine formation and could be rescued by Nrg1-ICD expression , emphasizing the physiological relevance of our observations . Our study indicates that the Aph1bc-γ-secretase complex controls the establishment of excitatory circuits under physiological conditions at least in part through the regulation of Nrg1 intracellular signalling . It should be kept in mind that both γ-secretase complexes and Nrg1 isoforms impinge on multiple signalling molecules that control many facets of neuronal development and function . As several putative γ-secretase substrates have been identified it is still possible that some of them may also contribute to the behavioural deficits of Aph1bc−/− mice ( Dejaegere et al . , 2008 ) . Nonetheless , we have previously shown that deletion of Aph1bc does not affect the processing of other major γ-secretase substrates such as Notch and ErbB4 in vivo ( Serneels et al . , 2005; Dejaegere et al . , 2008 ) . In addition , mutations of other γ-secretase substrates such as APP or N-Cadherin have not been previously shown to cause schizophrenia-like behavioural deficits in mice or to be associated with schizophrenia . Hence , the biochemical and cell biological evidences from the current work , and the available genetic evidence are consistent with the conclusion that Nrg1 is the major physiological target of Aph1bc-γ-secretase in vivo in the context of synapse formation . The pathological relevance of human APH1B and NRG1 genetic interaction and of NRG1 processing is further supported by recent studies in Schizophrenic patients linking polymorphisms in these genes to the disease ( Marballi et al . , 2012; Hatzimanolis et al . , 2013 ) . Nrg1 forward signalling controls the establishment of inhibitory circuits in the cortex by activating its specific receptor ErbB4 , which is primarily expressed in cortical interneurons ( Fazzari et al . , 2010; Wen et al . , 2010 ) . On the other hand , the physiological role of CRD-Nrg1 intracellular signalling via the Nrg1 intracellular domain could not be addressed unambiguously in vivo in available mutant mice since deletion of CRD or TM domain of Nrg1 protein would affect both forward and intracellular signalling . Previous in vitro studies proposed that γ-secretase dependent Nrg1 signalling may control the expression of genes that control neuronal survival ( Bao et al . , 2003 ) and dendritic growth during development ( Chen et al . , 2010a ) . Our in vivo experiments using Aph1bc-γ-secretase deficient brains do not demonstrate neuronal loss or impaired dendritic arborisation , similar to the observations in Nrg1 transmembrane domain mutant heterozygous mice ( Zhang et al . , 2009 ) . We speculate that these differences may be explained by differential effects of long term abrogation of Nrg1 intracellular signalling in Nrg1 constitutive null mice which is not the case in Aph1bc−/− conditional mice . On the other hand , our results show that Aph1bc-dependent Nrg1 intracellular signalling promotes dendritic spine formation . Consistent with these findings , schizophrenia-like deficits and impaired maturation of glutamatergic synapses have also been described for mice deficient in Nrg1 , ErbB4 and BACE1 , a protease that initiates Nrg1 processing by cleaving its extracellular domain ( Chen et al . , 2008; Barros et al . , 2009; Del Pino et al . , 2013 ) . Although these deficits were attributed to loss of ErbB4 activation in pyramidal neurons , this interpretation was challenged by the recent discovery that ErbB4 is almost exclusively expressed by cortical interneurons ( Fazzari et al . , 2010; Wen et al . , 2010 ) . More recently , it was proposed that altered glutamatergic wiring could be the result of a homeostatic response to alterations in ErbB4 expressing fast-spiking interneurons ( Del Pino et al . , 2013 ) . Here we propose a complementary , but not mutually exclusive , mechanistic model whereby dendritic spine maturation could directly , in a cell-autonomous manner , be promoted by Aph1bc-dependent Nrg1 intracellular signalling . Notably , it has been suggested that Nrg1 could be involved in activity-dependent regulation of the structural plasticity of glutamatergic circuits since neuronal activity enhances both Nrg1 expression ( Eilam et al . , 1998 ) and proteolytic processing by γ-secretase ( Bao et al . , 2003; Ozaki et al . , 2004 ) . Consistently , we show here that Aph1bc-γ-secretase-dependent Nrg1 intracellular signalling promotes spine formation . In conclusion , we provide a cellular and molecular mechanism for the cognitive deficits observed in Aph1bc-γ-secretase-deficient mice . Moreover , our study suggests that schizophrenia linked cSNPs in TM domain of NRG1 ( Walss-Bass et al . , 2006; Mei and Xiong , 2008 ) and mutations in APH1B gene ( Hatzimanolis et al . , 2013 ) may contribute to the alteration of dendritic spines density observed in schizophrenia ( Lewis and Sweet , 2009; Glausier and Lewis , 2013 ) , linking APH1B and NRG1 misprocessing firmly to this disorder . Aph1afl/fl and Aph1bcfl/fl were previously described ( Serneels et al . , 2005 ) . To obtain Aph1bc deficient brains for WB and IF analysis Aph1bcfl/fl mice were crossed with heterozygous Nestin driven Cre mice ( B6 . Cg-Tg ( Nes-cre ) 1Kln/J; Jackson Laboratory , Bar Harbor , ME ) , Cre negative littermates were taken as controls . To obtain homozygous and heterozygous Aph1bc null mice for electrophysiology , Aph1bcfl/fl were first crossed with Pgk driven Cre as described ( Serneels et al . , 2005 ) to obtain constitutive Aph1bc−/− that were backcrossed with wild type mice . Neuronal cultures were performed from conditional Aph1bcfl/fl embryos . All colonies were kept in C57BL/6J background and littermates were taken as controls . All experiments were approved by the Ethical Committee on Animal Experimenting of the University of Leuven ( KU Leuven ) . Mice were transcardially perfused with PBS followed by freshly prepared 4% PFA in PBS , cut with cryostat at 40 µm , and eventually processed for immunofluorescence on floating sections as described ( Fazzari et al . , 2010 ) or cut with Vibratome at 100 µm and processed with FD Rapid Golgistain kit ( PK401; FD NeuroTechnologies ) according to manufacturer instructions . Antibodies: mouse anti-NeuN ( MAB377; 1:500; Chemicon-Millipore ) ; rabbit anti-CDP ( Cux1 ) ( sc-13024; 1:200; Santa Cruz Biotechnology ) ; mouse anti-VGluT1 ( MAB5502; 1:200 in IF; Chemicon ) ; rabbit anti-Homer1 ( 160003; 1:500; synaptic systems ) ; rabbit anti-VGAT ( 131003; 1:500; synaptic systems ) ; mouse anti-Parvalbumin ( 235; 1:500; Swant ) chicken anti-GFP ( GFP-1020; 1:1000; Aves ) . All secondary antibodies were conjugated with Alexa Fluor® dyes ( Life technologies ) . Conventional imaging was performed with a Zeiss Axioplan2 upright microscope with 20x Plan-Apochromat ( NA = 0 . 5 ) and 100x Plan-Apochromat oil immersion ( NA = 1 . 4 ) objectives . For confocal imaging we used Olympus FV1000 IX2 Inverted Confocal microscope with 60x UPlanSapo ( NA = 1 . 35 ) oil immersion objective . For image analyses all pictures were processed and quantified with ImageJ software . For neuronal distribution , cortices were divided in 10 bins and NeuN fluorescence intensities in each bin were normalized for the total NeuN intensity . For VGluT1 , Homer1 , VGAT and PV puncta quantification confocal pictures ( 18 confocal planes out of three animals per condition ) were taken 5 μm beneath tissue surface as described ( Fazzari et al . , 2010; Iijima et al . , 2011 ) . Images received automatically thresholds with ImageJ algorithm ( Yen for VGluT1 and Homer1; Intermodes for VGAT and PV ) and resulting masks were redirected to the original image for automatic quantification of puncta intensity . For PV puncta we also applied a circularity filter 0 . 5–1 . 00 and a high size cut-off filter of 1 . 5 µm2 as previously described ( Fazzari et al . , 2010; Del Pino et al . , 2013 ) . For spine quantification in neuronal cultures we took image stacks of transfected neurons ( z = 0 . 5 μm ) and we counted spines in dendritic segments at 90 to 100 µm from the soma . Morphology of Golgi stained neurons was reconstructed with ImageJ Simple neurite tracer . Spine quantification was carried out in image stacks ( z = 0 . 5 μm ) in apical dendrites of CA1 pyramidal neurons at 100 μm from pyramidal layer . In electroporated mice , we selected neurons from layer II/III and we quantified spine density in confocal stacks of basal dendrites starting from 5 µm away from the first branching point . All statistics were performed with Graph Pad Prism software . Mice were sacrificed by cervical dislocation; brains were dissected in ice cold PBS and snap frozen in liquid nitrogen . Prefrontal cortices were homogenized in ice cold HEPES buffer ( 320 mM Sucrose; 4 mM HEPES , pH 7 . 3; EDTA with complete protease inhibitors from Roche ) ; cleared by centrifugation at 800×g for 10 min; equal amount of protein was loaded on NuPAGE 4–12% Bis-Tris gel ( Novex , NP0322 ) and blotted detected with HRP conjugated secondary antibodies using a ECL chemiluminescence detection kit ( NEL105; PerkinElmer Life Sciences ) . All antibodies were diluted in 1% BSA TBST buffer . Antibodies: mouse anti-VGluT1 ( MAB5502; 1:2000; Chemicon ) mouse anti-Synaptophysin ( S5768; 1:5000; Sigma ) ; mouse anti-PSD95 ( ADI-VAM-PS002-E; 1:5000; Stressgen ) ; rabbit Anti-Tubulin ( ab21058; 1:5000; abcam ) . The density of bands was quantified by densitometry using ImageJ software . For field recordings hippocampal slices were prepared as described ( Denayer et al . , 2008 ) . Briefly , 6–8 weeks old mice , were killed by cervical dislocation and the hippocampus was rapidly dissected into ice-cold artificial cerebrospinal fluid ( ACSF , pH 7 . 4 , saturated with carbogen , 95% O2/5% CO2 ) . Transverse slices ( 400 μm thick ) were prepared from the dorsal area and placed into a submerged-type slice chamber , where they were maintained at 33°C and continuously perfused with carbogen-saturated ACSF . After 90 min incubation , tungsten stimulating electrodes and glass recording electrodes were placed into the stratum radiatum of hippocampal area-CA1 to evoke field excitatory post-synaptic potentials ( fEPSP ) . To assess basic properties of synaptic responses , I/O curves were established by stimulation with 30–90 µA constant currents ( pulse width 0 . 1 ms ) . The stimulation strength was adjusted to evoke a fEPSP-slope of 35% of the maximum and kept constant throughout the experiment . Paired pulse facilitation ( PPF ) was examined by applying two pulses in rapid succession ( interpulse intervals of 10 , 20 , 50 , 100 , 200 , and 500 ms , respectively ) at 120 s intervals . 60 min thereafter , baseline recordings were started consisting of three single stimuli with 0 . 1 ms pulse width repeated at a 10-s interval and averaged every 5 min . LTP was induced by three TBS episodes separated by 10 min , with evoked responses monitored at 1 , 4 and 7 min between TBS episodes . 10 min after the last TBS episode , evoked responses were recorded every 5 min during 4 hr . In all experiments , the recording of slices from mutant mice was interleaved by experiments with wild type controls . Patch-clamp recordings of mEPSCs were performed in acute hippocampal slices obtained from control and Aph1bc mutant siblings . Animals were decapitated and the brain was quickly removed and placed in an ice-cold artificial cerebrospinal fluid ( ACSF ) containing ( in mM ) 124 NaCl , 4 . 9 KCl , 1 . 2 NaH2PO4 , 25 . 6 NaHCO3 , 2 MgSO4 , 2 CaCl2 , and 10 glucose , and saturated with 95% O2 and 5% CO2 ( pH 7 . 3–7 . 4 ) . Transverse hippocampal slices ( 400 μm thick ) were cut with a vibratome ( HM 650V; 'MIKROM' ) and stored at room temperature in a holding bath ( pre-chamber ) containing the same ACSF as above . After a recovery period of at least 1 hr , an individual slice was transferred to the recording chamber where it was continuously superfused with oxygenated ACSF at a rate of 2 . 5 ml/min . Pyramidal neurons in the CA1 region of the hippocampus were visually identified using infrared-differential interference contrast ( DIC ) microscopy . Whole-cell recordings were obtained using a patch-clamp amplifier ( MultyClamp 700B; Axon Instruments ) . Patch pipettes ( resistance 3–5 MΩ ) were pulled from borosilicate glass using a horizontal puller ( Model P-97; Sutter Instruments , Novato , CA ) and were filled with a solution containing: 135 mM CsMeSO4 , 4 mM NaCl , 4 mM MgATP , 0 , 3 mM Na-GTP , 0 , 5 mM EGTA , 10 mM K-HEPES; pH 7 . 24; 281 mOsm . Voltage-clamp recordings of mEPSCs were performed in ACSF supplemented with 1 µM tetrodotoxin ( TTX ) and 100 µM picrotoxin ( PicTX ) . Holding voltage −70 mV . Data were low-pass filtered at 2 kHz and acquired at 10 kHz using Digidata 1440 and pClamp 10 software . Off-line analysis of mEPSCs was performed using MiniAnalysis ( v . 6 . 0 . 7 , Synaptosoft , Decatur , GA ) software . Whole-cell recordings in neuronal cultures were performed at DIV20 similarly as described above . Briefly , cells were visualized on a Nikon Eclipse FN1 microscope with a Hamamatsu C10-600 camera; mEPSCs recordings were performed in low fluorescence Hibernate E neuronal culture medium ( HE-If , Brainbits UK ) supplemented with 1 µM tetrodotoxin ( TTX ) . Holding voltage −70 mV . Hippocampi from E17 . 5–18 . 5 mouse embryos were dissected , treated with trypsin and dissociated into single cells by gentle trituration . Cells were resuspended in MEM ( cat no 31095029; Invitrogen ) containing 10% Horse serum ( cat no 26050088; Invitrogen ) , penicillin–streptomycin and 0 . 6% glucose , and then plated at a density of 1000 cells per mm2 on coverslips coated with 1 mg ml−1 poly-L-lysine ( P2636-1g; Sigma ) and laminin 5 µg ml−1 ( cat no 3400-010-01; R&D systems ) . After 2 hr , the plating medium was replaced by Neurobasal medium ( Invitrogen ) , penicillin–streptomycin and B27 supplements ( cat no 17504-044; Invitrogen ) as described ( Fazzari et al . , 2010 ) . Transfection was performed with Lipofectamine 2000 ( Invitrogen ) according to manufacturer instructions . Briefly , neuronal culture medium was taken and replaced with Lipofectamine/DNA mix diluted in Neurobasal medium that was left on neuronal culture for 90–120 min; next , saved conditioned medium was put back on neuronal cultures . After DIV10 , half of the neuronal culture medium was refreshed every other day . Plasmids: pCMV-GFP ( Clontech ) , pCMV-GFP-ires-Cre ( Fazzari et al . , 2010 ) ; CRD-Nrg1 full length tagged with GFP was kindly provided by Prof . Bao Jianxin ( Bao et al . , 2003 ) , and cloned into pcDNA3 . 1 TOPO ( Invitrogen ) with forward primers 5′-AAA TAA GGC GCC ACT ATA GGG AGA CCC AAG CTG GC-3′ , 5′-AAA TAA GGC GCC ATG AAA ACC AAG AAA CAG CGG CAG AAG C-3′ and 5′-AAA TAA GGC GCC ATG CAG AGC CTT CGG TCA GAA CGA AAC-3′ for CRD-Nrg1-FL , Nrg1-ICD and Nrg1-ΔNLS-ICD respectively , adapted from Bao et al . ( 2003 ) and reverse primer 5′-AAT AAT GTC GAC CAA ACA ACA GAT GGC TGG CAA CTA GAA G-3′ for all constructs . The correct expression of all plasmids was tested by WB ( not shown ) . Immunofluorescence was performed as described ( Fazzari et al . , 2010 ) . In utero electroporation was performed as described ( Shariati et al . , 2013 ) . Pregnant mice were anaesthetized by intramuscular injections of 88 mg ketamine and 132 mg xylazine per gram of body weight . The uterine horns were exposed and the plasmids mixed with Fast Green ( Sigma ) were microinjected in the lateral ventricles of E14 . 5 embryos . Five current pulses ( 50 milliseconds pulse/950 milliseconds interval ) were delivered across the head of the embryos ( 36 V ) targeting the dorsal-medial part of the cortex . An equal amount of pCAG-ires-GFP ( 0 . 5 µg/µl ) was electroporated in all conditions to ensure an equal visualization of neuronal morphology . Plamids: pCAG-ires-GFP ( Add Gene , 11 , 159 ) ; pCMV-GFP-ires-Cre ( Fazzari et al . , 2010 ) ; Nrg1-ICD was subcloned from pcDNA3 . 1 ( see above ) to pCAGEN ( 11160; Add Gene ) . All animal experiments were approved by the Ethics Committee of the KU Leuven .	Schizophrenia affects around 1% of the world's population , with symptoms including hallucinations and delusions , apathy and cognitive impairments . Multiple genes and environmental factors interact to increase the risk of schizophrenia , making the causes of the disease—which can differ between individuals—difficult to disentangle . However , Schizophrenia is known to be associated with a reduction in the number of dendritic spines , the small protrusions that allow brain cells to receive inputs from other brain cells . One gene that has repeatedly been implicated in schizophrenia is neuregulin 1 ( NRG1 ) , which encodes a signalling protein with more than thirty different variants . One of these variants , type III NRG1 , is located on the cell membrane . An enzyme called γ-secretase can cleave the 'tail' of this protein , which means that the tail becomes free to move to the nucleus of the cell , where it can alter the expression of genes . Fazzari et al . have now studied how different γ-secretases interact with type III NRG1 by using genetic techniques to remove a specific part of the enzymes in the brains of mice . The brain cells of these mutant mice contained fewer dendritic spines than mice with normal γ-secretases . However , the number of dendritic spines in the mutant mice could be restored by introducing γ-secretase . These results are consistent with a model in which mutations that remove the ability of γ-secretases to cleave NRG1 lead to some of the structural and functional changes in the brain that are associated with schizophrenia . An improved understanding of the properties of the various γ-secretases could also lead to the design of safer versions of drugs called γ-secretase modulators that are used to treat Alzheimer's disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Cell autonomous regulation of hippocampal circuitry via Aph1b-γ-secretase/neuregulin 1 signalling
The histone variant H2A . Z is a universal mark of gene promoters , enhancers , and regulatory elements in eukaryotic chromatin . The chromatin remodeler SWR1 mediates site-specific incorporation of H2A . Z by a multi-step histone replacement reaction , evicting histone H2A-H2B from the canonical nucleosome and depositing the H2A . Z-H2B dimer . Binding of both substrates , the canonical nucleosome and the H2A . Z-H2B dimer , is essential for activation of SWR1 . We found that SWR1 primarily recognizes key residues within the α2 helix in the histone-fold of nucleosomal histone H2A , a region not previously known to influence remodeler activity . Moreover , SWR1 interacts preferentially with nucleosomal DNA at superhelix location 2 on the nucleosome face distal to its linker-binding site . Our findings provide new molecular insights on recognition of the canonical nucleosome by a chromatin remodeler and have implications for ATP-driven mechanisms of histone eviction and deposition . The histone variant H2A . Z , a universal component of nucleosomes flanking eukaryotic promoters , enhancers , and other genetic elements , has an important role in transcriptional regulation ( Santisteban et al . , 2000; Albert et al . , 2007; Barski et al . , 2007 ) . In Saccharomyces cerevisiae , H2A . Z is deposited by the ATP-dependent activity of the multi-component SWI/SNF-related SWR1 complex , which replaces nucleosomal histone H2A-H2B with H2A . Z-H2B in a coupled histone-dimer transfer ( Mizuguchi et al . , 2004; Luk et al . , 2010 ) . SWR1 recruitment to nucleosome-deficient or nucleosome-free regions ( NFRs ) of yeast promoters is due to its preference for nucleosomes adjoining long linker DNA ( Ranjan et al . , 2013 ) , but post-recruitment activation of the SWR1 complex requires binding of both its natural substrates—the canonical nucleosome and the H2A . Z-H2B dimer—which also serve as essential activators of SWR1 ( Luk et al . , 2010 ) . Progression of the SWR1-mediated reaction on the canonical ‘AA’ nucleosome generates the ‘AZ’ and ‘ZZ’ nucleosome states consecutively , which leads to repression of the ATPase and histone exchange activities of SWR1 by the H2A . Z-nucleosome end-product , thereby preventing futile expenditure of chemical energy ( Luk et al . , 2010 ) . Specific α-C helix residues of H2A . Z on the free H2A . Z-H2B dimer are critical for its SWR1-activating function ( Clarkson et al . , 1999; Wu et al . , 2005 ) . However , the key-activating elements of the canonical nucleosome that distinguish it from the non-activating H2A . Z-nucleosome have been obscure . Here , we show that the histone-fold , but not the α-C helix of histone H2A in the nucleosome , has a major role in the activation of the SWR1 complex . We also define a local DNA site on the nucleosome core particle that is critical for activating SWR1 . To identify elements of histone H2A on the canonical nucleosome that activate SWR1 , we constructed hybrids in which H2A segments were systematically interchanged with segments of histone H2A . Z . Hybrid nucleosome substrates were reconstituted and analyzed by a SWR1-mediated histone H2A . Z replacement assay . Interchanging the M6 and α-C regions of H2A with H2A . Z on the nucleosome had only a small effect on SWR1 activity as measured by nucleosomal incorporation of H2A . Z-3xFlag , which generates a native gel mobility upshift ( Figure 1A ) . Strikingly , an additional interchange of the M4 domain of nucleosomal H2A with H2A . Z caused a ∼90% decrease in H2A . Z replacement by SWR1 , and extension of the interchange to M2 and M3 domains further reduced SWR1's activity ( Figure 1A ) . Interchanging the M4 domain alone caused a large reduction in activity of SWR1 , and activity was also reduced by interchange of M5 and M3A domains individually , whereas M2 and M3B domain interchanges had minimal effects ( Figure 1B , C ) . Thus , residues contained entirely within the H2A histone-fold motif ( the α1 helix , α2 helix , and loop 2 ) contribute to activation of SWR1 . Underlying mechanism ( s ) could include improved enzyme binding , as the non-activating H2A . Z-nucleosome shows slightly decreased affinity for SWR1 ( Figure 1—figure supplement 1 ) , but activation at a post-recruitment step is required , because neither ATPase stimulation nor histone replacement occurs under saturating conditions ( [Luk et al . , 2010] and data not shown ) . Interchange of the M4 region of H2A for H2A . Z ( substitution of five residues ) in yeast causes lethality ( Figure 1—figure supplement 2A ) . This indicates that the N-terminal portion of the H2A α2 helix provides an essential function apart from regulating SWR1 activity , which itself is not essential for viability . Glycine 47 is the only surface accessible H2A-specific residue in the M4 region . Strains bearing single or double amino acid substitutions to corresponding H2A . Z residues-G47K and P49A are viable . The single P49A substitution showed no reduction ( even an increase ) of H2A . Z levels at gene promoters by ChIP-PCR . However , the G47K interchange in the M4 region shows reduced H2A . Z incorporation ( average 63% of WT ) , as does the double-substitution G47K , P49A ( average 54% of WT ) , suggesting that G47 facilitates activation of SWR1 in vivo ( Figure 1—figure supplement 2B ) . 10 . 7554/eLife . 06845 . 003Figure 1 . H2A histone regions in the canonical nucleosome that activate H2A . Z replacement by SWR1 . ( A ) Left: H2A/H2A . Z hybrid histones used for reconstituting nucleosomes . Right: Histone H2A . Z replacement assay . Hybrid ( red ) and WT ( green ) nucleosomes ( 2 . 5 nM ) were incubated with SWR1 ( 2 nM ) , H2A . Z-3F-H2B ( 22 nM ) , and ATP ( 1 mM ) for the indicated times , and nucleosomes containing zero ( AA ) , one ( AZ ) , or two copies ( ZZ ) of H2A . Z-3F were resolved by 6% native PAGE . Top: EMSA ( electrophoretic mobility shift assay ) and fluorescence imaging . Bottom: H2A . Z incorporation curves . ( B ) Sequence alignment of histone H2A and H2A . Z from budding yeast . ( C ) Histone replacement assay as above , with hybrid nucleosomes containing fine H2A/H2A . Z interchanges . Bottom: H2A . Z incorporation curves . DOI: http://dx . doi . org/10 . 7554/eLife . 06845 . 00310 . 7554/eLife . 06845 . 004Figure 1—figure supplement 1 . SWR1 binding to nucleosome core particles containing H2A or H2A . Z histone . EMSA shows SWR1 binding to Alexa 647-labeled H2A- and H2A . Z-nucleosome core particles ( 1 nM ) . Free and bound complexes are resolved on 1 . 3% agarose gel . Bottom: binding curves for H2A- and H2A . Z-nucleosome core particles . DOI: http://dx . doi . org/10 . 7554/eLife . 06845 . 00410 . 7554/eLife . 06845 . 005Figure 1—figure supplement 2 . Effect of H2A M4 on H2A . Z enrichment at gene promoters . ( A ) Viability of H2A mutants . In the plasmid shuffle experiment , yeast cells containing episomal copies of WT HTA1/HTB1 ( under URA selection ) and a second plasmid of WT or indicated mutants ( under HIS selection ) were plated on CSM-His/5-FOA plates . ( B ) ChIP-PCR for H2A . Z-HA for WT and mutant cells . The signal from gene promoters was normalized to a sub-telomeric region on chromosome 6 . Error bars are standard deviations from technical repeat . DOI: http://dx . doi . org/10 . 7554/eLife . 06845 . 00510 . 7554/eLife . 06845 . 006Figure 1—figure supplement 3 . Nucleosome structure showing critical H2A residues that effect SWR1 activity . ( A ) Left: The yeast nucleosome crystal structure 1ID3 in Protein Data Bank was modeled to show histones on one face of nucleosome . Histone H2A is yellow , H2B is black and H3 , H4 are gray . The domains of H2A that affect SWR1 activity-M3A ( cyan ) , M4 ( magenta ) , and M5 ( blue ) are marked . Center and right: Buried residues of histone H2A are shown by removing other histones and rotating on X-axis by 45° . ( B ) The H2A surface residue G47 in 1ID3 is shown in magenta . Bottom left: Zoom-in view shows that G47 is at the bottom of a cleft . Bottom right: Replacing Glycine for Lysine in H2A . Z histone shows the long side-chain of Lysine filling the cleft . DOI: http://dx . doi . org/10 . 7554/eLife . 06845 . 006 We next investigated the role of nucleosomal DNA . Previous studies have established the importance of specific DNA contacts by ATP-dependent chromatin remodelers ( Mueller-Planitz et al . , 2013; Bartholomew , 2014 ) . In vitro , SWR1 is known to bind preferentially to long linker DNA adjacent to a nucleosome core particle ( Ranjan et al . , 2013 ) . To favor SWR1 binding in one orientation , we reconstituted mono-nucleosomes bearing only one 60 bp linker and subjected bound nucleosomes to hydroxyl radical footprinting ( Figure 2A ) . Notably , we observed protection at superhelix locations ( SHLs ) SHL0 , SHL+1 , and SHL+2 on the linker-distal side of the nucleosome dyad ( Figure 2B , C; Figure 2—figure supplement 1 ) . Strongest protection was observed at SHL2 , where other ATP-dependent chromatin remodelers have been shown to interact with the nucleosome , but on the linker-proximal or both sides of dyad ( Bartholomew , 2014 ) . We also observed broad protection from hydroxyl radical cleavage of long linker DNA by SWR1 ( Figure 2D ) , consistent with previous findings ( Ranjan et al . , 2013 ) . 10 . 7554/eLife . 06845 . 007Figure 2 . Hydroxyl radical footprinting of SWR1 on nucleosomal DNA . ( A ) EMSA ( 1 . 3% agarose gel ) shows SWR1 ( 12 pmole; 240 nM ) binding to a fluorescent end-labeled asymmetric 60 bp long linker nucleosome ( 7 . 4 pmole; 150 nM ) after reaction with hydroxyl radical . ( B ) DNA samples resolved on 8% sequencing gel . Top strand is fluorescently labeled . The strongest protected area is shown as magenta bar . ( C ) Top strand intensity plots of free ( green ) and bound ( red ) nucleosome corresponding to B . ( D ) Bottom strand intensity plots for free and bound samples were normalized to signals at +2 and +3 SHL from dyad axis . DOI: http://dx . doi . org/10 . 7554/eLife . 06845 . 00710 . 7554/eLife . 06845 . 008Figure 2—figure supplement 1 . Position of SWR1 footprint on linker-distal face of nucleosome . The 601 DNA-containing nucleosome structure PDB 3MVD was modeled to highlight the position of the SWR1 footprint in blue on the linker-distal side of the dyad axis . The H2A on the linker-distal face is in yellow . DOI: http://dx . doi . org/10 . 7554/eLife . 06845 . 008 Furthermore , we examined how gaps in nucleosomal DNA interfere with histone replacement by SWR1 . Systematic introduction of two-nucleotide gaps on one DNA strand of a nucleosome showed that a single gap at −17 , −18 nt from the nucleosome dyad blocked the second round of H2A . Z replacement ( Figure 3A ) . Further scanning identified a 6 nt region ( −17 to −22 nt from the dyad ) , whose integrity is required for histone replacement ( Figure 3B ) . This gap-sensitive region overlaps with the hydroxyl radical footprint of linker-oriented SWR1 at nucleosome position SHL2 . Introduction of gaps on both sides of the nucleosome dyad caused a complete failure of H2A . Z replacement by SWR1 ( Figure 3C ) . Taken together , our findings indicate that close contact between SWR1 and nucleosomal DNA around SHL2 is critical for enzyme activation . This activation likely occurs post-recruitment , as SWR1 binding is not adversely affected on the gap-containing nucleosome substrate ( Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 06845 . 009Figure 3 . DNA gaps block SWR1 activity when positioned 17–22 bp on either side from dyad . All nucleosomes have a 20 bp linker DNA at both ends , and a two-nucleotide gap introduced at indicated positions . EMSA ( 6% native PAGE ) shows the H2A . Z replacement reaction , terminated at the indicated times , using fluorescently labeled nucleosomes ( 4 nM ) , SWR1 ( 2 nM ) , and H2A . Z-3F-H2B dimer ( 10 nM ) . Nucleosome products with 0 , 1 , and 2 H2A . Z-3FLAG molecules are resolved ( AA , AZ , ZZ ) . ( A ) Mapping of gap sites that block SWR1 activity . Left: Design of WT and gap nucleosomes . Right: ( + ) and ( − ) denote presence and absence of the AZ or ZZ species . ( B ) Fine mapping of the gap-sensitive region near two turns from nucleosome dyad . ( C ) Gaps within the sensitive region on both sides of nucleosome completely block SWR1 activity . DOI: http://dx . doi . org/10 . 7554/eLife . 06845 . 00910 . 7554/eLife . 06845 . 010Figure 3—figure supplement 1 . SWR1 binding to nucleosome core particle with gaps on both sides of dyad . Fluorescently labeled WT ( green ) and Gap ( red ) nucleosome core particles ( 5 nM ) were mixed with indicated amounts of SWR1 . Free and SWR1-bound nucleosome core particles were resolved on a 1 . 3% agarose gel . Bottom: binding curves for WT and Gap particles . DOI: http://dx . doi . org/10 . 7554/eLife . 06845 . 01010 . 7554/eLife . 06845 . 011Figure 3—figure supplement 2 . Nucleosomal histone and DNA elements critical for SWR1 activity and model for SWR1-mediated H2A-H2B displacement . ( A ) Yeast nucleosome structure PDB 1ID3 was modeled to show one face of the nucleosome and the histone-fold elements that are critical for SWR1 activation . The SWR1 footprint is shown in blue . The gap-sensitive region , 17–22 nt from dyad , is shown in cyan . Residues of H2A that affect SWR1 activity are shown in magenta . ( B ) Nucleosome model showing histone-DNA and histone–histone interactions that hold H2A-H2B within the nucleosome . Also shown is the gap-sensitive region , where SWR1 interacts with nucleosome DNA leading to eviction of H2A/H2B and concomitant deposition of H2A . Z/H2B . DOI: http://dx . doi . org/10 . 7554/eLife . 06845 . 011 To date , all chromatin remodelers examined are able to mobilize positioned nucleosome in vitro , including the strongly positioned 601 nucleosome ( Lowary and Widom , 1998; Becker and Workman , 2013; Bartholomew , 2014 ) . SWR1 quantitatively evicts both H2A-H2B dimers on this nucleosome , replacing them with H2A . Z-H2B , but whether nucleosome positioning was also altered was unknown . To examine this question , we used a 601 nucleosome bearing a 60 bp linker on one side , and a native PAGE mobility assay , which separates nucleosomes on the basis of different linker lengths and spatial orientation ( Hamiche et al . , 1999 ) . We found no substantial mobility shift indicative of a repositioned nucleosome after incorporation of ( untagged ) H2A . Z-H2B ( Figure 4A ) . Similar results were obtained for a center-positioned 601 nucleosome ( data not shown ) . By contrast , the INO80 remodeler mobilized the nucleosome from the end- to center-position , as shown by gel mobility shift ( Figure 4A ) ( Shen et al . , 2000; Udugama et al . , 2011 ) . For a more discerning technique , we mapped the precise position of AA , AZ , and ZZ nucleosomes after histone H2A . Z replacement by hydroxyl radical footprinting ( Figure 4B ) . Strikingly , at single nucleotide resolution , there was no net change of the 601 nucleosome position after it underwent one or two rounds of histone H2A . Z replacement ( Figure 4C ) . 10 . 7554/eLife . 06845 . 012Figure 4 . SWR1 mediates histone exchange without net change of nucleosome position . ( A ) Left: EMSA ( 6% native PAGE ) shows INO80-mediated nucleosome sliding . An asymmetrically positioned 601 nucleosome with a 43 bp and 0 bp DNA linker was used for the sliding assay . Right: SWR1-mediated incorporation of H2A . Z-H2B dimer ( without 3FLAG epitope tag ) . Incorporation of H2A . Z in nucleosome was confirmed by immunoblotting with anti-H2A . Z antibody . ( B ) Hydroxyl radical footprinting strategy . A canonical nucleosome with 60 bp and 0 bp linker DNA and fluorescence end-label ( bottom strand ) was used as substrate for histone replacement , followed by hydroxyl radical treatment and separation by 6% native PAGE . ( C ) Recovered DNA from gel slices containing AA , AZ , and ZZ states was analyzed on DNA sequencing gels . ( D ) Intensity plots for AA , AZ , and ZZ nucleosomes . DOI: http://dx . doi . org/10 . 7554/eLife . 06845 . 012 We have identified elements of the canonical nucleosome that activate the SWR1 complex for histone H2A . Z replacement . A DNA site at SHL2 on the nucleosome , in a region identified for DNA translocation by chromatin remodelers RSC , SWI/SNF , ISW2 , and ISW1 , is also important for histone exchange by SWR1 . The ATPase domains of the SNF2 and ISW2 nucleosome sliding complexes are known to interact with the nucleosome at SHL2 ( Dang and Bartholomew , 2007; Dechassa et al . , 2012 ) , and we envision that the catalytic Swr1 ATPase also contacts the SHL2 site . Footprinting experiments show that other chromatin remodelers contact the nucleosome core particle at either the linker-proximal side or both sides of the dyad axis; however , SWR1 contacts the core particle on the linker-distal side of the dyad . This distinction between SWR1 and other remodelers may reflect the unique requirements of dimer eviction and deposition as opposed to nucleosome sliding . Within the nucleosome core particle , each H2A-H2B dimer is stabilized by histone–DNA interactions ( at three minor groove locations SHL3 . 5 , SHL4 . 5 , and SHL5 . 5 ) and histone–histone interactions ( the α2 and α3 helices of H2B interact with α2 and α3 helices of H4 in a four-helix bundle ) . For histone exchange , SWR1 must disrupt either one or both of these interactions , coordinated with H2A . Z-H2B deposition . This might be initiated by transient , confined DNA translocation by the SWR1 ATPase essentially as indicated for other remodelers ( Clapier and Cairns , 2009; Mueller-Planitz et al . , 2013 ) , but without propagation as histone exchange is not accompanied by repositioning of the histone octamer on DNA . Alternatively , histone replacement could be initiated by a local , ATP-driven DNA conformational change near SHL2 that alters the path of the DNA superhelix , resulting in destabilization of H2A-H2B contacts with DNA or with the H3-H4 tetramer ( Figure 3—figure supplement 2 ) . In budding yeast , the +1 nucleosome flanked on one side by a NFR would orient SWR1 to interact with the linker-distal face . We speculate that this configuration favors replacement of the H2A-H2B dimer on the NFR-distal side ( Figure 2—figure supplement 1 , Figure 3—figure supplement 2 ) . Consistent with this possibility , recent genome-wide sub-nucleosomal mapping shows enrichment of H2A . Z at the NFR-distal face of the +1 nucleosome ( Rhee et al . , 2014 ) . Earlier work has shown that the structures of H2A- and H2A . Z-containing nucleosomes show prominent differences in the region C-terminal to the histone-fold domain ( Suto et al . , 2000 ) . This C-terminal region is important for binding of the free H2A . Z-H2B dimer to specific chaperones ( Luk et al . , 2007; Zhou et al . , 2008; Hong et al . , 2014 ) , and for effector interactions post-incorporation ( Clarkson et al . , 1999; Adam et al . , 2001 ) . For histone H2A . Z replacement , our analysis shows that SWR1 utilizes other unique and conserved features of the H2A nucleosome for substrate specificity . Of the three SWR1-activating regions of the H2A histone-fold , the α2 helix and loop 2 are exposed on the nucleosome surface for contact with SWR1 , whereas the α1 helix is buried and may act by allostery ( Figure 1—figure supplement 3A ) . Residue G47 of the H2A α2 helix is highly conserved and is located at the bottom of a cleft ( ∼8 Å deep ) on the H2A nucleosome surface ( Figure 1—figure supplement 3B ) . This cleft might serve as a structural feature for recognition by SWR1; the presence of a Lysine residue at this position in H2A . Z would fill it ( Figure 1—figure supplement 3B ) . It would be of interest to determine structural interactions of SWR1 with this local nucleosome surface . Our findings provide new insights on the structural basis by which canonical and H2A . Z-nucleosomes are recognized by SWR1 and should facilitate future studies of the histone H2A . Z replacement mechanism . Plasmid pZS66 used in this study was a gift from Zu-Wen Sun . It was made by cloning HTA1-HTB1/BamHI-SacII 2 . 6 kb fragment ( 913673–916283 sequence of chromosome IV ) into the same site of pRS313 ( HIS3 , CEN ) , and pZS66 was an intermediate for pZS145 ( HTA1-Flag-HTB1 ) ( Sun and Allis , 2002 ) . Yeast strain FY406 was a gift from Fred Winston and allowed mutating the sole copy of the gene-expressing histone H2A ( Hirschhorn et al . , 1995 ) . All strains used are listed in Supplementary file 1 . DNA for nucleosome preparations was PCR amplified from a plasmid containing the Widom's 601 DNA ( Lowary and Widom , 1998 ) . Primers labeled with Cy5 , Cy3 , or 6-FAM ( 6-carboxyfluorescein ) were used for PCR . For nucleosomes with DNA gaps: primers containing deoxyuridine residues at gap sites were used for PCR amplification , and the PCR product was treated with a mix of Uracil-DNA glycosylase and endonuclease III ( USER Enzyme from NEB , Ipswich , MA ) ( Zofall et al . , 2006 ) . DNA fragments with a gap have slower mobility on 6% native PAGE; and this was used to monitor completion of digestion . All DNAs , with and without gap , were PAGE-purified using a Mini Prep Cell ( Bio-Rad , Hercules , CA ) . Recombinant core histones from yeast H2A , H2B , H2A . Z , and fly H3 , H4 were purified following methods described earlier ( Luger et al . , 1999; Vary et al . , 2004 ) . Nucleosomes were reconstituted by salt gradient dialysis following a standard protocol ( Luger et al . , 1997 ) . The complex was purified as published ( Luk et al . , 2010; Ranjan et al . , 2013 ) . In brief , SWR1-3FLAG was affinity purified from 12-liter budding yeast cells and sedimented over a 20–50% glycerol gradient . Peak fractions were pooled and concentrated using Centricon filters ( 50 kDa cut-off ) , and the buffer was changed to 25 mM HEPES–KOH ( pH 7 . 6 ) , 1 mM EDTA , 2 mM MgCl2 , 10% glycerol , 0 . 01% NP-40 , 0 . 1 M KCl . Aliquots of purified SWR1 were flash frozen and stored at −80°C . Hydroxyl radical footprinting was performed with minor modifications according to ( Schwanbeck et al . , 2004 ) . Before setting up the reaction , an aliquot of purified SWR1 was thawed and buffer changed to mEX ( 5 mM HEPES–KOH pH 7 . 6 , 0 . 3 mM EDTA , 0 . 3 mM EGTA , 0 . 01% NP40 , 56 mM KCl , 5 . 6 mM MgCl2 ) . In a 1 . 5 ml tube , 12 pmole of SWR1 was mixed with 7 . 4 pmole 6-FAM labeled nucleosomes . Nucleosomes were in TE/50 buffer ( 10 mM Tris pH 7 . 5 , 1 mM EDTA , 50 mM NaCl , and 0 . 4 mg/ml BSA ) . Typically , 20 µl nucleosome and 20 µl SWR1 were mixed and volume made up to 50 µl with mEX . On the inner wall of the tube , at different spots , the following solutions were placed: ( i ) 0 . 5 µl of 20 mM ( NH4 ) FE ( II ) SO4 , 40 mM EDTA , ( ii ) 2 . 5 µl of 100 mM sodium ascorbate , ( iii ) 0 . 5 µl of 3% vol/vol hydrogen peroxide . Ammonium iron ( II ) sulfate powder ( light sensitive ) was freshly dissolved in water to make 20 mM ( NH4 ) FE ( II ) SO4 , 40 mM EDTA solution . The sodium ascorbate solution is light sensitive and it can be stored for few weeks at 4°C . Stock hydrogen peroxide purchased from Sigma is 30% vol/vol and it is diluted in water before use . Hydroxyl radical cleavage was initiated by spinning down reagents in microfuge and after 1 min at RT , the reaction was stopped by adding 5 µl of 100 mM thiourea , 0 . 5 µl of 500 mM EDTA and 8 µl sucrose loading buffer ( 50% sucrose in TE/50 ) . SWR1-bound and free nucleosomes were resolved on 1 . 3% agarose gel in 0 . 2× TB . DNA from free and SWR1-bound nucleosomes was excised from gel ( as shown in Figure 2A ) , and resolved on an 8% sequencing gel . FAM fluorescence signal from end-labeled DNA was scanned through sequencing glass plates on Typhoon scanner . The assay for SWR1-mediated H2A . Z-3FLAG/H2B incorporation in mono-nucleosomes is published ( Mizuguchi et al . , 2004; Ranjan et al . , 2013 ) . In brief , purified SWR1 , reconstituted nucleosomes , and recombinant H2A . Z-3FLAG/H2B dimer were mixed with ATP at RT . Reactions are terminated by adding excess lambda DNA at indicated times . Incorporation of H2A . Z-3FLAG/H2B slowed the mobility of nucleosomes on 6% native PAGE , and nucleosomes containing 0 , 1 , or 2 copies of H2A . Z-3FLAG/H2B are resolved . ChIP follows ( Venters and Pugh , 2009 ) . Briefly , yeast cells with wild-type H2A or mutant H2A were grown in CSM-His medium at 30°C to A600 = 0 . 7 and fixed with 1% formaldehyde at room temperature for 15 min . Chromatin was sheared by sonication and H2A . Z-HA bound chromatin was immunoprecipitated using anti-HA antibody ( Clone HA-7 , Sigma ) and Magna ChIP Protein G beads ( Millipore ) . Purified DNA was analyzed for enrichment of gene promoter sequences over a control sub-telomeric region on Chromosome 6 ( Kurdistani and Grunstein , 2003 ) by multiplex PCR . The H2A . Z replacement reaction was set up with 100 nM 6-FAM labeled nucleosome , 40 nM SWR1-3FLAG , 400 nM H2A . Z-3XFLAG/H2B dimer , and 1 mM ATP in 50 µl mEX buffer . After 1 hr at RT , reaction was stopped by adding 4 µg of competitor salmon sperm DNA . Hydroxyl radical cleavage was performed as described above , and nucleosomes were resolved in 6% native PAGE . DNA from nucleosomes containing 0 , 1 , and 2 copies of H2A . Z-3FLAG/H2B was eluted and resolved on an 8% DNA sequencing gel ( SequaGel 19:1 Acrylamide:BisAcrylamide , National Diagnostics ) . Supplemental information includes six figures and a list of strains used .	A DNA molecule can be several meters long and to fit this length inside a cell , it is wrapped around proteins called histones . This compacts the DNA to form a structure known as chromatin; complexes of DNA and histones , called nucleosomes , serve as the building blocks of chromatin . Cells regulate the organization of chromatin to switch genes ‘on’ and ‘off’ . Complexes of proteins , such as SWR1 , alter the packing of chromatin and are known as ‘chromatin modifiers’ . To express a gene , parts of the chromatin have to unpack to allow various proteins and other factors to access to the underlying DNA . Chromatin remodeling enzymes can loosen chromatin by sliding nucleosomes away from each other , removing them altogether , or replacing one type of histone with another . For example , a histone variant called H2A . Z appears to poise genes for expression and is enriched near the start sites of most genes in the genome . The SWR1 complex evicts the conventional , ‘canonical histone’ called H2A that is already present at these sites and replaces them with H2A . Z . H2A . Z is related to H2A , and the SWR1 complex can interact with both of these proteins . However , it remains poorly understood how SWR1 can discriminate between the two at the molecular level . Ranjan et al . have now addressed this in budding yeast cells , by constructing hybrids that contain parts of H2A combined with H2A . Z . The experiments revealed that the SWR1 complex recognizes key elements within the histone H2A protein itself that differ from H2A . Z . Binding to H2A activates SWR1 and causes it to replace H2A with H2A . Z . Ranjan et al . next looked to see if the SWR1 complex also interacts with the DNA present within a nucleosome and whether any gaps in the DNA interfere with histone replacement . The experiments revealed that gaps in DNA at a specific region of the nucleosome prevent SWR1 from depositing H2A . Z . Therefore , close contact between SWR1 and a nucleosome's DNA is another factor that is required for SWR1 activity . These findings provide new insights as to how SWR1 recognizes histone and DNA elements of a canonical nucleosome . Further work is needed to understand how SWR1 acts to replace H2A with H2A . Z .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods", "Supplemental", "information" ]	[ "chromosomes", "and", "gene", "expression", "short", "report", "biochemistry", "and", "chemical", "biology" ]	2015	H2A histone-fold and DNA elements in nucleosome activate SWR1-mediated H2A.Z replacement in budding yeast
Most behaviors such as making tea are not stereotypical but have an obvious structure . However , analytical methods to objectively extract structure from non-stereotyped behaviors are immature . In this study , we analyze the locomotion of fruit flies and show that this non-stereotyped behavior is well-described by a Hierarchical Hidden Markov Model ( HHMM ) . HHMM shows that a fly's locomotion can be decomposed into a few locomotor features , and odors modulate locomotion by altering the time a fly spends performing different locomotor features . Importantly , although all flies in our dataset use the same set of locomotor features , individual flies vary considerably in how often they employ a given locomotor feature , and how this usage is modulated by odor . This variation is so large that the behavior of individual flies is best understood as being grouped into at least three to five distinct clusters , rather than variations around an average fly . There are many approaches to the study of neural underpinnings of behavior: One large body of work is rooted in the psychophysical literature where an animal is forced to choose between a few discrete behaviors ( Green and Swets , 1974 ) . This approach allows stimulus control and rigorous analysis of behavior based on an established framework ( Gold and Shadlen , 2000 ) , but sacrifices a full analysis of behavioral dynamics , leaving critical issues unexplored . Other studies have focused on behaviors that are reflexive ( albeit with some flexibility ) such as saccades ( Laurutis and Robinson , 1986 ) and collision avoidance in insects ( Tammero and Dickinson , 2002 ) . Another large body of work has focused on the control processes involved in goal-directed behaviors such as reaching movements and has revealed many fundamental principles of motor control . Yet , another popular behavioral motif that has received much attention is behaviors that require meticulous sequencing ( Graybiel , 2008 ) . Finally , much work has been done to elucidate the workings of central pattern generators that underlie the rhythmic motor activity during walking and running ( Grillner , 1979 ) . Although many of these relatively stereotypical behavioral motifs are at play during most behaviors , they are not helpful in describing the structure underlying most everyday activities such as making a cup of coffee or a peanut-butter sandwich or walking to a car which consist of a sequence of actions , but neither the sequence nor each sub-action is stereotyped . These activities and their underlying sub-actions cannot be described either as sensorimotor reflexes or as behaviors that arise out of meticulous sequencing . An important example of such a behavior is an animal’s locomotion . While tracks of a mouse or a fly exploring a chamber are not stereotypical there is an obvious structure to it . Uncovering the structure within non-stereotyped behaviors such as locomotion requires sophisticated analytical tools . These tools can be applied to two complementary representations of an animal’s behavior which can be described in the shape/posture space or in the world coordinate system . Recently much progress has been made in employing analytical tools that describe behaviors as a sequence of transformations in the shape/posture space using both supervised ( Branson et al . , 2009; Kabra et al . , 2013 ) , and unsupervised ( Berman et al . , 2016; Berman et al . , 2014; Wiltschko et al . , 2015; Vogelstein et al . , 2014 ) algorithms . These studies provided remarkable insights by showing that much of an animal’s behavioral repertoire can be described as transitions between few postures . While behavioral descriptions as transformations in posture space accurately classify the type of behavior ( such as locomotion vs . grooming ) , because an animal’s position in the world coordinate system is ignored , the structure within an animal’s trajectory in world coordinates remains relatively unexplored . Much of the work in extracting structure from an organism’s trajectory is derived from the ‘run and tumble model’ which was originally employed in the context of bacterial chemotaxis ( Berg and Brown , 1972 ) . In this model , the organism is assumed to travel in relatively straight lines ( runs ) of exponentially distributed run lengths until they make sharp turns ( tumble ) to choose another direction at random . It is tempting to consider the motion of larger animals as roughly approximated by a run-and-tumble model , and many studies ( explicitly and implicitly ) employ a run-and-tumble framework with increasing sophistication as an analytical framework for locomotion ( Kim and Dickinson , 2017; Pierce-Shimomura et al . , 1999; Schulze et al . , 2015 ) . One obvious well-documented limitation of this framework is that animals do not turn at discrete times ( Stephens et al . , 2010; Ohashi et al . , 2014; Iino and Yoshida , 2009; Jung et al . , 2015; Gomez-Marin and Louis , 2014; Straub and Heisenberg , 1990 ) , and therefore , their locomotion cannot be described using a run-and-tumble framework . More generally , larger animals are likely to exert greater control over speed and direction of their locomotion and a better model is necessary to understand the resulting structure in their trajectory . The lack of a model for locomotion makes it difficult to quantify the effect of stimuli on locomotion and is a critical missing piece in understanding the underlying sensorimotor transformations . For example , in many studies of odor modulation of locomotion , odors are primarily described as attractive or repulsive; this description is based on the end result , and does not consider the navigational maneuvers that underlie these end-results . Ignoring the underlying navigational maneuvers has led to a fundamental misunderstanding of odor modulation of locomotion . In a recent detailed analysis of a fly’s locomotion , we demonstrated that it's navigational maneuvers in response to similarly attractive odors are quite distinct ( Jung et al . , 2015 ) ; the analysis used was based on an ad-hoc parametrization of locomotion , and not on a generative model of locomotion , making it difficult to determine whether our chosen parameter set was appropriate . A model of locomotion also makes it possible to compare how locomotion is affected by a given stimuli , and also how different individuals differ in their locomotion and in their response to stimuli . In this study , we employ a hierarchical statistical model , Hierarchical Hidden Markov Model ( HHMM ) to describe the structure in the fly’s locomotion ( Fine et al . , 1998 ) . We show that fly locomotion is well-structured and an HHMM is an elegant representation of this structure . HHMM provides a simple and intuitive description of both a fly's locomotion and the effect of odors on the same . Surprisingly , different flies employ different strategies in their locomotion both before odor onset and in response to odors . Our data are , thus , inconsistent with the idea that the behavior of different flies represent variations around an ‘average’ fly . Rather , our data are most consistent with the idea that flies employ three to five different strategies , at a minimum , to explore a small circular arena and a similar number in their response to odors . We model the locomotion of wild-type flies exploring a circular arena ( Jung et al . , 2015 ) whose center ( odor-zone ) consists of a fixed concentration of odor ( Figure 1A ) . The arena and the experimental procedure was previously described ( Jung et al . , 2015 ) . Briefly , locomotion of each of the 34 flies in our dataset was measured 3 min before an odor ( apple cider vinegar or ACV ) was turned on , and 3 min during the presence of ACV . Sample trajectories are shown in Figure 1B . We first attempted to model the fly’s locomotion using Hidden Markov Model ( HMM ) ( Gallagher et al . , 2013; Isakov , 2016 ) . HMMs create discrete states based on a time series of observables such as position , speed or acceleration . The advantage of using HMMs in modeling locomotion is well described in earlier studies ( Gallagher et al . , 2013 ) ( see Materials and methods ) . In this study , we use observables that describe the change of position as a function of time , and hence our analysis will focus on behavioral states in the velocity space . Speed and angular speed are commonly used measures of velocity . But , because it is difficult to measure angular speed accurately at low speeds ( Gallagher et al . , 2013 ) ( see Materials and methods section 2 for details ) , we fit the model to the component of speed parallel ( v^\|\| ) and perpendicular ( v^⊥ ) to its movement during the previous time point ( Figure 1C , Materials and methods section 2 ) . If the fly walks straight ahead , v^⊥ would be zero , therefore , v^\|\| and v^⊥are closely related to speed and angular speed . We fit a time series of v^\|\| and v^⊥ to an HMM . The HMM architecture is shown in Figure 1D . We employed models with 24 to 50 states . HMM was only modestly successful because it was never able to classify >70% of the tracks into one of the states with >85% confidence . The transition probability matrix for the HMM was sparse ( Figure 1—figure supplement 1 ) suggesting that from each state there are transitions to only a handful of other states . One method to improve upon HMM performance is to cluster the states obtained by HMM according to the transition probability matrix . A common approach to clustering is to block-diagonalize the transition probability matrix ( Berman et al . , 2016 ) . Tracks corresponding to the 10 states obtained by clustering are shown in Figure 1—figure supplement 1 . Some of these states appear to describe recognizable features in the data such as left ( state 10 ) or right turn ( state 9 ) . But efforts to block-diagonalize the transition probability matrix were only partially successful . The most obvious failure corresponds to the states with little movement . These states – describing the absence of movement – can occur in many different contexts , such as the fully stopped state or intermittent runs . When the same state is used in different contexts , an approximate block diagonalization of the transition probability matrix fails because the same state belongs to two blocks . In these cases , the different states that correspond to the absence of movement did not cluster , and appeared alone even after block-diagonalization ( Figure 1—figure supplement 1 ) . Thus , the existence of the same state in different contexts is one important reason for the modest success of HMM ( Marco et al . , 2017; Wonjoon et al . , 2010; Michele Weiland and Nelson , 2005; Nguyen et al . , 2005; Murphy and Paskin , 2001; Chou , 2006 ) . We employed a two-layered Hierarchical Hidden Markov model ( HHMM ) to model the data ( Figure 2A ) . We reasoned that the low-level states ( LL states ) would be represented by Gaussian distributions on the observables , and the high-level states ( HL states ) would therefore be a mixture of Gaussians and would be able to model the experimental data better . Moreover , these HL states would have longer duration than the states discovered by HMM allowing it to more naturally model composite states . Indeed , in a Bayesian model comparison ( see Materials and methods section 4 ) , HHMMs outperformed HMMs with the same number of states . Since HMMs rarely used more than 35 states ( even when models with higher number of states were fit ) , to perform model comparisons , we used models with less number of states than the particular HHMM we eventually employed . We compared a two-level HHMM with 6 HL states and 4 LL states to a single-level HMM with the same number of states ( 4 × 6 = 24 states ) . We labeled the non-hierarchical model as the null model , and were able to reject the null model using Bayesian model comparison at p < 0 . 0001 , implying that a hierarchical model is necessary . Another model comparison – a two-level model with 8 HL and 4 LL states compared to a single-level model with 32 states – also yielded similar results . The objectively better performance of an HHMM compared to an HMM suggests that a model that includes a hierarchical structure is more consistent with a fly’s locomotion . It is important to note that , HHMMs are actually simpler than HMMs with the same number of states . This simplicity comes from the fact that any HHMM – which puts very specific constraints on the transition probability matrix – can be represented by an HMM but not vice-versa . HHMMs with the same number of states has far fewer parameters . Thus , for the two comparisons above , the HHMM has 62 + 642 + 645 = 252 , and 82 + 842 + 845 = 352 parameters , and the HMM has 242 + 245 = 696 and 322 + 325 = 1184 parameters respectively; therefore , HHMM has fewer parameters . When a simpler model better characterizes the data , we can conclude that the additional structure contained in that model provides a more accurate characterization of the structure within the data . The model we chose has 10 HL states ( Figure 2A ) and 5 LL states for each HL state . The model was fit to the entire dataset – both before and during the presence of the odors . The fitting process initializes by fitting each fly’s tracks to its own HHMM and then clusters these 34 HHMMs – one for each fly – using a Gaussian mixture model , resulting in a smaller number of models . Remarkably , a single HHMM is an excellent fit for all the data suggesting that the behavior of wild-type flies is composed of similar components . The model was able to successfully assign an HL state ( defined as >85% confidence ) for >80% of the data points ( Figure 2B , median 81% ) . This percentage was consistently high for all flies in our dataset ( Figure 2B ) . In comparison , an HMM with 50 states can only classify 68% of the data with the same level of confidence . Tracks of a fly with each HL state labeled with a different color are shown in Figure 2C . To make the difference between HHMM and HMM clearer , we compare the HHMM above to a HMM . As expected the time a fly spends in a HMM state is shorter than that in a HHMM state ( Figure 2—figure supplement 1A ) . The longer time a fly spends in a HHMM state results from its hierarchical structure , and allows a HHMM to more accurately assign states , and is illustrated with two examples . First , consider a track that is assigned as a left turn by the HHMM , the HMM only classifies parts of the track as a left turn because of the inability of HMMs to consider longer duration trends in the observables ( Figure 2—figure supplement 1B1 ) . Short-term inhomogeneity in the data throws the HMM off; as soon as the v^⊥ decreases , the HMM exits its left turn state . Another example ( Figure 2—figure supplement 1B2 ) shows that the HMM exits the stopped state as soon as there is a small movement . The net result is that the HHMM can classify all long stops into a single state while HMM needs four different stop states . HHMM also assigns more of the left turn as such ( 6% compared to only 2% by HMM ) . The duration of the LL states of a HHMM is shorter than the duration of the HL state state ( Figure 2—figure supplement 2A ) . Moreover , as the duration of HL states becomes longer , the mean number of transitions increase . The shorter duration of LL states compared to the HL states , and increased number of transitions between LL states within each HL state transition support the idea that there is structure at multiple timescales , and some of this structure is captured by the HHMM . The limitations of HMM in describing phenomenon which have hierarchical and shared structure because of the short duration of its states is well documented ( Marco et al . , 2017; Wonjoon et al . , 2010; Michele Weiland and Nelson , 2005; Nguyen et al . , 2005; Murphy and Paskin , 2001; Chou , 2006 ) . Therefore , an HHMM is an objectively a better model of fly walking data than an HMM . Both the organized transitions between HL states , and the narrow range of observables associated with each state shows that the fly’s locomotion is structured: The transition probability matrix is sparse – a vast majority of transitions from each HL state were to 2-3 other HL states . When we reordered the states ( see Materials and methods section 5 ) from low-speed-high-turn-states to high-speed-low-turn states , we found that from any state the flies transitioned to the neighboring states with a high probability ( Figure 2—figure supplement 3 ) suggesting a gradual transition from low-speed-high-turn states to high-speed-low-turn states . This gradual transition is not because flies cannot make large transitions due to biomechanical limitations because 47/81 1 possible transitions between HL states have a non-zero probability . Rather , transitions to states with similar kinematics show that under our experimental conditions – locomotion in a dark , small circular arena - flies locomote at similar v^\|\| and v^⊥ for extended periods of time , and represent one way in which locomotion is organized . More important to the organization is the narrow distribution of observables - v^\|\| and v^⊥- associated with each HL-state . The distribution of observables for a HL state is a composite of the distributions of its LL states ( Figure 3A ) . Both the model ( solid line in Figure 3A1 ) and a random sample of observables drawn from the time points assigned to a given LL state ( gray markers ) show that during each LL state within HL state 10 , the observables are limited to a narrow range of values . In each LL state , v^\|\| is large and v^⊥ is negative implying that in HL state 10 , flies turn counter-clockwise at high speeds as observed in sample tracks corresponding to a single transition to HL state 10 ( Figure 3B ) . The sample tracks also show that within each transition to a HL state there are multiple transitions between LL states , a signature of the hierarchical organization in our data . Fast , counter-clockwise turns represent a locomotor feature which describes a fly’s locomotion in HL state 10 . To better visualize this feature , we translated each track such that it began at the origin and rotated the tracks so that the initial velocity vector pointed along the y-axis ( Figure 3C1 , see Materials and methods ) . These transformations make it apparent that the locomotor feature for state 10 is turning left at high speeds ( Figure 3C2 ) . Rotated and translated ( as in Figure 3C ) tracks for each of the 10 HL states are shown in Figure 4 . The distribution of the observables for each HL state is also plotted . HL state 1 represents very slow walking with frequent changes in direction . In state 2 , flies are either completely stopped or they walk at a speed about twice the speed of the fly in state 1; state 2 represents stop and start locomotion . The subtle , but important differences between state 1 and state 2 show an instance in which the HHMM is successful at extracting an unexpected feature in the velocity profile in a fly’s locomotion . During state 3 , the fly is exhibiting a sharp turn that is reflected in the increase in v^⊥ with a concomitant decrease in v^\|\| . These three states together represent slow locomotion . In states 4-7 , flies are walking at a medium-speed . In contrast to the clear drop in v^\|\| with increases in v^⊥ in state 3 , v^\|\| remains strikingly constant irrespective of v^⊥ . These states are different from each other because v^\|\| is slightly different . States 8–10 are high-speed states; each of these states is also characterized by their turn direction . During states 8 and 9 , the fly turns right; the fly’s speed is higher during state 9 than during state 8 . During state 10 , the fly turns left . States 9 and 10 are mirror-symmetric versions of each other . Flies spend 60% of time performing a locomotor feature for >300 ms , and >10% of their time performing a single locomotor feature for >3 s ( Figure 2—figure supplement 2 ) . Thus , flies spend extended time in the same state . In the absence of ACV , the state occupancy inside and outside the odor-zone are quite similar: The fly spends 30% of its time in state 2 and roughly equal time in all other HL states . Introducing ACV changes the fly’s locomotor behavior both inside and outside the odor-zone , but with opposite effects on the HL state occupancy in the two zones . Inside the odor-zone , in the presence of ACV , the fly spends more time in HL states 1 and 3 at the expense of time spent in HL states 7–10 ( Figure 5A ) . These changes from high-speed states to low-speed states suggest that in the presence of ACV the fly is performing a local search , presumably to find food . Outside the odor-zone ( Figure 5B ) , the fly spends more time in the high-speed states ( HL states 8–10 ) , with a decrease in the occupancy of HL state 2 ( which includes stopping ) . Decreased stopping and increased high-speed walking with turning is likely to represent a different search strategy , wherein the fly might be attempting to re-find the odor it has recently lost . We also investigated whether there were changes in the LL state composition of the HL states and found no changes ( Figure 5—figure supplement 1 ) . Overall , these results showed that odors affect locomotion not by creating new locomotor features , but by altering the frequency with which existing locomotor features are used . The divergent effect of ACV on the probability of HL states inside and outside the odor-zone is consistent with our previous analysis ( Jung et al . , 2015 ) and shows that the effect of ACV can be described by the change in the probability of the fly occupying HL states . To assess whether there is a more fine-grained spatial structure to the effect of ACV on a fly’s behavior , we divided the arena into a 60-by-60 grid and measured the ACV-induced changes in the probability of occupying a given HL state at each of the 3600 locations ( Figure 6 ) . The probability that a fly is in HL state 1 increases dramatically only at the edge of the odor-zone ( Figure 6A ) , and not throughout the odor-zone where the odor concentration is uniform throughout , showing that the effect of odor on locomotion has a fine-grained spatial structure . Location-specific change in the probability of each HL state are shown in Figure 6B . The fine-grained modulation of locomotion is observed in other states as well - increases in state 2 are largest in an annular region just inside the odor-zone and increases in state 3 are largest at the very center of the arena . A similar specificity is observed in the increase in the probability of HL states outside the odor-zone . Increases in the occupancy of state 8 are uniform across the entire chamber outside the odor-zone; in contrast , the occupancy of states 9 and 10 increases in the region close to the odor-zone . The structure that we observe represents the time-average over the entire duration of the odor period , and ignores the time evolution of the behavior ( Figure 6—figure supplement 1 and discussion ) . We were unable to explore the spatio-temporal evolution of behavior because of substantial fly-to-fly differences in locomotion and how it is modulated by odor . Given that ACV affects the occupancy of HL states , it should be possible to do the reverse , that is decode the presence of ACV based on the distribution of HL states . Surprisingly , a variety of different decoding techniques failed to decode the presence of ACV based on the distribution of HL states . One such method ( Figure 7—figure supplement 1 ) in which we employed logistic regression to classify each one second of every fly’s track into ‘ACV present’ or ‘ACV absent’ failed . Even more surprisingly , population decoding based on HL states did not perform any better than decoding based on the observables ( Figure 7—figure supplement 1 ) . One possibility that the logistic regression approach failed is because the average behavior represented in Figure 5 does not accurately encapsulate individual fly behavior . Large fly-to-fly differences , where different individuals have fundamentally different basal locomotion or response to odor , might doom decoding methods aimed at discovering a single set of regressors that captures individual fly behavior . Consistent with large fly-to-fly differences in behavior , we found that the distances between empirical flies are much larger than the distances between the synthetic flies ( Figure 7A ) . Synthetic flies were generated as described in Figure 7—figure supplement 7–2 . It is statistically impossible ( p < 10−131 ) that the observed Euclidean distance represents variations around the same average fly . The same conclusion applied to the fly’s behavior in the other three conditions ( before-inside , during-outside and during-inside: Figure 7—figure supplement 3 ) . Because the data in Figure 7A is inconsistent with individual flies being variations around a single average fly . We assessed whether the observed variability can be approximated based on a small number of discrete locomotor-types . X-means clustering ( see Materials and methods section 6 ) showed that there are 4 clusters of flies based on their locomotion outside the odor-zone , before odor onset ( Figure 7B ) . Although the identity of flies that cluster together changed , a similar number of clusters was found in each of the four conditions ( Before odor/inside odor-zone , during odor/inside odor-zone , before-outside and during-outside , Figure 7 and Figure 7—figure supplement 3 ) . Importantly , X-means clustering on a set of 34 randomly sampled points from a uniform distribution in the probability simplex space that the data reside in found no clusters . The Euclidean distances between synthetic flies drawn from the four different clusters were similar to the distances between empirical flies ( Figure 7B and Figure 7—figure supplement 3 ) . Since X-means clustering tends to underestimate the actual number of clusters in data ( Pelleg and Moore , 2000 ) , the analyses in Figure 7 and Figure 7—figure supplement 3 suggest that there are at least three to four fly-types based on the frequency with which they use the HL states . Consistent with the analysis with Euclidean distances above , the KL divergences ( Figure 7C , left set of data points ) show a large range indicating that some , but not all , flies are well-represented by the population average while others are not . Employing three to four fly-types defined as average distribution of their respective cluster decreased the information loss ( Figure 7C , right ) . How was the behavior of the flies in the four clusters different from each other ? The average occupancy of the HL states for flies in each cluster and one example from each cluster is shown in Figure 7D . Cluster two was distinct from the others because flies move at markedly slower speed and spent >60% of their time in State 2 ( Figure 7D ) during which the fly was often stopped . Locomotion of flies in the largest cluster ( cluster 1 ) was characterized by an alternation between medium- and slow-speed states . Flies in this cluster employed states 5 and 7 with a high frequency while making radially inward forays into the center of the arena . Similar behavior was observed in cluster 3 , except that the flies employed the slower medium-speed states ( states 4 and 5 ) . Finally , the fourth cluster of flies demonstrated a different locomotor strategy . Flies in cluster four traversed the arena in concentric circles using the high-speed states ( states 8–10 ) . Thus , X-means performed on the HL state distributions appear to identify different locomotor strategies employed by the fly . How similar is the locomotion of a fly at different times during a trial ? To investigate this issue we first examined whether the mean behavior of flies from different clusters are different enough that they can be accurately clustered based on a small sample of the HL states ( Figure 7—figure supplement 4A ) . We limited the analysis to before-outside and during-inside scenarios because the fly spent much of its time in these two scenarios . Only a one-second chunk of data is sufficient for better than chance clustering , and just 30 s of sampling is enough to accurately classify >85% of the flies into their respective clusters . We performed two analyses to test whether a fly’s behavior is persistent: First , we divided the tracks into bins of different length , and asked whether cluster assignments based on small bin sizes is stable ( Figure 7—figure supplement 4B ) . We found that state distribution within each bin was highly predictive of the cluster they belonged to . Second , we repeated the same analysis , but with bins of varying size starting from the first data point or ending at the last data point ( Figure 7—figure supplement 4C ) . These analyses show that within the admittedly short timeframe of our experiments , the cluster assignments are stable . Apart from the fly-to-fly variability , another reason why decoding based on the average fly fails is that the behavior of the fly before odor onset is only weakly predictive of its behavior during the presence of odor . Some flies exhibited similar behavior in the before odor/outside odor-zone , but were divided into separate clusters in the presence of odor because their locomotion differed ( e . g . flies 11 and 33 , and 17 and 27;Figure 7—figure supplement 5A , see the distributions of HL states ) . A similar trend is observed inside the odor-zone ( Figure 7—figure supplement 5B ) . These examples imply that behavior before odor onset is unlikely to be strongly predictive of behavior during the presence of odor . This conclusion is supported by the weak correlation between Euclidean distances between pair of flies in the before and during periods ( Figure 7—figure supplement 5A and Figure 7—figure supplement 5B ) . The analysis presented above suggests that individual differences explain why the logistic regression approach based on the average HL state distribution across flies failed to decode the presence of ACV from its absence based on the HL states usage by individual flies . If so , individualized logistic regression should be more successful . Logistic regression based on individual flies to decode the presence or absence of odor based on HL state occupancy during a 1s-interval was able to classify odor-no odor at better than chance level for every single fly ( Figure 8A , see Materials and methods section seven and Figure 8—figure supplement 1 for details ) . Moreover , as expected , logistic regression using the HL states performed significantly better than did the observables ( Figure 8B ) which indicate that HL states are more predictive of the presence of ACV than the observables . The analysis in Figure 8A shows that the occupancy of HL states is predictive of the presence or absence of odor when the analysis is performed at the level of individual flies . Does this mean that each fly follows an individualistic strategy ? To evaluate whether a fly’s response to ACV cluster into a small number of response types , we once again started with X-means clustering based on the change in state occupancy before and during odor . X-means clustering found five clusters inside the odor-zone and four clusters outside the odor-zone ( Figure 8C ) . Using these clusters as a starting point , we could reconfigure the clusters such that the logistic regression on flies in each cluster performed at a better than chance level for each fly in the cluster ( Figure 8C , see Materials and methods section seven for details ) , thus implying that a fly’s response to odors can be approximated as a choice between few response-types . Based on their behavior inside the odor-zone , the flies were divided into five clusters , four of these clusters have more than three flies ( Figure 9A ) . Flies in cluster 5 , just like the average fly ( in Figure 5 ) , slow down inside the odor-zone . Flies in cluster 3 also demonstrate a strategy similar to flies in cluster five except that ACV causes a large decrease in the time a fly spends in the medium-speed states rather than the high-speed states . In contrast to clusters 3 and 5 during which the fly slows down inside the odor-zone , flies in cluster two demonstrate a fundamentally different strategy in which there is a large decrease in state 2 in favor of states 1 , 3 and 4 . The flies in this cluster go from stop-start locomotion to locomotion in which they either meander at slow speeds or walk slowly with many sharp turns . Finally , for the flies in cluster 1 , there is no dramatic change in state . These different strategies represent diametrically different effects of ACV on some HL states – the most striking example is the opposing effects of the odor on HL state 2 occupancies in different clusters – a large decrease in cluster 2 , and an increase in clusters 3 and 5 . These differences explain the odor-induced increase in usage of all the slow states in the average fly inside the odor-zone , except state 2 ( Figure 6A ) . Based on how odors modulate their behavior outside the odor-zone , there were four clusters . Behaviors that represent three of these clusters are shown in Figure 9B . Flies in cluster 1 decrease the time they spend in slow-states ( state 1–2 ) and instead spend time in the fast states ( states 8–10 ) resembling the behavior of the average fly . Cluster 2 ) shows a large decrease in HL state 2 occupancy similar to the behavior of flies in inside odor-zone cluster 2 while exhibiting a large increase in medium-speed states ( state 4–6 ) . Cluster 3 showed no dramatic change in state . The model presented here is a model of locomotion and not the model of locomotion . The choice of observables and model strongly influences the features of the structure that is discovered . Our particular model reveals the structure of locomotion in the velocity space . In choosing the observables , we employ a common method for describing locomotion , that is we treat the fly as a point object and measure the instantaneous change in the position of this point object; therefore , much of the insights from the model relate to how the fly changes its position in time . Apart from v^\|\| and v^⊥ , another similar and more commonly used representation of the change in fly’s position: instantaneous speed and angular speed yielded similar locomotor features ( data not shown ) . Ultimately , we used v^\|\| and v^⊥ because this representation is more closely related to movement representation within the insect brain ( Green et al . , 2017; Heinze , 2017; Turner-Evans and Jayaraman , 2016 ) , and because the measurement errors associated with angular speed are particularly large when flies moves slowly ( Gallagher et al . , 2013 ) . A fly’s position can also be described using the actual position of the animal as observables rather than the change in the position , as employed in a recent study in rats ( Shan and Mason , 2017 ) Using the instantaneous position as an observable would reveal different aspects of the structure underlying an animal’s locomotion . Consider the trajectories of flies in Clusters 1 and 3 ( Figure 7 ) . They cross similar spatial positions , but are classified into different states because the flies travel at different speeds . In terms of the sequence of position in space , flies in both clusters have a similar behavior – they explore the outer arena border and make occasional radial forays inside the odor-zone . An analysis based on position would likely place these two clusters of flies together whereas our analysis , in the velocity space , placed them in different clusters . Model architecture is also important . A hierarchical model performed better than a non-hierarchical model . The current model has state durations of <3 s . It is clear to human observers that there is structure in the data that is >3 s long . Flies sometime explore the outer border of the arena using characteristic paths that can last up to a minute . The short duration of states in our model cannot capture structure on these long-time scales . One possibility is choosing a deeper-layered architecture . Given the structured transitions between the HL states in our model , it is likely that if we used a deeper-layered architecture , we would likely uncover structure on a longer timescale . During both HL states 1 and 2 , the fly’s locomotion is quite slow , but in state 2 the fly stops and runs intermittently while in state 1 , the fly is continuously in motion , albeit slowly . Similarly , in each of the HL states 4 , 5 and 7 , v^\|\| lies within a narrow range , which is distinct for each of these three states , implying a tight control over forward speed . These locomotor characteristics can persist over 3 seconds ( Figure 2—figure supplement 2 ) – a time period during which a fly takes 30 steps on average ( given a step frequency of 10 Hz; Mendes et al . , 2013 ) . This tight control over v^\|\| over many steps strongly suggests that locomotion unfolds in blocks . The HHMM presented here or the states revealed by it may not reflect the actual states employed by neurons in the brain . In fact , there is an ongoing debate whether behavior and its control is better represented as a continuum than by discrete states . The presence of long-lasting states that are employed repeatedly by all flies in our dataset implies that either locomotion does consists of transition between discrete states , or that these states represent fixed points or peaks of a dynamical system around which the animal spends most of its time ( Berman , 2018 ) . Another surprising result is that the same set of locomotor features describes the behavior of all the flies in the dataset . This result is particularly surprising given that our model explicitly allows each fly its own set of locomotor features . The fact that all flies can be reasonably modeled by the same model implies that within a given environment all flies construct their locomotion from the same building blocks , and differences in locomotion amongst flies or the effect of sensory stimulation can be quantified as changes in the frequency with which these building blocks are employed . An important avenue for future research is to assess whether these locomotor features are fixed or flexible . In nature , animals encounter odors in a cluttered and dynamic sensory environment ( Riffell et al . , 2014 ) . Discriminating between odors , navigating towards the chosen odor source and pinpointing the odor source requires a flexible deployment of multiple different motor programs . It is difficult to replicate the complex natural environment in the laboratory . Therefore , laboratory studies are typically aimed at different subsets of the complex environment experienced by animals . In insects , much research has focused on an environment in which it experiences odors in a highly structured odor plume often within a high-contrast visual environment ( Budick and Dickinson , 2006; Kennedy , 1983; Vickers , 2000; van Breugel and Dickinson , 2014; Álvarez-Salvado et al . , 2018 ) . Recently , similar experiments have been repeated for flies walking towards an odor source ( Bell and Wilson , 2016 ) . These experiments model an insect’s behavior under one specific condition wherein the fly tries to locate an odor source at a distance using strong directional information from wind and vision . The experiments described here explore a fly’s behavior in a small , dark circular arena . At most locations in the arena , the air speed was 0 . 07 m/s; the highest wind-speed was 0 . 11 m/s ( Jung et al . , 2015 ) , a value lower than has been employed in most studies . Therefore , non-olfactory directional cues from vision or wind are minimized ( Jung et al . , 2015 ) . Consistent with this idea , there was no change in the distribution of the flies when wind was completely eliminated . We find that this behavior near the odor source can be described by changes in the HL states . The clearest evidence that changes in HL states are a good description of the fly’s behavior is the analysis in which we measured the spatial distribution of odor-evoked changes in HL states ( Figure 6 ) , and observed a pattern that strongly resembles the odor-zone . This analysis shows that the HL state description is accurate enough to facilitate discovery of arena structure . However , because we averaged HL state distribution over the entire 3 min of odor exposure , the analysis misses some details . The flies first detect odor slightly outside the odor-zone as defined in this study ( Jung et al . , 2015 ) , and their behavior during the first 10 s after odor encounter differs from their behavior during the rest of the odor period ( Figure 6—figure supplement 1A ) . Moreover , at least some of the spatial structure results from the change in the radial density of the fly as a function of time ( Figure 6—figure supplement 1B ) . A fly’s interaction with odor is dynamic and that HL states are a good analytical framework to extract the spatiotemporal patterning of fly’s behavior by odor . This spatiotemporal pattern likely differs among flies; a full description of this pattern requires a larger dataset and represents an important avenue for future research . Even single-cell organisms and animals with simple nervous systems display substantial individuality ( Jordan et al . , 2013; Gallagher et al . , 2013 ) . Animals with larger nervous systems are likely to display even greater individuality , in the case of adult flies this individuality was demonstrated in the context of locomotor handedness ( Buchanan et al . , 2015 ) in a choice assay . The nature and extent of individuality is harder to assess in more complex behaviors because of the difficulty in assessing individuality in a large behavioral space; differences in behavior can simply be different instantiation of the same behavior , or reflect fundamental differences in behavior . In this study , we find that different flies employ the same locomotor features but use them in vastly different proportions . The observed variability between the flies is inconsistent with a single type of locomotor behavior but can be approximated by invoking 3–4 clusters of flies . Because the clustering framework we employed ( X-means ) underestimates the number of clusters , and because there were only 34 flies in our dataset , it remains to be seen whether there are a few locomotor-types or a whole continuum of locomotor-types . Another limitation of this study is that we have not yet ascertained whether a fly’s behavior persists over a longer time frame . Despite these limitations , this study makes two important contribution to the study of individuality . First , we develop a statistical framework to study complex behaviors . This method can be extended to examine whether the differences we observe truely represent individuality . Second , in many behavioral studies , researchers focus on the effect of some stimuli on behavior and make conclusions based on the average fly . In this study , we provide a framework for testing whether the description based on an average fly is appropriate , and ways to proceed if such an approach is inadequate . The diversity of odor responses observed here is consistent with work done on moths where ( Willis and Arbas , 1998 ) but in sharp contrast to recent work on walking Drosophila ( Bell and Wilson , 2016 ) in which the authors reported that attraction to odors results from a stereotypical motor pattern . The authors of that study claimed that the relatively simple response to odors in their study is likely a result of their simple behavioral arena . Another possibility is that in their study there is a strong , directional wind cue . In the presence of a steady wind cue in a narrow arena , it is likely that the flies’ locomotor behavior is dominated by upwind walking and suppresses other elements of their behavior . It is well-established that a fly’s response to odor is strongly influenced by context , as was demonstrated recently by comparing the response to odors in different visual and air flow conditions ( Saxena et al . , 2018 ) . When considered from the viewpoint of an individual animal , this variability is hard to understand: A hungry fly in search of food should respond with a singular , hardwired behavior which represents an optimal strategy for locating food . However , species evolve as large populations of individuals , and a successful species should be able to adapt to fluctuating environmental condition . Recent work has shown that - bet hedging - a process by which the same genotype shows considerable phenotypic variation is important for adaptation to fluctuating environments ( Kain et al . , 2015 ) . Having a diversity of phenotypes ensures that some individuals would thrive in any condition and behavioral variability is a feature not a bug and its careful consideration is critical . Studying individual behavioral responses is also important to understand the mechanism underlying both the control of locomotion and how odors control locomotion . Analyzing behavior at the population level can provide important insights into an animal’s response but does not provide the resolution necessary for understanding the neural mechanism underlying the moment-by-moment control of behavior at the level of individual flies . In this context , it is instructive to take a closer look at the average response to odors in the light of clusters of response to odors . The average response ( Figure 5A ) was surprising: the occupancy of all the slow states except state 2 is increased . The lack of increase in occupancy of state 2 results from a cluster of flies in which the occupancy of state 2 is strongly decreased ( Figure 9A ) . A similar effect is observed in the response of the average fly to the medium-speed states – states 4–7 . The occupancy of these states decreases in some flies and increases in others . Thus , the average fly is an aggregate of these different clusters of flies , each of which has a distinct response to odor . Disaggregation is an essential first step to understanding neural control of behavior . The methods used to collect the behavioral data were reported in a previous study ( Jung et al . , 2015 ) . Briefly , flies were raised in a sparse culture . Flies that were 3–5 days post-eclosion were starved for 14–18 hr . Locomotion of a single fly was recorded for a 3 min period before odor was introduced ( before period ) and a 3-min period during odor ( during period ) , using a video camera at a rate of 30 frames per second . The coordinates of the fly were extracted using a custom MATLAB program ( https://github . com/bhandawat/fly-walking-behavior/tree/master/tracking; Bhandawat , 2017; copy archived at https://github . com/elifesciences-publications/fly-walking-behavior ) . The behavioral arena was normalized to a unit circle centered at the origin . The raw coordinates of the centroid of the fly were smoothed using wavelet denoising followed by a locally weighted ( lowess ) filter . Speed and curvature were defined exactly as in the previous study . To quantify the behavior of the 34 flies , we computed the speed of the fly along the original direction of movement ( v^\|\| ) and the speed of the fly perpendicular to the original direction of movement ( v^⊥ ) . v^\|\| at time t was defined as the component of the velocity at time t in the direction of velocity of the fly at time t−1 . v^⊥ was defined as the component of the velocity at time t perpendicular to velocity of the fly at time t−1 ( Figure 1C ) . These values were calculated as follows:v^⊥ ( t ) =dyt−1dxt−dytdxt−1dxt−12+dyt−12v^\|\| ( t ) = dxtdxt−1+dytdyt−1dxt−12+dyt−12 Values of v^\|\| and v^⊥ found to be further than 4 standard deviations away from the average were set to values drawn from a normally distributed distribution ( σ=1 ) centered at the 4 standard deviation mark . The resulting v^\|\| and v^⊥ were then set as the observables used in fitting a 2-level Hierarchical Hidden Markov model ( HHMM ) . We employed v^\|\| and v^⊥ instead of speed and curvature because curvature is very noisy at low speeds because the calculation of curvature requires division by the third-power of speed . v^\|\| and v^⊥ are directly related to speed and curvature as followsSpeed=v^⊥2+v^\|\|2Curvature= v^⊥dxt−12+dyt−12 HMMs are widely used in a variety of fields for modeling time series data . An HMM is a Markov model which assumes that a given sequence of observations may be explained by a set of states that are not observed ( or hidden states ) , and the time independent probability of transitioning between these states . The model processes which produce the observations in an HMM are hidden to the researcher and thus , the goal of fitting an HMM is to uncover the highest likelihood probability model parameters that can generate the data . Baum and others developed the core theory of HMMs ( Baum and Petrie , 1966 ) . Since then there has been much exploration of model architecture , and fitting procedure . HMMs have been shown to be effective in modeling behavior because instantaneous measures of an observable are variable; therefore , behavioral states inferred by the application of simple thresholding to instantaneous measures of the observables are likely to be erroneous . HMM remedies this problem by inferring states based not only on the value of the observable at the current time point but also on the previous and following time points , and allows a more accurate determination of state ( this idea is well-explained in Figure 2 in ref 17 ) . Specifically , the assumption of Markov dynamics with a sparse prior on state transitions penalizes the consideration of unlikely state transitions based upon recent history ( forward filtering ) and future destinations ( backward smoothing ) . The HHMM is an extension to the HMM which applies hierarchical structure in the form that higher level state is itself an HMM composed of its lower level states ( Fine et al . , 1998 ) . The approach we take in exploring and fitting the HHMM closely follows the approach developed by Matt Beal in which he applied variational algorithms to fit HHMM to a time series of observables ( Beal , 2003 ) . This section is divided into three parts . First is the description of the model , second is the details of the process by which the model is fit , and third is the thought process behind our model selection . Time points were sampled using four methods based on bin duration for each fly for the two scenarios with the most data ( Bo and Di ) separately ( Figure 7—figure supplement 4 ) . In method one , for a given scenario , segments of continuous repeated HL states were randomly sampled and stitched together until the duration of the time bin was fulfilled . In the second method , a window lasting the time bin was sampled from the HL states for the scenario . In method three , the segments were sampled starting with the first time point of the scenarios . In method four , the segments were sampled starting with the last time point of the scenarios and moving backwards in time . After sampling , the average HL state distributions were calculated for each time bin and the Euclidean distance from the distribution to the centroid of each X-means cluster for the given scenario were calculated . The closest cluster was compared to the X-means cluster assignment based on all time points . This process was repeated 100 times to generate a mean percentage of correctly labeled flies based on the subsampling duration . Chance was calculated as the probability of choosing a fly for a given cluster and being in the Voronoi cell of the cluster . This translates to:E ( X ) =∑i=1Kxipiwhere xi is the probability of observing cluster i based on the number of flies in each cluster , pi is a weighting based on the size of Voronoi cell in the simplex space , and K is the total number of clusters .	Many behaviors that we perform everyday , including something as familiar as making a peanut-butter sandwich , consist of a sequence of recognizable acts . These acts may include , for example , holding a knife and opening a jar . Yet often neither the sequence nor the individual acts are always performed in the exact same way . For example , there are many ways to hold a knife and there are many ways to open a jar , meaning neither of these actions could be called “stereotyped” . A lack of stereotypy makes it difficult for a computer to automatically recognize the individual acts in a sequence . This same problem would apply to other common behaviors , such as walking around somewhere you have not visited before . While we easily recognize it when we see it , walking is not a stereotyped behavior . It consists of a series of movements that differ between individuals , and even in the same individual at different times . So how can someone automatically recognize the individual acts in a non-stereotyped behavior like walking ? To begin to find out , Tao et al . developed a mathematical model that can recognize the walking behavior of a fruit fly . Existing recordings of fruit flies walking were analyzed using a type of mathematical model called a Hierarchical Hidden Markov Model ( often shortened to HHMM ) . Such models assume that there are hidden states that influence the behaviors we can see . For example , someone’s chances of going skiing ( an observable behavior ) depend on whether or not it is winter ( a hidden state ) . The HHMM revealed that the seemingly random wanderings of a fly consist of ten types of movement . These include the “meander” , the “stop-and-walk” , as well as right turns and left turns . Exposing the flies to a pleasant odor – in this case , apple cider vinegar – altered how the flies walked by changing the time they spent performing each of the different types of movement . All flies in the dataset used the same ten movements , but in different proportions . This means that each fly showed an individual pattern of movement . In fact , the differences between flies are so great that Tao et al . argue that there is no such thing as an average walk for a fruit fly . The model represents a complete description of how fruit flies walk . It thus provides clues to the processes that transform an animal’s sensory experiences into behavior . But it also has potential clinical applications . Similar models for human behaviors could help reveal behaviors that are abnormal because of disease . Normal behaviors also show variability , and some diseases increase or decrease this variability . By making it easier to detect these changes , mathematical models could support earlier diagnosis of medical conditions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2019	Statistical structure of locomotion and its modulation by odors
Neurophysiological studies in humans and nonhuman primates have revealed movement representations in both the contralateral and ipsilateral hemispheres . Inspired by clinical observations , we ask if this bilateral representation differs for the left and right hemispheres . Electrocorticography was recorded in human participants during an instructed-delay reaching task , with movements produced with either the contralateral or ipsilateral arm . Using a cross-validated kinematic encoding model , we found stronger bilateral encoding in the left hemisphere , an effect that was present during preparation and was amplified during execution . Consistent with this asymmetry , we also observed better across-arm generalization in the left hemisphere , indicating similar neural representations for right and left arm movements . Notably , these left hemisphere electrodes were centered over premotor and parietal regions . The more extensive bilateral encoding in the left hemisphere adds a new perspective to the pervasive neuropsychological finding that the left hemisphere plays a dominant role in praxis . A primary tenet of neurology is the contralateral organization of movement . The vast majority of the fibers from the corticospinal tract cross to the opposite side of the body ( Nyberg-Hansen and Rinvik , 1963 ) and functionally , hemiparesis resulting from cortical stroke is manifest on the contralateral side of the body ( Bourbonnais and Vanden Noven , 1989 ) . Although direct control of arm movements is primarily mediated through contralateral projections , unimanual arm movements elicit bilateral activity in the primary motor cortex ( M1 , Babiloni et al . , 1999; Ghacibeh et al . , 2007 ) , indicating that neural activity in the ipsilateral hemisphere contains information relevant to ongoing movement . Correspondingly , kinematic and movement parameters of the ipsilateral limb can be decoded from ipsilateral hemisphere intracortical recordings in monkeys ( Ganguly et al . , 2009; Ames and Churchland , 2019 ) and from electrocorticography ( ECoG ) in humans ( Bundy et al . , 2018; Ganguly et al . , 2009; Wisneski et al . , 2008 ) . Ipsilateral signals represent an intriguing source of neural activity , both for understanding how activity across the two hemispheres results in coordinated movement and because this information might be exploited for rehabilitative purposes . While it is established that information about unimanual movements is contained within the ipsilateral hemisphere , there remains considerable debate about what this signal represents . Previous studies have centered on the question of whether ipsilateral representations overlap or are independent of contralateral representations , leading to mixed results . Consistent with the overlap hypothesis , neural activity for the contralateral and ipsilateral limb movements shows several similarities , including shared target tuning preferences and the ability to cross predict kinematic features from a model trained on the opposite arm ( Bundy et al . , 2018; Cisek et al . , 2003; Steinberg et al . , 2002; Willett et al . , 2020 ) . Consistent with the independence hypothesis , intracortical recordings in monkeys have revealed that the lower dimensional representations of the two arms lie in orthogonal subspaces ( Ames and Churchland , 2019; Heming et al . , 2019 ) . These hypotheses are not mutually exclusive . For example , the degree of overlap or independence may depend on the gesture type ( e . g . , overlapping representations for grasping but not arm movement , Downey et al . , 2020 ) , or brain region ( e . g . , premotor cortex displays stronger preservation of tuning preferences across the two arms than primary motor cortex , Cisek et al . , 2003 ) . One factor that has received little attention in this literature is the recording hemisphere . This is surprising given the marked asymmetries between the two hemispheres in terms of praxis ( Corballis et al . , 2012; Rothi et al . , 1997 ) . Tracing back to the early 20th century , marked hemispheric asymmetries have been defined by the behavioral deficits observed following unilateral brain injury ( Schaefer et al . , 2007; Liepmann , 1908 , cited in De Renzi and Lucchelli , 1988 ) . Apraxia , an impairment in the production of coordinated , meaningful movement in the absence of muscle recruitment deficits , is much more common after left compared to right hemisphere insult ( Haaland et al . , 2000; De Renzi and Lucchelli , 1988 ) . Moreover , left hemisphere stroke will frequently result in apraxic symptoms for gestures produced with either hand , as well as impairments in action comprehension ( De Renzi and Lucchelli , 1988 ) . Hemispheric asymmetries are also evident in neuroimaging activation patterns in healthy participants , with the left hemisphere having stronger activation during ipsilateral movement than the right hemisphere , especially with increasing task difficulty ( Chettouf et al . , 2020; Verstynen et al . , 2005; Verstynen and Ivry , 2011; Schäfer et al . , 2012 ) . These patterns raise the possibility that the ipsilateral cortical representation differs between the left and right hemispheres . In the present study , we use intracranial recordings from the cortical surface ( ECoG ) to examine the degree of cortical overlap for ipsilateral and contralateral upper limb movement in the left and right hemispheres . The data were collected from six patients , three with left hemisphere implants and three with right hemisphere implants , while they engaged in an instructed-delay reaching task . We focus on predicting the temporal dynamics of high-frequency activity ( HFA; 70–200 Hz ) , a surrogate for infragranular single-unit activity and supragranular dendritic potentials ( Leszczyński et al . , 2020 ) , which tracks local activation of the cortex ( Muthukumaraswamy , 2010 ) . Going beyond previous studies that use decoding models which combine multiple neural features from multiple electrodes to predict kinematics , we employed an encoding model which uses kinematic features to predict neural activity for each electrode , allowing us to retain the high spatial and temporal resolution of the ECoG signal . This approach allows us to create high-resolution topographic maps depicting encoding strength on the surface of the cortex for movements produced with the contralateral and ipsilateral arms . This is preferable to projecting the weights obtained from decoding models since these models have difficulty disambiguating between informative and uninformative electrodes ( Kriegeskorte and Douglas , 2019 ) . Moreover , our approach provides a way to map kinematics to neural activity in a time-resolved manner ( rather than as single weights ) , allowing us to identify time ranges of representational overlap and divergence across the two arms for each electrode . We used a delayed response , out-and-back reaching task . On each trial , a cue indicating the target location was presented on a touchscreen followed , after a short delay , by an imperative signal . Participants were instructed to prepare to move to the target during the delay period . The participant was free to move at their own pace , with the instructions emphasizing that the participant should focus on touching the screen near the target and then returning to the start position . Left and right arm reaches were tested in separate blocks , with the position of the nonresponding hand fixed throughout the block . Table 1 summarizes the total number of successful trials , along with the reaction time and movement time data . A trial was considered unsuccessful if the reach was initiated before the go cue or if contact with the touchscreen was outside the boundary of the target . The percentage of unsuccessful trials was low , ranging between 0% and 12 . 5% across individuals . The movements had roughly , bell-shaped velocity profiles for the outbound and the inbound segments ( Figure 1C , E ) and the outbound reaches were , on average , faster than inbound reaches . The marked interindividual differences in reaction time and movement time reflect the fact that the instructions emphasized accuracy and smoothness . At a more fine-grained level of spatial accuracy , we calculated the distance from the center of each target to the touch location for each trial . On average , the mean distance from the center of the 2 . 5 cm circle was 0 . 80 cm ( SD = 0 . 10 cm ) for right-handed reaches and 0 . 90 cm ( SD = 0 . 17 cm ) for left-handed reaches ( Figure 1D ) . These values did not differ from one another [t = 1 . 538 , p = 0 . 22] . We examined the extent to which movement kinematics were encoded for contralateral and ipsilateral reaches in individual electrodes . To do this we fit a kinematic encoding model that maps continuous kinematic features to the HFA signal ( Figure 1E ) for the 665 electrodes meeting our inclusion criteria . This procedure was done separately for contralateral and ipsilateral reaches . We quantified the cross-validated model fit by generating HFA predictions using the kinematic features from held-out trials of the same condition and calculating prediction performance as the square of the linear correlation ( R2 ) between the predicted and actual HFA signal ( Figure 2B ) . Figure 2A displays R2 values for each electrode for the contralateral and ipsilateral conditions , presented on the individual patient MRIs . Electrodes with high prediction performance were primarily located in arm areas of sensorimotor cortex . In line with previous research ( Downey et al . , 2020 ) , a sizeable percentage of the electrodes were able to predict the HFA at or above our criterion of R2 > 0 . 05 ( examples shown in Figure 2B ) . This degree of prediction was observed not only when the data were restricted to contralateral movement ( 31% of electrodes ) , but also when the data were from ipsilateral movement ( 25% ) . A number of electrodes ( 24% ) were predictive in both the contralateral and ipsilateral models . Electrodes that did not meet this criterion for either arm are represented as small dots in Figure 2A and were excluded from further analysis , leaving a total of 216 predictive electrodes ( 32% , 141 = left hemisphere , 75 = right hemisphere ) . Before comparing prediction performance across the two hemispheres , we first evaluated the distribution of the electrodes in the right and left hemispheres . Electrode placement for ECoG is determined solely for clinical purposes; as such , hemispheric asymmetries could arise from differences in coverage rather than functional differences . To evaluate coverage , we categorized the position of all electrodes based on a cortical parcellation ( Desikan et al . , 2006; Figure 2—figure supplement 1 ) . When the categorization data were pooled across the three left hemisphere and the three right hemisphere patients , the proportion of electrodes was similar across the eight parcellations that encompass premotor , sensorimotor motor , and parietal regions [averages: premotorright = 11% , premotorleft = 10% , sensorimotorright = 17% , sensorimotorleft = 16% , parietalright = 5% , parietalleft = 3%; χ2 ( 7 ) = 0 . 057 , p = 0 . 99; Figure 2—figure supplement 1] . Given the similar distributions , we next asked whether prediction performance for the two arms differed across the two hemispheres . Figure 2C compares the predictive performance of each electrode for the contralateral and ipsilateral conditions for patients with left or right hemisphere grids . Values close to the unity line yield similar predictions for the two conditions; values off the unity line indicate that encoding is stronger for one arm compared to the other . To compare prediction performance at the group level , we fit a permutation-based mixed-effects model with fixed factors of Arm and Hemisphere and a random factor of Participant . We found a main effect of Arm with contralateral reaches being encoded more strongly than ipsilateral reaches [ χ2 ( 1 ) = 29 . 34 , p < 0 . 001] . We found no effect of Hemisphere [ χ2 ( 1 ) = 0 . 46 , p > 0 . 50] , but we found a significant interaction between Arm and Hemisphere [ χ2 ( 1 ) = 12 . 03 , p < 0 . 001] . To further explore this interaction , we calculated the difference between the R2 values for the contralateral and ipsilateral conditions for each electrode , using this as a proxy of an encoding bias between the two arms ( Figure 2C , upper right corner of each scatterplot ) . Values close to zero indicate similar encoding across the two arms , whereas positive values correspond to stronger contralateral encoding and negative values stronger ipsilateral encoding . The distribution for each condition was positively skewed indicating that , overall , there was a bias for better encoding for contralateral reaches [permutation test: pright < 0 . 001 , pleft < 0 . 001] . However , there was a significant difference in the distributions for the two hemispheres: The bias scores were lower in the left hemisphere compared to the right hemisphere [permutation test: p < 0 . 001] indicating stronger bilateral encoding in the left hemisphere . We also found that the contralateral bias becomes weaker the further the electrodes are from the primary motor cortex , an effect observed in both hemispheres [rleft = −0 . 48 , pleft < 0 . 001 , rright = −0 . 45 , pright < 0 . 001; Figure 2—figure supplement 3] . As neural activity unfolds from preparation to movement , the underlying computations may change substantially ( Elsayed et al . , 2016 ) . To examine if hemispheric asymmetries in encoding depend on task state , we repeated the mixed-effects model described in the previous section , but now added a factor Task Phase , separating the data to test the held-out predictions during the instruction and movement phases ( Figure 3A ) . The effect of Arm was again significant , with contralateral reaches more strongly encoded than ipsilateral reaches [χ2 ( 1 ) = 16 . 19 , p < 0 . 001] . The main effects of Hemisphere [ χ2 ( 1 ) = 0 . 72 , p > 0 . 40] and Task Phase were not significant [χ2 ( 1 ) = 0 . 01 , p > 0 . 90] . Importantly , there was a three-way interaction between Arm , Hemisphere , and Task Phase indicating that the level of encoding for the two arms varied across the two hemispheres for the two task phases [χ2 ( 4 ) = 22 . 47 , p < 0 . 001] . To explore this interaction we again examined the distribution of difference scores for each electrode . The right hemisphere electrodes show a positive skew in both the planning and movement phase [permutation test: pright_move < 0 . 001 , pright_planning < 0 . 001] . However , this pattern is only seen in the left hemisphere during the planning phase [pleft_planning < 0 . 001]; the mean difference score was not statistically different from zero for the left hemisphere electrodes in the movement phase [pleft_move = 0 . 482] . Analyzing simple effects within each hemisphere , we found that the difference score was smaller ( i . e . , more bilateral encoding ) in the left hemisphere during the movement phase compared to the planning phase [R2left_move = 0 . 01 , R2left_planning = 0 . 04 , p < 0 . 001] . In contrast , the opposite pattern was observed in the right hemisphere , with encoding being more bilateral during the instruction phase [R2right_move = 0 . 13 , R2right_planning = 0 . 10 , p < 0 . 001] . Taken together , these results suggest that the left and right hemispheres may have different roles in bilateral encoding with regard to task phase . The preceding analyses focused on an encoding analysis for within-arm prediction . We next evaluate the overlap between the neural representations for contralateral and ipsilateral movements . To this end , we examined across-arm prediction performance by training the kinematic encoding model with the data from movements produced with one arm and testing prediction performance using the data from movements produced with the other arm . Figure 4A shows the traces for two representative electrodes , one that shows good generalization across the two arms and the other that shows poor generalization . For the electrode that shows good generalization ( E1 ) , prediction performance for held-out contralateral reaches is comparable when the model is trained on data from either the contralateral or ipsilateral arm . This suggests that there is overlap between the neural representations for reaches performed with either upper limb for this electrode . In contrast , the electrode showing poor generalization ( E2 ) showed good prediction for contralateral reaches when trained with contralateral data , but poor prediction when trained with ipsilateral data . Here , the neural representations for the arms do not overlap . Note that E2 showed relatively strong within-arm ipsilateral encoding ( R2 = 0 . 25 ) ; thus , the inability of this electrode to generalize across arms is not a result of poor encoding of the ipsilateral arm . Rather , E2 encodes movement produced by either arm , but the manner in which they are encoded differs . Figure 4B summarizes the comparison of within-arm prediction ( y-axis ) against across-arm prediction ( x-axis ) , with the data separated for the instruction and movement phases . In this depiction , electrodes close to the unity line have overlapping neural representations during contralateral and ipsilateral movements , whereas electrodes off the unity line encode the two arms differentially . We again used a permutation-based mixed-effects model , this time with fixed factors of Generalization , Task Phase , and Hemisphere and a random factor of Participant . As expected , we found a main effect of Generalization with within-arm predictions outperforming across-arm predictions [χ2 ( 1 ) = 103 , p < 0 . 001] . There was also a main effect of Task Phase [χ2 ( 1 ) = 30 . 51 , p < 0 . 001] with better prediction accuracy during the movement phase compared to the instruction phase . The main effect of Hemisphere was not significant [χ2 ( 1 ) < 0 . 01 , p > 0 . 90] , but there was a significant three-way interaction between the three factors [ χ2 ( 4 ) = 19 . 56 , p < 0 . 001] . To evaluate this interaction , we used a difference score , calculated by subtracting the across-arm R2 from the within-arm R2 for each electrode ( Figure 4B , upper right corner of each scatterplot ) . As such , the larger ( more positive ) the score , the poorer the electrode is in predicting activity when the movement is produced with the other arm . Overall , across-arm generalization was better during the instruction phase compared to the movement phase [R2instruction = 0 . 05 , R2movement = 0 . 08; p < 0 . 001] . In terms of laterality differences , the left hemisphere showed stronger across-arm generalization ( lower difference values ) than the right hemisphere [R2left = 0 . 04 , R2right = 0 . 11; p < 0 . 001] . Analyzing simple effects within each task phase , the left hemisphere had better across-arm generalization compared to the right hemisphere for both the instruction and movement phase [simple effect analysis: pinstruct < 0 . 001 , pmove < 0 . 001] . In addition , better across-arm generalization was found during the instruction phase in both the left and right hemispheres [simple effect analysis: pleft < 0 . 001 , pright < 0 . 001] . In sum , the three-way interaction reflects the fact that electrodes in the left hemisphere exhibit better generalization across arms than right hemisphere electrodes , an effect observed in both the instruction and movement phases . In addition , the difference in across-arm generalization between phases is more pronounced in the right hemisphere . Considered together , the within- and across-arm encoding results indicate that left hemisphere electrodes show stronger bilateral encoding and more similar encoding during left and right arm movements . Interestingly , although overall encoding was stronger during movement , across-arm generalization was stronger during the instruction phase . To examine how generalization varied across the cortex , we categorized each electrode as showing either good across-arm generalization ( decrease of up to 20% relative to within-arm performance ) or poor across-arm generalization ( decrease of more than 50%; Figure 5A ) . We focused on the extremes of the generalization distribution based on the assumption that these electrodes were more likely to share similar underlying neural profiles . This also allowed us to have similar numbers of electrodes in each group . Figure 5B plots ipsilateral encoding strength ( R2 ) for electrodes that either generalize well or generalize poorly . Electrodes that generalize poorly have lower ipsilateral encoding strength than electrodes that generalize well [t = 5 . 921 , p < 0 . 001] . Despite this difference , 61% ( 23/38 ) of electrodes that generalize poorly meet our encoding criteria of R2 > 0 . 05 . To examine the dynamics of representational overlap and divergence , we averaged the time-resolved HFA amplitude across electrodes , restricted to those showing good or poor across-arm generalization . Figure 5C displays the average time series for contralateral ( solid line ) and ipsilateral ( dashed line ) predictions for electrodes that generalize well ( white ) or poorly ( magenta ) . The temporal profile of HFA is similar for electrodes that generalize well , showing a single peak in the movement phase . A cluster-based permutation test identified two periods where the HFA amplitude differed for contralateral and ipsilateral reaches , one during instruction and one well into the movement period . In contrast , the temporal profiles are radically different for those that generalize poorly . The lack of correspondence likely arises , at least in part , because of the weak modulation for these electrodes during ipsilateral reaches . It is possible that similarity in temporal structure is obscured in the preceding analysis by the differences in HFA amplitude for the electrodes that showed poor generalization . To control for this , we standardized the time series data by dividing each sample by the overall standard deviation ( insets: Figure 5C ) . Using the standardized traces , we calculated the linear correlation coefficient between the contralateral and ipsilateral traces , separately for instruction and movement . As expected , electrodes that generalized well across arms showed strong across-arm correlations for both task phases ( inset: Figure 5C , left ) . In contrast , for electrodes that generalize poorly across arms , the correlation between arms was negative during instruction and then rose to a moderate positive correlation during movement ( inset: Figure 5C , right ) . Thus , the poor generalization of these electrodes is due in part to the temporal divergence of the two arms during instruction , where the ipsilateral trace becomes inhibited compared to the contralateral trace . Interestingly , although the ipsilateral trace remains inhibited during movement , the temporal structure between the two arms reemerges . We next examined the relationship between generalization , hemisphere and spatial position on the cortical surface . As can be seen in Figure 5D , electrodes that generalize well were predominantly found in the left hemisphere ( proportion of electrodes that generalize across patients: L1 = 22% , L2 = 39% , L3 = 29% , R1 = 0% , R2 = 5% , R3 = 4%; χ2 ( 2 ) = 48813 , p < 0 . 001 ) . In contrast , electrodes showing poor generalization were observed in both hemispheres ( proportion of electrodes that generalize poorly: L1 = 21% , L2 = 5% , L3 = 6% , R1 = 14% , R2 = 15% , R3 = 37%; χ2 ( 2 ) = 0 . 400 , p = 0 . 818 ) . Moreover , in both hemispheres , electrodes showing poor generalization were clustered near the dorsal portion of the central sulcus , a region corresponding to the arm area of motor cortex . Electrodes showing strong generalization ( mostly limited to the left hemisphere ) tended to be in dorsal and ventral premotor cortices ( PMd and PMv ) , along with a few in superior parietal cortex . This pattern was also observed when we analyzed all electrodes , rather than restrict the analysis to those showing extreme values . Here we used a continuous measure , correlating the amount of across-arm generalization with the distance ( absolute value ) from the dorsal aspect of the central sulcus . The correlation was significant in the left hemisphere [rleft = 0 . 46 , pleft < 0 . 001] but did not reach significance in the right hemisphere , although the trend was in the same direction [rright = 0 . 22 , pright = 0 . 068; Figure 5—figure supplement 1] . To examine the extent of target modulation for the contralateral and ipsilateral arms , we calculated the modulation depth ( MD ) of each electrode during the instruction and movement phases . The modulation index reflects the amount of variability in the signal captured by target tuning ( or target specificity ) : A modulation index of 0 . 1 means 10% of the variance is captured by the difference between the response to the four target locations . The modulation values overall were relatively low ( Figure 5D ) . However , it should be noted that the reaches were all within the frontoparallel plane which comprise a considerably smaller range of movement compared to studies that use a center-out reaching task . For both electrode types ( showing good or poor across-arm generalization ) , there was a main effect of arm , with ipsilateral modulation lower than contralateral modulation [permutation test: pGeneralize_well < 0 . 001; pGeneralize_poorly < 0 . 005] . Both subgroups of electrodes also displayed a main effect of task phase , with the depth of modulation greater during the movement phase compared to the instruction phase [permutation test: pGeneralize_well < 0 . 001; pGeneralize_poorly < 0 . 005] . No significant interactions were found for either group . We also examined the representational overlap between the two arms in terms of their tuning profiles . We computed a tuning similarity index ( SI ) , defined as the sum of squared errors ( SSEs ) for average HFA predictions to the same target between the contralateral and ipsilateral arms . An SI of 1 would correspond to identical tuning preferences for the arms whereas an SI of 0 would indicate completely disparate tuning preferences . The similarity data were analyzed with a mixed design permutation test , including the factors task phase and electrode type ( good vs . poor generalizers ) . Electrodes that generalize well across the two arms ( predominately found in the left hemisphere ) showed more overlap of tuning preferences compared to electrodes that generalized poorly [main effect of generalizability: p < 0 . 001] . While there was no effect of phase [p > 0 . 70] , the interaction was significant [p < 0 . 005] , with electrode types showing more comparable tuning similarity during instruction and tuning similarity diverging during movement . Simple effects analysis revealed that for electrodes that generalize poorly , tuning similarity was higher during the instruction phase compared to the movement phase [permutation test: p < 0 . 001] . In contrast , for electrodes that generalize well , tuning similarity was higher during movement compared to instruction [permutation test: p < 0 . 001] . These analyses demonstrate that a number of electrodes in the left hemisphere encode kinematic variables for both arms , including similar tuning preferences across the two arms , which was especially pronounced during the movement phase . In the final analysis , we examined how the kinematic features of the movements contribute to the encoding model used to predict HFA . Each of the four kinematic features includes 400 time lags and thus 400 weights that contribute to the model . To obtain a metric of the relative contribution of the features , we calculated the total contribution of each feature and normalized these values by dividing by the total contribution of the four features . The calculation was done for each patient separately and then averaged , with error bars representing the standard deviation across patients ( Figure 6A ) . The relative contribution of the four kinematic features was similar for contralateral and ipsilateral reaches . We next examined the temporal profile of the weights ( Figure 6B ) and found that this was also similar for the two conditions , although the average weights for ipsilateral reaches are substantially lower , consistent with the observation of lower performance metrics for ipsilateral reaches across all predictive electrodes ( Figure 2—figure supplement 2 ) . As can be seen in Figure 6A , speed and position , kinematic features which are associated with timing and movement initiation make a strong contribution to the encoding model ( relative contribution: contra = 68% , ipsi = 63% ) . This is in contrast with the smaller contribution of theta and phi , features which provide information about movement direction ( relative contribution: contra = 32% , ipsi = 37% ) . This result is similar to that observed in single-unit and population activity recorded in premotor and motor cortex of nonhuman primates . Kaufman et al . , 2016 observed that the largest response component was associated with movement timing/initiation rather than features such as movement direction . Similarly , this direction-independent signal occurs twice during sequential movements ( Zimnik and Churchland , 2021 ) ; in our data , speed has two prominent peaks , one occurring before the reach and the second occurring before the return movement . We were surprised to see the markedly differential weighting for the vertical ( theta ) and horizontal ( phi ) directional features . We assume this is likely idiosyncratic to the layout of our targets . We also examined the correspondence between HFA and the kinematic features as a function of whether electrodes generalize well or poorly ( Figure 7 ) . For electrodes that generalize well , position most closely corresponds to HFA for both contralateral and ipsilateral movements . The maximum cross-correlation for contralateral and ipsilateral movements was found at a lag of 200 and 150 ms , respectively , with HFA leading hand position . For electrodes that generalize poorly , the kinematic feature that most closely corresponds to HFA for both contralateral and ipsilateral movements is speed . For these electrodes , the maximum cross-correlation for contralateral and ipsilateral movements were both at a lag of 200 ms , with HFA again leading the kinematic feature . Although the ipsilateral HFA signals are considerably lower in amplitude , the pattern between HFA and speed is quite similar for both ipsilateral and contralateral movements . The fact that the neural activity from electrodes that generalize poorly ( primarily located over M1 ) correlates well with speed provides additional evidence that a strong component of the HFA ECoG signal is related to timing and movement initiation ( Kaufman et al . , 2016 ) . We observed a striking asymmetry between the two hemispheres for ipsilateral movement encoding . While contralateral movements were encoded similarly across the two hemispheres , ipsilateral encoding was much stronger in the left hemisphere , an effect that was especially pronounced during movement execution . In addition , there was greater overlap between the representation of contralateral and ipsilateral movement in the left hemisphere compared to the right hemisphere . The bilateral encoding effect size is quite substantial ( Cohen’s d = 1 . 34 ) , exceeding the conventional criterion for a large effect ( d = 0 . 80 ) . Given the size of the hemispheric asymmetry effect , it is surprising that this asymmetry has not been described in previous reports . This may in part reflect the smaller sample size in ECoG studies . For example , in Bundy et al . , 2018 , three of the four patients had left hemisphere grids , leaving a hemisphere analysis dependent on the data from a single right hemisphere patient . Studies with nonhuman primates tend to ignore hemispheric differences , perhaps because these animals do not show consistent patterns of hand dominance across individuals . One exception here is a study by Cisek et al . , 2003 who reported no hemispheric differences in neural recordings obtained from M1 and PMd during ipsilateral and contralateral arm reaches . In addition to examining hemispheric differences in the encoding of unimanual movement , we also asked if kinematic features were encoded differently for contralateral and ipsilateral movements by testing across-arm generalization . We categorized electrodes as showing either good across-arm generalization ( decrease of up to 20% relative to within-arm performance ) or poor across-arm generalization ( decrease of more than 50% ) . This categorization scheme revealed a striking anatomical division , with electrodes showing good across-arm generalization clustering in the left premotor and superior parietal regions and electrodes that generalized poorly clustering in left and right M1 . This result does not appear to reflect a sampling bias . There are more left hemisphere electrodes compared to right hemisphere electrodes because two of the left hemisphere patients had high density grid implants . However , the electrodes in our left and right hemisphere samples were similarly distributed over premotor , sensorimotor , and parietal regions ( see Figure 2—figure supplement 1 ) . Examining the temporal profile of electrodes that generalize poorly , we found that the divergence between the two arms occurs during instruction , with the ipsilateral trace becoming inhibited relative to the contralateral trace . TMS studies have shown reduced corticospinal excitability of the nonselected hand during movement preparation ( e . g . , Leocani et al . , 2000; Liepert et al . , 2001 ) , an effect that has been hypothesized to reflect inhibition from prefrontal regions to facilitate response selection ( Duque and Ivry , 2009 ) . We further examined the spatial tuning of the electrodes . Target tuning in the HFA band was found for both contralateral and ipsilateral movements , although ipsilateral tuning was significantly shallower . Interestingly , electrodes that generalized well across arms had similar spatial/target tuning for each arm . This suggests that for these electrodes , ipsilateral signals are not just encoding generic movement , but encoding movement direction in a similar manner to contralateral signals . A similar overlap in tuning has been observed in single-unit recordings from PMd ( Cisek et al . , 2003 ) and can be inferred from the across-arm generalization decoding results reported by Bundy et al . , 2018 . In contrast , electrodes that failed to generalize , located primarily in M1 in either left or right hemisphere , exhibited disparate tuning for contralateral and ipsilateral reaches . By using a delayed response task , we were able to segregate activity into an instruction phase during which the participant was presented with the target location for the forthcoming movement and a movement phase , defined at the onset of the reach . With this design , we found that the encoding model could predict neural activity during the instruction phase based on the kinematics of the forthcoming reach , evidence that the participants were indeed planning the upcoming movement . This task phase analysis also revealed robust asymmetries between the two hemispheres . There was a main effect of hemisphere , with the left hemisphere displaying stronger bilateral encoding overall compared to the right hemisphere . However , there was also an interaction: In the left hemisphere bilateral encoding was stronger during the movement phase whereas in the right hemisphere bilateral encoding was stronger during the instruction phase . Surprisingly , in the left hemisphere the contralateral bias completely disappeared during the movement phase , with both the contralateral and ipsilateral arms being encoded to the same extent . Stronger bilateral encoding during movement ( compared to instruction ) is surprising given the spatial distribution of electrodes that encode ipsilateral movement was primarily outside of M1 , regions typically associated more with planning than execution ( e . g . , premotor cortices and parietal cortex ) . The asymmetry observed here is in accord with the long-standing recognition of hemispheric asymmetries in praxis . Starting with the classic observations of Liepmann at the turn of the 20th century on the association of the left hemisphere and apraxia ( Liepmann , 1908 , cited in De Renzi and Lucchelli , 1988; see also Schaefer et al . , 2007 ) and continuing with functional imaging studies in neurotypical populations , a large body of evidence points to a dominant role for the left hemisphere in skilled movement ( Corballis et al . , 2012; Przybylski and Króliczak , 2017 ) . This asymmetry is most pronounced in tasks involving functional object use ( Buxbaum et al . , 2006 ) , symbolic gestures ( Xu et al . , 2009 ) , and intransitive pantomimes ( Bohlhalter et al . , 2009 ) . While the neuropsychological and neuroimaging work have highlighted the involvement of left premotor and parietal cortex in praxis , corresponding asymmetries have also been noted in subcortical structures such as the basal ganglia , cerebellum , and thalamus ( Swinnen et al . , 2010 ) . We note that for our patient population , we are limited to regions of the brain where we have sufficient electrode coverage . Apraxia , following left hemisphere damage can be manifest in movements produced with either limb ( De Renzi and Lucchelli , 1988 ) , and are usually associated with lesions that encompass premotor and parietal cortices ( Haaland et al . , 2000 ) . While this asymmetry may be linked to hand dominance ( Ochipa et al . , 1989 ) , functional imagining studies with relatively large sample sizes have shown that handedness only influences the strength of the left hemisphere bias for skilled movement but does not produce a reversal in left handers ( Vingerhoets et al . , 2012; Verstynen et al . , 2005; Chettouf et al . , 2020; Vingerhoets et al . , 2013 ) . Of the six patients tested in the current study , five are right handed and the remaining patient reported being ambidextrous with a slight preference for using the left hand . We note that the results from this patient ( L3 ) did not qualitatively differ from the other two left hemisphere patients . Ipsilateral encoding was most prominent in the premotor and parietal cortex of the left hemisphere , overlapping with the neural regions implicated in praxis . However , two features of our results do not map on readily to an interpretation that focuses on hemispheric asymmetries in praxis . First , our task involved simple reaching movements , whereas praxis generally encompasses more complex learned movements associated with tool use or symbolic gestures . Second , ipsilateral encoding became more pronounced during movement execution . Assuming that ipsilateral encoding is indicative of bilateral motor representations , one might have expected , a priori , that hemispheric asymmetries in ipsilateral encoding would be more prevalent during movement planning . We recognize that , even with delayed response tasks , it is overly simplistic to assume that the activity cleanly separates into planning and execution phases . This is especially true with sequential movements where planning effects are observed both prior to and during movement execution ( Ariani and Diedrichsen , 2019; Zimnik and Churchland , 2021 ) . While the experimental task was to reach to a cued target , the participants made out-and-back movements , returning to the home position in a relatively smooth manner ( see Figure 1—figure supplement 1 ) . It is reasonable to assume that some component of the activity during the primary outward movement was related to planning the return movement . As such , we are hesitant to draw strong inferences about the differences between ipsilateral encoding during action planning and movement execution . To this point , we have focused on how ipsilateral activity may be reflective of control processes associated with movements of the ipsilateral arm . However , it is possible that this activity is related to postural stabilization of the body during reaching . Indeed , extrapyramidal pathways such as the reticulospinal track have a prominent ipsilateral projection associated with postural control ( Cleland and Madhavan , 2021 ) , and these pathways receive cortical input . We aimed to reduce postural demands in our task by having the participants seated in an upright hospital bed with the back fully supported; nonetheless , there are surely postural shifts associated with the reaching movement . The clinical setting precluded the use of EMG or video , measurements that would allow a quantitative assessment of the relationship of ipsilateral activity to postural adjustments . We do note a few points that are at odds with a posture-based account of ipsilateral activity . First , we are unaware of evidence suggesting a left hemisphere specialization in postural control . In fact , the evidence suggests that postural instability is more frequently associated with right hemisphere lesions ( Bohannon et al . , 1986; Spinazzola et al . , 2003 ) . Second , postural adjustments typically precede movement onset ( Belenkii et al . , 1967 , cited in Guiard , 1987 ) , whereas ipsilateral encoding in our data increased with movement onset . Third , and perhaps most convincing , the encoding model shows that the ipsilateral signals predict reaching kinematics with relatively high precision . The asymmetry of ipsilateral encoding may be reflective of a prominent role of the left hemisphere in bimanual coordination ( Jäncke et al . , 2000; Toyokura et al . , 1999; Maki et al . , 2008; Serrien et al . , 2003; Fujiyama et al . , 2016 ) . For example , Schaffer et al . , 2020 observed greater impairments in bimanual coordination following left hemisphere stroke compared to right hemisphere stroke . Interestingly , the impairment was manifest prior to peak velocity , a finding interpreted as a disruption in predictive control . It may be that the left hemisphere makes an asymmetric contribution to interlimb coordination by tracking or predicting where both limbs are in space . As such , the encoding of ipsilateral arm movement might be a form of state representation , a means to keep track of the state of the ipsilateral arm given that many actions require the coordinated activity of the two limbs . This hypothesis , derived from the current data , is consistent with the increased ipsilateral encoding during the movement phase . The need to monitor the state of the other limb should hold for unimanual gestures performed with either limb . An important question for future work is to examine how ipsilateral representations in the left hemisphere are affected during more complex movements , including those that involve both limbs . Using fMRI , Diedrichsen et al . , 2013 compared ipsilateral movement representations during unimanual and bimanual movements . Within the primary motor cortex , ipsilateral representations could only be discerned during unimanual movement . However , caudal premotor and anterior parietal regions retained similar ipsilateral representation during uni- and bimanual movement . If the left hemisphere tracks both limbs to facilitate bimanual coordination , we would predict that ipsilateral representations in premotor cortex are retained more strongly in the left hemisphere compared to the right hemisphere when both arms are engaged in the task . Using a kinematic encoding model , we observed a striking hemispheric asymmetry , with the left hemisphere more strongly encoding the ipsilateral arm than the right hemisphere , a finding that was apparent during preparation and amplified during movement . This asymmetry was primarily driven by electrodes positioned over premotor and parietal cortices , with strong contralateral encoding for electrodes positioned over sensorimotor cortex . We propose that these networks monitor the state of each arm , a prerequisite for most skilled actions . Intracranial recordings were obtained from six patients ( two female; five right handed; Table 1 ) implanted with subdural grids as part of their treatment for intractable epilepsy . Data were recorded at three hospitals: University of California , Irvine ( UCI ) Medical Center ( n = 2 ) , University of California , San Francisco ( UCSF ) Medical Center ( n = 2 ) , and California Pacific Medical Center ( CPMC ) , San Francisco ( n = 2 ) . Electrode placement was solely determined based on clinical considerations and all procedures were approved by the institutional review boards at the hospitals , as well as the University of California , Berkeley . All patients provided informed consent prior to participating in the study . Participant performed an instructed-delay reaching task while sitting upright in their hospital bed . The participant rested their arms on a horizontal platform ( 71 cm × 20 cm ) that was placed over a standard hospital overbed table . The platform contained two custom-made buttons , each connected to a microswitch . At the far end of the platform ( 13 cm from the buttons , approximately 55 cm from the participant’s eyes ) , a touchscreen monitor was attached , oriented vertically . Visual targets could appear at one of six locations , four for each arm ( Figure 1A ) . The two central locations were used as targets for reaches with either arm; the two eccentric targets varied depending on the arm used . Stimulus presentation was controlled with Matlab 2016a . A photodiode sensor was placed on the monitor to precisely track target presentation times . The analog signals from the photodiode and the two microswitches were fed into the ECoG recording system and were digitized into the same data file as the ECoG data with identical sampling frequency . Testing of the contralateral and ipsilateral arms ( relative to the ECoG electrodes ) was conducted in separate experimental blocks that were counterbalanced . To start each trial , the participant placed their left and right index fingers on two custom buttons to depress the microswitches ( this indicated they were in the correct position and ready to start the trial ) . If both microswitches remained depressed for 500 ms , a fixation stimulus was presented in the middle of the screen for 750 ms , followed by the target , a circle ( 1 . 25 cm diameter ) which appeared in one of the four locations . Another hold period of 900 ms followed in which the participant was instructed to prepare the required movement while the target remained on the screen . If the microswitch was actuated during this hold period , an error message appeared on the screen and the program would advance to the next trial . If the start position was maintained , a compound imperative stimulus was presented at the end of the hold period . This consisted of an auditory tone and an increase in the size of the target ( 2 . 5 cm diameter ) . The participant was instructed that this was the signal to initiate and complete a continuous out-and-back movement , attempting to touch the screen at the target location before returning back to the platform . The target disappeared when the touchscreen was contacted . The imperative was withheld on 5% of the trials ( ‘catch’ trials ) to ensure that the participant only responded after the onset of the imperative . Once back at the home position , the screen displayed the word ‘HIT’ or ‘MISS’ for 750 ms to indicate if the touch had occurred within the target zone . The target zone included the 2 . 5 diameter circle as well as a 1 cm buffer around the target . After the feedback interval , the screen was blank for 250 ms before the reappearance of the fixation stimulus , signaling the start of the next trial . The participants were informed to release either of the buttons at any time they wished to take a break . Each block consisted of 40 trials ( 10/target ) , all performed with a single limb . Blocks alternated between contralateral and ipsilateral arms ( relative to the ECoG electrodes ) , with the order counterbalanced across participants . Each block took approximately 5–6 min to complete . All participants completed at least two blocks with each per arm ( Table 1 ) . We used two methods to analyze the movements . For the first method , we recorded key events defined by the release of the microswitch at the start position , time and location of contact with the touchscreen , and return time to the home position , defined by the time at which they depressed the home position microswitch . For the second method , we used the Leap Motion 3D movement analysis system ( Weichert et al . , 2013 ) to record continuous hand position and the full movement trajectory ( sampling rate = 60 Hz ) . Although the Leap system is a lightweight video-based tracking device that is highly mobile , the unpredictable environment of the ICU led to erratic recordings from the Leap system . For example , patients frequently had intravenous lines in one or both hands which obstructed the visibility of the hand and interfered with the ability of the Leap system to track the hand using their built-in hand model . This resulted in lost samples and therefore satisfactory kinematic data were obtained from only a subset of conditions collected from patients using the Leap system . Given the limitations with the Leap data , we opted to use a simple algorithm to reconstruct the time-resolved hand trajectory in each trial , estimating it from the event-based data obtained with the first method . We used a beta distribution to estimate the velocity profile of the forward and return reach based on reach times and the travel distance ( sampling rate = 100 Hz ) . We opted to use a beta distribution because this best matched the velocity profiles of the data obtained with the Leap system . For conditions that had clean kinematic traces ( no lost samples ) from the Leap system , we compared the estimated kinematic profiles with those obtained with the Leap system . There was a high correlation between the two datasets ( r = 0 . 98 for position in the Z dimension; r = 0 . 93 for velocity in the Z dimension; Figure 1—figure supplement 1 ) . We note that our method of estimating the trajectories results in a smoothed version of the movement , one lacking any secondary or corrective movements that are sometimes observed when reaching to a visual target ( Suway and Schwartz , 2019 ) . We believe this is still a reasonable estimation given the high correlation with the continuous Leap data , and the fact that participants had ample time to prepare the movements and were instructed and observed to make ballistic movements by the experimenter who was present for all recording sessions ( CMM ) . Grid and strip electrode spacing was 1 cm in four patients and 4 mm in the two other patients . The electrode locations were visualized on a three-dimensional reconstruction of the patient’s cortical surface using a custom script that takes the postoperative computed tomography scan and coregisters it to the preoperative structural magnetic resonance scan ( Stolk et al . , 2018 ) . Intracranial EEG data and peripheral data ( photodiode and microswitch traces ) were acquired using a Nihon Kohden recording system at UCI ( 128 channel , 5000 Hz digitization ) and CPMC ( 128 channel , 1000 Hz digitization rate ) , and two Tucker Davis Technologies recording systems at UCSF ( 128 channel , 3052 Hz digitization rate ) . Offline preprocessing included the following steps . First , if the patient’s data were not sampled at 1000 Hz ( UCI and UCSF recording sites ) , the signal from each electrode was low-pass filtered at 500 Hz using a Butterworth filter as an antialiasing measure before downsampling to 1000 Hz . Electrodes were referenced using a common average reference . Each electrode was notch filtered at 60 , 120 , and 180 Hz to remove line noise . The signals were then visually inspected and electrodes with sustained excessive noise were excluded from further analyses . The signals were also inspected by a neurologist ( RTK ) for epileptic activity and other artifacts . Electrodes that had pathological seizure activity were also excluded from the main analyses . Out of 752 electrodes , 82 were removed due to excessive noise and 5 were removed due to epileptic activity , resulting in a final dataset of 665 electrodes . Catch trials and unsuccessful reaches were not included in the analyses . From the cleaned dataset , we extracted the HFA instantaneous amplitude using a Hilbert transform . To account for the 1/f power drop in the spectrum , we divided the broadband signal into five narrower bands that logarithmically increased from 70 to 200 Hz ( i . e . , 70–86 , 86–107 , 107–131 , 131–162 , and 162–200 Hz ) , and applied a band-pass filter within each of these ranges . We then took the absolute value of the Hilbert transform within each band-pass , performed a z-score transformation , and averaged the five values . z-Scoring was performed after concatenating all the blocks for each patient , ensuring that we did not obscure possible amplitude differences across the two arms . As a final step , the data were downsampled to 100 Hz to reduce computational load ( e . g . , number of parameters in the encoding model , see below ) . HFA amplitude fluctuations ( envelope; are evident at lower frequencies Canolty et al . , 2006; Pei et al . , 2011 ) . Four estimated kinematic features were used to predict HFA ( Figure 1B , left ) . The first two features were position and speed in the Z dimension . This dimension captures variability related to movement that is relatively independent of target location ( i . e . , along the axis between the participant and touchscreen ) . The second pair of features were spherical angles that define the specific target locations ( Figure 1A , right ) . Features were selected to reduce collinearity and redundancy in the encoding model . Because we include time lags for each kinematic feature , derivatives can emerge from the linear model ( e . g . , velocity and acceleration can be created from position ) ; thus , velocity and acceleration were not included as additional features . Speed is a nonlinear transformation of position and is added as a separate feature . The estimated kinematic features were used to predict the HFA for each electrode ( Figure 1F ) . We created a 4 × 400 feature matrix by generating a time series for each feature by time lagging the values of the selected feature relative to the neural data , with lags extending from 2 s before movement onset to 2 s after movement onset ( sampling rate at 100 Hz ) . This wide range of lags serves two purposes . First , it provides a way to compensate for the anticipated asynchrony between neural data and movement kinematics . Second , it allowed us to evaluate HFA activity during the instructed-delay ( beginning ~1 . 5 s before movement onset ) period as well as during movement . HFA at each time point [HFA ( t ) ] was modeled as a weighted linear combination of the kinematic features at different time lags , resulting in a set of beta weights , b1 , … , b400 per kinematic feature . To make the beta weights scale-free , the kinematic features and neural HFA were z-scored before being fit by the model . Regularized ( ridge ) regression ( Hoerl and Kennard , 1970 ) was used to estimate the weights that map each kinematic feature ( X ) to the HFA signal ( y ) for each electrode , with λ being the regularization hyperparameter:β^= ( XTX+λI ) −1XTy For within-arm model fitting , the total dataset consisted of all clean , successful trials performed with either the ipsilateral or contralateral arm ( each arm was fit separately ) . Nested fivefold cross-validation was used to select the regularization hyperparameter on inner test sets ( validation sets ) and assess prediction performance on separate , outer test sets . At the outer level , the data were partitioned into five mutually exclusive estimation and test sets . For each test set , the remaining data served as the estimation set . For each outer fold , we further partitioned our estimation set into five mutually exclusive inner folds to train the model ( 80% of estimation set ) and predict neural responses across a range of regularization values on the validation set ( 20% of estimation set ) . For each inner fold , the regularization parameter value was selected that produced the best prediction as measured by the linear correlation of the predicted and actual HFA . The average of the selected regularization parameters across the five inner folds was computed and used to calculate the prediction of the HFA on the outer test set . This procedure was done at the outer level five times . Our primary measure is held-out prediction performance ( R2 ) , which we quantified as the squared linear correlation between the model prediction and the actual HFA time series , averaged across the five mutually exclusive test sets . To be considered as predictive , we established a criterion that an electrode must account for at least 5% of the variability in the HFA signal ( R2 > 0 . 05 ) for either ipsilateral or contralateral reach ( Downey et al . , 2020 ) . Electrodes not meeting this criterion were not included in subsequent analyses . For across-arm model fitting , the same procedure was used except the test set was partitioned from the total dataset of the other arm . We partitioned the data in this manner ( 80% estimation , 20% test ) to make the fitting procedure for the across-arm model comparable to that employed in the within-arm model . MD of target tuning was calculated as the standard deviation of the mean HFA predictions for each of the four target locations:MD=∑i=1n ( xi−x¯ ) 2n where x is the average HFA prediction , xi is the average HFA prediction for each of the four target positions . i iterates through the target locations and n is the total number of target locations . This process was done separately for contralateral and ipsilateral HFA predictions . To assess similarity in tuning across the two arms , we computed the SSEs for average HFA predictions to the same target between the contralateral and ipsilateral arms . This calculation was computed for each electrode as follows:SSEe=∑i=1n ( contrai−ipsii ) 2 where contra and ipsi are average HFA predictions for a given target location reached with either the contralateral or ipsilateral arm . Note that for this calculation n is limited to the two positions that served as target locations for both arms . This metric was only calculated for the two central targets , the targets common to both arms ( the two eccentric target locations varied depending on the arm used ) . These values were scaled from 0 to 1 based on the minimum and maximum values of SSE across all electrodes . SSE represents a metric of dissimilarity; to calculate a similarity index ( SI ) , we subtracted the scaled SSE values from 1:SI=1-SSEe-minSSEemaxSSEe-minSSEe Thus , higher SI represents more similar average predictions . The encoding model was run to predict the full HFA time course . To compare model prediction performance during different phases of the task , the data were epoched into instruction and movement phases , using event markers recorded in the analog channel ( i . e . , cue onset and movement onset ) . Epochs of the same task phase were concatenated together , and prediction performance was operationalized as the square of the Pearson’s correlation between the predicted and actual HFA for each task phase . Linear mixed-effects models were carried out in RStudio using software packages lme4 and permlmer ( Bates et al . , 2014; Lee and Braun , 2012 ) . Each mixed-effects model used participant as a random effect and experimental variables ( e . g . , reaching arm , hemisphere ) as fixed effects . The models were used to predict performance from the kinematic encoding model for all predictive electrodes . Nested models were created to assess the effect of the fixed factors , with the null model using patient as a random effect to predict encoding values . For the nesting , fixed factors were added to the model to assess if each new factor improved prediction above that obtained with the null model using a permutation-based method ( Anderson and Braak , 2003 , Lee and Braun , 2012 ) . Interactions were tested by comparing a model in which the fixed effects were restricted to have additive effects to a model that could have both multiplicative and additive effects . Average weights for each patient were calculated for all predictive electrodes for each kinematic feature . Because each feature has 400 time lags , there are 400 weights per feature in the full model . To assess the relative contribution of each feature , we calculated the sum of the weights across time lags after taking the absolute value of each weight ( since negative weights are as informative as positive weights ) . The sum of each feature was then plotted as a proportion against the total sum of all weights ( after taking the absolute value ) to assess the relative contribution . To capture the temporal profile of the weights , the average was taken for all predictive electrodes for all patents . There are several methods to assess prediction performance for encoding models ( Lage-Castellanos et al . , 2019; Nunez-Elizalde et al . , 2019; Schoppe et al . , 2016 ) . A drawback with the Pearson’s product-moment correlation coefficient is that it does not distinguish between explainable variability and response variability ( Hsu et al . , 2004 ) . To better capture this distinction , we calculated CCnorm , a measure which normalizes by signal power ( SP ) to account for response variability ( Schoppe et al . , 2016 ) :CCnorm=Cov ( y , y^ ) Var ( y^ ) 1SP where y is neural activity and y^ are predictions from the model , and SP is:SP=Var∑n=1NRn-∑n=1NVarRnNN-1 where Rn is the neural time series for the nth trial and N is the total number of trials . We plot the density estimates of all predictive electrodes for both the Pearson’s correlation coefficient ( CCabs ) and CCnorm ( see Figure 2—figure supplement 2 ) . We find similar values for the two metrics , suggesting that the number of trials in our study and SP is sufficient for the encoding model to capture the majority of explainable variance . ( Calculations of Pearson’s correlation coefficient were always taken from held-out trials . ) For each patient , 30 discrete ( x , y ) coordinates were manually demarcated along the central sulcus on individual MRI scans . The 30 points were then interpolated to create a line traversing the central sulcus for each individual . The dorsal aspect of the central sulcus was defined as all points dorsal to the midpoint of the central sulcus . We then calculated the absolute distance between each electrode and the closest point on the dorsal aspect of the central sulcus ( our interpolated line ) .	The brain is split into two hemispheres , each playing the leading role in coordinating movement for the opposite side of the body: lesions on the left hemisphere therefore often result in difficulties moving the right arm or leg , and vice versa . In fact , very few anatomical connections exist between a given hemisphere and the body parts on the same ( or ‘ipsilateral’ ) side . Yet , movements produced with only one limb still engage both sides of the brain , with the hemisphere which does not control the action production , still encoding the direction and speed of the movement . Previous evidence also indicate that the two hemispheres may not have equal roles when coordinating ipsilateral movements . Merrick et al . aimed to shed light on these processes; to do so , they measured electrical activity from the surface of the brain of six patients as they moved their arms to reach a screen . The results revealed that , while the right hemisphere only encoded information about the opposite arm , the left hemisphere contained information about both arms . Finer analyses showed that , for both hemispheres , moving the opposite arm was strongly associated with activity in the primary motor cortex , a region which helps to execute movements . However , in the left hemisphere , movements from the ipsilateral arm were related to activity in brain areas involved in planning and integrating different types of sensory information . These findings contribute to a better understanding of how the motor system works , which could ultimately help with the development of brain-machine interfaces for patients who need a neuroprosthetic limb .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2022	Left hemisphere dominance for bilateral kinematic encoding in the human brain
Optogenetics is a powerful technique to control cellular activity by light . The light-gated Channelrhodopsin has been widely used to study and manipulate neuronal activity in vivo , whereas optogenetic control of second messengers in vivo has not been examined in depth . In this study , we present a transgenic mouse model expressing a photoactivated adenylyl cyclase ( bPAC ) in sperm . In transgenic sperm , bPAC mimics the action of the endogenous soluble adenylyl cyclase ( SACY ) that is required for motility and fertilization: light-stimulation rapidly elevates cAMP , accelerates the flagellar beat , and , thereby , changes swimming behavior of sperm . Furthermore , bPAC replaces endogenous adenylyl cyclase activity . In mutant sperm lacking the bicarbonate-stimulated SACY activity , bPAC restored motility after light-stimulation and , thereby , enabled sperm to fertilize oocytes in vitro . We show that optogenetic control of cAMP in vivo allows to non-invasively study cAMP signaling , to control behaviors of single cells , and to restore a fundamental biological process such as fertilization . Almost every eukaryotic cell contains a specialized surface protrusion called the primary cilium ( Singla and Reiter , 2006 ) . Primary cilia serve as sensory antennae that register physical and chemical cues from the environment . Cilia are not only involved in for example , mechano- , chemo- , and photo-sensation but also in embryonic and neuronal development ( Praetorius and Spring 2001; Singla and Reiter , 2006; Insinna and Besharse , 2008; Goetz and Anderson , 2010; Satir et al . , 2010 ) . Motile cilia , also called flagella , are used both as sensory antenna and motors that move fluids or propel cells ( Salathe , 2007; Bloodgood , 2010; Lindemann and Lesich , 2010; Pichlo et al . , 2014 ) . Signaling in cilia and flagella is tightly regulated , spatially confined , and relies on a cilia-specific transport machinery ( Rosenbaum and Witman , 2002; Delling et al . , 2013; Chung et al . , 2014; Pichlo et al . , 2014 ) . Cyclic nucleotide-dependent signaling plays a central role in both primary and motile cilia . In ciliary structures of olfactory neurons and photoreceptors , it regulates chemosensation and photoreception , respectively ( Johnson and Leroux , 2010 ) . In cilia of endothelial and mesenchymal cells , cAMP signaling controls cilia length in response to extracellular stimuli ( Besschetnova et al . , 2010 ) . Defects in cilia of the renal epithelium cause polycystic kidney disease ( Pazour et al . , 2000 ) . During embryonic development , ciliary cAMP signaling is involved in neural tube development by regulating the Sonic hedgehog ( Shh ) pathway ( Mukhopadhyay et al . , 2013 ) . A case in point for the importance of cAMP signaling in flagella is the mammalian sperm cell . Cyclic AMP signaling is essential for sperm development , motility , and maturation in the female genital tract ( Visconti et al . , 1995b; Wennemuth et al . , 2003; Krähling et al . , 2013 ) . However , the underlying signaling pathways are not well understood . CRIS , a cAMP-binding protein , controls spermatogenesis and sperm motility ( Krähling et al . , 2013 ) . An increase in bicarbonate ( HCO3− ) during the transit from the epididymis to the female genital tract activates a soluble adenylyl cyclase ( SACY ) and , thereby , accelerates the flagellar beat and enhances progressive sperm motility ( Esposito et al . , 2004; Xie et al . , 2006 ) . HCO3−-induced cAMP synthesis by SACY stimulates protein kinase A ( PKA ) ( Wennemuth et al . , 2003; Nolan et al . , 2004 ) and controls capacitation , a maturation process of sperm in the female genital tract that is essential for fertilization ( Chang , 1951; Austin , 1952 ) . Acceleration of the flagellar beat and capacitation proceed on quite different time scales: the beat frequency increases within seconds , whereas PKA-dependent tyrosine phosphorylation , a hallmark of sperm capacitation , takes about an hour to become noticeable ( Visconti et al . , 1995a ) . The study of cAMP-dependent signaling pathways in mammalian sperm is further complicated by an ill-defined interplay between cAMP and Ca2+ signaling; for example , Ca2+ controls SACY activity ( Carlson et al . , 2007 ) and cAMP reportedly evokes a Ca2+ influx ( Kobori et al . , 2000; Ren et al . , 2001; Xia et al . , 2007; Wertheimer et al . , 2013 ) . Previous attempts to delineate these signaling pathways relied on common pharmacological tools: membrane-permeable cAMP analogs and modulators of cAMP signaling components ( Kobori et al . , 2000; Xia et al . , 2007; Wertheimer et al . , 2013 ) . Recent studies in human sperm , however , refute that cAMP controls [Ca2+]i ( Strünker et al . , 2011 ) and disclose serious shortcomings of these pharmacological tools that directly act on the sperm Ca2+ channel CatSper and , thereby , artifactually change Ca2+ levels ( Brenker et al . , 2012 ) . In mice , however , it is equivocal whether cAMP indeed evokes a Ca2+ signal . Therefore , it is required to study the interconnection of cAMP and Ca2+ signaling in mouse sperm by novel , non-invasive techniques . Optogenetics is a powerful tool to spatially and temporally control second messenger-dependent signaling , uncompromised by pharmacological side effects . To this end , we present here a transgenic mouse model expressing the photoactivated adenylyl cyclase bPAC ( Ryu et al . , 2010; Stierl et al . , 2011 ) to manipulate cAMP levels in sperm and , thereby , control sperm motility by light . We show that bPAC mimics the action of HCO3− in sperm . Furthermore , bPAC functionally replaces the endogenous SACY activity and remedies cAMP signaling defects , demonstrating that in vitro , fertility can be restored by optogenetics . Finally , an increase of cAMP levels by light does not elevate Ca2+ levels in sperm . Thus , the Prm1-bPAC mouse model is a powerful tool to analyze cAMP-dependent signaling in ciliary structures with high spatio-temporal resolution uncompromised by pharmacological artifacts . We engineered a targeting vector to express the beta subunit of photo activated adenylyl cyclase from the soil bacterium Beggiatoa ( bPAC ) under the control of the protamine 1 promoter ( Prm1 , Figure 1A ) that is exclusively active in post-meiotic spermatids ( Zambrowicz et al . , 1993 ) . Transgenic mice were generated by pronuclear injection using standard procedures ( Ittner and Götz , 2007 ) . Genomic insertion of the transgene was confirmed by PCR ( Figure 1B ) . Protein expression of bPAC in testis lysates varied between different founder lines ( Figure 1C ) . This variation in transgene expression reflects differences in integration site and/or copy number . For further analysis , founder lines 1 and 5 were chosen due to stable inheritance of the transgene to the next generation . The bPAC protein was exclusively expressed in sperm ( Figure 1D–F ) . Prm1-bPAC males were fertile and did not show any defects during spermatogenesis , demonstrating that bPAC expression does not affect sperm development or function ( Figure 1—source data 1 ) . 10 . 7554/eLife . 05161 . 003Figure 1 . Characterization of the Prm1-bPAC mouse . ( A ) Scheme of the Prm1-bPAC targeting vector . Expression of hemagglutinin ( HA ) -tagged bPAC is driven by the protamine 1 promoter ( Prm1 ) ; arrows indicate the position of genotyping primers . ( B ) Genotyping by PCR . In Prm1-bPAC mice , a 213-bp fragment is amplified . The targeting vector served as a positive control ( + ) . ( C ) Western blot analyzing bPAC-HA expression in testis lysates from different founder lines . Lysates from HEK cells expressing bPAC-HA served as positive control ( + ) , wild-type testis lysates as negative control ( − ) . ( D ) Western blot analyzing bPAC-HA expression in tissue lysates from male and female Prm1-bPAC mice . ( E ) Western blot analyzing bPAC-HA expression in testis and sperm . ( F ) Immunohistochemical analysis of bPAC-HA expression ( left panel: transmission , right panel: fluorescence ) . Pictures at the bottom show a higher magnification ( see white box ) . Sperm flagella are indicated ( arrow ) . Cryosections of mouse testis were probed with anti-HA antibody and fluorescent secondary antibody ( green ) , DNA was stained with DAPI ( blue ) . Loading control for Western blots: β-tubulin . DOI: http://dx . doi . org/10 . 7554/eLife . 05161 . 00310 . 7554/eLife . 05161 . 004Figure 1—source data 1 . The Prm1-bPAC mouse model shows no change in fertility parameters . Data are given as mean ± s . d . ; n = number of experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 05161 . 004 To scrutinize whether bPAC allows optogenetic control of sperm cAMP signaling , we determined cAMP levels in whole sperm before and after light stimulation . Basal cAMP levels in bPAC sperm were slightly enhanced ( bPAC: 15 . 3 ± 3 . 6 fmol/105 sperm versus wild-type: 10 . 0 ± 4 . 4 fmol/105 sperm; Figure 2A ) , which might be due to basal bPAC activity ( Stierl et al . , 2011 ) . In wild-type sperm , light stimulation did not alter cAMP levels , whereas in bPAC sperm , cAMP levels increased by about 2 . 5-fold ( Figure 2A ) . During continuous light stimulation , cAMP levels reached a maximum within 2 min ( Figure 2B ) ; after switching off the light , cAMP returned to baseline within 10 min ( Figure 2B ) . The light-induced cAMP increase was similar to the HCO3−-induced cAMP increase ( Figure 2A ) , indicating that activation of bPAC mimics the activation of SACY by HCO3− . 10 . 7554/eLife . 05161 . 005Figure 2 . Manipulation of cAMP signaling and sperm motility by light . ( A ) Light stimulation of bPAC sperm for 10 min increases cAMP levels . Stimulation of SACY activity with HCO3− ( 1 min ) evokes the same response in wild-type and bPAC sperm . ( B ) cAMP levels in bPAC sperm during prolonged light stimulation and after switching off the light . ( C ) Western blot analyzing SACY and PKA expression in wild-type and bPAC sperm . ( D ) Phosphorylation of PKA targets in wild-type and bPAC sperm detected with an anti-phospho- ( Ser/Thr ) PKA substrate antibody ( p-PKA ) . Wild-type sperm were stimulated with 25 mM HCO3− , bPAC sperm with HCO3− or light . ( E ) db-cAMP- ( 1 mM ) and HCO3−-induced tyrosine phosphorylation in wild-type and bPAC sperm detected with an anti-phospho-tyrosine antibody ( pY ) . ( F ) Acrosome-reaction assay . Percentage of sperm that has undergone the acrosome reaction under non-capacitating conditions ( no HCO3− ) before and after light-stimulation and under capacitating conditions ( 25 mM HCO3− ) . ( G ) Basal flagellar beat frequency in wild-type and bPAC sperm ( individual values and mean ± s . d . ) . ( H , I ) Light-induced change in flagellar beat frequency of individual bPAC sperm . ( J ) Average change in flagellar beat frequency after light stimulation . ( K ) Change in flagellar beat frequency of bPAC sperm after global or local light stimulation of the flagellum for 100 ms; see also Video 3 and 4 . Data are plotted as mean ± s . d . ; ( n ) = number of experiments , n ≥ 5 if not stated otherwise; p values calculated using Student's t test . Loading control for Western blots: β-tubulin . DOI: http://dx . doi . org/10 . 7554/eLife . 05161 . 005 Next , we studied whether bPAC expression interferes with cellular events controlled by cAMP , such as capacitation . Sperm are capacitated in vitro by incubation with a cholesterol acceptor ( e . g . , serum albumin ) and HCO3− to stimulate cAMP synthesis via SACY ( Visconti et al . , 1995a; Xie et al . , 2006 ) . cAMP in turn activates PKA , the principal cAMP target in sperm . PKA activation indirectly promotes tyrosine phosphorylation of a subset of proteins , which is considered to be a hallmark of capacitation ( Visconti et al . , 1995a; Nolan et al . , 2004 ) . The presence of bPAC did not affect SACY expression level ( Figure 2C ) , consistent with the observation that HCO3−-stimulated cAMP synthesis is similar in wild-type and bPAC sperm ( Figure 2A ) . The expression level of PKA also remained unchanged in bPAC sperm ( Figure 2C ) . The phosphorylation of PKA targets under non-stimulated conditions was also not majorly different in bPAC sperm compared to wild-type sperm ( Figure 2D ) . Furthermore , stimulation of bPAC sperm with either light or HCO3− both evoked phosphorylation of PKA targets ( Figure 2D ) . Finally , we analyzed whether the capacitation-associated phosphorylation of tyrosine residues was altered: incubation with HCO3− or the cAMP analog dibutyryl-cAMP ( db-cAMP ) enhanced tyrosine phosphorylation to a similar extent in wild-type and bPAC sperm ( Figure 2E ) . In summary , bPAC expression does not interfere with known cAMP-signaling events in sperm . Next , we studied whether a cAMP increase evoked by light-stimulation of bPAC results in sperm capacitation . As a functional read-out for sperm capacitation , we used an acrosome-reaction assay ( Figure 2F ) . Under non-capacitating conditions ( without HCO3− ) in the dark , the percentage of sperm undergoing the acrosome reaction was similar between wild-type and bPAC sperm ( wild-type: 34 . 1% versus bPAC: 41 . 7%; Figure 2F ) . After light-stimulation , the number of sperm undergoing the acrosome reaction changed dramatically for bPAC sperm but remained unchanged for wild-type sperm ( wild-type: 38 . 4% bPAC: 70 . 5%; Figure 2F ) . This demonstrates that a light-stimulated increase in cAMP levels results in sperm capacitation and that bPAC functionally mimics the activation of SACY by HCO3− . In mammalian sperm , cAMP controls the flagellar beat ( Wennemuth et al . , 2003; Esposito et al . , 2004; Nolan et al . , 2004 ) . The basal beat frequency of bPAC sperm was slightly lower compared to wild-type sperm ( 5 . 5 ± 2 . 5 Hz versus 8 . 0 ± 3 . 7 Hz; Figure 2G ) , which might be due to a cAMP-dependent feedback mechanism activated by elevated basal cAMP levels ( Burton and McKnight , 2007 ) . Light stimulation of bPAC sperm evoked a transient acceleration of the flagellar beat: a 200-ms light pulse increased the beat frequency within 10 s about 2 . 5-fold from 7 Hz to 18 Hz ( Figure 2H; Videos 1 , 2 ) . The acceleration persisted for at least 10 s and sperm returned to basal beat frequency within 30 s ( Figure 2H ) . The light-stimulated acceleration of the flagellar beat was dose-dependent: the longer the light exposure , the faster the onset and higher the beat frequency ( Figure 2I , J ) . Local illumination of a small part of the flagellum also enhanced the beat frequency , albeit the increase was smaller and its kinetics slower compared to global illumination ( Figure 2K , Videos 3 , 4 ) . Altogether , we conclude that ( 1 ) light activation of bPAC mimics the transient beat acceleration evoked by HCO3− stimulation of SACY ( Wennemuth et al . , 2003; Carlson et al . , 2007 ) , ( 2 ) the behavioral response of single cells can be modulated in a graded fashion , and that ( 3 ) bPAC is able to control cAMP signaling in the sperm flagellum with spatial precision . 10 . 7554/eLife . 05161 . 006Video 1 . Wild-type sperm before and after UV flash . The cell was tethered to the glass surface by lowering the BSA concentration ( 0 . 3 mg/ml ) . The recording was performed using an epifluorescent microscope ( IX71; Olympus ) equipped with a dark-field condenser and a 10x objective ( UPlanFL , NA 0 . 3; Olympus ) and an additional 1 . 6× lenses . Frames were acquired at 200 fps using a CMOS camera ( Dimax; PCO ) . Stimulation with UV light for 200 ms was achieved using a UV LED . For clarity , the video is played with 100 fps ( half original speed ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05161 . 00610 . 7554/eLife . 05161 . 007Video 2 . bPAC sperm before and after UV flash . The recording was performed as described for Video 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05161 . 00710 . 7554/eLife . 05161 . 008Video 3 . Local illumination of bPAC sperm . The recording was performed as described for Video 1 using local illumination of the sperm flagellum for 100 ms; the video is played with 100 fps ( half of the original speed ) . In this setting , the light flash is not visible and is , therefore , visualized with a circle . The analysis of the beat frequency is presented in Figure 2K . DOI: http://dx . doi . org/10 . 7554/eLife . 05161 . 00810 . 7554/eLife . 05161 . 009Video 4 . Local illumination of bPAC sperm . Video 3 shown at higher intensity for the frames containing the light flash . This allows to visualize the light flash . The profile of the light flash is indicated in an inset . For clarity , the video speed was reduced to 10% of the frames containing the light flash . DOI: http://dx . doi . org/10 . 7554/eLife . 05161 . 009 We tested whether a light-induced increase of cAMP evokes a Ca2+ influx into bPAC sperm . The use of a red-shifted fluorescent Ca2+ indicator enabled us to orthogonalize indicator excitation and bPAC activation ( Stierl et al . , 2011 ) : after recording basal Ca2+ levels at 520 nm , we repeatedly switched between excitation of the Ca2+ indicator at 520 nm and activation of bPAC at 485 nm; finally , we challenged sperm with 8-Br-cNMPs or db-cAMP while exciting the Ca2+ indicator only ( Figure 3A–C ) . To scrutinize our experimental conditions , we determined sperm cAMP levels before and after activation of bPAC ( Figure 3D ) . We observed that cAMP synthesis via light stimulation of bPAC did not evoke a Ca2+ signal ( Figure 3A–D ) , consistent with the observation that a HCO3−-induced cAMP increase does not affect [Ca2+]i in mouse ( Wennemuth et al . , 2003; Carlson et al . , 2007 ) and human sperm ( Strünker et al . , 2011 ) . In contrast , 8-Br-cAMP , 8-Br-cGMP , and db-cAMP evoked a Ca2+ signal in wild-type and bPAC sperm ( Figure 3A–C , E ) that was abolished in CatSper-null sperm ( Figure 3F ) ; similar results have been reported by others ( Ren et al . , 2001; Xia et al . , 2007 ) . In human sperm , 8-Br-cNMPs directly activate CatSper via an extracellular site ( Brenker et al . , 2012 ) . Experiments analyzing the action of 8-Br-cNMPs on membrane currents in mouse sperm yielded inconsistent results ( Kirichok et al . , 2006; Cisneros-Mejorado et al . , 2014 ) . Thus , this effect needs to be addressed by future studies . 10 . 7554/eLife . 05161 . 010Figure 3 . cAMP does not evoke a Ca2+ influx in mouse sperm . ( A , B ) Ca2+ signals induced by 8-Br-cAMP ( 10 mM; left ) and db-cAMP ( 10 mM; right ) in wild-type ( A ) and bPAC sperm ( B ) loaded with the fluorescent Ca2+ indicator FluoForte . Arrows indicate the addition of compounds . Light stimulation does not evoke a Ca2+ signal . Signals were measured in 96 multi-well plates in a fluorescence plate reader . ( C ) Mean signals evoked by light stimulation , cyclic nucleotide analogs , and ionomycin . ( D ) Illumination according to the protocol in ( A , B ) stimulates cAMP synthesis in bPAC sperm . Data are plotted as mean ± s . d . ; ( n ) = number of experiments; p values calculated using Student's t test . ( E , F ) Ca2+ signals of capacitated sperm induced by cyclic nucleotide analogs ( 10 mM ) in wild-type ( E ) and CatSper-null ( F ) sperm loaded with the fluorescent Ca2+ indicator Cal-520 in a stopped-flow device . Inset in ( F ) : ionomycin control ( 2 µM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05161 . 010 The loss of SACY function abolishes cAMP signaling in sperm , rendering sperm immotile and resulting in male infertility ( Esposito et al . , 2004; Xie et al . , 2006 ) . We explored the potential of optogenetics to compensate for the functional loss of SACY . Removal of extracellular Ca2+ attenuated HCO3−-induced cAMP synthesis by SACY ( Figure 4A ) ( Jaiswal and Conti , 2003; Carlson et al . , 2007 ) . Thereby , tyrosine phosphorylation in wild-type and bPAC sperm was largely abolished ( Figure 4B ) . Elevation of basal cAMP levels by the phosphodiesterase inhibitor IBMX restored tyrosine phosphorylation in wild-type sperm ( Figure 4A , B ) . Similarly , in Ca2+-free medium , the light-stimulated increase of cAMP levels restored tyrosine phosphorylation in bPAC sperm ( Figure 4A , B ) , demonstrating that bPAC activation compensates for the loss of SACY activity . Furthermore , we examined whether in sperm lacking functional SACY , bPAC activation can restore motility and the ability to fertilize the egg . To this end , we crossed Prm1-bPAC mice with mice lacking the sperm-specific Na+/H+ exchanger sNHE , encoded by the Slc9a10 gene . Male Slc9a10-null mice are infertile , because their sperm are largely immotile ( Wang et al . , 2003 ) . In Slc9a10-null mice , the HCO3−-stimulated SACY activity is abolished as well , probably because sNHE and SACY form a functional signaling complex that is disrupted in these mice ( Wang et al . , 2007 ) . The motility of Slc9a10-null sperm is restored by db-cAMP ( Wang et al . , 2003 ) ; however , db-cAMP not only elevates cAMP levels but also evokes a Ca2+ influx via CatSper ( Figure 3E ) . Thus , we tested whether motility can be restored by solely increasing cAMP levels with light . Sperm were immotile in darkness ( Figure 4C , left ) ; light stimulation restored both the flagellar beat and forward motility ( Figure 4C , right; Video 5 ) , demonstrating that restoring cAMP levels alone is sufficient to rescue motility . We went one step further and tested whether Slc9a10-null/bPAC sperm also regain their fertilization potential upon light stimulation . Indeed , light-stimulated Slc9a10-null/bPAC sperm were able to fertilize oocytes in vitro ( Figure 4D ) , demonstrating that optogenetics can restore fertility . 10 . 7554/eLife . 05161 . 011Figure 4 . bPAC restores fertility in mice lacking functional SACY . ( A ) Nominally Ca2+-free medium attenuates HCO3−-induced cAMP synthesis by SACY in wild-type and bPAC sperm . Inhibition of phosphodiesterases by IBMX in wild-type sperm and light stimulation in bPAC sperm increases cAMP levels without extracellular Ca2+ . Data are plotted as mean ± s . d . ; ( n ) = number of experiments; p values calculated using Student's t test . ( B ) In the absence of extracellular Ca2+ , HCO3−-induced tyrosine phosphorylation ( pY ) is strongly attenuated . In bPAC sperm , light stimulation is sufficient to restore tyrosine phosphorylation . ( C ) Light stimulation restores flagellar beating of Slc9a10-null/bPAC sperm . Flagellar waveform of Slc9a10-null/bPAC sperm before ( left ) and after light stimulation ( right ) . Successive , aligned , and superimposed images creating a ‘stop-motion’ image , illustrating one flagellar beating cycle . Scale bar: 30 µm . ( D ) Upon light stimulation , Slc9a10-null/bPAC sperm fertilize oocytes in vitro ( mean ± s . d . ; ( n ) = total number of oocytes from three independent experiments ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05161 . 01110 . 7554/eLife . 05161 . 012Video 5 . Slc9a10-null/bPAC sperm before and after UV flash . The recording was performed as described for Video 1 , but without the additional 1 . 6× magnification , an acquisition frequency of 90 fps , and 500 ms UV stimulation . The video is shown in real time . DOI: http://dx . doi . org/10 . 7554/eLife . 05161 . 012 In this study , we report on a transgenic mouse model designed to control cAMP signaling in sperm using optogenetics . We demonstrate that transgenic expression of bPAC in sperm does not interfere with the endogenous HCO3−-stimulated SACY activity . In fact , bPAC mimics the effect of SACY activity: the light-stimulated increase in cAMP levels is similar to that activated by HCO3− and activation of bPAC increases the flagellar beat frequency in a dose-dependent manner . Thus , the Prm1-bPAC mouse model is an ideal optogenetic tool to study cAMP signaling in sperm . Furthermore , we show that optogenetics can be used to restore fertility . Sperm from infertile Slc9a10-null mice are devoid of cAMP signaling , because they lack HCO3−-stimulated SACY activity , which is essential for sperm motility ( Wang et al . , 2003 , 2007 ) . Expression of bPAC in Slc9a10-null sperm allowed restoring motility after light stimulation and , most importantly , restoring fertilization in vitro . Of note , in vitro fertilization ( IVF ) using light-stimulated sperm was successful using zona pellucida-intact oocytes , demonstrating that a cAMP increase , at least in vitro , is sufficient to restore fertility in Slc9a10-null mice . In contrast , incubation of Slc9a10-null sperm with cAMP analogs only restored fertilization of zona pellucida-free but not zona pellucida-intact oocytes ( Wang et al . , 2003 ) . This discrepancy could be explained by ( 1 ) cAMP analogs not reaching sufficiently high levels in sperm , or ( 2 ) cAMP analogs being not as efficient as cAMP to stimulate downstream targets , or ( 3 ) non-specific effects compromising the interpretation of their action ( Brenker et al . , 2012 ) . The Prm1-bPAC mouse model holds great promise studying cAMP signaling uncompromised by side effects and allowed us to solve a long-standing controversy by showing that cAMP , in fact , does not stimulate a Ca2+ influx into mouse sperm ( Figure 3B ) . These results demonstrate that optogenetics can disentangle cellular events such as Ca2+ and cAMP signaling in vivo . Now that we have established the Prm1-bPAC mouse model and demonstrated that it can be used to analyze cAMP signaling in sperm , the next step will be to implement the spatio-temporal precision of this tool to study cAMP signaling in sperm . For example , HCO3− stimulation evokes a rapid cAMP increase and changes the flagellar beat frequency within seconds . However , PKA-dependent activation of tyrosine phosphorylation is barely detectable within the first hour of HCO3− incubation ( Visconti et al . , 1995a ) . Thus , HCO3− exhibits short-term and long-term effects and probably controls cellular events in addition to cAMP synthesis via SACY . The Prm1-bPAC mouse model allows deciphering these signaling events , if bPAC expression does not interfere with other signaling pathways . Signaling molecules in sperm flagella are highly compartmentalized ( Chung et al . , 2014 ) . The flagellar beat is controlled by cAMP-dependent phosphorylation of PKA targets in the axoneme ( Salathe , 2007 ) ; however , it is unknown whether cAMP signaling is restricted to microdomains or whether changes in cAMP propagate along the flagellum . Compartmentalization of cAMP signaling confers specificity to this ubiquitous cellular messenger , for example , in cell-cycle progression or modulation of gene expression . ( Michel and Scott , 2002; Zaccolo and Pozzan , 2002; Willoughby and Cooper , 2007 ) . Our results show that local illumination of the flagellum of bPAC sperm is sufficient to increase its beat frequency . Optogenetics using bPAC provides the spatial and temporal precision to increase cAMP levels in subcellular regions and specialized signaling compartments such as cilia and flagella , which allows to study complex signaling networks - not only in mammalian sperm but also any other ciliated cell type . Finally , the optogenetic toolkit has been recently expanded by a light-activated phosphodiesterase ( LAPD ) that hydrolyses cyclic nucleotides upon stimulation by red light ( Gasser et al . , 2014 ) . A combination of LAPD and bPAC with fluorescent cAMP biosensors holds great promise to map the dynamics of cAMP signaling in live cells in precise spatio-temporal and quantitative terms . These findings encourage future studies to explore the full potential of controlling cellular messengers by optogenetics . The bPAC cDNA ( beta subunit of photo-activated adenylyl cyclase from Beggiatoa sp . PS , BGP_1043; modified for expression in mammalian systems according to Stierl et al . ( 2011 ) , GU461307 ) sequence was amplified via PCR . A hemagglutinin ( HA ) tag was fused to the C terminus , an EcoRI restriction site was added to the 5′ end and a BamHI and XhoI restriction site to the 3′ end by nested PCR . The PCR product was cloned into a pBluescript SK ( - ) vector ( Agilent Technologies , Santa Clara , USA ) using EcoRI and XhoI ( pB-bPAC ) . After sequencing , the bPAC-HA insert was excised and cloned into pPrCExV ( a kind gift from Robert Braun , Jackson Laboratory ) using EcoRI/BamHI . Transgenic mice were generated via pronuclear injection following standard procedures ( Ittner and Götz , 2007 ) at the HET animal facility ( University Hospital Bonn , Germany ) . Mice were genotyped by PCR using bPAC-specific primers ( see also Figure 1A ) . Mice used in this study were 2–5 months of age . All animal experiments were in accordance with the relevant guidelines and regulations . Slc9a10-null mice were purchased from the Jackson Laboratory ( B6;129S6-Slc9a10tm1Gar/J , stock number: 007661 ) . For heterologous expression , HEK293 cells were transfected using Lipofectamine 2000 ( Life Technologies , Carlsbad , USA ) . Lysates were obtained by homogenizing cells or tissues in lysis buffer ( 10 mM Tris/HCl , pH 7 . 6 , 140 mM NaCl , 1 mM EDTA , 1% Triton X-100 , mPIC protease inhibitor cocktail 1:500 ) followed by trituration through a 18-gauge needle . Samples were incubated for 30 min on ice and centrifuged at 10 , 000×g for 5 min at 4°C . Prior to separation by SDS-PAGE , samples were mixed with 4× SDS loading buffer ( 200 mM Tric/HCl , pH 6 . 8 , 8% SDS ( wt/vol ) , 4% β-mercaptoethanol ( vol/vol ) , 50% glycerol , 0 . 04% bromophenol blue ) and heated for 5 min at 95°C . To activate bPAC , sperm samples were illuminated for 90 min in suspension in a reaction tube using a halogen fiber optics illuminator ( Olympus , Tokio , Japan ) . Sperm samples used for SDS-PAGE were washed with 1 ml PBS and sedimented by centrifugation at 5000×g for 5 min . 1–2 × 106 cells were resuspended in 4× SDS loading buffer and heated for 5 min at 95°C . For Western blot analysis , proteins were transferred onto PVDF membranes ( Merck Millipore , Billerica , USA ) , probed with antibodies , and analysed using a chemiluminescence detection system . For re-probing , membranes were incubated in stripping buffer ( 62 . 5 mM Tris/HCl pH 6 . 7 , 2% SDS ) for 30 min at 65°C and washed with PBS before incubation with a new antibody . For the study of protein phosphorylation , sperm were incubated in capacitating buffer containing 25 mM NaHCO3 for 10 min ( PKA-dependent phosphorylation ) or 90 min ( tyrosine kinase-dependent phosphorylation ) . Similarly , bPAC sperm were illuminated for either 10 or 90 min using a halogen fibre optics illuminator ( 150 W , Intralux 6000-1; Volpi , Schlieren , Switzerland ) . Primary antibodies: anti-HA 3F10 ( 1:5000; Roche , Basel , Switzerland ) , anti-SACY R21 ( 1:2000; CEP Biotech , Tamarac , USA ) , anti-phosphotyrosine 4G10 ( 1:1000; Merck Millipore ) , anti-PKA[C] 5B ( 1:4000; BD Transduction Laboratories , San Jose , USA ) , anti-phospho-PKA substrate ( 1:1000; Cell Signaling , Danvers , USA ) , anti-α-tubulin ( 1:5000; Sigma-Aldrich , Seelze , Germany ) ; secondary antibodies: goat-anti-rat , HRP conjugated ( 1:5000 , Dianova , Hamburg , Germany ) , sheep-anti-mouse , HRP conjugated ( 1:5000 , Dianova ) . Testes were fixed in 4% paraformaldehyde/PBS overnight , cryo-protected in 10 and 30% sucrose , and embedded in Tissue-Tek ( Sakura Finetek , Alphen aan den Rijn , Netherlands ) . To block unspecific binding sites , cryosections ( 16 µm ) were incubated for 1 hr with blocking buffer ( 0 . 5% Triton X-100 and 5% ChemiBLOCKER ( Merck Millipore ) in 0 . 1 M phosphate buffer , pH 7 . 4 ) . The primary anti-HA antibody ( rat monoclonal; Roche ) was diluted 1:1000 in blocking buffer and incubated for 2 hr . Fluorescent secondary antibodies ( donkey anti-rat Alexa488; Dianova ) was diluted 1:400 in blocking buffer containing 0 . 5 mg/ml DAPI ( Life Technologies ) and pictures were taken on a confocal microscope ( FV1000; Olympus ) . Sperm were isolated by incision of the cauda followed by a swim-out in modified TYH medium ( 135 mM NaCl , 4 . 8 mM KCl , 2 mM CaCl2 , 1 . 2 mM KH2PO4 , 1 mM MgSO4 , 5 . 6 mM glucose , 0 . 5 mM sodium pyruvate , 10 mM L-lactate , 10 mM HEPES , pH 7 . 4 ) . For capacitation , the medium contained 3 mg/ml BSA and 25 mM of the NaCl was substituted with 25 mM NaHCO3 . The pH was adjusted at 37°C . After 15–30 min swim-out at 37°C , sperm were collected and counted . Red light was used during the isolation of bPAC sperm . Sperm were adjusted to a concentration of 1 . 25 × 107 cells per ml with TYH buffer containing different compounds with the following final concentrations: 25 mM NaHCO3 , 1 mM N6 , 2′-O-dibutyryladenosine-3′ , 5′-cyclic monophosphate ( db-cAMP; Sigma-Aldrich ) , 0 . 5 mM IBMX ( Sigma-Aldrich ) . To activate bPAC , sperm samples were illuminated in suspension in a reaction tube using a halogen fiber-optics illuminator ( 150 W , Intralux 6000-1; Volpi , Schlieren , Switzerland ) . After stimulation with light or compounds for 1 to 10 min , the reaction was quenched with HClO4 ( 1:3 ( vol/vol ) ; 0 . 5 M final concentration ) . Samples were frozen in liquid N2 , thawed , and neutralized by addition of K3PO4 ( 0 . 24 M final concentration ) . The salt precipitate and cell debris were sedimented by centrifugation ( 15 min , 15 , 000 g , 4°C ) . The cAMP content in the supernatant was determined by a competitive immunoassay ( Molecular Devices , Sunnyvale , USA ) , including an acetylation step for higher sensitivity . Calibration curves were obtained by serial dilutions of cAMP standards . 96-well plates were analysed by using a microplate reader ( FLUOstar Omega; BMGLabtech , Ortenberg , Germany ) . 1 × 106 sperm were capacitated in TYH buffer supplemented with 3 mg/ml BSA and 25 mM NaHCO3 . Non-capacitated sperm were incubated in BSA buffer only . To induce capacitation by light , sperm were illuminated with a halogen fiber-optics illuminator ( 150 W , Intralux 6000-1; Volpi ) at 37°C for 90 min . The acrosome reaction was induced by incubating sperm with 2 µM ionomycin for 10 min , incubation with 1% DMSO served as control . Sperm were collected by centrifugation and resuspended in 100 µl PBS buffer . Samples were air dried on microscope slides and fixed with 100% ethanol for 30 min at room temperature . For acrosome staining , sperm were labeled with 5 μg/ml PNA-FITC ( L7381 , Sigma-Aldrich ) and 2 μg/ml DAPI in PBS for 30 min . Images were acquired using a confocal laser scanning microscope ( FV1000; Olympus ) , a minimum of 600 cells was analyzed per condition . Sperm motility was studied in shallow observation chambers ( depth 150 µm ) . Cells were tethered to the glass surface by adjusting the BSA concentration in the buffer to 0 . 3 mg/ml . For analysis , cells that had their head attached to the glass surface and that had a free beating flagellum were chosen . Sperm motility was recorded under an inverted microscope ( IX71; Olympus ) equipped with a dark-field condenser , a 10x objective ( UPlanFL , NA 0 . 3; Olympus ) , and additional 1 . 6× lenses ( 16× final magnification ) . The temperature of the microscope incubator ( Life Imaging Services , Basel , Switzerland ) was adjusted to 37°C . To obtain sharp images of moving sperm , stroboscopic illumination ( 2 ms light pulses ) was achieved using a red LED ( M660L3-C1; Thorlabs , Newton , USA ) with a custom-made power supply . The camera and the LED light pulses were synchronized using a function generator ( 33220A; Agilent ) . Images were collected at 200 frames per second using a CMOS camera with a pixel size of 11 µm ( Dimax; PCO , Kelheim , Germany ) . Stimulation of bPAC was achieved using a collimated UV LED ( 365 nm; ∼12 mW ) coupled to the epifluorescence port of the microscope ( M365L2-C1; Thorlabs ) and a TTL-controlled custom-made power supply . Irradiation time was set using a TTL pulse generator ( UTG100; ELV ) . For local activation of the flagellum of bPAC sperm , we partially focused the UV-LED onto a 100-µm diaphragm ( Linos Photonik ) by detuning its corresponding collimator lens . The diaphragm aperture was then imaged onto the objective focal plane by a plano-convex lens ( f = 100 mm; LA1509A , Thorlabs ) . This resulted in a Gaussian-shaped UV spot ( R2 of the fit 0 . 96 ) with a 12-µm width at the focal plane . The total UV-light power delivered was 2 . 3 mW . Quantification of the flagellar beat was performed using custom-made programs written in MATLAB ( MathWorks . Natick , USA ) . The software can be made available upon request . The program identified the best threshold for binarization followed by a skeleton operation to identify the flagellum . The flagellar beat parameters were determined within a time window of 0 . 5 s before and after each frame . For frames at the boundaries ( the beginning or the end of the video or flanking the UV flash ) , the time windows were asymmetric but contained the same number of frames . We monitored the angle between the straight line connecting the middle of the flagellum with the sperm head and the axis of symmetry of the cell . This angle varied in a sinusoid-like manner in time . The beat frequency was obtained by fitting a sinus to this wave . For alignment of the flagellar beat envelopes , we used custom-made programs written in LabVIEW . Using defined thresholds , the image was binarized . From a user-defined region-of-interest centred at the cell head , the program determined the location of the head on subsequent frames using a registering procedure . The neck of the cell was identified by applying a mask with the shape of an annulus centred into the sperm head . The annulus had an internal diameter of 16 µm to cover the sperm head and a 4 µm longer external diameter , enough to resolve the first pixels of the neck . All frames were then rotated and superimposed with a rotation angle equal to the azimuth of the neck region on a reference system centred at the sperm head . For better visualization , representative datasets were smoothed using Graph Pad Prism 5 . 02 ( factor 20 for global illumination , factor 100 local illumination ) . Changes in [Ca2+]i were recorded using the fluorescent Ca2+ indicators Cal-520 , AM ( AAT Bioquest , Sunnyvale , USA ) and FluoForte ( ENZO Life Sciences , Lörrach , Germany ) in 96 multi-well plates in a fluorescence plate reader ( Fluostar Omega; BMGLabtech ) and in a rapid-mixing device in the stopped-flow mode ( SFM400; Bio-Logic , Claix , France ) . For Ca2+ measurements in the plate reader , sperm were loaded with FluoForte ( 20 µM ) in the presence of Pluronic F-127 ( 0 . 02% vol/vol ) at 37°C for 45 min . After incubation , excess dye was removed by three centrifugation steps ( 700×g , 7 min , room temperature ) . The pellet was resuspended in TYH and equilibrated for 5 min at 37°C . Each well was filled with 100 µl ( 106 sperm ml−1 ) of the sperm suspension . To record changes in [Ca2+]i upon activation of bPAC , samples were alternately excited at 485 nm and 520 nm . As a control , fluorescence was recorded before and after the addition of 10 µl of different compounds to the final concentration of 10 mM ( db-cAMP , 8-Bromo-cAMP , 8-Bromo-cGMP; Sigma-Aldrich ) . Changes in fluorescence are depicted as ΔF/F0 ( % ) , indicating the percentage change in fluorescence ( ΔF ) with respect to the mean basal fluorescence ( F0 ) before application of buffer or compounds . To record changes in [Ca2+]i in a rapid-mixing device ( SFM-400; Bio-Logic ) in the stopped-flow mode , sperm were loaded with Cal-520 ( 5 μM ) as described above for FluoForte . Changes in [Ca2+]i were measured as previously described ( Strünker et al . , 2011 ) with minor modifications . In brief , the sperm suspension ( 5 × 106 sperm/ml ) was rapidly mixed 1:1 ( vol/vol ) with the respective stimulants at a flow rate of 0 . 5 ml/s . Fluorescence was excited by a SpectraX Light Engine ( Lumencor , Beaverton , USA ) , whose intensity was modulated with a frequency of 10 kHz . The excitation light was passed through a BrightLine 475/28 nm filter ( Semrock , Rochester , USA ) onto the cuvette . Emission light was passed through a BrightLine 536/40 filter ( Semrock ) and recorded by photomultiplier modules ( H10723-20; Hamamatsu Photonics ) . The signal was amplified and filtered through a lock-in amplifier ( 7230 DualPhase; Ametek , Paoli , USA ) . Data acquisition was performed with a data acquisition pad ( PCI-6221; National Instruments . Austin , USA ) and Bio-Kine software v . 4 . 49 ( Bio-Logic ) . Ca2+ signals are depicted as the percent change in fluorescence ( ΔF ) with respect to the mean of the first three data points recorded immediately after mixing ( F0 ) , that is , when a stable fluorescence signal was observed . The control ( buffer ) ΔF/F0 signal was subtracted from compound-induced signals . Superovulation in females was induced by intraperitoneal injection of 10 I . U . Intergonan ( SimposiumVet , Lisbon , Portugal ) 3 days before the experiment . 14 hours before oocyte preparation , mice were injected with 10 I . U . Ovogest ( SimposiumVet ) . KSOM medium ( EmbryoMax Modified M16 Medium; Merck Millipore ) was mixed 1:1 with mineral oil ( Sigma-Aldrich ) and equilibrated overnight at 37°C . On the day of preparation , 100 µl drops of KSOM medium were covered with the medium/oil mixture and 100 , 000 sperm were added to each drop . Sperm were capacitated for 90 min in TYH medium supplemented as indicated above . For stimulation of bPAC , the medium contained no HCO3− and reaction tubes containing sperm were placed in a custom-made rack equipped with blue LEDs for 90 min . Cumulus-enclosed oocytes were prepared from the oviducts of superovulated females and added to the drops . After 4 hr at 37°C and 5% CO2 , oocytes were transferred to fresh KSOM medium . The number of 2-cell stages was evaluated after 24 hr .	Tiny hair-like structures called cilia on the outside of cells play many important roles , including detecting physical and chemical signals from the environment . Special cilia—called flagella—help cells to move around and perhaps the most well-known of these are sperm flagella , which propel sperm in their quest to fertilize the egg . A chemical messenger called cAMP is essential for the movement of sperm flagella . When a sperm cell enters the female reproductive tract , an enzyme called SACY is activated . Within seconds , SACY produces cAMP and , thereby , causes the flagella to beat faster so that the sperm cell speeds toward the egg . cAMP also controls sperm maturation , which is needed to penetrate the egg . However , the precise details of the role of cAMP in sperm cells are not clear . Here , Jansen et al . have investigated this role using a cutting-edge technique—called optogenetics—that was originally developed to study brain cells in living organisms . Jansen et al . genetically engineered a mouse so that exposing sperm to blue light activates a light-sensitive enzyme called bPAC that increases cAMP levels in sperm . In these mice , the activation of bPAC by light accelerated the beating of the flagella so the sperm moved faster , in a way that was similar to the effects that are normally observed after the activation of the SACY enzyme . In mice lacking among other things the SACY enzyme—whose sperm cells are unable to move or fertilize an egg—activating the light-sensitive bPAC enzyme restored sperm motility and enabled the sperm to fertilize an egg . These results show that optogenetics may be a useful tool for studying how flagella and other types of cilia work .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2015	Controlling fertilization and cAMP signaling in sperm by optogenetics
The cohesin complex topologically encircles DNA to promote sister chromatid cohesion . Alternatively , cohesin extrudes DNA loops , thought to reflect chromatin domain formation . Here , we propose a structure-based model explaining both activities . ATP and DNA binding promote cohesin conformational changes that guide DNA through a kleisin N-gate into a DNA gripping state . Two HEAT-repeat DNA binding modules , associated with cohesin’s heads and hinge , are now juxtaposed . Gripping state disassembly , following ATP hydrolysis , triggers unidirectional hinge module movement , which completes topological DNA entry by directing DNA through the ATPase head gate . If head gate passage fails , hinge module motion creates a Brownian ratchet that , instead , drives loop extrusion . Molecular-mechanical simulations of gripping state formation and resolution cycles recapitulate experimentally observed DNA loop extrusion characteristics . Our model extends to asymmetric and symmetric loop extrusion , as well as z-loop formation . Loop extrusion by biased Brownian motion has important implications for chromosomal cohesin function . Cohesin is a member of the Structural Maintenance of Chromosomes ( SMC ) family of ring-shaped chromosomal protein complexes that are central to higher order chromosome organization ( Hirano , 2016; Uhlmann , 2016; Yatskevich et al . , 2019 ) . Cohesin holds together replicated sister chromatids from the time of their synthesis in S phase , until mitosis , to ensure their faithful segregation during cell divisions ( Guacci et al . , 1997; Michaelis et al . , 1997 ) . In addition , budding yeast cohesin participates in mitotic chromosome condensation , while higher eukaryotic cohesin impacts gene regulation by defining boundary elements during interphase chromatin domain formation ( Parelho et al . , 2008; Wendt et al . , 2008 ) . Cohesin is also recruited to sites of double-stranded DNA breaks to promote DNA repair by homologous recombination and to stalled DNA replication forks to aid restart of DNA synthesis ( Ström et al . , 2004; Unal et al . , 2004; Tittel-Elmer et al . , 2012 ) . Understanding how cohesin carries out all these biological functions requires the elucidation of the molecular mechanisms by which cohesin interacts with DNA , as well as how cohesin establishes interactions between more than one DNA . Cohesin’s DNA binding activity is contained within its unique ring architecture ( Gligoris et al . , 2014; Huis in 't Veld et al . , 2014 ) . Two SMC subunits , Smc1Psm1 and Smc3Psm3 , form long flexible coiled coils that are connected at one end by a dimerization interface known as the hinge ( generic gene names are accompanied by fission yeast subunit names in superscript; fission yeast cohesin was used for the experiments and structural analyses in this study , see Figure 1A ) . At the other end lie ABC transporter-type ATPase head domains whose dimerization is regulated by ATP binding . The two SMC heads are further connected by a kleisin subunit , Scc1Rad21 . The kleisin N-terminus reversibly engages with Smc3Psm3 coiled coil next to the ATPase head , forming the kleisin N-gate through which DNA enters the cohesin ring ( Figure 1A; Higashi et al . , 2020 ) . The kleisin C-terminus in turn binds to the Smc1 head domain . These kleisin terminal domains are connected via a long unstructured region , to which two HEAT repeat subunits bind that promote topological cohesin loading onto DNA . Scc3Psc3 interacts with the middle of the kleisin unstructured region . Scc2Mis4 , together with its binding partner Scc4Ssl3 , transiently associates with the kleisin between the kleisin N-gate and Scc3Psc3 . Once cohesin loading onto DNA is complete , Scc2Mis4 is replaced by a related HEAT repeat subunit , Pds5 ( Murayama and Uhlmann , 2015; Petela et al . , 2018 ) . Because of its transient role , Scc2Mis4-Scc4Ssl3 is often thought of as a cofactor , termed ‘cohesin loader’ ( Ciosk et al . , 2000; Murayama and Uhlmann , 2014 ) . Following topological loading , cohesin is free to linearly diffuse along DNA in vitro ( Davidson et al . , 2016; Kanke et al . , 2016; Stigler et al . , 2016 ) , while RNA polymerases push cohesin along chromosomes toward sites of transcriptional termination in vivo ( Lengronne et al . , 2004; Ocampo-Hafalla et al . , 2016; Busslinger et al . , 2017 ) . Cohesin promotes sister chromatid cohesion following DNA replication by topologically entrapping two sister DNAs ( Haering et al . , 2008; Murayama et al . , 2018 ) . In addition to topologically entrapping DNA , in vitro experiments have revealed the ability of human cohesin to translocate along DNA in a directed motion , as well as its ability to extrude DNA loops ( Davidson et al . , 2019; Kim et al . , 2019 ) . These activities are reminiscent of those previously observed with a related SMC complex , condensin , a central mediator of mitotic chromosome condensation ( Terakawa et al . , 2017; Ganji et al . , 2018 ) . Like topological loading onto DNA , loop extrusion by cohesin depends on its ATPase , as well as on the human Scc3Psc3 homolog SA1 and the cohesin loader ( NIPBL-MAU2 ) . In contrast to topological loading , cohesin is able to extrude DNA loops if all three cohesin ring interfaces are covalently closed ( Davidson et al . , 2019 ) . This suggests that loop extrusion does not involve topological DNA entry into the cohesin ring . Several models have been proposed as to how SMC complexes extrude DNA loops . These include a tethered-inchworm model in which a scissoring motion of the ATPase heads translates into movement along DNA ( Nichols and Corces , 2018 ) . The DNA-segment-capture model instead suggests that a pumping motion between open and closed configurations of the SMC coiled coils constrains DNA loops ( Marko et al . , 2019 ) . Finally , a scrunching model proposes that the SMC hinge reaches out to capture and reel in DNA loops ( Ryu et al . , 2020a ) . A characteristic of experimentally observed loop extrusion is that very small counterforces ( <1 pN ) stall loop growth ( Ganji et al . , 2018; Golfier et al . , 2020 ) . Both the DNA-segment-capture and the scrunching models therefore suggest that diffusive DNA motion contributes to loop growth . However , molecular details how SMC complexes enact these loop extrusion mechanisms remain elusive . Another important open question remains whether SMC complexes can move along and extrude physiological chromatin substrates , densely decorated by histones and other DNA binding proteins and folded into higher order structures ( Nozaki et al . , 2017; Xu et al . , 2018 ) . We recently solved a cryo-EM structure of fission yeast cohesin with its loader in a nucleotide-bound DNA gripping state ( Higashi et al . , 2020 ) . Together with DNA-protein crosslinking and biophysical experiments , this allowed us to trace the DNA trajectory into the cohesin ring by sequential passage through a kleisin N-gate and an ATPase head gate . We noticed , however , that kleisin N-gate passage might not be a strict prerequisite for DNA to reach the gripping state . We speculated that an alternative gripping state arises in which DNA has not passed the kleisin N-gate . While topological DNA entry is barred , this state might constitute an intermediate during loop extrusion . Here , we show how two DNA binding modules of the cohesin complex , formed by its HEAT repeat subunits ( Murayama and Uhlmann , 2014; Li et al . , 2018; Collier et al . , 2020; Kurokawa and Murayama , 2020 ) , are juxtaposed in the gripping state but swing apart following ATP hydrolysis . We illustrate how this swinging motion promotes completion of topological DNA entry , or alternatively generates DNA movements . Computational simulations of the latter scenario demonstrate how a Brownian ratchet forms that can drive loop extrusion . Our study provides a molecular proposal for both topological entry into the cohesin ring as well as for DNA loop extrusion . Two cryo-EM structures of fission yeast and human cohesin in the presence of non-hydrolyzable ATP analogs Higashi et al . , 2020; Shi et al . , 2020 have revealed how the cohesin complex components come together during topological loading onto DNA ( Figure 1A ) . ATP binding to the ATPase heads opens up a kleisin N-gate , allowing DNA to enter and reach the top of the engaged ATPase heads . The Scc2Mis4 cohesin loader subunit then embraces the DNA , thereby closing the kleisin N-gate again and forming the DNA ‘gripping state’ that is captured by the cryo-EM structures ( Figure 1B , left ) . The DNA is held in place by numerous positively charged surface residues , contributed by both the SMC heads and Scc2Mis4 . The cohesin loader has undergone a marked conformational change , compared to its crystallographically observed ‘extended’ form . Its N-terminal handle embraces the DNA in the gripping state , thereby adopting a ‘bent’ conformation . We refer to the composite DNA interaction site , consisting of Scc2Mis4 and the SMC ATPase heads , as the ‘Scc2-head module’ . A second DNA contact in the gripping state , situated behind the cohesin loader , is made by Scc3Psc3 in conjunction with the kleisin middle region ( Figure 1A and Figure 1C , left ) . The SMC hinge touches down to bridge cohesin loader and Scc3Psc3 , enabled by SMC coiled coil inflection at their elbows . Scc3Psc3 and the kleisin middle region bind DNA in a fashion similar to that previously seen in a crystal structure of budding yeast Scc3 , with an Scc1 peptide , bound to DNA ( Li et al . , 2018 ) . The DNA interaction is again provided by an array of positively charged amino acids that line the combined Scc3Psc3 and kleisin surface . We refer to this second DNA binding site as the ‘Scc3-hinge module’ ( see Figure 1—figure supplement 1A for details of the structural model ) . We now consider the consequences of ATP hydrolysis on the two DNA binding modules outlined above . Upon ATP hydrolysis , the ATPase heads disengage , leading to loss of at least some of the DNA contacts within the Scc2-head module . If we assume that Scc2Mis4 returns to its extended crystal structure form , further DNA contacts are lost as the gripping state opens up ( Figure 1B , right ) . We can see how , in this state , DNA is free to leave the Scc2-head module during topological DNA entry ( a structural model for the full DNA trajectory during topological entry is provided in Figure 3 ) . During loop extrusion , we propose that DNA fails to exit the Scc2-head module through the ATPase head gate following ATP hydrolysis . This could be because an alternative kleisin path obstructs the head gate , or due to persisting electrostatic interactions between Scc2Mis4 and the DNA ( discussed below in Figure 3 ) . DNA movements are instead limited to transverse DNA sliding , that is in and out of the image plane in Figure 1B , akin to experimentally observed cohesin loader sliding along DNA ( Stigler et al . , 2016 ) . Because of the moveable nature of the DNA interaction , we refer to this shape of the Scc2-head module as its ‘slipping state’ . A bent Scc2Mis4 conformation that embraces DNA , compared to its extended crystal structure form , is shared between the fission yeast , human and budding yeast gripping states ( Figure 1—figure supplement 1B; Collier et al . , 2020; Higashi et al . , 2020; Shi et al . , 2020 ) . This commonality opens up the possibility that the gripping to slipping state conformational transition is a conserved feature of the Scc2-head module . In contrast to the Scc2-head module , the Scc3-hinge module and its DNA binding site do not undergo an obvious conformational change when comparing its gripping state and free crystal structure forms . Human Scc3SA1 in the gripping state shows an almost perfect overlap with the crystal structure conformation of free Scc3SA2 ( RMSD = 2 . 4 Å , Figure 1—figure supplement 1C; Hara et al . , 2014; Shi et al . , 2020 ) . In the gripping state , Scc3Psc3 interacts with the cohesin loader both along the N-terminal Scc2Mis4 handle , as well as the central Scc2Mis4 hook . Scc2Mis4 rearrangement into its extended form disrupts at least some of these contacts , thereby terminating Scc3Psc3 - Scc2Mis4 juxtaposition ( Figure 1C , right ) . A conformational change within the Scc2Mis4 handle is furthermore likely to weaken its interaction with the SMC hinge . We therefore hypothesize that , as a consequence of Scc2Mis4 structural rearrangements following ATP hydrolysis , the interaction between the Scc2-head and Scc3-hinge modules resolves . While the Scc2-head module turns from the DNA gripping to the slipping state , the DNA binding characteristics of the Scc3-hinge module remain unaltered . Above , we predicted positional changes of the Scc3-hinge module relative to the Scc2-head module , when comparing the gripping and ATP post-hydrolysis states . To experimentally observe the positions of module components , we designed FRET reporters inserted at the hinge within Smc1Psm1 , at the C-terminus of Scc3Psc3 and at the N-terminus of Scc2Mis4 ( N191 ) ( Figure 2A ) . Scc2Mis4 ( N191 ) is an N-terminally truncated Scc2Mis4 variant missing the first 191 amino acids . The truncation abrogates Scc4Ssl3 interaction , a factor important for in vivo cohesin loading onto chromatin . In vitro , using naked DNA as a substrate , Scc2Mis4 ( N191 ) retains full biochemical capacity to promote gripping state formation , topological cohesin loading , as well as loop extrusion ( Chao et al . , 2015; Higashi et al . , 2020; Shi et al . , 2020 ) . Based on our structural model , these locations are within distances that should allow FRET signal detection in the gripping state . CLIP or SNAP tags , inserted at these positions , served as fluorophore receptors . We labeled these tags during protein purification with Dy547 and Alexa 647 dyes as donor and acceptor fluorophores , respectively ( Figure 2B and Figure 2—figure supplement 1A ) . The tagged and labeled proteins retained the ability to topologically load onto DNA in vitro , albeit at slightly reduced efficiencies ( Figure 2—figure supplement 1B ) . We then mixed labeled cohesin , cohesin loader , a 3 kb circular double stranded plasmid DNA and ATP in the indicated combinations . To create the gripping state , we included all components but substituted ATP for the non-hydrolyzable nucleotide ground state mimetic ADP · BeF3- . Following Dy547 excitation , we measured the relative FRET efficiency , defined as the Alexa 647 emission divided by the sum of Dy547 and Alexa 647 emissions . We first recorded FRET between the fluorophore pair at the Smc1Psm1 hinge and the Scc3Psc3 C-terminus . The FRET efficiency measured with the cohesin complex alone was 0 . 22 and displayed only negligible changes following the addition of one or more of the different cofactors . Even under conditions of gripping state formation , the FRET efficiency remained unchanged ( Figure 2C ) . As a control , we prepared a mixture of singly Smc1Psm1 hinge and singly Scc3Psc3 C-terminus labeled cohesin complexes . This mixture provides a baseline for the apparent background FRET value due to spectral overlap . At 0 . 17 the measurement remained substantially below the FRET values observed when both fluorophores were incorporated within the same cohesin complex . This observation supports the idea that the SMC hinge and Scc3Psc3 lie in proximity of each other to form an Scc3-hinge module , consistent with biochemically observed Scc3Psc3-hinge binding ( Murayama and Uhlmann , 2015 ) . Module formation was observed under all tested conditions , irrespective of the stage during cohesin’s ATP binding and hydrolysis cycle . Next , we investigated the positioning of the Scc3-hinge module relative to the Scc2-head module . We first measured FRET between a donor fluorophore at the Scc2Mis4 ( N191 ) N-terminus and an acceptor fluorophore at the Smc1Psm1 hinge . We observed FRET at relatively low values under most conditions . Strikingly , the FRET efficiency markedly increased under conditions of gripping state formation ( Figure 2C ) . This observation confirms that the Scc3-hinge and Scc2-head modules come into proximity in the ATP-bound gripping state , as seen in the cryo-EM structures . We also measured FRET between the Scc2Mis4 ( N191 ) N-terminus and an acceptor fluorophore at the Scc3Psc3 C-terminus . Again , FRET showed a relative increase under conditions of gripping state formation . The absolute FRET efficiency at this fluorophore pair remained lower when compared with the acceptor fluorophore at the Smc1Psm1 hinge . This is expected from a longer predicted Euclidean distance between Scc2Mis4 ( N191 ) and the Scc3Psc3 C-terminus in the gripping state , compared to Scc2Mis4 ( N191 ) and the Smc1Psm1 hinge ( Figure 2A ) . Together , these observations suggest that the Scc3-hinge and Scc2-head modules come close to each other in the gripping state but separate from each other in other conditions . When using Scc2Mis4 ( N191 ) as the FRET donor , its transitory interaction with cohesin becomes a confounding factor . Higher FRET efficiency in the gripping state could have been due to increased cohesin-Scc2Mis4 ( N191 ) complex formation , rather than a conformational change . To examine this possibility , we monitored the cohesin-Scc2Mis4 ( N191 ) interaction by co-immunoprecipitation . This revealed equal interaction efficiencies under all of our incubation conditions ( Figure 2—figure supplement 2 ) . Therefore , the observed FRET differences cannot be explained by different cohesin-Scc2Mis4 ( N191 ) complex stabilities . Rather , the FRET changes indeed point to conformational transitions within the cohesin complex . What are the consequences of the Scc3-hinge module , and its movement relative to the Scc2-head module , for the DNA trajectory during topological DNA entry ? Our earlier results suggested that DNA arrives from the top of the ATPase heads and usually passes the kleisin N-gate before reaching the gripping state ( Figure 3A , panel a ) ( Higashi et al . , 2020 ) . The kleisin N-gate initially opens as the consequence of ATP-dependent SMC head engagement ( Muir et al . , 2020 ) . A positively charged kleisin N-tail then guides DNA through this gate en route to the gripping state . In the gripping state , the DNA together with the Scc2Mis4 cohesin loader shut the gate , while the Scc3-hinge module docks onto the Scc2-head module . The straight DNA path through both DNA binding modules in turn requires that the DNA bends where it arrives between the Smc1Psm1 and Smc3Psm3 coiled coils . The DNA path shown in Figure 3A , panel a , highlights the position of the bend , based on our DNA-protein crosslink mass spectrometry data ( Higashi et al . , 2020 ) . The notion of DNA bending in the gripping state finds further support from magnetic tweezer experiments , in which condensin introduced a discrete DNA shortening step under gripping state conditions ( Ryu et al . , 2020b ) . A stable DNA gripping state forms only in the presence of non-hydrolyzable ATP . Usually , gripping state formation triggers ATP hydrolysis , resulting in ATPase head gate opening and Scc3-hinge and Scc2-head module uncoupling . This uncoupling allows a swinging motion of the Scc3-hinge module and proximal coiled coil , with a pivot point at the elbow ( Figure 3A , panel b ) . No force needs to be transmitted along the SMC coiled coil for this swinging motion to initiate . Rather , following release , Brownian motion can take the Scc3-hinge module only in one direction , away from the Scc2-head module . When we consider the consequence of the Scc3-hinge swinging motion on the DNA path , we make two observations . Firstly , the bent DNA straightens , an effect that might favor the swinging motion . Secondly , the movement effectively steers the DNA through the ATPase head gate to complete topological entry into the cohesin ring . Following head gate passage , we expect that DNA retains Scc3-hinge module association only for a limited time . DNA affinity to Scc3Psc3 and the kleisin middle region in isolation has been measured at around 2 μM ( Li et al . , 2018 ) , a relatively low affinity that implies a fast off-rate once Scc3Psc3 has left the gripping state . DNA consequently finds itself in a cohesin chamber delineated by Scc3Psc3 , the Smc3Psm3 coiled coil , as well as the unstructured part of the kleisin between the kleisin N-gate and the kleisin middle region ( Figure 3A , panel c ) . We refer to this space as cohesin’s Scc3-Smc3-kleisin-N chamber . Two separase recognition sites in Scc1Rad21 , whose cleavage liberates DNA from the cohesin ring to trigger anaphase ( Tomonaga et al . , 2000; Uhlmann et al . , 2000 ) , are situated within this part of the kleisin unstructured region ( Figure 3—figure supplement 1A ) . Single molecule imaging of cohesin , topologically loaded onto DNA , showed that its diffusion is blocked by obstacles smaller than those expected to be accommodated by cohesin’s SMC compartment ( Davidson et al . , 2016; Kanke et al . , 2016; Stigler et al . , 2016 ) . This observation is consistent with the possibility that DNA resides in a sub-chamber of the cohesin ring following topological loading . How durable the Scc3Psc3-SMC hinge association is , whether DNA permanently resides inside the Scc3-Smc3-kleisin-N chamber , or whether subunit rearrangements take place following successful topological loading , for example when the cohesin loader is replaced by Pds5 , remains to be further ascertained . The structured components of the gripping state do not by themselves contain information about whether DNA has in fact passed the kleisin N-gate . While mechanisms are in place to ensure kleisin N-gate passage , for example the kleisin N-tail , DNA might under certain conditions reach the gripping state without having passed this gate ( Figure 3B , panel a ) . What will be the consequence of ATP hydrolysis in such an alternative gripping state ? The Scc2-head module turns into its DNA slipping state , but the kleisin path prevents DNA from passing between the ATPase heads . The Scc3-hinge module again uncouples from the Scc2-head module , but now its diffusion-driven swinging motion cannot steer DNA through the head gate . The only way for the Scc3-hinge module to launch its swinging motion is to further bend the DNA , turning it into a loop , while DNA slips through the Scc2-head module ( Figure 3B , panel b ) . The directed diffusive motion of the Scc3-hinge module has created a Brownian ratchet , allowing DNA motion only in one direction ( Figure 3B , panel c ) . The entropy gain from gripping state disassembly in turn helps to offset the energetic cost of DNA loop formation . Once a DNA loop is initiated , the extent of loop growth per reaction cycle is limited by how far the Scc3-hinge and Scc2-head modules separate from each other . The maximum separation is likely dictated by the kleisin unstructured regions that link Scc3Psc3 to the Scc2-head module . Their lengths of 135 amino acids ( between the Scc2Mis4 and Scc3Psc3 binding sites ) and 109 amino acids ( between the Scc3Psc3 binding site and the kleisin C-terminal domain ) gives a conservative estimate of ~40 nm ( Figure 3—figure supplement 1B; Ainavarapu et al . , 2007 ) . This distance allows considerable , but perhaps not complete , extension of the ~47 nm long SMC proteins . As we will see below , the actual amount of loop growth is likely less and depends on the SMC elbow angle reached by stochastic diffusive motion at the time when DNA dissociates ( Figure 3B , panel d ) . After DNA dissociation from the Scc3-hinge module , there is a time when there is only loose cohesin-DNA contact with the Scc2-head module . Thermal fluctuations now lead to random loop size changes , depending on the probability of diffusion and on external forces that might apply . As long as the Scc2Mis4 cohesin loader remains part of the Scc2-head module , the local proximity of all components means that a return of the Scc3-hinge module ( Figure 3B , panel e ) and the establishment of a new DNA gripping state following nucleotide binding ( Figure 3B , panel f ) are very likely events . The next loop extrusion cycle begins . The above model makes a specific molecular proposal for DNA loop extrusion by the cohesin complex . At its core lies a Brownian ratchet built from the Scc3-hinge and Scc2-head modules , juxtaposed during gripping state formation but allowing unidirectional DNA diffusion following ATP hydrolysis . This robust core of a loop extrusion mechanism tolerates variations in its surrounding features . One point of uncertainty regards the Scc3Psc3 interaction with the SMC hinge . While our FRET observations suggest proximity of Scc3Psc3 and the SMC hinge under all tested solution conditions , others have observed Scc3Psc3 close to the ATPase heads and distant from the hinge ( Anderson et al . , 2002; Huis in 't Veld et al . , 2014 ) . It is possible that the Scc3Psc3-hinge interaction has a limited lifetime following gripping state disassembly . As long as Scc3Psc3 retains hinge association during initial SMC coiled coil unfolding , the Brownian ratchet has served its purpose . Scc3Psc3 might then dissociate from both the DNA and the hinge without impacting on loop extrusion efficiency ( Figure 3—figure supplement 1C ) , as long as all subunits rejoin during gripping state reassembly . Another open question in our molecular proposal is whether DNA indeed fails to pass the kleisin N-gate before the initiation of loop extrusion . The finding that loop extrusion is not prevented by covalent fusions between cohesin ring subunits ( Davidson et al . , 2019 ) does not strictly answer the question whether or not DNA passes the kleisin N-gate , which lies at a distinct distance from the kleisin N-terminus . If DNA did pass the kleisin N-gate , as it would during topological entry into the cohesin ring , the kleisin path would not obstruct DNA exit from the Scc2-head module . Instead , under the low-salt conditions typical for loop extrusion experiments , DNA might retain electrostatic contact with the positive charges of the Scc2-head module in the slipping state ( Figure 3—figure supplement 1D ) . Whether or not such interactions are sufficient to sequester DNA within the Scc2-head module during loop extrusion remains to be learned . The above molecular model for loop extrusion is based on experiments using the fission yeast cohesin complex and its loader . To date , loop extrusion by cohesin has only been observed using human proteins , while attempts to observe loop extrusion by budding yeast cohesin have remained unsuccessful ( Ryu et al . , 2021 ) . We therefore asked whether fission yeast cohesin is able to extrude DNA loops . We employed a single-molecule assay similar to those previously used to observe DNA loop extrusion by budding yeast condensin and human cohesin ( Ganji et al . , 2018; Davidson et al . , 2019 ) . Individual molecules of λ-phage DNA were tethered via both ends to a cover glass surface of a microfluidic flow cell , stained with Sytox Orange and stretched by continuous buffer flow while being imaged using total internal reflection ( TIRF ) microscopy . In the presence of 5 nM fission yeast cohesin , the cohesin loader and ATP , over 40% of the DNA molecules were seen to display active DNA loop extrusion ( Figure 4A , Video 1 ) . Loop extrusion proceeded in a symmetrical manner at a mean rate of ~1 kbp s−1 ( maximal rate 2 . 4 kbp s−1 , Figure 4B ) resembling DNA loop extrusion by human cohesin ( Davidson et al . , 2019; Kim et al . , 2019 ) . This confirms that fission yeast cohesin can both , topologically load onto DNA and extrude DNA loops . In our proposed model of loop extrusion , two DNA binding modules within the cohesin complex generate a Brownian ratchet . The ratchet is operated by repeated cycles of ATP-dependent DNA gripping state formation and its unidirectional dissolution following ATP hydrolysis . To evaluate whether such a mechanism is physically plausible , we constructed a structure-based molecular-mechanical model of the cohesin-DNA interaction and carried out computational simulations to explore its behavior . We modeled DNA as a discrete stretchable , shearable wormlike chain ( dssWLC ) , which describes DNA with persistence length as the only parameter ( Figure 5A; Koslover and Spakowitz , 2014 ) . We assumed the persistence length to be Lp = 50 nm ( Wang et al . , 1997; Bustamante et al . , 2000 ) . The cohesin coiled coil segments as well as the linkage between the two SMC heads were modeled using the same approach . Each coiled coil was represented as three beads that interact via a dssWLC ( Figure 5B ) . This again leaves persistence length as the sole parameter that we chose such that it leads to a head-to-hinge distance distribution that matches experimentally measured head-to-hinge distances in a freely diffusing eukaryotic SMC complex ( Ryu et al . , 2020a ) . Based on our structural and biochemical observations , we define two states of the cohesin complex . In the gripping state , the Scc3-hinge and Scc2-head modules are engaged and the coiled coil elbows are folded . Both modules make stable contact with DNA ( Figure 5C , Gripping state ) . In the second state , the slipping state , the Scc3-hinge and Scc2-head modules do not interact , allowing an unfolded cohesin conformation . In this state , the Scc2-head module permits free transverse DNA motion . DNA association with the Scc3-hinge module , defined by its equilibrium dissociation constant , is manually controlled in our model ( Figure 5C , Slipping state ) . First , we explored the dynamics of the transition between cohesin’s gripping and slipping states . As a starting point we assume that a small DNA loop is inserted into the cohesin ring in the gripping state . The first panel in Figure 5D shows a snapshot of this initial state after equilibration by the Metropolis Monte-Carlo algorithm ( see Materials and methods ) . Our 3D model has no explicit chemical kinetics , and to simulate transitions between chemical states we prescribed parameter changes corresponding to a new state , then sampled a sufficient number of iterations to reach a new equilibrium . To simulate transition to the slipping state , we impose parameter changes that disconnect the Scc3-hinge from the Scc2-head module and switch the Scc2-head module to its slipping state , while the Scc3-hinge module remains bound to DNA . We then sampled conformations with the new parameters until a new equilibrium was reached . As cohesin unfolds , DNA binding to the Scc3-hinge module limits DNA movement at the Scc2-head module to only one direction , toward an increased loop size ( Figure 5D ) . The average increase in loop size during multiple repeats of this transition is ~30 nm ( Figure 5E ) . When we then prescribe DNA detachment from the Scc3-hinge module and switch cohesin back to the gripping state , the system readily resets and primes itself for the next cycle ( Figure 5D ) . Our simulations reveal that repeated rounds of the states: ‘gripping - > slipping - > DNA detachment from the Scc3-hinge module - > gripping’ results in continuous extrusion of DNA with an average loop size increase of ~30 nm per cycle ( Video 2 ) . In the above computational model , all state transitions were manually prescribed to obtain cycles of DNA loop extrusion . In the following , we consider requirements for such cycles to arise based on the chemical kinetics of the cohesin complex . We assume , based on the structural data , that both the Scc2-head and Scc3-hinge modules bind DNA in the ATP-bound gripping state . Following ATP hydrolysis , the Scc2-head module switches to its slipping state , while DNA remains bound to the Scc3-hinge module . To achieve processive cycles of loop extrusion , DNA binding must persist for long enough to ensure biased DNA diffusion toward loop growth while cohesin unfolds ( Figure 5F ) . An upper limit for the time it takes cohesin to unfold is given by the time of a diffusive process that separates the Scc3-hinge and Scc2-head modules . Assuming molecular masses of both modules in the 200 kDa range , it takes ~0 . 1 ms for them to diffuse ~50 nm apart ( see Materials and methods ) . This time is an upper estimate . If cohesin opening was driven not merely by diffusion , but assisted by internal stiffnesses of the coiled coils , this could speed up opening . Based on this estimate , our model predicts that the DNA off-rate at the Scc3-hinge module should be lower than 1/0 . 1 ms = 10 , 000 s−1 in order for DNA to maintain Scc3-hinge module association until cohesin fully unfolds . After cohesin has opened , two further scenarios are possible . Firstly , DNA could dissociate from the Scc3-hinge module before cohesin transitions back into the next gripping state . In this case a loop length gain is made and the ensuing gripping state primes cohesin for the next round of loop extrusion ( Figure 5F ) . Alternatively , cohesin could switch back to the gripping state before DNA is released from the Scc3-hinge module . In this situation , DNA ends up in the same position as before and there is no net loop size gain , resulting in an unproductive cycle . Based on these considerations , loop extrusion in our model is most effective when the DNA lifetime at the Scc3-hinge module is longer than the time required for diffusion-driven cohesin unfolding , but shorter than it takes cohesin to transition back into the next gripping state . Such a lifetime would ensure that most reaction cycles result in net loop growth . The ATP-bound DNA gripping state is an unstable state , so we can expect cohesin to spend the majority of its time in the post-hydrolysis slipping state . We can then approximate the lifetime of the slipping state based on cohesin’s ATP hydrolysis rate . This rate has been measured with a lower limit of ~ 2 s−1 ( Davidson et al . , 2019; Ganji et al . , 2018; Murayama and Uhlmann , 2014 ) . As two ATPs are coordinately hydrolyzed by the two ATPase heads , this equates to a cycle rate of ~ 1 s−1 . Thus , our model predicts that efficient loop extrusion is achieved when the DNA off-rate from the Scc3-hinge module is in the range of 1 – 10 , 000 s−1 . While we do not know the actual off-rate , the equilibrium dissociation constant Kd between DNA and Scc3 has been measured at ~ 2 μM ( Li et al . , 2018 ) . Assuming an association rate kon typical for biomolecular interactions of ~ 107 M−1 s−1 ( Howard , 2001 ) , we arrive at a corresponding off-rate koff=Kd×kon of ~ 20 s−1 . This value sits well within the range predicted to support processive loop extrusion . The number ensures processivity even if cohesins that are actively engaged in loop extrusion undergo conformational cycles and hydrolyze ATP up to twenty times faster than measured in bulk solution experiments . Having established that transitions between cohesin’s gripping and slipping states can drive directional DNA movements , we explored how this mechanism compares to available experimental observations of loop extrusion at realistic time scales . To do this , we constructed a simplified model of loop development . We assume that both DNAs that enter cohesin at the Scc2-head module and exit cohesin at the Smc3Psm3 elbow can randomly diffuse in and out of the ring with rates depending on a DNA diffusion coefficient D . We then use a Monte-Carlo algorithm to simulate DNA loop dynamics as a function of time . If we adopt a diffusion coefficient of ~1 μm2/s , as measured for cohesin movements on DNA following topological loading ( Davidson et al . , 2016; Kanke et al . , 2016; Stigler et al . , 2016 ) , we see that both strands randomly diffuse back and forth , leading to stochastic loop size changes ( Figure 6A ) . Within a few minutes , a typical time frame used to microscopically observe DNA loop extrusion , the loop size changes over a range of several kilobases . However , these random movements do not show a preferred direction and cannot drive loop extrusion . This situation changes when the Scc3-hinge module engages with DNA in the gripping state and disengages predominantly in the slipping state . The Scc3-hinge module restricts DNA diffusion at the Scc2-head module to only one direction – toward loop growth . This effect applies only to the DNA that enters the loop at the Scc2-head module , but not to the DNA that exits cohesin . We assume that the latter DNA continues to diffuse randomly in both directions irrespective of the cohesin state . If we simulate directed DNA motion at the Scc2-head module of 30 nm per cohesin turnover cycle , we see how , overlaid over stochastic diffusive loop size fluctuations , the loop length steadily increases over time ( Figure 6B ) . We next explored how the variables in this model affect the outcome of loop extrusion . There are three independent variables: the two diffusion coefficients that describe the two DNAs that enter and exit cohesin , as well as the ATPase turnover rate , that is the lifetime of each slipping state . We simulated multiple 10 min intervals of cohesin-DNA dynamics and compared loop extrusion rates extracted from these simulations to those determined in our and in published experiments ( Davidson et al . , 2019 ) . The simulations revealed that the average loop extrusion rate is unaffected by the DNA diffusion coefficients ( Figure 6C ) . Indeed , thermal movement of DNA has no net direction and therefore should not contribute to directed loop growth . Instead , the average loop extrusion rate v is simply a product of the step size during cohesin’s state transitions L and the frequency γ of these events: ( 1 ) v=γ*L Using the value of L = 30 nm = 0 . 088 kb , we find that there must be around 10 successful cohesin state transitions per second to reach experimentally observed average loop extrusion speeds of ~ 1 kb s−1 . The required rate of cohesin state transitions necessitates an equal rate of ATPase cycles . This means that a cohesin complex that is actively engaged in loop extrusion hydrolyzes ATP ~ 10 times faster than average bulk solution ATP hydrolysis rates suggested . A striking feature of experimentally observed loop extrusion is a high variation in loop growth rates ( Figure 4B; Ganji et al . , 2018; Davidson et al . , 2019 ) . To obtain insight into the origin of these variations , we compared scatter in our modeled traces with experimental data . We quantified the scatter as the interquartile range , that is the range that contains 50% of datapoints around the median . This analysis revealed that extrusion rate variations strongly correlated with the DNA diffusion coefficient . The bigger the diffusion coefficient , the greater is the variation ( Figure 6D ) . Of the two DNAs that enter and exit cohesin , additional simulations showed that only the DNA with the higher diffusion coefficient determines the amount of scatter in extrusion speed ( Figure 6—figure supplement 1 ) . A diffusion coefficient of ~1 . 5 µm2/s resulted in a good match to the experimentally observed variation ( Figure 6E ) , matching the upper range of experimentally measured values ( Davidson et al . , 2016; Kanke et al . , 2016; Stigler et al . , 2016 ) . We imagine that the outward pointing DNA , which is not constrained by the Scc2-head module , might not interact strongly with cohesin and show the greater diffusion coefficient amongst the two DNAs . In addition to the high variations of loop extrusion rates , the low friction of the outward pointing DNA could also explain why DNA can be readily pulled from a condensin complex undergoing loop extrusion ( Kim et al . , 2020 ) . Finally , we explored how Brownian ratchet-driven loop extrusion in our model is affected by external force . If cohesin unfolding in the slipping state is driven by thermal motion , its rate k in response to external force is given by: ( 2 ) k=koe-δ∙FkBTwhere k0 is the rate in the absence of external force , F is the external force and δ = 30 nm from our simulations ( Howard , 2001 ) . Introducing this dependency into our model , we find good agreement between simulations in the presence of a range of applied external forces and the experimentally observed decay of the force-velocity relationship ( Figure 6F; Ganji et al . , 2018 ) . Both the similarity between the simulated and experimentally observed responses to external force , as well as the high variation of experimentally observed loop extrusion rates , support the idea of a largely diffusion-driven molecular mechanism of loop extrusion . In our molecular model of DNA loop extrusion , the Brownian ratchet acts only on the DNA that enters the cohesin ring through the Scc2-head module . No directional effect is exerted on the DNA that exits cohesin . This results in asymmetric loop extrusion ( Figure 7A , Asymmetric loop extrusion ) , a scenario that is seen in the case of the condensin complex ( Ganji et al . , 2018; Golfier et al . , 2020 ) . In contrast , both our and the published experimental observations ( Davidson et al . , 2019; Kim et al . , 2019 ) suggest that the cohesin complex symmetrically extrudes DNA loops . How could this difference be explained ? In our model , the cohesin loader is a stable part of the cohesin complex . However , Scc2Mis4 only weakly binds to the cohesin complex . Suggestive of subunit turnover , the continuous presence of cohesin loader in the incubation buffer is a requirement for processive loop extrusion by human cohesin ( Davidson et al . , 2019 ) . If we picture a situation in which Scc2Mis4 dissociates from the cohesin complex , DNA will be released from the Scc2-head module ( Figure 7A , Symmetric loop extrusion ) . The DNAs that enter and exit the cohesin ring are now indistinguishable and , once Scc2Mis4 rebinds , both DNAs have an equal chance to associate with the Scc2-head module during gripping state formation . Every round of cohesin loader dissociation and reloading thereby results in a one-in-two chance that the extruded DNA strand switches . Averaged over time , this takes the appearance of symmetric loop extrusion . In addition to loop extrusion , single-molecule studies have reported ATP-dependent unidirectional cohesin and condensin translocation along DNA ( Terakawa et al . , 2017; Davidson et al . , 2019 ) . The experimental conditions under which both complexes move along DNA , or extrude DNA loops , are largely similar . A difference lies in the DNA substrates on which translocation was observed . These substrates were stretched , either by liquid flow or by being double tethered to a flow cell surface . We have seen above that a Brownian ratchet is able to extrude DNA loops only against very small externally applied forces ( Figure 7B , Loop extrusion ) . If DNA is longitudinally stretched , loop extrusion will thus be limited to a small loop size . As the Brownian ratchet continues to deliver DNA to enlarge the loop , the stretching force begins to extract the DNA from the opposite side . This will especially be the case if , as suggested above , the diffusion coefficient of the outward pointing DNA is higher than that of the inward moving DNA ( Figure 7B , Loop translocation ) . Instead of promoting loop growth , the Brownian ratchet now fuels a motor that moves along the DNA . Consistent with this scenario , a small DNA loop was observed to precede initiation of condensin translocation along stretched DNA ( Ganji et al . , 2018 ) . We can imagine an alternative scenario by which cohesin could turn into a Brownian ratchet-driven motor . Following successful topological loading onto DNA , cohesin might be able to return to a gripping state , for example if Scc2Mis4 is not replaced by Pds5 . This scenario finds support from the observation that the cohesin loader retains the ability to stimulate cohesin’s ATPase following completion of topological loading ( Çamdere et al . , 2015 ) . Repeated gripping to slipping state transitions could then result in cohesin translocation along DNA ( Figure 7B , Cohesin translocation ) . This second model for directed cohesin movement is not mutually exclusive with the ‘loop translocation’ model . Both models make the prediction that , similar to what is observed during loop extrusion , cohesin is a weak translocating motor that can be stalled by very small forces . DNA loop extrusion by cohesin and condensin has so far only been observed in vitro and only using naked DNA substrates . In vivo , DNA is densely decorated by histones and other DNA binding proteins related to transcription , DNA replication and other forms of DNA metabolism . If we portray a loop extruding cohesin complex in its slipping state next to a nucleosome ( Figure 8A , left ) , it becomes apparent that a nucleosome is too big to pass through the channel formed between Scc2Mis4 and the SMC ATPase heads . A possible path for nucleosome bypass opens up when the Scc2Mis4 cohesin loader transiently dissociates . DNA now passes in and out of cohesin through the Scc3-Smc3-kleisin-N chamber ( Figure 8A , Nucleosome bypass , and Figure 8—figure supplement 1A ) . If Scc3Psc3 disengages from the SMC hinge in the slipping state , the clearance available for obstacle bypass further increases ( Figure 8—figure supplement 1B ) . In the case of topologically loaded cohesin , the same channel is in principle wide enough to allow nucleosome bypass , albeit denser nucleosome arrays block purely diffusive cohesin sliding ( Stigler et al . , 2016 ) . During loop extrusion , on one hand , Brownian ratchet-driven directional cohesin movement will facilitate nucleosome bypass . On the other hand , only one DNA lies in the Scc3-Smc3-kleisin-N chamber following topological loading , but both in and outward pointing DNAs must be accommodated during loop extrusion . If both DNAs are histone-bound , especially if these histones form part of higher order structures , considerable steric constraints will be encountered that likely slow down or stop loop extrusion . An alternative outcome of cohesin-nucleosome collisions is therefore that loop extrusion at least temporarily stalls . As the cohesin loader dissociates , DNA is released from the Scc2-head module and free to move within the Scc3-Smc3-kleisin-N chamber . As Scc2Mis4 rejoins the complex , there is a new chance for kleisin N-gate passage during gripping state formation , as originally foreseen during topological loading ( Figure 8A , Kleisin N-gate passage ) . If this occurs , the DNA loop resolves following ATP hydrolysis , resulting in cohesin topologically embracing one DNA . There might be yet another possible outcome of cohesin-nucleosome encounters . Given the thrust of a diffusion-mediated collision between cohesin and an obstacle , the closed kleisin N-gate might rupture ( Figure 8A , Kleisin N-gate rupture ) . If this happens , the extruded DNA loop will again resolve , resulting in cohesin topologically embracing one DNA . One can imagine that frequent stalling on a nucleosome-dense chromatin template prevents processive loop extrusion . The encountered obstacles could be seen as triggering a ‘proofreading’ mechanism , prompting recurring attempts at kleisin-N gate passage , eventually resulting in successful topological cohesin loading . Chromosomal cohesin loading sites lie at considerable distances from each other ( Schmidt et al . , 2009 ) . Nevertheless , we cannot outright dismiss the possibility that loop extruding cohesin complexes might collide in head-on encounters . In vitro observations of colliding , loop-extruding condensins have revealed that these complexes are able to traverse each other to form distinctive z-loop structures ( Kim et al . , 2020 ) . Can this behavior be explained by our molecular model of SMC complex function ? Upon encounter , condensins have been observed to pause for a period of time . This is consistent with the behavior of two Brownian ratchets that collide ( Figure 8B , Loop collision ) . A way out of this conflict is provided by one of the above-mentioned loop resolution pathways , kleisin N-gate passage or N-gate rupture . Both allow one of the colliding condensins to resolve their loop and turn into a topologically loaded condensin ( Figure 8B , Loop resolution ) . The newly gained freedom of movement of the now singly tethered condensin allows it to diffuse . If , akin to how cohesin entraps a second single-stranded DNA ( Murayama et al . , 2018 ) , condensin engages with a second double-stranded DNA that lies beyond the colliding condensin complex , a z-loop is formed ( Figure 8B , second DNA capture ) . Various outcomes have been observed during z-loop formation in vitro , including one- and two-sided z-loop growth . It is possible to explain both behaviors if we assume that second DNA capture during z-loop formation results in either topological capture of the second DNA or results in the second DNA entering loop extrusion mode . Any model of cohesin function must explain how the energy from ATP binding and hydrolysis is used to fuel or regulate its activities . ATP binding leads to SMC head engagement and to a conformational change at the Smc3Psm3 neck that favors kleisin N-gate opening ( Higashi et al . , 2020; Muir et al . , 2020 ) . However , kleisin N-gate opening might not be hard-wired and it is possible that head engagement occurs sometimes without kleisin N-gate opening ( see below ) . Head engagement creates a composite DNA binding surface on top of the ATPase heads . The next steps in cohesin’s reaction cycle are now likely driven by the binding energy of DNA itself . First , DNA establishes contact with the extensive positively charged surface on the ATPase heads . Next , the DNA engages Scc2Mis4 , which turns into its gripping state conformation as its own positive charges embrace the DNA . Together , the DNA and cohesin loader also establish contact with and close the kleisin N-gate . The DNA-induced Scc2Mis4 conformational change creates a docking interface for Scc3Psc3 . The latter subunit joins the complex by concurrently binding the DNA , as well as by recruiting the SMC hinge . The binding energy released from these additional interactions compensates for the energetic cost of introducing a DNA bend , required to reach this configuration . Assembly of the energy-loaded gripping state is now complete; DNA has entered the cohesin ring through the kleisin N-gate . ATP hydrolysis now ensues , which dissolves the gripping state . However , dissolution equates to more than mere dispersal of the components . The geometric arrangement of the Scc3-hinge and Scc2-head modules means that diffusion-driven gripping state dissolution generates a swinging motion of the Scc3-hinge module that guides DNA through the head gate to complete topological entry into the cohesin ring . If DNA did not pass the kleisin N-gate , all other considerations for gripping state assembly remain unchanged . Again , gripping state dissolution initiates Brownian Scc3-hinge swinging motion . However , the kleisin prevents head gate passage , instead resulting in DNA slippage along the Scc2-head module and the initiation of a DNA loop . Based on the observed DNA path , DNA binding to the Scc3-hinge and Scc2-head modules has already introduced a roughly 120° DNA bend in the gripping state . This bend must mature into a 180° turn that fits through the approximately 23 nm clearance of the Scc3-Smc3-kleisin-N chamber . DNA spontaneously adopts narrow bends ( Vafabakhsh and Ha , 2012 ) and DNA thermal fluctuations might suffice for loop extrusion to commence , possibly after a certain delay . We cannot exclude that additional mechanisms or DNA-protein contacts contribute to loop initiation , a process that remains to be experimentally examined . Once a DNA loop is formed , its extension is energetically favorable as the loop radius increases . The energetic cost of reducing the loop radius again decreases the chance that a DNA loop slips back , once formed . A key feature that determines the processivity of the Brownian ratchet is the half-life of the Scc3-hinge module interaction with DNA following the gripping to slipping state transition . This interaction should last long enough to guide directed diffusion but short enough so that DNA is released before the next gripping state forms . Given the stochastic nature of Scc3-hinge module dissociation from , and possible re-association with the DNA , we expect that not all loop extrusion cycles result in productive loop size gain , a fact that might contribute to the wide spread of observed loop extrusion rates . The corresponding component of the Scc3-hinge module in the condensin complex is its putative Ycg1-hinge module . Experiments with condensin harboring a DNA binding site mutation in the Ycg1-hinge module revealed greatly compromised loop extrusion ( Ganji et al . , 2018 ) , consistent with an important role of this element . Biological motors typically couple ATP-dependent conformational changes to their motion , allowing for robust movements in the presence of counteracting forces ( Howard , 2001 ) . We cannot exclude the possibility that cohesin also uses an ATP-dependent mechanism to control the relative positions of the Scc3-hinge and Scc2-head modules . For example SMC coiled coil stiffness could store torsional energy during gripping state formation that is released following ATP hydrolysis to add a power stroke to loop extrusion . However , SMC coiled coils appear very flexible ( Eeftens et al . , 2016; Ryu et al . , 2020a ) and as yet there is no evidence for energy coupling between the ATPase heads and hinge . Our computational simulations suggest that a purely diffusion-driven Brownian ratchet recapitulates experimental observations well . In this scenario , the energy from ATP binding and hydrolysis operates a Brownian ratchet , while thermal energy moves the DNA . Bulk solution measurements suggested that cohesin undergoes on average approximately one ATP hydrolysis cycle per second in the presence of the cohesin loader and DNA . In contrast , our simulations predict that ATP hydrolysis cycles must happen an order of magnitude faster for the Brownian ratchet to reach experimentally observed loop extrusion speeds . These observations are not necessarily incompatible if complex formation between cohesin , the cohesin loader and DNA is a rate-limiting step . In a typical bulk cohesin loading reaction , approximately 20% of DNA is captured by cohesin within an hour of incubation ( Murayama and Uhlmann , 2014 ) . This illustrates that gripping state formation is likely a slow process in solution . ATP hydrolysis rates might well be substantially greater than average , once a protein-DNA assembly has formed . A conserved kleisin N-tail interacts with DNA to ensure kleisin N-gate passage during gripping state formation ( Higashi et al . , 2020 ) . Why then might kleisin N-gate passage sometimes fail , resulting in loop extrusion ? A potentially relevant observation is that the biochemical reconstitution of both topological cohesin loading onto DNA ( Murayama and Uhlmann , 2014 ) , as well as DNA loop extrusion ( Ganji et al . , 2018 ) , are helped by unphysiologically low-salt concentrations . Electrostatic interactions between DNA and cohesin , which characterize the gripping state , are stronger when less salt competes with them in the incubation buffer . While enhanced DNA-protein contacts likely promote the biochemical reactions , the low ionic strength will also affect protein-protein interactions . For instance , electrostatic interactions contribute to keeping the kleisin N-gate shut , which are augmented in a low salt buffer and could impede N-gate opening . The boosted efficiency of gripping state formation at low-salt concentrations might thus come at the cost of an increased fraction of failed topological entry events . Even if kleisin N-gate passage is disfavored in a low-salt environment , recurrent cohesin loader dissociation and reassociation events during loop extrusion might eventually permit successful N-gate passage . This outcome would result in topological cohesin loading onto DNA and loop resolution , similar to what we envision might happen when loop extrusion meets an obstacle . How easily the fate of DNA with respect to the kleisin N-gate changes once loop extrusion has started will be important to explore . How does our Brownian ratchet model compare to previously proposed models for DNA loop extrusion ? The first proposed , tethered inchworm model ( Nichols and Corces , 2018 ) , also features the two HEAT repeat subunits as DNA binding elements that perform a scissoring motion while remaining connected by a flexible kleisin linker . Instead of forming head and hinge modules , the HEAT subunits associate with one of the two ATPase heads , each . The HEAT subunit DNA affinity is postulated to change during the ATP hydrolysis cycle such that force from ATP head engagement and resolution is transferred to move DNA . The SMC hinge in turn is tacitly assumed to act as a second DNA anchor , thereby achieving asymmetric DNA loop extrusion . The authors suggest that obstacle bypass is possible in a stepping motion , as the HEAT subunits alternatingly reduce their DNA affinity . In retrospect , the tethered inchworm model shares several components of our Brownian ratchet . A main difference is that its deterministic nature should achieve a defined loop extrusion speed , even against small counteracting forces . The DNA segment capture model makes use of one major DNA binding site at the ATPase heads ( Marko et al . , 2019 ) , not too dissimilar to the Scc2-head module of our Brownian ratchet . A key difference is that the head module in the segment capture model exerts a power stroke that alters the angle at which the DNA intersects with the SMC complex , thereby initiating DNA looping or promoting loop growth . After the power stroke , DNA affinity at the head module changes following ATP hydrolysis , again not too dissimilar to our Brownian ratchet . Evidence for the proposed power stroke , as well as for the required mechanism by which DNAs change place between hinge and heads between cycles , remains to be sought . Finally , the scrunching model is similar to our Brownian ratchet in that thermal fluctuations between DNA binding sites at the head and hinge form the basis for loop formation and loop growth ( Ryu et al . , 2020a ) . However , the DNA trajectories in the two models show opposite polarity . The scrunching model foresees DNA capture by the SMC hinge in the unfolded state and DNA release in the folded state . This requires at least two regulated DNA binding sites , in addition to which the HEAT repeat subunits are thought to form a static DNA anchor . This contrasts with our proposed ratchet where DNA is caught in the folded state and released after unfolding , and where the HEAT subunits are the two dynamic components that control DNA motion . Our model for cohesin function also makes a molecular proposal for how DNA topologically enters the cohesin ring by sequential passage through the kleisin N-gate and then the ATPase head gate in a top-down direction ( Figure 8—figure supplement 2 ) . DNA arrival from the top is experimentally supported by DNA-protein crosslink mass spectrometry data and by the fact that covalent closure of the ATPase head gate does not block gripping state formation ( Higashi et al . , 2020 ) . Despite this , an alternative model for DNA entry into the cohesin ring was proposed , in which DNA entry starts by bottom-up passage through the ATPase head gate ( Collier et al . , 2020; Shi et al . , 2020 ) . While this latter model left open the question how topological entry might be completed , an apparent argument for bottom-up DNA passage through the head gate came from experiments to locate the DNA using chemical crosslinkers . Immediately upon cohesin addition to DNA , before topological loading becomes detectable , the DNA takes up a position in which crosslinkers can trap it within what have been termed engaged-SMC and engaged-kleisin compartments ( Figure 8—figure supplement 2; Collier et al . , 2020 ) , a position that can be reached by bottom-up passage through the ATPase head gate . Our model offers an alternative explanation for this positioning . If we imagine DNA approaching from the top , between the SMC coiled coils , it might frequently pass the disengaged ATPase heads , top-down , before the relatively infrequent ATP-dependent series of head engagement and kleisin N-gate opening commences . This approach results in an equivalent DNA topology following crosslinking . Could SMC complexes have evolved to be loop extruding Brownian ratchets ? If the primordial function of SMC complexes was that of loop extruders , we should expect the loop extrusion mechanism to be conserved in evolutionary ancient SMC complexes . Our model of loop extrusion suggests that cohesin’s DNA-interacting HEAT repeat subunits are key components of the Brownian ratchet . These HEAT repeat subunits are relatively modern additions to SMC complexes . Evolutionarily older SMC complexes that are reflected in today’s prokaryotic SMC complexes , as well as in the Smc5-Smc6 complex , contain in place of HEAT subunits two smaller kleisin interacting tandem winged helix elements ( Kite ) ( Palecek and Gruber , 2015 ) . While Kite subunits interact with DNA ( Zabrady et al . , 2016 ) , they show important differences from how HEAT subunits are incorporated into SMC complexes . Kite subunits bind in close proximity to each other to a relatively short kleisin unstructured region ( Woo et al . , 2009; Bürmann et al . , 2013; Jo et al . , 2021 ) . This observation makes it hard to imagine that a similar ratchet mechanism , which requires the DNA binding modules to separate from each other , operates in SMC-kite complexes . In vitro DNA loop extrusion by SMC-kite complexes has not yet been observed , while the molecular basis for SMC-dependent chromatin proximities in B . subtilis , attributed to loop extrusion ( Wang et al . , 2018 ) , remains to be fully understood . Further biochemical investigations of SMC-Kite complexes ( Kanno et al . , 2015; Niki and Yano , 2016 ) will provide necessary insight into the question whether these complexes act by topologically loading onto DNA or by DNA loop extrusion . Cohesin topologically entraps DNA to promote sister chromatid cohesion ( Haering et al . , 2008 ) . It is possible that kleisin N-gate passage is an error-prone event and that accidental failure of kleisin N-gate passage during a topological DNA loading attempt initiates DNA looping . A close-by obstacle would soon stall loop extrusion and allow proofreading in the form of kleisin N-gate passage or N-gate rupture to reinstate topological loading . This scenario portrays loop extrusion as an unwanted , and possibly rare , side effect of topological cohesin loading . Alternatively , was it a cunning evolutionary twist that allowed cohesin to add loop extrusion to its repertoire of DNA acrobatics ? An obvious challenge to DNA loop extrusion in vivo is the presence of histones and other DNA binding proteins . Our model predicts that obstacle bypass by loop extruding SMC complexes is possible , but also that obstacles reduce the speed and processivity of extrusion , especially when present at a high density on both the in- and outward pointing DNAs . Recent studies reported DNA compaction of histone-bound DNA by cohesin and condensin , but whether SMC complexes indeed bypassed histones in these experiments is not yet known ( Kim et al . , 2019; Kong et al . , 2020 ) . DNA loop extrusion was also observed in Xenopus egg extracts , but loop extrusion was detectable only following histone depletion ( Golfier et al . , 2020 ) . Obstacle encounters during in vitro loop extrusion are an obvious and important area for experimental investigation . Many cellular DNA transactions make use of histone chaperones and chromatin remodellers to navigate the nucleosome landscape . Indeed , in vitro reconstituted chromosome assembly using purified histones and condensin depends on the histone chaperone FACT ( Shintomi et al . , 2015 ) . While FACT loosens histone-DNA interaction , this chaperone does not possess catalytic activity . FACT typically acts together with much slower-moving RNA or DNA polymerases . Whether FACT can facilitate nucleosome eviction by a Brownian ratchet remains to be explored . Other than FACT , ATP-dependent chromatin remodelers could aid in vivo loop extrusion . When studying the contribution of histone chaperones and chromatin remodelers , we have to keep in mind that topological cohesin and condensin loading onto chromosomes also requires free DNA access that these enzymes provide ( Toselli-Mollereau et al . , 2016; Garcia-Luis et al . , 2019; Muñoz et al . , 2019 ) . DNA loop extrusion by SMC complexes is a captivating molecular event that provides an at first sight intuitive explanation for chromosome loop formation . However , DNA extrusion is not the only explanation for loop formation . Cohesin and condensin could alternatively generate loops simply by sequential topological capture of two DNAs that come into proximity by Brownian motion , a mechanism that we refer to as diffusion capture ( Cheng et al . , 2015; Gerguri et al . , 2021 ) . When cohesin is depleted and re-supplied to human cells , small and large loops form with similar kinetics ( Rao et al . , 2017 ) , a behavior that is more readily explained by a diffusion-mediated process than by gradual loop growth . In addition to cohesin and condensin , weak diffusion-driven motors , chromosomes harbor abundant , strong DNA translocases in the form of RNA polymerases . These are known to push SMC complexes as they move along chromosomes during gene transcription ( Lengronne et al . , 2004; Davidson et al . , 2016; Ocampo-Hafalla et al . , 2016; Busslinger et al . , 2017 ) . We have suggested that , following loop formation by diffusion capture , RNA polymerases provide extrinsic motor activity that promotes loop expansion ( Uhlmann , 2016 ) . Such transcription-dependent extrinsic loop growth could explain chromatin domain features in an analogous fashion to cohesin moving on its own accord ( Bailey et al . , 2020 ) . The role of RNA polymerase-dependent SMC complex movements in chromosome architecture deserves further attention . Our molecular proposal for SMC complex function informs the evaluation how topological loading onto DNA and loop extrusion by SMC complexes contribute to chromosome function . While topological loading and loop extrusion share many reaction steps , the two mechanisms also differ . The next challenge will be to exploit these differences to engineer SMC complexes that can topologically load onto DNA but not loop extrude , and vice versa . This will eventually enable genetic experiments that clarify the respective physiological contributions of topological loading and loop extrusion by SMC complexes . The molecular model of cohesin in this study is based on our cryo-EM structure of the fission yeast cohesin complex together with its loader in the nucleotide-bound DNA gripping state ( PDB: 6YUF ) ( Higashi et al . , 2020 ) . A molecular model of the fission yeast SMC hinge domain was obtained based on a mouse cohesin hinge crystal structure as a template ( PDB:2WD5 ) ( Kurze et al . , 2011 ) using SWISS-MODEL ( Waterhouse et al . , 2018 ) . Scc3Psc3 with the kleisin middle region , bound to DNA , was modeled based on a crystal structure of the corresponding budding yeast components ( PDB: 6H8Q ) ( Li et al . , 2018 ) . The hinge and Scc3Psc3 were manually placed so that they align with the respective positions of the SMC hinge and Scc3SA1 in the cryo-EM structure of human cohesin in the gripping state ( PDB: 6WG3 ) ( Shi et al . , 2020 ) . The indicative position of Scc3Psc3 in the fission yeast structure , shown for comparison , was estimated based on distance constraints from a protein crosslink mass spectrometry dataset and guided by the negative staining EM density ( Higashi et al . , 2020 ) . The coiled coils emanating from the ATPase heads were extended toward the SMC hinge using modeled Smc1Psm1 and Smc3Psm3 coiled coil segments , built based on their amino acid sequence using CCbuilder2 . 0 ( Wood and Woolfson , 2018 ) . The elbow positions in Smc1Psm1 and Smc3Psm3 were previously identified ( Higashi et al . , 2020 ) . To build a molecular model of cohesin in the post-hydrolysis slipping state , Scc2Mis4 and the Smc3Psm3 head domain were replaced with models of the same elements in new conformational forms , corresponding to previously observed free crystal structure states , as described ( Gligoris et al . , 2014; Kikuchi et al . , 2016; Higashi et al . , 2020 ) . To model Brownian motion of the Scc3-hinge module relative to the Scc2-head module , Scc3Psc3 with the kleisin middle region and DNA together with the SMC hinge were considered to be one rigid body . The position of this Scc3-hinge module was then developed using a swinging motion of the SMC coiled coils with the inflection point at their elbows . The molecular model of a nucleosome is based on the crystal structure of a human nucleosome ( PDB: 3AFA ) ( Tachiwana et al . , 2010 ) . All structural figures were prepared using Pymol ( Schrödinger ) and ChimeraX ( Goddard et al . , 2018 ) . For the construction of cohesin complexes harboring FRET reporters , SNAP- and CLIP-tag sequences were introduced into the YIplac211-Psm1-Psm3 and YIplac128-Rad21-Psc3 budding yeast expression vectors , as well as the pMis4 ( N191 ) -PA fission yeast expression vector that were previously described ( Murayama and Uhlmann , 2014; Chao et al . , 2015; Higashi et al . , 2020 ) . For labeling the Psm1 hinge , the SNAP tag sequence was inserted between Psm1 amino acids R593 and P594 . Psc3 was fused to the SNAP or CLIP tag sequence at its C-terminus . For labeling Mis4 , the CLIP tag sequence preceded the Mis4 ( N191 ) N-terminus . All proteins were purified and labeled with BG-surface Alexa 647 and BC-surface Dy547 dyes ( New England Biolabs ) as previously described ( Murayama and Uhlmann , 2014; Higashi et al . , 2020 ) . The absorbance spectra of the labeled proteins were recorded between 220–800 nm in 1 nm increments using a V-550 Spectrophotometer ( Jasco ) . The concentrations of protein , Dy547 and Alexa 647 were determined from the absorbance at 280 , 550 , and 650 nm , respectively , and the labeling efficiencies calculated ( assuming molar extinction coefficient ( ε ) for Dy547 and Alexa 647 of 150 , 000 M−1 cm−1 and 270 , 000 M−1 cm−1 , respectively ) . Topological DNA loading assays to confirm the biochemical activity of the cohesin fluorophore fusions were performed as previously described ( Higashi et al . , 2020 ) . Bulk FRET measurement were performed as previously described ( Higashi et al . , 2020 ) , with minor modifications . All fluorescence measurements were carried out in reaction buffer ( 35 mM Tris-HCl pH 7 . 5 , 0 . 5 mM TCEP , 25 mM NaCl , 1 mM MgCl2 , 15% ( w/v ) glycerol and 0 . 003% ( w/v ) Tween 20 ) . Forty μl of reaction mixtures containing 37 . 5 nM of the respective labeled cohesin , 100 nM Mis4-Ssl3 or 12 . 5 nM Dy547-labeled Mis4 ( N191 ) and 10 nM pBluescript KSII ( + ) as the DNA substrate were mixed and the reaction was started by addition of 0 . 5 mM ATP . Alternatively , 0 . 5 mM ADP and 0 . 5 mM BeSO4 +10 mM NaF were included to generate the DNA gripping state . The reactions were incubated at 32°C for 20 min . The samples were then applied to a 384-well plate and fluorescence spectra were recorded at 25°C on a CLARIOstar Microplate Reader ( BMG LABTECH ) . Samples were excited at 525 nm and emitted light was recorded between 560–700 nm in 0 . 5 nm increments . To evaluate FRET changes caused by cohesin’s conformational changes across different experimental conditions , we report relative FRET efficiency , IA/ ( ID + IA ) , where ID is the donor emission signal intensity at 565 nm resulting from donor excitation at 525 nm and IA is the acceptor emission signal intensity at 665 nm resulting from donor excitation . To obtain a baseline of apparent FRET due to spectral overlap , we mixed singly Dy547 and Alexa 647-labeled cohesin at concentrations of 12 . 5 nM and 37 . 5 nM , respectively , to reflect the fluorophore stoichiometry of the double labeled complex . 100 nM cohesin labeled at the Psm1 hinge with Alexa 647 , 100 nM Mis4∆N191 labeled at the N-terminus with Dy547 , 10 nM pBluescript KSII ( + ) DNA and 0 . 5 mM ATP or 0 . 5 mM ADP/0 . 5 mM BeSO4/10 mM NaF were incubated in 15 μl of reaction buffer at 32°C for 20 min , then diluted with 100 μl of reaction buffer . 5 μl of anti-Pk antibody ( Bio-Rad , MCA1360 ) -bound Protein A conjugated magnetic beads were added to the reactions and rotated at 25°C for 10 min . The beads were washed twice with 500 μl of reaction buffer and the recovered proteins were analyzed by SDS-PAGE followed by in gel fluorescence detection , as well as immunoblotting with the indicated antibodies . Flow cells were assembled using a piece of parafilm in which a microfluidic channel ( 1 . 5 × 40 mm ) was pre-cut , sandwiched between a glass slide and a salinized cover slip ( Marienfeld , High-precision , 24 × 60 mm ) . Two holes were drilled in a glass slide and a metal tubing ( New England Small Tube Corp ) was glued using an epoxy glue ( Devcon ) to form an inlet and an outlet of the flow cell . Slides were cleaned by sequential 5 min sonication in ethanol , 1 M NaOH and purified water , then baked at 100°C for 10 min on a heating plate and finally plasma cleaned for 5 min . Cover slips were first cleaned by sequential sonication in acetone , ethanol , 0 . 1 M NaOH and purified water ( 5 min for each step ) , then blow-dried using compressed air and plasma cleaned for 5 min . Subsequently the cover slip surface was silanized by incubating for 1 hr in 5% dichlorodimethylsilane ( Sigma-Aldrich ) dissolved in heptane . Finally , cover slips were washed by 10 min sonication in chloroform followed by 10 min sonication in purified water and blow-dried using compressed air . Flow cells were incubated with anti-Digoxigenin antibodies ( Roche , 150U ) diluted 30 times in PBS for 20 min and washed with 400 µl PBS . To prevent non-specific binding of DNA and proteins , the surface of the flow cell was passivated by incubating with Pluronic F127 ( Sigma-Aldrich , 1% solution in PBS ) for 10 min followed by PBS wash and incubating with β-Casein ( Sigma-Aldrich , 10 mg/ml in PBS ) for at least 1 hr . Subsequently the flow cell was washed with PBS and equilibrated with 50 µl of buffer W ( 0 . 1 mg/ml β-Casein in PBS ) . Forty µl of 10 pM of Digoxigenin-labeled λ-phage DNA in buffer W were introduced into the flow cell , incubated for 15 min and washed with 60 µl of buffer W at 4 µl/min using a syringe pump ( Harvard Apparatus , Pico Plus Elite 11 ) . The flow cell was further equilibrated with 150 µl of buffer R ( 40 mM Tris-HCl pH 7 . 5 , 50 mM NaCl , 2 mM MgCl2 , 5 mM ATP , 10 mM DTT , 250 nM Sytox Orange , 0 . 2 mg/ml glucose oxidase , 35 µg/ml catalase and 4 . 5 mg/ml dextrose ) at 15 µl/min . Fission yeast cohesin and cohesin loader were introduced at 5 nM and 10 nM concentration , respectively , in buffer R at 15 µl/min . Individual DNA molecules stained with Sytox Orange were imaged at two frames per second using a Nikon Eclipse Ti2 commercial TIRF microscope . Flow cells were illuminated with a 561 nm laser through a Nikon SR HP Apo TIRF 100x/1 . 49 oil immersion objective . Images were collected using a sCMOS camera ( Photometrics , Prime 95B ) and saved as TIFF files without compression for further analysis using FIJI ImageJ . All experiments were performed at room temperature . λ-phage DNA was terminally labelled with Digoxigenin in a 50 µl reaction containing 0 . 25 mg/ml λ-phage DNA ( NEB ) , DNA Taq polymerase ( NEB ) , dATP , dCTP , dGTP ( Promega ) , dUTP-Digoxigenin ( Jena Bioscience ) and 1X Standard Taq buffer ( NEB ) . The mixture was incubated at 72°C for 30 min and cleaned up to remove unincorporated nucleotides using a spin-column ( Bio-Rad , Micro Bio-Spin P30 ) . The concentration of the final product was assessed by measuring absorbance at 260 nm using a Nanodrop ND1000 spectrophotometer . Digoxigenin-labelled λ-phage DNA was aliquoted and stored at −20°C . Loop extrusion rates were determined using FIJI ImageJ software as described ( Davidson et al . , 2019 ) . The length of each DNA molecule was manually measured before loop extrusion to convert distance in pixels into kbp . During loop extrusion the length of DNA not contained inside the loop was measured manually , converted into kbp and subtracted from the full length of Lambda DNA ( 48 . 5 kbp ) to calculate the size of the DNA loop . The loop extrusion rate was determined as the slope of a linear fit to the experimental loop growth data . Numerical calculations were carried out with the Metropolis method for Monte-Carlo simulation ( Heermann , 1990 ) . Briefly , random beads representing a DNA segment or a part of cohesin is chosen at each step and its position or orientation vectors are randomly modified . Position vectors are modified by adding a random 3D vector with each coordinate distributed evenly on a [-0 . 75:0 . 75] nm interval . Orientation vectors are rotated in random direction in 3D by an angle randomly distributed on [-0 . 8:0 . 8] . The new full energy of the system Enew is calculated . If the new energy is lower than the previous value Enew , the new state is accepted . If it is larger , the new state Boltzmann factor is calculated: r=e-Enew-EoldkBT and the new state is accepted if r>p , where p is drawn from a standard uniform distribution on the interval ( 0 , 1 ) . In the simplified model for DNA loop extrusion , a DNA loop inside a cohesin molecule consists of N segments , each 5 nm in length . On each iteration of the algorithm , we consider all possible events that can happen to the system consisting of cohesin and a DNA loop inside . There are five types of events: 1 . Random thermal movement of the inbound DNA at the Smc3 head ( only possible in the slipping state ) , 2 . Random thermal movement of the outbound DNA , 3 . If the hinge is bound to DNA , DNA can move with the hinge if cohesin changes its state from the folded gripping to the unfolded slipping state , or the other way around , 4 . The hinge can bind/unbind DNA , 5 . Cohesin can change state , from gripping to slipping state , or vice versa . At each iteration , rate constants are calculated for all possible events . The rate constant of DNA diffusion is the rate at which one DNA segment moves one step forward or backward . It is given by: ( 8 ) k±=Da2where D is the diffusion coefficient and a=5nm length of the DNA segment . Off-rate of hinge DNA unbinding: ( 9 ) kHinge=1tDNA-hingewhere tDNA-hinge is the lifetime of DNA-hinge interaction defined in the main text . We assume that the rates of slipping - > gripping and gripping - > slipping state conversion are the same: ( 10 ) kCohesin=1textendedwhere textended is the lifetime of the slipping state defined in the main text . At a given iteration of the Monte Carlo simulation , we calculate a set of times at which each possible event stochastically occurs: ( 11 ) ti ( j ) =-ln⁡ ( 1-ri ) /kiwhere j is the current iteration , i is the given event . ri is a uniformly distributed number on the interval ( 0 , 1 ) , and ki the rate for event i . Then we implement the event which has the smallest ti ( j ) and modify the system based on which event occurred . The total time of the simulation is extended by ti ( j ) . The simplest assumption about the nature of the cohesin transition from the folded gripping to the unfolded slipping state is that it is driven by diffusion only . The diffusion coefficient can be estimated as: ( 12 ) D=kBT6πηrwhere η = 10−3 Pa·s is the viscosity of water and r is the effective radius of the diffusing protein . The estimate of the time it takes to diffuse distance x is: ( 13 ) t=<x2>2D The size of the protein is related to its molecular weight . Given cohesin’s irregular shape , the diffusion can be better estimated based on the actual size of the protein . Using a size of ~ 20 nm for the Scc3-hinge and Scc2-head modules , we get a conservative estimate of ~ 0 . 1 ms for the time it will take the modules to separate by diffusion ~ 50 nm . All data generated in this study are included in the manuscript . Materials and resources described in this manuscript will be made available by the authors upon reasonable request . The code used for the molecular-mechanical simulation of behavior and DNA loop extrusion is available in the GitHub repository ( https://github . com/FrancisCrickInstitute/CohesinModel , copy archived at swh:1:rev:62d9088238066810feb6a98e56cc0940fd4943d1 , Higashi , 2021 ) .	When a cell divides , it has to ensure that each of its daughter cells inherits one copy of its genetic information . It does this by duplicating its chromosomes ( the DNA molecules that encode the genome ) and distributing one copy of each to its daughter cells . Once a cell duplicates a chromosome , the two identical chromosomes must be held together until the cell is ready to divide in two . A ring-shaped protein complex called cohesin does this by encircling the two chromosomes . Cohesin embraces both chromosome copies , as they emerge from the DNA replicating machinery . The complex is formed of several proteins that bind to a small molecule called ATP , whose arrival and subsequent breakdown release energy . Cohesin also interacts with DNA in a different way: it can create loops of chromatin ( the complex formed by DNA and its packaging proteins ) that help regulate the activity of genes . Experiments performed on single molecules isolated in the laboratory show that cohesin can form a small loop of DNA that is then enlarged through a process called DNA loop extrusion . However , it is not known whether loop extrusion occurs in the cell . Although both of cohesin’s roles have to do with how DNA is organised in the cell , it remains unclear how a single protein complex can engage in two such different activities . To answer this question , Higashi et al . used a structure of cohesin from yeast cells gripping onto DNA to build a model that simulates how the complex interacts with chromosomes and chromatin . This model suggested that when ATP is broken down , the cohesin structure shifts and DNA enters the ring , allowing DNA to be entrapped and chromosomes to be bound together . However , a small change in how DNA is gripped initially could prevent it from entering the ring , creating a ratchet mechanism that forms and enlarges a DNA loop . This molecular model helps explain how cohesin can either encircle DNA or create loops . However , Higashi et al . ’s findings also raise the question of whether loop extrusion is possible inside cells , where DNA is densely packed and bound to proteins which could be obstacles to loop extrusion . Further research to engineer cohesin that can only perform one of these roles would help to clarify their individual contributions in the cell .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "computational", "and", "systems", "biology" ]	2021	A Brownian ratchet model for DNA loop extrusion by the cohesin complex
Perceptual learning is often orientation and location specific , which may indicate neuronal plasticity in early visual areas . However , learning specificity diminishes with additional exposure of the transfer orientation or location via irrelevant tasks , suggesting that the specificity is related to untrained conditions , likely because neurons representing untrained conditions are neither bottom-up stimulated nor top-down attended during training . To demonstrate these top-down and bottom-up contributions , we applied a “continuous flash suppression” technique to suppress the exposure stimulus into sub-consciousness , and with additional manipulations to achieve pure bottom-up stimulation or top-down attention with the transfer condition . We found that either bottom-up or top-down influences enabled significant transfer of orientation and Vernier discrimination learning . These results suggest that learning specificity may result from under-activations of untrained visual neurons due to insufficient bottom-up stimulation and/or top-down attention during training . High-level perceptual learning thus may not functionally connect to these neurons for learning transfer . Visual perceptual learning is the process in which the observers improve their discrimination of fine differences of basic visual features , such as contrast , orientation , motion direction , etc . , through practice . For several decades visual perceptual learning has been regarded as a distinct format of learning because it is specific to the orientation and retinal location of the trained stimulus ( Schoups et al . , 1995; Ahissar and Hochstein , 1997; Shiu and Pashler , 1992; Dosher and Lu , 1998; Poggio et al . , 1992; Fiorentini and Berardi , 1980; Yu et al . , 2004 ) . Such learning specificity has inspired theories that interpret visual perceptual learning as a result of training induced neural plasticity in the early visual areas ( Schoups et al . , 1995; Karni and Sagi , 1991; Teich and Qian , 2003; Bejjanki et al . , 2011 ) . For example , it has been proposed that training could sharpen neuronal orientation tuning in the primary visual cortex ( V1 ) , so that neurons become more sensitive to the fine changes of orientation differences ( Teich and Qian , 2003; Schoups et al . , 2001 ) . The learning specificity has also constrained alternative reweighting theories of visual perceptual learning ( Dosher and Lu , 1998; Poggio et al . , 1992; Yu et al . , 2004; Mollon and Danilova , 1996; Petrov et al . , 2005; Law and Gold , 2008; 2009 ) . These theories propose that training may not change the tuning properties of sensory neurons . Rather , the inputs from activated neurons are reweighted through training to improve readout at a later decision stage . Reweighting theories are supported by neurophysiological evidence . For example , motion direction learning in monkeys is correlated with changes in motion-driven responses of neurons in the lateral intraparietal area ( LIP ) that is related to the transformation of motion information into decisions ( saccadic choices ) , but not with changes of neurons in the medial temporal area ( MT ) that represents motion direction signals ( Law and Gold , 2008 ) . However , in a series of 'double training' studies we demonstrated that visual perceptual learning of various tasks can transfer significantly , and often completely , to new orientations or locations ( Xiao et al . , 2008; Wang et al . , 2012 , 2014; Zhang et al . , 2010 , 2014 ) . In a double training experimental design the observers are additionally exposed to the new orientation or location via practicing an irrelevant task besides the primary learning task . For example , perceptual learning of foveal orientation discrimination ( e . g . , which of two consecutively presented gratings is more clockwise-tilted ? ) initially shows little transfer to an orthogonal orientation . However , if the observers are exposed to an orthogonal orientation via an irrelevant contrast discrimination task ( e . g . , which of two gratings has higher contrast ? ) , learning transfers completely to the orthogonal orientation ( Zhang et al . , 2010 ) . Similarly , perceptual learning of Vernier discrimination ( e . g . , whether a lower grating is placed to the left or right of an upper grating ) can also transfer significantly and often completely to a new retinal location when the observers are additionally exposed to the transfer location via an irrelevant contrast or orientation discrimination task ( Wang et al . , 2012 , 2014 ) . These results suggest two important insights regarding visual perceptual learning . First , visual perceptual learning is mainly a high-level rule-based learning process that occurs beyond the retinotopic and orientation selective visual areas , so that learning is in principle transferrable to untrained conditions ( Zhang et al . , 2010 ) . Second , learning specificity may be related to the untrained conditions , rather than the trained conditions as the field has been assuming ( i . e . , plasticity with the trained early visual cortical neurons or reweighting of the inputs from these neurons ) . The second insight , which stands completely different from the interpretations of specificity by the field , forms the basis of the current study . During training at a specific orientation or location in a typical visual perceptual learning experiment , most mental resources are devoted to the trained stimuli . For example , an observer has to focus the attention on the near-threshold difference between the reference orientation and the target orientation at a specific location . As a result the untrained orientations and locations , and thus the visual neurons representing these orientations and locations , are neither bottom-up stimulated nor top-down attended during training . We thus suspect that this insufficient bottom-up stimulation of , and/or top-down attention to , the untrained conditions are responsible for the orientation and location specificity . In other words , because visual neurons representing untrained conditions are not properly activated as a result , high-level perceptual learning may not be able to functionally connect to these neurons for learning transfer . Our previous double training experiments are unable to separately identify the potential bottom-up and/or top-down contributions because suprathreshold stimuli are used in the secondary exposure task . In the current study we applied a continuous flash suppression ( CFS ) technique ( Tsuchiya and Koch , 2005 ) to suppress the exposure stimulus into sub-consciousness ( see Materials and methods ) . We further manipulated the subconscious stimulus conditions to make the exposure task to be bottom-up only or top-down only . The results show that either bottom-up stimulation of the untrained condition , or top-down attention to it , is sufficient to enable substantial and often complete transfer of learning . These results provide a solution to the mystery of learning specificity that has dominated the history of perceptual learning research . With learning specificity considered as a by-product of training , the field should move on to study the brain mechanisms of perceptual learning without much of specificity-related constraints . Moreover , more efficient training paradigms can be designed to generate perceptual learning without the unwanted specificity in practical settings . Vernier learning is highly location specific . In our previous study ( Wang et al . , 2014 ) that used the same stimulus configuration as here in Figure 4 , Vernier learning at one visual quadrant location showed zero transfer to a diagonal location , with the transfer index TI = -0 . 1 ± 0 . 16 . However , additional training at the transfer location with an orientation discrimination task successfully enabled Vernier learning transfer ( TI = 0 . 98 ± 0 . 22 ) . The strong location specificity as well as the complete learning transfer after double training formed the baselines for the current experiments . 10 . 7554/eLife . 14614 . 009Figure 4 . The effects of bottom-up stimulation of the transfer location on Vernier learning transfer . ( a ) CFS configurations for the simultaneous Vernier training and bottom-up stimulation of the diagonal transfer location . Instruction stimuli: The observer was instructed before the experiment to perform the Vernier task in the dominant eye , while flashing noise patterns covered the opposite visual hemifield across the horizontal meridian . A blank screen was shown to the non-dominant eye . Actual stimuli: A horizontal Gabor was simultaneously flashed in the non-dominant eye at the diagonal transfer location . The observers were neither told , nor were they aware of , the presence of the Gabor stimulus . ( b ) The mean and individual learning and transfer data with training and bottom-up stimulation of the transfer location . ( c ) Control experiment . Same as 4b except that there was no Vernier training . ( d ) . Control experiment . Same as 4b except that there was no presence of the Gabor stimulus at the transfer location . ( e ) . A summary of learning and transfer in the training plus bottom-up stimulation condition , the bottom-up stimulation alone condition , and the training plus noise-only condition . ( f ) . A summary of the transfer indices in the training plus bottom-up stimulation condition and the training plus noise-only condition . Error bars indicate ± 1 standard error of the mean . DE - dominant eye . p<0 . 05; p<0 . 01; **p<0 . 001 . See Figure 4—source data 1 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 14614 . 00910 . 7554/eLife . 14614 . 010Figure 4—source data 1 . The first data sheet summarizes the mean and individual data presented in figure panels 4b , 4c , and 4d . The other data sheets contain raw staircase data for 4b , 4c , and 4d . DOI: http://dx . doi . org/10 . 7554/eLife . 14614 . 010 A major focus of perceptual learning research has been on the links between learning specificity and training-induced neural changes in the brain . The common explanations include learning specificity as a product of neural plasticity in early visual areas that are most feature selective and retinotopic ( Karni and Sagi , 1991; Teich and Qian , 2003; Schoups et al . , 2001 ) , or of improved readout of inputs from early visual areas ( Dosher and Lu , 1998; Poggio et al . , 1992; Mollon and Danilova , 1996; Law and Gold , 2009 ) . Here our results paint a completely opposite picture: It is the actions ( or the absence of actions ) with the untrained conditions that decide the learning specificity and transfer . We show that for orientation and location specificity , the absence of bottom-up stimulation of neurons representing the untrained conditions , as well as top-down attention to these neurons , prevent perceptual learning from transferring to untrained conditions . In one ERP study ( Zhang et al . , 2013 ) , we discovered that Vernier learning and its transfer to an untrained hemisphere accompanies significant occipital P1-N1 changes when the Vernier task is performed at either the trained or the untrained location after training . However , if learning does not transfer , as shown in about half the observers , the P1-N1 changes are limited to the trained location . We interpret P1-N1 changes as possible indications of top-down connections between high-level Vernier learning and visual neurons at the trained location , as well as at untrained locations when learning transfers . We also interpret learning specificity as a result of absent functional connections . Our current data suggest that both bottom-up stimulation of , and top-down attention to , untrained conditions may activate visual neurons representing the untrained conditions , so as to foster the connections to enable learning transfer . In another ERP study ( Zhang et al . , 2015 ) , we also found that orientation discrimination learning and its transfer to an untrained hemisphere accompanies significant occipital C1 changes when the orientation discrimination task is performed at the trained and untrained locations , respectively , after training ( Zhang et al . , 2015 ) . Such C1 changes are not evident with an untrained shape-discrimination task ( Zhang et al . , 2015 ) or with passive viewing ( Bao et al . , 2010 ) . These results further indicate that the functional connections are task specific , consistent with the observations that perceptual learning is task specific ( Shiu and Pashler , 1992; Ahissar and Hochstein , 1993; Cong et al . , 2016 ) . The task-specific functional connections associated with learning and its transfer are also supported by fMRI evidence , in that the generalized orientation discrimination learning is accompanied with task-specific enhancement of orientation-selective responses in the early visual areas including V1 , V2 and V3 ( Byers and Serences , 2014 ) . These results together may also explain fMRI results that orientation specificity is accompanied with refined representation of the trained orientation in early visual areas ( Jehee et al . , 2012 ) . This is because top-down modulation by high-level orientation learning may not reach the early cortical representations of untrained orientations as a result of absent functional connections . As we pointed out earlier , when a near-threshold task is practiced , most brain resources are devoted to the training orientation and retinal location . The untrained orientations and retinal locations , and thus the relevant visual neurons , are neither bottom-up stimulated nor top-down attended . At least in the case of location specificity , there is evidence that spatial attention could suppress other unattended locations even with no presence of competing stimuli ( Smith et al . , 2000; Shmuel et al . , 2006 ) , as is typical in a perceptual learning study . We suspect that some or all of these factors could contribute to under-activations of neurons at untrained conditions , which in turn could lead to missing functional connections of these neurons to high-level learning to prevent learning transfer . Our previous double training studies have revealed significant and often complete learning transfer to untrained conditions ( Xiao et al . , 2008; Zhang et al . , 2010 ) . These transfer results prompted us to propose a rule-based perceptual learning theory . Specifically , we suggest that visual perceptual learning is a high-level process . The brain learns the rules of reweighting visual inputs to achieve better visual performance , and these rules are potentially applicable to new orientations and retinal locations . More recent evidence also suggests that these rules are conceptual , in that learning can transfer between physically distinct stimuli that are initially encoded by different neural mechanisms ( e . g . , between local and global orientations defined by gratings and symmetric dot patterns , or between first- and second-order motion directions ) ( Wang et al . , 2016 ) . The current findings add to this theory by elucidating why perceptual learning , if high-level and rule-based , can be specific in the first place . These results also suggest that double training schemes function by providing bottom-up and top-down forces to activate untrained neurons , which in turn initiate functional connections for learning transfer . A recent computational model explains how such functional connections can be built with double training ( Solgi et al . , 2013 ) . In the model , the secondary training task activates the untrained neurons , which could be recalled when the brain is off-task , so that high-level 'concept neurons' that have learned the task can connect to these untrained neurons in an off-task self-organization manner for learning transfer . Apparently the activations of untrained neurons due to unconsciousness stimulation , or top-down attention without actual stimulation , can also be recalled off-task in the context of this computational account . We once had observers practice a peripheral Vernier task identical to the current one , while flashing a Gabor simultaneously at a diagonal location ( without noise suppression ) for the purpose of stimulating neurons at this latter location ( Wang et al . , 2012 ) . However , we found no evidence for learning transfer . A more recent psychophysical study also replicated the null-transfer results ( Mastropasqua et al . , 2015 ) . Similarly , an earlier monkey-recording study ( Schoups et al . , 2001 ) reported that orientation discrimination learning does not transfer to a different location where a same grating stimulus is simultaneously presented . These null-transfer effects may be caused by attentional competition , in that concentrated attention to the training location would suppress the high-contrast stimuli at a different location ( Watanabe et al . , 2001 ) . The attentional suppression effect is avoided in double training when the primary training and the secondary training that stimulates the untrained location are performed in alternating blocks of trials ( Xiao et al . , 2008 ) , or when the stimulation of the untrained locations is below awareness as in our current study . In the latter case the bottom-up stimulation of transfer location through one eye may create an input contrast in the eye-of-origin feature , and this contrast , although invisible perceptually , has been shown to be very salient because it more strongly activates V1 neurons than the surrounding stimuli ( Zhaoping , 2008 ) . Perceptual learning can also be task irrelevant . Watanabe and colleagues ( Watanabe et al . , 2001; Seitz and Watanabe , 2003 ) reported that training improves discrimination of a nearby feature that is task irrelevant and sub-threshold ( to avoid attentional suppression ) , and that the learning occurs only when the irrelevant feature is temporally paired with rewards with the trained task . They explained this task irrelevant perceptual learning ( TIPL ) as a result of interactions between spatially diffusive rewards and bottom-up exposure of the task irrelevant feature ( Seitz and Watanabe , 2005 ) . Note that our double-training enabled learning transfer are distinct from TIPL in two important ways . First , unlike TIPL , the training and exposure do not require temporal pairing and spatial proximity of two stimuli . Indeed the training phase can precede the exposure phase with a time gap of up to 8 weeks ( Zhang et al . , 2010 ) , and the training and exposure locations can be at diagonal quadrants of the visual field separated by 10 degrees ( Wang et al . , 2012 ) , while double training is still effective . In the current study , the training and sub-threshold exposure are performed in separate blocks of trials , rather than paired , in Figure 1 , and are spatially separated by 10 degrees in Figure 4 . Second , our recent evidence suggests that double-training enabled learning transfer is still task specific ( Cong et al . , 2016 ) . We found that orientation discrimination learning cannot transfer to a contrast discrimination task using the same Gabor stimulus , even after the observers receive additional exposure of the transfer task through easy trials in separate blocks of trials . The same is true at a reverse direction when the observers learn contrast discrimination and receive exposure of orientation discrimination , the transfer task , through easy trials . One hundred and thirteen ( 113 ) undergraduate students with normal or corrected-to-normal vision participated in this study . All were inexperienced in psychophysical observations and were naïve to the purpose of the study . Informed consent , and consent to publish was obtained from each observer before testing . This study was approved by the Peking University Institution Review Board . The stimuli were generated by a Matlab-based WinVis program ( Neurometrics Institute , Oakland , CA ) and presented on a 21-inch Sony G520 CRT monitor ( 1600 × 1200 pixel , 0 . 25 × 0 . 25 mm per pixel , 75 Hz frame rate , 43 . 5 cd/m2 mean luminance ) . The luminance of the monitor was linearized by an 8-bit look-up table . A chin-and-head rest helped stabilize the head of the observer . Experiments were run in a dimly lit room . The Gabor stimuli ( Gaussian-windowed sinusoidal gratings ) for foveal orientation discrimination ( Figure 1–3 ) had a standard deviation at 0 . 48° , a spatial frequency at 1 . 5 cpd , a contrast at 0 . 47 , a base orientation at 36° or 126° , and a phase randomized for every presentation . The CFS configuration consisted of a central flashing white noise pattern in the dominant eye , and sometimes a Gabor stimulus orthogonal to the trained orientation in the non-dominant eye ( Figure 1b ) . The noise pattern , refreshed at 9 . 4 Hz , consisted of 25 × 25 randomly generated black or white blocks ( 0 . 17° × 0 . 17° each ) for a total size of 4 . 30° × 4 . 30° . The dichoptic stimulus presentations were realized with a stereoscope . The noise pattern was presented in the dominant eye to suppress the perception of the Gabor stimulus presented in the non-dominant eye ( Tsuchiya and Koch , 2005 ) . The Vernier stimuli for peripheral Vernier discrimination ( Figure 4–5 ) were identical to those used in a previous study ( Wang et al . , 2014 ) , which consisted of an upper and a lower vertical Gabor on a mean luminance screen background . The two Gabors had an identical standard deviation at 0 . 29° , a spatial frequency at 3 cycles per degree , a contrast of 0 . 47 , a phase fixed at 90° , and a center to center distance at 4λ ( Figure 4a ) . The vertical position of each Gabor shifted away in opposite directions to form a specific Vernier offset . The Vernier stimuli were presented in the upper left quadrant ( or lower right quadrant , balanced among observers ) at 5° retinal eccentricity . The CFS configuration consisted of a flashing white noise pattern ( 10 Hz ) , which covered either the lower or upper visual hemifield opposite to the Vernier stimulus location in the dominant eye , and a horizontal Gabor in the non-dominant eye in a quadrant diagonal to the Vernier quadrant . The latter Gabor was identical to those forming the Vernier stimuli . The noise pattern consisted of 50 × 38 randomly generated black or white blocks ( 0 . 25° × 0 . 25° each ) for a total size of 12 . 44° × 9 . 46° . The viewing distance was 1 m . In an orientation discrimination trial , the fixation cross was first presented for 320 ms , and was then followed by a blank gap of 267 ms before the onset of the first stimulus interval . The reference orientation Gabor ( 36° or 126° , counterbalanced among observers ) and the test orientation Gabor ( reference + ∆ori ) were presented in two stimulus intervals ( 106 ms each ) in a random order , which were separated by a 533 ms inter-stimulus interval . The observers judged in which stimulus interval the Gabor was more clockwise . Auditory feedback was given on incorrect responses . In a Vernier discrimination trial , the fixation cross was first presented for 200 ms , and was then followed by a blank gap of 200 ms before the onset of the first stimulus interval . The Vernier stimuli were then presented at one visual quadrant for 200 ms . The observers’ task was to judge whether the lower Gabor was shifted to the left or the right in comparison to the upper Gabor . Auditory feedback was given on incorrect responses . Orientation and Vernier discrimination thresholds were measured with a 2AFC staircase procedure using a classical 3-down-1-up staircase rule that resulted in a 79 . 4% convergence level . Each staircase consisted of four preliminary reversals and six experimental reversals ( approximately 50–60 trials ) . The step size of the staircase was 0 . 05 log units . The geometric mean of the experimental reversals was taken as the threshold for each staircase run . Each experiment consisted of seven sessions including the pre- and post-training sessions on seven different days . The pre-training session measured the orientation thresholds at the training and transfer orientations ( Figures 1–3 ) , or Vernier thresholds at the training and transfer locations ( Figures 4–5 ) , for six staircases each . The post-training session measured the same thresholds for five staircases each . The geometric mean was taken as the pre- or post-training threshold with each condition . The five training sessions each consisted of 10 staircases of orientation discrimination training ( Figures 1–3 ) or Vernier discrimination ( Figures 4–5 ) , as well as 10 blocks of bottom-up and/or top-down trials with the transfer condition ( 50 trials per block in Figures 1–3 , or the same number of trials as in 10 training staircases in Figures 4–5 ) , with the training and bottom-up/top-down tasks switched every five blocks of trials . Each training session lasted about 1 . 5 hr .	People can become more sensitive to small changes in what they are seeing – such as detecting a slight change in the angle of a particular line – with practice . This process is called perceptual learning , but the improvement is often specific such that it is typically lost if the line moves to a new place , or a different line angle is used . Previous work does show that it is possible to transfer the learning to a new location or angle if the individual also practices another , seemingly irrelevant , task at the same or a later time – such as judging how bright the line is . To understand what might be happening to produce these seemingly conflicting results , Xiong et al . used a technique called “continuous flash suppression” with human volunteers . This approach meant that the volunteers were shown an object ( such as an angled line ) in one eye , while their other eye viewed white noise similar to the “snowflakes” seen on an old-fashioned un-tuned television screen . The flashing snowflakes in one eye meant that the volunteers were not consciously aware of the presence of the angled line in the other eye . The experiments revealed that perceptual learning at the new location or line angle happened when a subconsciously-observed object was shown in the new location or angle , or when the volunteers were asked to pay attention to the “subconscious object” when no object was actually there . This suggests that perceptual learning can happen in new conditions both through ‘bottom-up’ processes , which rely entirely on information coming in from the senses , and ‘top-down’ processes , which are influenced by what a person is aware of and paying attetion to . What is more , the results suggest that the classical observations of specificity in perceptual learning are likely to be a result of the lack of bottom-up and top-down influences in the untrained condition , when the volunteers work hard to improve their performance with the trained condition . Future studies could directly look at what is going on in the brain when perceptual learning becomes less specific , for example by using a technique like functional magnetic resonance imaging to measure brain activity .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Bottom-up and top-down influences at untrained conditions determine perceptual learning specificity and transfer
The capacity to remember temporal relationships between different events is essential to episodic memory , but little is currently known about its underlying mechanisms . We performed time-lapse imaging of thousands of neurons over weeks in the hippocampal CA1 of mice as they repeatedly visited two distinct environments . Longitudinal analysis exposed ongoing environment-independent evolution of episodic representations , despite stable place field locations and constant remapping between the two environments . These dynamics time-stamped experienced events via neuronal ensembles that had cellular composition and activity patterns unique to specific points in time . Temporally close episodes shared a common timestamp regardless of the spatial context in which they occurred . Temporally remote episodes had distinct timestamps , even if they occurred within the same spatial context . Our results suggest that days-scale hippocampal ensemble dynamics could support the formation of a mental timeline in which experienced events could be mnemonically associated or dissociated based on their temporal distance . All procedures were approved by the Weizmann Institute IACUC . Five male C57BL/6 mice aged 8-12 weeks at the start were used in this study . Mice were housed with 1-4 cage-mates in cages with running wheels , and underwent two surgical procedures under isoflurane anesthesia ( 1 . 5-2% volume ) . First , we injected into the CA1 , 400 nL of the viral vector AAV2/5-CaMKIIα-GCaMP6s or AAV2/5-CaMKIIα-GCaMP6f ( ~2 × 1013 particles per ml , packed by University of North Carolina Vector Core ) ( Chen et al . , 2013 ) . Stereotatic coordinates were: -1 . 9 mm anterio-posterior , -1 . 4 mm mediolateral , -1 . 6 mm dorsoventral from bregma . The second surgery , which took place at least one week after the viral injection , was the implantation of a glass guide tube directly above the CA1 . We used a trephine drill to remove a circular part of the skull centered posterio-lateral to the viral injection site . We removed the dura and cortex above the CA1 by suction with a 29 gauge blunt needle while constantly washing the exposed tissue with sterile PBS . We then implanted an optical guide tube with its window just dorsal to , but not within , area CA1 , and sealed the space between the skull and guide tube using 1 . 5% agarose in PBS . The exposed areas of the skull were then sealed with Metabond ( Parkell , Edgewood , NY ) and dental acrylic . For time-lapse imaging in freely behaving mice using an integrated miniature fluorescence microscope ( nVistaHD , Inscopix ) , we followed a previously established protocol ( Ziv et al . , 2013 ) . Briefly , at least three weeks after guide tube implantation , we imaged water restricted mice under isoflurane anesthesia using a two-photon microscope ( Ultima IV , Bruker , Germany ) , equipped with a tunable Ti:Sapphire laser ( Insight , Spectra Physics , Santa Clara , CA ) . We inserted a ‘microendoscope’ consisting of a single gradient refractive index lens ( 0 . 44 pitch length , 0 . 47 NA , GRINtech GmbH , Germany ) into the guide tube , and examined Ca2+ indicator expression and tissue health . We selected for further imaging only those mice that exhibited homogenous GCaMP6 expression and healthy appearance of the tissue . For the selected mice , we then affixed the microendoscope within the guide tube using ultraviolet-curing adhesive ( Norland , NOA81 , Edmund Optics , Barrington , NJ ) . Next , we attached the microscope’s base plate to the dental acrylic cap using light cured acrylic ( Flow-It ALC , Pentron , Orange , CA ) . After a few days , we began training the mice to run back and forth on two elevated linear tracks ( Environments A and B ) . Environment A was a straight 96 cm long track and Environment B was an L-shaped track consisting of two 48 cm long arms . Each environment had distinct sets of visual and tactile cues , overhead lights , flavored liquid rewards , and odor cues . Before the beginning of each pre-training or imaging session we wiped the tracks with differently scented paper towels ( 0 . 5% acetic acid for environment A and 10% ethanol for environment B ) . We trained the mice to run back and forth along the track by giving them a measured amount of water sweetened with commercial fruit juice concentrate , lemon flavored for track A and raspberry flavored for track B , with 2% added sugar by weight . The water reward was dispensed using a custom-made computer controlled device . To record mouse behavior , we used an overhead camera ( DFK 33G445 , The Imaging Source , Germany ) , which we synchronized with the integrated microscope . Ca2+ imaging was performed at 20Hz . Before beginning with Ca2+ imaging , we pre-trained the mice for 8–11 days , until the mice ran at least 60 times the entire length of each track in two consecutive days . Pre-training and imaging sessions consisted of five 3-min-long trials , with an inter-trial interval of 3 min . We imaged a total of 5 mice ( two that were injected with AAV2/5-CaMKIIα-GCaMP6f and three that were injected with AAV2/5-CaMKIIα-GCaMP6s ) every other day for 15 days , making for 8 recording days . Each day of the experiment consisted of two sessions ( AM and PM ) separated by 4–5 hr . Remapping tests within a single session were performed on days 16–17 . At the end of the experiment we removed the base plate by drilling away the top acrylic cap and re-examined the health of the CA1 neurons by imaging the mice under isoflurane anesthesia using a two-photon microscope as described above . We processed imaging data using commercial software ( Mosaic , Inscopix ) and custom MATLAB routines as previously described ( Ziv et al . , 2013 ) . To increase computation speed , we spatially down-sampled the data by a factor of two in each dimension . To correct for non-uniform illumination both in space and time , we normalized the images by dividing each pixel by the corresponding value in a smoothed version . The smoothed version was obtained by applying a Gaussian filter with a radius of 40 pixels on the videos . Normalization enhanced the appearance of the blood vessels , which were later used as stationary fiducial markers for image registration . We used rigid body image registration to correct for lateral displacements of the brain . This procedure was performed on a high contrast subregion of the normalized movies for which the blood vessels were most prominent . The registered movies were transformed to relative changes in fluorescence , ∆F' ( t ) F0= ( F' ( t ) −F'0 ) /F'0 , where F'0 is the value for each pixel averaged over time . For the purpose of cell identification the movies were downsampled in time by a factor of five . We identified spatial filters corresponding to individual cells using an established cell-sorting algorithm that applies principal and independent component analyses ( PCA and ICA ) ( Mukamel et al . , 2009 ) . For each spatial filter , we used a threshold of 50% of the filter’s maximum intensity and each pixel that did not cross the threshold was set to zero . After the cells were identified , further cell sorting was performed to find the spatial filters that follow a typical cellular structure . This was done by measuring the filters’ area and circularity and discarding those whose radius was smaller than 5 μm or larger than 14 μm , or which had a circularity smaller than 0 . 8 . In some cases , the output of the PCA/ICA algorithm included more than one component that corresponded to a single cell . To eliminate such incidents , we examined all cells whose centroids were less than 18 μm apart and whenever their traces had correlation > 0 . 9 , the cell with the lower average peak amplitude was discarded . Ca2+ activity was extracted by applying the thresholded spatial filters to the full temporal resolution ( 20Hz ) ∆F' ( t ) /F0 videos . Baseline fluctuations were removed by subtracting the median trace ( 20 s sliding window ) . The Ca2+ traces were smoothed with a low-pass filter with a cutoff frequency of 2Hz . Ca2+ candidate events were detected whenever the amplitude crossed a threshold of 4 or 5 median absolute deviations ( MAD ) , for GCaMP6s or GCaMP6f , respectively . Cellular Ca2+ events are characterized by fast rise and slow decay times . To capture these characteristics in our data we considered for further analysis only candidate Ca2+ events that followed typical indicator decay time , and decay-to-rise time ratios . In order to avoid the detection of several peaks for a single Ca2+ event , only peaks that were 4 or 5 MAD higher than the previous peak ( within the same candidate event ) and 2 or 2 . 5 MAD higher than the next peak for GCaMP6s or GCaMP6f , respectively , were regarded as true events . We set the Ca2+ event occurrence to the time of the peak fluorescence . To mitigate the effects of crosstalk ( i . e . , spillover of Ca2+ fluorescence from neighboring cells ) , we adopted a conservative approach , allowing only one cell of a group of neighbors ( cells whose centroids are less than 18 μm apart ) to register a Ca2+ event in a 200 msec time window . If multiple Ca2+ events occurred within ~200 msec in neighboring cells , we retained only the events with highest peak ∆F' ( t ) /F0 value . If two neighboring cells had correlation > 0 . 9 in their events , the cell with the lower average peak amplitude was discarded . For each session we projected centroids of all thresholded filters onto a single image . We computed the spatial cross-correlation among the projections from all sessions to align them according to a reference session . Because changing the reference did not change the alignment output , we chose the first session as the reference . This step corrected slight translations and rotation changes between sessions and yielded each cell’s location in the reference coordinate system . Next , we searched for cells from different sessions that might be the same neuron . This was performed using two separate methods based on either spatial correlations or centroids distances ( Figure 1—figure supplement 2 ) . Figures 1 and 2 , Figure 1—figure supplement 3 , and Figure 2—figure supplements 2–6 show longitudinal data for which we used the spatial correlations-based registration method . Figure 2—figure supplement 5 shows longitudinal data for which we used the centroids distances-based registration method . Within each session , the nearest neighbors spatial correlations were always < 0 . 6 ( Figure 1—figure supplement 2A , C ) and the centroids distances were always > 6 μm ( Figure 1—figure supplement 2B , D ) . Between sessions , however , a large amount of cell pairs had spatial correlations > 0 . 6 and centroid distances < 6 μm . Pairs with spatial correlation > 0 . 7 or distance < 5 μm were registered as the same neuron . In cases with more than one candidate , the cells with the minimal distance or maximal correlation were assigned to be the same neuron . Analyzing the data using a range of different thresholds demonstrated the robustness of our registration process to the choice of the threshold ( Figure 2—figure supplement 5 ) . We analyzed mouse behavior videos using a custom MATLAB ( Mathworks ) routine that detected the mouse’s center of mass in each frame , calculated its velocity and applied a rectangular smoothing window of 250 msec . For place field analysis , we considered periods when the mouse ran >1 cm s−1 . We divided each track into 24 bins ( 4 cm each ) and excluded the last 2 bins at both ends of the tracks where water rewards were consumed and the mouse was generally stationary ( Ziv et al . , 2013 ) . We computed the time spent in each bin , and the number of Ca2+ events per bin , and smoothed these two maps ( ‘occupancy’ and ‘Ca2+ event number’ ) using a truncated Gaussian kernel ( σ = 1 . 5 bins , size = 5 bins ) . We then computed the place field map for each neuron by dividing the two smoothed maps of Ca2+ event number and occupancy . We separately considered place fields for left and right running directions and normalized each place field by its maximum value . We defined each place field's position at its peak value . For each place field with >5 events for a given session , we computed the spatial information ( in bits per event ) using the unsmoothed events-rate map of each cell , as previously described ( Markus et al . , 1994 ) : Spatial Information=∑ipi ( ri/r¯ ) log2 ( ri/r¯ ) Where ri is the Ca2+ event rate of the neuron in the ith bin; pi is the probability of the mouse being in the ith bin ( time spent in ith bin/total session time ) ; r̄ is the overall mean Ca2+ event rate; and i running over all the bins . We then performed 1000 distinct shuffles of animal locations during Ca2+ events , accounting for the spatial coverage statistics at the relevant session and direction , and calculated the spatial information for each shuffle . This yielded the p value of the measured information relative to the shuffles . Place fields with p ≤ 0 . 05 were considered significant . To determine the level of similarity between representations of the different environments , we calculated the mean population vector correlation between them ( Leutgeb et al . , 2005 ) . For each spatial bin ( excluding the last 2 bins at both ends of the tracks ) we defined the population vector as the mean event rate for each neuron given that bin’s occupancy . We computed the correlation between the population vector in one environment with that of the matching location in the other environment , and averaged the scores over all positions . Since there are two edges to each of the two linear tracks there are two possible transformations between them . Therefore , we used the one that resulted in higher global population vector correlation . We generated the null hypothesis for place fields' displacements between a pair of days by taking the measured centers of place fields in the same environment on the two days and shuffling cells' identities on each of the days . We calculated the distribution of all displacements and averaged them over 10 , 000 distinct pairs of shuffles . Figure 1F shows the mean null hypothesis for the displacement curve found by averaging over all pairs of days for a given elapsed time . For the analyses shown in Figure 1G , H , and Figure 1—figure supplement 3A-E we used analysis of variance ( ANOVA ) with repeated measures . Greenhouse-Geisser estimates of sphericity were used for degrees of freedom adjustment . To capture the temporal information encoded in the hippocampal neural representations of different episodes we constructed three types of time decoders: ( 1 ) ordinal time decoder , ( 2 ) within-environment time decoder , and ( 3 ) across environments time decoder . The time decoders estimated the true order of the recording days from sets of eight episodes ( sessions or trials ) from the different days in the experiment . Decoding analyses were performed separately for five mice . Vectors of ensemble activity patterns were constructed where each element corresponded to the total number of events of one neuron within an episode . We notated the full-session ensemble activity pattern in day d in environment E as VdE , and the ensemble activity pattern in the tth trial on day d in environment E as vd , tE . To quantify the similarities between ensemble activity patterns of different episodes we calculated the Pearson correlation between the activity vectors . For the within and between environments time decoders , we normalized each correlation value between a test-data pattern and a training-data pattern by subtracting the average correlation of the training-data vector over all the vectors . We used two measures of divergence between the representations of the two environments: ‘activity divergence’ and ‘peak displacement’ . To investigate cell-level dynamics we analyzed the changes in event rates over time for each cell . We applied these analyses separately for each of the two environments .	The ability to recall the timing of events is an important feature of long-term memory . Episodic memory , the mental account of “what” happened , “where” and “when” , depends on a region of a brain called the hippocampus . Certain neurons in the hippocampus , called place-cells , are known to capture information about the locations an animal has visited so that a specific pattern of place cell activity marks each location an animal visits . However , it is not clear how the brain can mark the relationship between the timing of different events . Some studies have documented gradual changes in the activity patterns of the place cells over time , which could help mark time . If these changes are specific to a particular environment then they would not allow animals to associate in memory events that occurred close in time ( for instance , in the same day ) if these events occurred in different environments . To do that , a certain component of the changes in the activity patterns would have to be independent of any specific environment or context in which events occur . Now , Rubin , Geva et al . have captured time-lapse images of the activity of thousands of hippocampal cells in mice as they explored two different environments on repeated occasions over a two-week period . The environments had different shapes , textures , visual cues , and odors . The mice were allowed to explore each environment daily for more than a week prior to the time-lapse filming so that they would be very familiar with the two environments . During the filming portion of the experiments , the mice visited one environment in the morning , and then the other in the afternoon . The analysis of the images revealed what appeared to be unique patterns of cell activity for specific days , which gradually changed over the course of the experiment . The patterns persisted even when the animals switched to a new environment during the same day , but were different for visits to the same environment on different days . Next , Rubin , Geva et al . used the patterns of activity collected from the mice while they were in one environment to create a timeline of events . From this timeline , it was possible to accurately deduce which day each visit to the other environment occurred based on the patterns of hippocampal cell activity alone . One challenge that stems from this work is to understand the biological mechanisms that drive the patterns in neuronal activity over timescales that are relevant for long-term memory .	[ "Abstract", "Materials", "and", "methods" ]	[ "short", "report", "neuroscience" ]	2015	Hippocampal ensemble dynamics timestamp events in long-term memory
Chromosome alignment in the middle of the bipolar spindle is a hallmark of metazoan cell divisions . When we offset the metaphase plate position by creating an asymmetric centriole distribution on each pole , we find that metaphase plates relocate to the middle of the spindle before anaphase . The spindle assembly checkpoint enables this centering mechanism by providing cells enough time to correct metaphase plate position . The checkpoint responds to unstable kinetochore–microtubule attachments resulting from an imbalance in microtubule stability between the two half-spindles in cells with an asymmetric centriole distribution . Inactivation of the checkpoint prior to metaphase plate centering leads to asymmetric cell divisions and daughter cells of unequal size; in contrast , if the checkpoint is inactivated after the metaphase plate has centered its position , symmetric cell divisions ensue . This indicates that the equatorial position of the metaphase plate is essential for symmetric cell divisions . During mitosis , chromosomes are bound to microtubules emanating from both poles of the mitotic spindle via sister-kinetochores and aligned on the metaphase plate precisely in the middle of the spindle . The equatorial position of the metaphase plate is a distinctive feature of metazoan , plant , and many fungal cells . A centered metaphase plate is established even in asymmetrically dividing cells , where daughter cells of unequal size are obtained by an asymmetric positioning of the spindle prior to anaphase ( e . g . , in Caenorhabditis elegans embryos ) or an asymmetric elongation of the spindle in anaphase ( e . g . , in Drosophila melanogaster embryonic neuroblasts [Kaltschmidt et al . , 2000; Schneider and Bowerman , 2003] ) . However , the reason why the metaphase plate is located in the middle of the spindle is not known . One hypothesis is that the centered position facilitates the synchronous arrival of chromosomes at spindle poles during anaphase to prevent chromosomes from being caught on the wrong side of the cytokinetic furrow ( Nicklas and Arana , 1992; Goshima and Scholey , 2010 ) . Elegant work in meiotic praying mantis cells demonstrated that the equatorial positioning of the metaphase plate is not a mere consequence of bipolar kinetochore–microtubule attachments , as trivalent sex-chromosome align in the middle of the spindle , even though trivalent attachment does not favor an equatorial position ( Nicklas and Arana , 1992 ) . Moreover , previous studies in Chlamydomonas rheinhardtii and C . elegans showed that an asymmetry in centriole numbers at spindle poles led to an asymmetric metaphase plate position , even though chromosomes established bipolar attachments ( Greenan et al . , 2010; Keller et al . , 2010 ) . While in algae , longer half-spindles were associated with the pole containing fewer centrioles , in nematodes , longer half-spindles emanated from the pole containing more centrioles . However , whether cells react to asymmetrically located metaphase plates and the long-term consequences of this asymmetry are not known . Here , we investigated these questions in human tissue culture cells . We find that cells correct metaphase plate position before anaphase onset , we demonstrate that a centered metaphase plate position relies on the spindle assembly checkpoint ( SAC ) to provide sufficient time for this correction mechanisms , and we show that a failure to correct plate position leads to asymmetric cell divisions . To monitor the relative position of the metaphase plate in the spindle over time , we recorded by time-lapse imaging HeLa cells stably expressing eGFP-centrin1 ( centriole marker ) and eGFP-CENPA ( kinetochore marker ) and automatically tracked centrosomes and the metaphase plate using an in-house developed software ( Jaqaman et al . , 2010; Vladimirou et al . , 2013 ) . Metaphase or late prometaphase cells were recorded over a short period of 5 min in 3D at a resolution of 7 . 5 s under conditions of low phototoxicity compatible with anaphase entry ( Jaqaman et al . , 2010 ) . By plotting the ratio R of the half-spindle lengths of metaphase cells at the onset of our recordings ( first three time points ) , we found a broad distribution centered around median R = 0 . 98 , which represents nearly equal half-spindle lengths . When analyzing the subset of cells that entered anaphase during our recordings 30 s before anaphase , we found a sharp R distribution in the middle of the spindle ( median R = 1 . 02; Figure 1A ) : less than 10% of the R values were smaller than 0 . 85 or larger than 1 . 15 at anaphase onset , while in the metaphase population over 24 . 2% were outside of these boundaries . This suggested a centering mechanism for the metaphase plate as cells progressed towards anaphase . To test this hypothesis , we aimed to create asymmetric spindles by generating cells with an asymmetric centriole distribution , using small interfering ( si ) RNAs against Sas-6 , a protein required for centriole duplication ( Leidel et al . , 2005 ) . This procedure was used on a set of HeLa eGFP-centrin cells that co-expressed either eGFP-CENPA , α-tubulin-mRFP ( spindle marker ) , or Histone H2B-mRFP ( chromosome marker ) . Every wild-type mitotic cell contains four centrioles: one oldest ( grandmother ) centriole , one older ( mother ) centriole , and their two respective daughter centrioles ( Nigg and Raff , 2009 ) , which all have different eGFP-centrin signal intensities ( Kuo et al . , 2011 ) . A 24-hr Sas-6 depletion led to a mix of cells with two centrioles per pole , one centriole per pole , or one pole with one centriole and the other pole with two centrioles ( called from here on 2:2 , 1:1 , or 2:1 cells; Figure 1B , C ) . Our intensity measurements revealed that in 2:1 cells it was most often the oldest centriole that gave rise to a daughter centriole , probably due to limiting levels of Sas-6 ( data not shown ) . Tracking of HeLa eGFP-centrin1/CENPA cells indicated that the distribution of half-spindle ratios during metaphase was broad in 2:2 or 1:1 cells , but that on average the plate was located in the middle of the cell ( median R = 1 . 04 ( 2:2 ) and 1 . 03 ( 1:1 ) ; note that 2:2 cells served as control transfection for all subsequent experiments; to be consistent in 2:2 or 1:1 cells , R was calculated as the length of the half-spindle associated to the grandmother centriole divided by the opposite half-spindle length , while in 2:1 cells , R represents the ratio of the half-spindle associated to the 2-centriole pole over the opposite half-spindle length; Figure 1D ) . As in wild-type cells , the range of R had narrowed by the time cells were about to enter anaphase , consistent with a centering process ( Figure 1E ) . In 2:1 cells , we found two half-spindles of unequal length , with a shorter half-spindle associated with the pole containing one centriole ( median R = 1 . 12 , p < 0 . 0001 in Mann–Whitney U test compared to symmetric distribution; R > 1 . 15 in 42 . 9% of the cells; Figure 1D ) . This asymmetry was corrected before anaphase onset ( median R = 1 . 03; Figure 1F ) , consistent with the existence of a centering process . Since our videos were much shorter than the overall duration of metaphase , we were unable to directly visualize the centering process . To circumvent this difficulty , we recorded longer time series of 2:1 cells , monitoring them for 15 min at 30-s intervals . By plotting R over time in 15 cells that spend at least 7 min in metaphase before entering anaphase , we could directly observe how cells centered the plate position over time , while reducing the variability in metaphase plate position ( Figure 1G , H ) . 10 . 7554/eLife . 05124 . 003Figure 1 . Cells center the position of the metaphase plate before anaphase onset . ( A ) Distribution of spindle ratio R in metaphase cells in wild-type eGFP-centrin1/CENPA HeLa cells during metaphase in general ( black curve ) or 30 s before anaphase onset ( red curve ) . The spindle ratio R was calculated by dividing the half-spindle length L1 associated with the grandmother centriole ( brightest eGFP-centrin1 ) by the other half-spindle length L2 ( for all cell numbers in all experiments see Table 1 ) . ( B ) Depletion of Sas-6 but not control depletion leads to the gradual loss of centrioles after 24 , 48 , or 72 hr , resulting in a mixed population of cells with different number of centrioles as indicated . Centrioles were visualized based on images of eGFP-centrin1/mRFP-α-tubulin cells as shown in C ( n = 50 cells per experiment , in 3 experiments , error bars indicate s . e . m . ) . ( C ) Immunofluorescence images of eGFP-centrin1 ( green ) /mRFP-α-tubulin ( red ) cells stained with DAPI ( blue ) with different centriole configurations as indicated . Scale bar indicates 2 μm . ( D ) Distribution of spindle ratio R in metaphase cells in siSas-6-transfected eGFP-centrin1/CENPA cells with different centriole configurations ( 2:2 , 2:1 , or 1:1 ) . The spindle ratio was calculated by dividing the half-spindle length L1 associated with the grandmother centriole ( 2:2 and 1:1 cells ) or the half-spindle length associated with 2 centrioles ( 2:1 cells ) by the other half-spindle length L2 . The spindle ratio of 2:1 cells was significantly different from the ratios seen in 2:2 or 1:1 cells ( T-test with Welch's correction , 2:1 > 2:2 , p = 0 . 018 ) . ( E , F ) Distribution of spindle ratio R in Sas-6-depleted eGFP-centrin1/CENPA 2:2 ( E ) or 2:1 ( F ) cells in metaphase and before anaphase onset . 2:1 cells have a significantly more asymmetric plate position in metaphase when compared to cells just before anaphase ( Mann–Whitney U test , p = 7 . 16 × 10−7 ) . ( G ) Plot of half-spindle ratio R over time in 15 individual eGFP-centrin1/CENPA 2:1 cells that entered anaphase during live-cell imaging ( black lines ) . The timeline was synchronized to anaphase onset ( t = 0 ) ; the red curve indicates the median of R , and red surface the 95% confidence interval . Note how median R approaches 1 over time and how its variability decreases . ( H ) Time-lapse images of a eGFP-centrin1/CENPA 2:1 cell as analyzed in G . Half-spindle ratio R and time before anaphase are indicated for each frame . Number of centrioles was determined in IMARIS in 3D ( see 3D-insets in green ) , * denotes the pole with 1 centriole . DOI: http://dx . doi . org/10 . 7554/eLife . 05124 . 00310 . 7554/eLife . 05124 . 004Table 1 . Number of cells in every experimentDOI: http://dx . doi . org/10 . 7554/eLife . 05124 . 004ConditionN° of cellsWT40WT at anaphase onset42WT + MPS1-IN19SiSas-6 2:2 cells41SiSas-6 2:2 cells at anaphase onset29SiSas-6 2:2 cells + MPS1-IN19SiSas-6 2:1 cells59siSas-6 2:1 cells + MG13218SiSas-6 2:1 cells at anaphase onset33SiSas-6 2:1 cells in long term videos14SiSas-6 2:1 cells + MPS1-IN26SiSas-6 1:1 cells36SiKif2a + siMCAK20SiKif2a + siMCAK at anaphase onset41SiKif2a + siMCAK + siSas-6 2:1 cells29SiKif2a + siMCAK + siSas-6 2:1 cells at anaphase onset24SiSas-6 2:1 cells + ZM110SiSas-6 2:2 + DMSO15SiSas-6 2:2 + taxol11SiSas-6 2:1 + DMSO16SiSas-6 2:1 + taxol15Centriole laser-ablation ( 2:1 ) 11Control laser-ablation8Centriole laser-ablation ( 2:1 ) + Mps1 inhibitor6Control laser-ablation + Mps1 inhibitor6 To investigate how cells center the metaphase plate , we analyzed the timing of 2:1 cells in comparison to 2:2 or 1:1 cells by monitoring either HeLa eGFP-centrin2/H2B-mRFP or HeLa eGFP-centrin1/mRFP-α-tubulin cells . A 2:2 , 1:1 , or a 2:1 centriole configuration affected neither the timing of bipolar spindle assembly nor the time it took to align all chromosomes on the metaphase plate ( Figure 2A , B ) . Anaphase onset , however , was delayed by 12 min ( Figure 2C , p = 0 . 003 in Mann–Whitney U test ) in 2:1 cells , when compared to 2:2 or wild-type cells , resulting in 2:1 cells spending twice the amount of time in metaphase ( Figure 2D , p = 0 . 015 in Mann–Whitney U test ) . Anaphase time in 2:1 cells was also delayed in eGFP-centrin1/mRFP-α-tubulin cells 2:1 cells ( Figure 2E ) , indicating that it is a robust phenomenon ( p = 0 . 003 in Mann–Whitney U test ) . In contrast , anaphase was not delayed in 1:1 cells , indicating that the longer metaphase timing seen in 2:1 cells was not caused by the loss of daughter centrioles , but the consequence of centriole asymmetry ( Figure 2C , E ) . Anaphase onset is under the control of the SAC , which inhibits the anaphase-promoting complex/cyclosome and delays anaphase onset if kinetochores are not properly attached to spindle microtubules ( Khodjakov and Pines , 2010; Foley and Kapoor , 2013 ) . To test whether the observed metaphase delay depended on the SAC , we co-depleted the SAC protein Mad2 or inhibited the SAC kinase Mps1 in Sas-6-depleted cells ( Figure 2—figure supplement 1 ) . In the absence of Mad2 , 2:2 , 2:1 , 1:1 , or control-treated cells had the same anaphase timing , indicating that the anaphase delay in 2:1 cells depends on the SAC ( Figure 2F ) . The same result could be found in Mps1-inhibited cells ( Figure 2—figure supplement 2 ) . To test whether metaphase plate position is corrected as a result of the additional time provided by the SAC , we prolonged metaphase in siSas-6-treated cells by 1 hr by adding the proteasome inhibitor MG132 and found a symmetric plate position in 2:1 cells ( median R = 0 . 98; Figure 2G ) . In contrast , when we abrogated the SAC in metaphase cells by adding an inhibitor of the SAC kinase Mps1 , MPS1-IN ( Kwiatkowski et al . , 2010 ) , 2:1 cells showed a bimodal distribution: 53% entered anaphase with symmetric spindles , with a median R centered around 1 , however , the other 47% had asymmetric spindles with a distribution centered around R = 1 . 15 ( Figure 2H , Mann–Whitney U test , p = 0 . 032 when compared to untreated 2:1 cells at anaphase onset ) . We conclude that the centering of the metaphase plate is not a mechanical consequence of cells preparing for anaphase , but that it depends on the SAC , which provides 2:1 cells with sufficient time to build a symmetric spindle . Mps1 inhibition in metaphase also led to a higher rate of segregation errors in 2:1 than in 2:2 or wild-type cells; this increase was reproducible , but not statistically significant ( Figure 2I , J ) . 10 . 7554/eLife . 05124 . 005Figure 2 . The SAC delays anaphase in cells with asymmetric spindles allowing the centering of the metaphase plate . ( A ) Boxplots of the spindle assembly time ( NBD until bipolar spindle formation ) in wild-type and Sas-6-depleted 2:2 , 1:1 , or 2:1 eGFP-centrin1/mRFP−α-tubulin cells . Numbers indicate the median value , n = 49–59 cells in 2–6 experiments . ( B–D ) Boxplots for the time between NBD and metaphase ( B ) ; the time between NBD and anaphase onset ( C ) ; and the time between metaphase and anaphase ( D ) in wild-type and Sas-6-depleted 2:2 , 1:1 , or 2:1 eGFP1-centrin2/H2B-mRFP cells . * indicates statistically significant difference in C ( Mann–Whitney U test , p = 0 . 003 ) , and D ( Mann–Whitney U test , p = 0 . 015 ) , n = 36–100 cells in 6–13 experiments . ( E ) Boxplot for the time between NBD and anaphase B in wild-type and Sas-6-depleted 2:2 , 1:1 , or 2:1 eGFP1-centrin1/ mRFP−α-tubulin cells . * indicates statistically significant difference ( Mann–Whitney U test , p = 0 . 003 ) . ( F ) Boxplots for the time between NBD and anaphase onset in Mad2-depleted or Mad-2/Sas-6-depleted 2:2 , 1:1 or 2:1 eGFP1-centrin2/H2B-mRFP cells . 2:1 cells are not delayed ( Mann–Whitney U test , p = 0 . 836 ) . ( G ) Distribution of spindle ratios R in 2:1 eGFP-centrin1/CENPA cells treated with DMSO or MG132 . For cell numbers see Table 1 . ( H ) Distribution of spindle ratio R in Sas-6-depleted eGFP-centrin1/CENPA 2:1 cells in metaphase , at anaphase onset , or at anaphase onset when treated in metaphase with the Mps1 inhibitor MPS1-IN-1 . Values for metaphase and anaphase without Mps1-IN treatment were taken from Figure 1E for comparison . MPS1-IN treated 2:1 anaphase cells are significantly more asymmetric ( Mann–Whitney U test , p = 0 . 032 ) . ( I ) Quantification of chromosome segregation errors in eGFP-centrin1/CENPA cells under the indicated conditions ( n = 17–30 cells; N = 4–13 experiments ) . Error bars indicate s . e . m . ( J ) Illustrative live-cell imaging stills of eGFP-centrin1/CENPA cells in anaphase with ( right panel ) or without ( left panel ) chromosome segregation errors . SAC , spindle assembly checkpoint . DOI: http://dx . doi . org/10 . 7554/eLife . 05124 . 00510 . 7554/eLife . 05124 . 006Figure 2—figure supplement 1 . Validation of Sas-6 and Mad2 co-depletion . ( A , B ) Immunoblots of wild-type , siControl- , siSas-6 , or siSas-6/siMad2 treated eGFP1-centrin2/H2B-mRFP HeLa cells probed with ( A ) Sas-6 and α-tubulin antibodies or ( B ) with Mad2 and α-tubulin antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 05124 . 00610 . 7554/eLife . 05124 . 007Figure 2—figure supplement 2 . Mps1 inhibition suppresses the anaphase delay in 2:1 cells . Boxplots for the time between NBD and anaphase onset in wild-type , siSas-6-treated 2:2 , 1:1 , or 2:1 eGFP1-centrin2/H2B-mRFP cells treated with an Mps1 inhibitor . Numbers indicate average anaphase times . Note that 2:1 cells are not delayed when compared to 2:2 cells ( Mann–Whitney U test , p = 0 . 9673 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05124 . 007 To understand the mechanism behind metaphase plate centering , we investigated how the loss of a daughter centriole affects the spindle poles and why it may affect the position of the metaphase plate . We first tested whether loss of centrioles reduces the polar ejection force ( Rieder et al . , 1986 ) , as a possible explanation for the shorter half-spindle attached to the pole with fewer centrioles . Wild-type eGFP-centrin1/mRFP−α-tubulin cells were compared with cells treated with RNAis against Sas-6 or the chromokinesin Kid , the main driving force of the polar ejection force ( Wandke et al . , 2012 ) . Cells were subjected to a half-hour treatment with the Eg5 inhibitor monastrol to obtain monopolar spindles ( Mayer et al . , 1999 ) , fixed and stained for HEC1 to label kinetochores . The average distance between kinetochores and centrosomes in monopolar spindles reflects the polar ejection force ( Wandke et al . , 2012 ) , as seen by a reduced distance in Kid-depleted cells ( Figure 3A ) . Loss of 2 daughter centrioles in Sas-6-depleted cells ( one in each pole ) , however , did not reduce this distance , indicating that daughter centriole loss does not reduce the polar ejection force ( Figure 3A ) . In C . elegans , half-spindle size has been linked to the abundance of TPX-2 , an activator of Aurora-A , a critical regulator of centrosome maturation ( Nigg and Raff , 2009; Greenan et al . , 2010 ) ; however , by quantitative immunofluorescence we found no significant differences for TPX-2 or for phosphorylated Aurora-A ( the active form of Aurora-A ) between the two spindle poles in Sas-6-depleted 2:1 cells ( Figure 3B ) . Next , we quantified the levels of centrobin , which stabilizes microtubule minus-ends , since it localizes only to daughter centrioles ( Zou et al . , 2005; Jeffery et al . , 2010 ) . Consistent with the daughter-specific localization , centrobin was only present on the pole with two centrioles in 2:1 cells; in contrast , α-tubulin density in each half-spindle and the abundance of the centrosome proteins γ-tubulin , pericentrin , ninein , and p150glued ( a marker for the dynein/dynactin motor complex ) on the two spindle poles in 2:1 cells were as symmetric as in wild-type or 2:2 cells ( Figure 3C–F ) . Since centrobin depletion leads to a SAC-dependent mitotic arrest due to unstable microtubule minus-ends and unstable kinetochore–microtubules ( Jeffery et al . , 2010 ) , we hypothesized that the asymmetric localization of centrobin in 2:1 cells may lead to a difference in microtubule stability between the two poles . To test this , we subjected 2:2 , 2:1 , and 1:1 eGFP-centrin1/CENPA cells to a 7-min cold treatment on ice to depolymerize the astral microtubules and stained the remaining cold-resistant kinetochore–microtubules ( Salmon and Begg , 1980 ) . The stability of kinetochore–microtubules varied from cell to cell , but in 75% of the 2:2 or 1:1 cells the microtubule minus-ends showed the same stability at both spindle poles ( Figure 3G , H ) . In contrast , in 2:1 cells , an unequal stability of the minus-ends was visible in over 50% of the cells , and in most cases it was the pole with 2 centrioles that had more stable minus-ends ( Figure 3G , H ) . This difference in kinetochore–microtubule minus-end stability suggested that microtubule minus-ends might depolymerize faster in the absence of a daughter centriole , which would explain the shorter half-spindles associated with the 1-centriole pole . We also noted in our live-cell imaging experiments that >70% of the spindles in 2:1 eGFP-centrin1/mRFP-α-tubulin cells ( but not 2:2 or 1:1 cells ) were rotating , a phenomenon that could reflect a difference in minus-end stability of the two astral microtubule populations , resulting in an imbalance of cortical forces in the two half-spindles ( Figure 3I and Videos 1–3 ) . To confirm this hypothesis , we first tested whether a 1-hr MG132 treatment , which leads to symmetric spindles in 2:1 cells , is accompanied by an equalization in microtubule stability at the two spindle poles . While in control-treated cells , we saw an unequal stability of the minus-ends in 49 ± 4% of the 2:1 cells , this was only the case in 22 ± 6% of the MG132-treated 2:1 cells ( p = 0 . 0275 in unpaired t-test; Figure 3J ) . In contrast , in 2:2 cells , MG132 treatment did not affect the percentage of cells with unequal minus-end stability ( 14% each; Figure 3J ) . Next , we tested whether the asymmetry of the spindle and the spindle rotation phenotype could be suppressed by increasing microtubule stability . We therefore co-depleted the microtubule-depolymerases KIF2a and MCAK , a condition known to increase microtubule stability ( Ganem and Compton , 2004 ) . KIF2a/MCAK depletion per se did not affect the position of the metaphase plate ( median R = 1 . 01 ) or the ratio of cells with rotating spindles ( Figure 3K , L ) . However , when KIF2a/MCAK was co-depleted in 2:1 cells , the spindle asymmetry was statistically less pronounced ( median spindle ratio R = 1 . 08 and 19 . 2% R > 1 . 15 compared to median R = 1 . 12 and 42 . 9% R > 1 . 15 in Sas-6 deplete cells , p = 0 . 039 Mann–Whitney test ) , and spindles did not rotate ( Figure 3L , M ) . To corroborate these results , we also stabilized microtubules by applying a brief , 30 min , 10 nM taxol treatment to Sas-6-depleted cells . While control ( DMSO ) -treated 2:1 cells were still asymmetric ( median R = 1 . 23 and 56 . 3% R > 1 . 15 ) , taxol-treated cells were much closer to symmetry ( median R = 1 . 02 and 13 . 3% R > 1 . 15 , p = 0 . 0062; Figure 3N ) . We conclude that loss of a single centriole leads to a difference in minus-end microtubule stability between the two spindle poles , that this difference plays a large role in the asymmetric positioning of the metaphase plate , and that cells reduce this difference as they center the metaphase plate before anaphase onset . 10 . 7554/eLife . 05124 . 008Figure 3 . 2:1 cells have half-spindles with different microtubule stability and fail to mature kinetochore–microtubule attachments . ( A ) eGFP-centrin1 cells ( green ) were treated with monastrol and stained for HEC1 ( red; right panel ) to calculate the distance between kinetochores and the closest centriole . The left panel shows the distribution of centriole-HEC1 distances from n =16−28 cells , >1000 kinetochores . Scale bar indicates 2 μm . ( B , C ) The difference in centrosomal levels of TPX-2 and phospho-Aurora-A ( B ) , and centrobin between each spindle pole was quantified in HeLa eGFP-centrin1 by immunofluorescence using the indicated formula , and plotted as boxplots for each centriole configuration; n = 12–63 cells . Scale bar indicates 2 μm . ( D ) Representative image of wild-type and siSas-6-treated 2:1 eGFP-centrin1 ( green ) /mRFP−α-tubulin ( white ) cells stained with anti-centrobin sera ( red ) . Insets show magnified centrioles . Scale bar indicates 2 μm . ( E ) Quantification of the difference in α-tubulin ( red signal ) levels between the two half-spindles , according to the formula shown in the box . Results were plotted in the left panel using a boxplot . n = 19–25 cells , N = 2 experiments . Scale bar indicates 2 μm . ( F ) The difference in centrosomal levels of γ-tubulin , pericentrin , ninein and p150glued between each spindle pole was quantified as in ( B ) and plotted as boxplots for each centriole configuration; n = 18–68 cells . ( G ) Immunofluorescence images of 2:2 and 2:1 siSas-6 eGFP-centrin1/CENPA ( green ) cells treated for 7 min with ice-cold medium and stained with anti-α-tubulin sera ( magenta ) . Subsetted images are maximum intensity projections of 10 stacks ( z = 0 . 2 μm ) around centrioles . Scale bar indicates 2 μm . ( H ) Quantification of kinetochore–microtubule minus-end stability at poles . Bar graph indicates percentage of cells that have asymmetric levels of kinetochore–microtubule minus-ends at the poles after cold-treatment; n = 26–49 cells; * indicates that within the 2:1 cell population the minus-end stability was significantly higher at the pole with 2 centriole ( p = 0 . 00082 exact binomial test ) . ( I ) Quantification of spindle rotation in control- and Sas-6-depleted 2:2 , 2:1 , or 1:1 eGFP-centrin1 ( green ) /mRFP−α-tubulin ( red ) cells based on time-lapse images as shown in the right panels . Times indicate minutes after NBD . Scale bar indicates 5 μm . A spindle was counted as rotating if it had turned by more than 90° in X/Y . n = 32–122 cells in 2–6 experiments . * indicates significant difference; Fisher's exact test p = 8 . 39e-09 . ( J ) Quantification of kinetochore–microtubule stability at poles . Bar graph indicates percentage of cells that have more stable kinetochore–microtubule minus-ends either at the pole with the grandmother centriole ( 2:2 ) cells or at the 2-centriole pole ( 2:1 cells ) ; n = 20–40 cells in N = 3 independent experiments; * indicates that the MG132 treatment significantly reduced the percentage of cells with more stable minus-ends at the 2-centriole pole ( p = 0 . 0275 in unpaired t-test ) . ( K ) Distribution of spindle ratios R in wild-type and siKIF2a/MCAK-treated HeLa eGFP-centrin1/CENPA cells in metaphase . For cell number see Table 1 . ( L ) Quantification of spindle rotation in wild type , Sas-6-depleted 2:1 , KIF2a/MCAK-depleted , or KIF2a/MCAK/Sas-6-depleted 2:1 eGFP-centrin2/H2B-mRFP cells , n = 18–37 cells in 1–3 experiments . * indicates significant difference; p = 0 . 024 in Fisher's exact test . See also Videos 1–3 . ( M ) Distribution of spindle ratios R in Sas-6-depleted 2:1 and KIF2a/MCAK/Sas-6-depleted 2:1 eGFP-centrin1/CENPA cells in metaphase . For cell numbers see Table 1 . ( N ) Distribution of spindle ratios R in metaphase in Sas-6-depleted 2:2 and 2:1 cells treated either with DMSO or 10 nM taxol . For cell numbers see Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05124 . 00810 . 7554/eLife . 05124 . 009Video 1 . Sas-6-depleted 2:2 HeLa cell expressing eGFP-centrin1 ( centriole marker; green ) and mRFP-α-tubulin ( microtubules; red ) in mitosis . Time is indicated in minutes . DOI: http://dx . doi . org/10 . 7554/eLife . 05124 . 00910 . 7554/eLife . 05124 . 010Video 2 . Sas-6-depleted 2:1 HeLa cell expressing eGFP-centrin1 ( centriole marker; green ) and mRFP-α-tubulin ( microtubules; red ) in mitosis . Time is indicated in minutes . Note the spindle rotation movements . DOI: http://dx . doi . org/10 . 7554/eLife . 05124 . 01010 . 7554/eLife . 05124 . 011Video 3 . Sas-6-depleted 1:1 HeLa cell expressing eGFP-centrin1 ( centriole marker; green ) and mRFP-α-tubulin ( microtubules; red ) in mitosis . Time is indicated in minutes . DOI: http://dx . doi . org/10 . 7554/eLife . 05124 . 011 In the next step , we investigated whether the difference in minus-end stability in 2:1 cells translated into kinetochore–microtubule attachment defects sensed by the SAC . The SAC responds to unattached kinetochores or to sister-kinetochores with insufficient tension ( measured as the distance between the two sister-kinetochores ) that become transiently detached due to the kinase activity of Aurora-B ( Foley and Kapoor , 2013 ) . Depletion of centrobin on both spindle poles leads to unstable kinetochore–microtubules and reduced inter-kinetochore distances that result in a permanent mitotic arrest ( Jeffery et al . , 2010 ) . When we tracked kinetochores in 2:1 eGFP-centrin1/CENPA cells , however , we found no difference in inter-kinetochore distances or the oscillatory sister-kinetochore movements along the spindle axis ( Figure 4A , B; Jaqaman et al . , 2010 ) . Since 42 . 9% of the 2:1 cells have asymmetric spindles , we restricted our analysis to only include cells that have a spindle ratio R > 1 . 15 , but again found no change in inter-kinetochore distances ( Figure 4A ) . As SAC satisfaction is also linked to intra-kinetochore stretching ( Maresca and Salmon , 2009; Uchida et al . , 2009 ) , we further measured the intra-kinetochore distance between CENPA and the outer kinetochore protein HEC1 and found no change in 2:1 cells compared to 2:2 or 1:1 cells ( Figure 4C ) ; the observed distance of 110 nm is consistent with previous studies ( Wan et al . , 2009 ) . Overall , this indicated that sister-kinetochores in 2:1 cells formed bipolar attachments with mechanical behaviors that were indistinguishable from normal cells . Since a single unattached kinetochore is sufficient to elicit a SAC response ( Rieder et al . , 1995; Collin et al . , 2013 ) , we next investigated whether 2:1 cells have rare unattached kinetochores , by measuring the proportion of metaphase cells with Mad2-positive kinetochores ( marker for unattached kinetochores ) . When compared to 2:2 , 1:1 , or wild-type cells , we found a small but non-significant increase in the proportion of Mad2-positive 2:1 cells ( Figure 4D ) , which could point to a minor or transient attachment defect caused by the imbalance of microtubule stability within the spindles of 2:1 cells . Such imbalance could lead to kinetochores that are not fully attached , that is , not bound by the full complement of microtubules ( up to 25 microtubules per kinetochore in vertebrate cells; Rieder , 1982 ) . To detect immature attachments , metaphase eGFPcentrin1/CENPA cells were stained with antibodies against SKAP , a marker for kinetochores fully bound by stable microtubules ( Schmidt et al . , 2010 ) , and analyzed for the presence of SKAP on kinetochores: cells in which more than 30% of the kinetochore-pairs were SKAP-negative were considered to have partially unstable kinetochore–microtubules ( Figure 4E , F ) . Such cells were rare in wild-type , 2:2 , or 1:1 cells ( <10% ) , but formed a substantial proportion of 2:1 cells ( 22%; Figure 4F ) . These SKAP-negative kinetochores were distributed in a symmetric manner ( were present both on the 1- or 2-centriole side; data not shown ) , and they all disappeared if 2:1 cells were briefly treated with 10 nmol taxol ( Figure 4F ) . Moreover a 1-hr treatment with the proteasome inhibitor MG132 also abolished the subpopulation of 2:1 cells with SKAP-negative kinetochores ( Figure 4F ) . Overall this suggested that in 2:1 cells , sister-kinetochores form amphitelic microtubule attachments , but that differences in minus-end stability delay stabilization of kinetochore–microtubule attachments in both half-spindles , as reflected by higher levels of SKAP-negative kinetochores . 10 . 7554/eLife . 05124 . 012Figure 4 . 2:1 cells have immature kinetochore–microtubule attachments . ( A , B ) Analysis of inter-kinetochore distances and sister-kinetochore oscillations in wild-type , Sas-6-depleted 2:2 , 1:1 , 2:1 , or the subset of 2:1 eGFP1-centrin1/CENPA metaphase cells with an asymmetric plate position based on our in-house kinetochore tracking assay ( Jaqaman et al . , 2010 ) , n = 620–889 kinetochores in 36–48 cells . The distribution of inter-kinetochore distances ( CENPA to CENPA distance ) is shown in A ( no significant difference , t-test , p = 0 . 99 ) , and the autocorrelation of the sister-kinetochore movements in B . The first minima of the autocorrelation curve indicate the half-period of the chromosome oscillations , and their depth the regularity of the oscillations . ( C ) Distribution of intra-kinetochore distances in wild-type and Sas-6-depleted 2:2 , 1:1 , or 2:1 eGFP1-centrin1/CENPA metaphase cells . Cells were stained with antibodies against the N-terminus of HEC1 . Using the tracking assay , we determined for each sister-kinetochore pair the CENPA–CENPA and the HEC1-HEC1 distances , and calculated the CENPA-HEC1 distances by halving the difference , n = 701–790 kinetochores in 26–30 cells in 3 experiments ( no significant difference , Mann–Whitney test , 2:1 vs 2:2 , p = 0 . 203 ) . ( D ) Quantification of Mad2-positive kinetochores in wild-type or Sas-6-depleted 2:2 , 1:1 , or 2:1 cells . eGFP-centrin1 ( green ) metaphase cells were stained with anti-Mad2 ( green ) , and CREST sera ( magenta; left panel ) and the number of Mad2-positive kinetochores quantified in the right panel ( n = 50–102 cells in 2 ( wt ) or 8 ( siSas-6 ) experiments; no significant difference was found; Fisher's exact test , p = 0 . 17 ) . Scale bar indicates 2 μm . ( E , F ) Quantification of SKAP-negative kinetochores in wild-type , siSas-6-depleted 2:2 , 2:1 , 2:1 + DMSO , 2:1 , 2:1 + MG132 , and 2:1 + taxol-treated eGFP-centrin1/CENPA cells . Cells were stained with antibodies against the kinetochore protein SKAP ( magenta ) , as shown in E ( maximum-intensity projection of 8 stacks [z = 0 . 3 μm] ) . Using the eGFP-CENPA ( green ) signal , we quantified the number of sister-kinetochore pairs with at least one SKAP-negative kinetochore ( as shown in inlets ) . Quantification in F shows the percentage of cells where more than 30% of the sister-pairs were SKAP-negative . Fisher's exact test for 2:1 > 2:2 , p = 0 . 0013 . n = 46–72 cells in N = 4–7 experiments . Scale bar indicates 2 μm . ( G ) Example of an eGFP-centrin1/CENPA ( green ) cell in which a single daughter centriole was ablated ( white arrow indicates the location of the laser pulse ) . Cells were fixed and stained with anti-centrin sera ( magenta ) to confirm the loss of a centriole , as opposed to the mere bleaching of eGFP-centrin1 . Scale bar indicates 5 μm . ( H ) Plot of half-spindle ratio R over time in 11 single eGFP-centrin1/CENPA cells in which a single centriole was ablated . The time point of laser ablation is t = 0 . The thick red curve indicates the median of R of laser ablated 2:1 cells ( * denotes when median R is asymmetric [p < 0 . 01] ) , the thick black curve indicates the median R distribution of 8 control-ablated cells . ( I ) Average inter-kinetochore distances in eGFP-centrin1/CENPA cell before or after a single daughter centriole was ablated as determined by the kinetochore tracking assay . Error bars indicate s . e . m . n = 11 cells , 2 time points before and after ablation and on average 20 kinetochores per cell . * denotes a statistically significant difference ( p = 0 . 001 in two-tailed paired t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05124 . 012 Since our results contrasted with recent hypotheses that suggest that correct force generation at kinetochores is locally regulated and does not require a direct connection between kinetochores and centrosomes ( Sikirzhytski et al . , 2014 ) , we aimed to exclude a possible off-target effect of Sas-6 depletion on kinetochore–microtubule attachments . For this purpose , we used laser-microsurgery to ablate one daughter centriole in cells that had already reached metaphase . As in Sas-6-depleted cells , 2:1 cells generated by microsurgery resulted in an asymmetric plate location , as , on average , R increased from 1 . 06 to 1 . 16 after laser-ablation ( Figure 4G , H and Video 4 ) . These cells displayed a reduced inter-kinetochore distance ( mean of 1 . 03 μm before ablation to 0 . 95 μm after the ablation , p = 0 . 001 in paired two-tailed t-test ) , which led to a prolonged metaphase arrest ( Figure 4I ) ; in contrast , a control laser pulse in the vicinity of the spindle pole did not affect half-spindle lengths or inter-kinetochore distances ( data not shown ) . This confirmed that loss of a single daughter centriole can directly affect force generation at sister-kinetochores , consistent with our findings that Sas-6-depleted 2:1 cells show defects in the quality of kinetochore–microtubule attachments . We also conclude that an acute laser-ablation of a daughter centriole in metaphase has a more severe effect on the forces acting on kinetochore than Sas-6 depletion . Possible explanations for this difference could be that Sas-6-depleted 2:1 cells might have more time to adapt to the lack of a missing centriole as they progressively build up the spindle , or that the laser-ablation destroys not just a centriole but also part of enzymatic activities in the vicinity of the centriole , such as minus-end depolymerases , which in normal cells are known to exert a pulling force on kinetochore-fibers ( Meunier and Vernos , 2011 ) . 10 . 7554/eLife . 05124 . 013Video 4 . Laser-ablated 2:1 HeLa cell expressing eGFP-centrin1 ( centriole marker ) and eGFP-CENPA ( kinetochore marker ) in metaphase . Note the asymmetric metaphase plate position after the ablation of a single centriole . DOI: http://dx . doi . org/10 . 7554/eLife . 05124 . 013 To confirm the presence of partially destabilized kinetochore–microtubule attachments in siSas-6-treated 2:1 cells and test whether they are the cause of the SAC response , we used the Aurora-B inhibitor ZM1 . Aurora-B inhibition does not overcome a SAC-dependent mitotic arrest caused by unattached kinetochores ( nocodazole treatment ) , but overcomes a mitotic arrest caused by insufficient tension at sister-kinetochores in monopolar spindles ( monastrol treatment ) , as Aurora-B inhibition stabilizes kinetochore–microtubules and prevents loss of kinetochore–microtubule attachment ( Figure 5A; Ditchfield et al . , 2003; Lampson et al . , 2004 ) . When we acutely inhibited Aurora-B in metaphase 2:1 cells , cells entered anaphase with asymmetric spindles ( median R = 1 . 18 , Mann–Whitney U test , p = 7 . 57 × 10−6 when compared to anaphase onset without Aurora-B inhibitor ) , indicating that a stabilization of kinetochore–microtubules satisfies the SAC in 2:1 cells ( Figure 5B ) . To confirm these findings in an independent manner , we also tested whether depletion of the microtubule depolymerases KIF2a/MCAK would allow 2:1 cells to enter anaphase with asymmetric spindles . Depletion of KIF2a/MCAK per se did not affect plate position at anaphase onset ( median R = 1 . 03 ) ; however , it allowed 2:1 cells to enter anaphase with asymmetric spindles ( median R 1 . 08 , p < 0 . 0001 compared to symmetric distribution in Mann–Whitney U test; Figure 5C ) . KIF2a/MCAK depletion also rescued the anaphase timing difference in 2:1 cells compared to 2:2 cells and led to the correct loading of SKAP on kinetochores ( Figure 5D , E; note that KIF2a/MCAK depletion alone led to a small anaphase delay when compared to wild-type cells , as previously reported [Ganem et al . , 2005] ) . This showed that co-depletion of KIF2a/MCAK overrides the SAC response in 2:1 cells , leaving them no time to center the metaphase plate , indicating that unstable kinetochore–microtubules cause the SAC-dependent anaphase delay . KIF2a/MCAK co-depletion , however , did not suppress a mitotic arrest caused by the presence of insufficient tension ( monastrol treatment ) or unattached kinetochores ( nocodazole treatment; Figure 5A ) . Since the spindle checkpoint response has been shown to be subject to off-target effects ( Hübner et al . , 2009; Westhorpe et al . , 2010 ) , we repeated these experiments with an alternative set of KIF2a and MCAK siRNAs . These experiments validated our initial findings , confirming that KIF2a/MCAK suppress the anaphase delay seen in Sas-6-depleted 2:1 cells ( Figure 5—figure supplement 1A ) . Validation of Sas-6 , KIF2a , and MCAK depletion siRNA treatments by immunoblotting also showed that in the triple MCAK/KIF2a/Sas-6 depletion , KIF2a was only partially depleted ( Figure 5—figure supplement 1B ) . We therefore tested whether the depletion of MCAK alone in 2:1 cells is sufficient to suppress the mitotic delay . As this was not the case , we conclude that depletion of both MT-depolymerases is necessary to overcome the instability of kinetochore–microtubules in 2:1 cells ( Figure 5—figure supplement 1C ) . 10 . 7554/eLife . 05124 . 014Figure 5 . Depleting KIF2a and MCAK overcomes the SAC response in 2:1 cells . ( A ) Mitotic index of untreated , ZM-1-treated , MPS1-IN-treated , control-depleted , or KIF2a/MCAK-depleted cells treated for 16 hr with nocodazole ( unattached kinetochores ) or monastrol ( lack of tension ) , n ≥ 400 cells in 3–4 experiments , error bars indicate s . e . m . * ZM1 and MPS1-IN overcome a monastrol arrest ( t-test p < 0 . 0001 ) , and MPS1-IN overcomes a nocodazole arrest ( t-test p = 0 . 0044 ) . ( B ) Distribution of spindle ratio R in Sas-6-depleted eGFP-centrin1/CENPA 2:1 cells at anaphase onset treated with or without the Aurora-B inhibitor ZM1 . Data from Figure 1E without Aurora-B inhibition are shown for comparison . Aurora-B inhibition allows cells to enter anaphase with asymmetric spindles ( n = 12 cells; Mann–Whitney U test , p = 7 . 57 × 10−6 ) . ( C ) Distribution of spindle ratios R in Sas-6-depleted 2:1 , KIF2a/MCAK-depleted , and KIF2a/MCAK/Sas-6-depleted 2:1 eGFP-centrin1/CENPA cells in metaphase or at anaphase onset . ( D ) Boxplots of anaphase timing of wild-type , KIF2a/MCAK-depleted , or KIF2a/MCAK/Sas-6 depleted 2:2 and 2:1 eGFP1-centrin2/H2B-mRFP cells . n = 23–61 cells in 1–3 experiments . ( E ) Quantification of SKAP-negative kinetochores as in Figure 4F in Sas-6-depleted 2:1 and KIF2a/MCAK/Sas-6-depleted 2:1 eGFP-centrin1/CENPA cells . SAC , spindle assembly checkpoint . DOI: http://dx . doi . org/10 . 7554/eLife . 05124 . 01410 . 7554/eLife . 05124 . 015Figure 5—figure supplement 1 . Validation of Sas-6 , KIF2a , and MCAK co-depletion . ( A ) Boxplots of anaphase timing of siControl , siSas-6 2:2 , siSas-6 2:1 , siCtrl/KIF2a/MCAK , siKIF2a/MCAK/Sas-6 2:2 , and siKIF2a/MCAK/Sas-6 2:1 eGFP1-centrin1/CENPA cells . n = 14–78 cells in 3 independent experiments . Note that siSas-6 2:1 cells are delayed compared to siSas-6 2:2 cells ( p < 0 . 00001 in Mann–Whitney test ) , but that siKIF2a/MCAK/Sas-6 2:1 are not delayed compared to siKIF2a/MCAK/Sas-6 2:2 cells . ( B ) Immunoblots of eGFP1-centrin2/H2B-mRFP cells treated with the indicated siRNA and probed with anti-Sas6 , anti-MCAK , anti-KIF2A and anti-α-tubulin ( loading control ) antibodies . The relative ratio with the α-tubulin signal is quantified for each condition on the right panels . Note the KIF2a siRNA only led to a partial depletion . ( C ) Boxplots of anaphase timing of siControl , siCtrl/MCAK , siMCAK/Sas-6 2:2 , siMCAK/Sas-6 2:1 eGFP1-centrin1/CENPA cells . n = 9–169 cells in 3 independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 05124 . 015 Since the mere depletion of KIF2a/MCAK allowed 2:1 cells to enter anaphase with asymmetric spindles , we could investigate the functional importance of a centered metaphase plate position beyond anaphase onset ( Aurora-B inhibition could not be used , since it blocks anaphase B and cytokinesis [Ditchfield et al . , 2003] ) . We found that KIF2a/MCAK/Sas-6-depleted 2:1 cells had slightly higher rates of chromosome segregation errors ( statistically insignificant difference , p = 0 . 4713 , Fischer's exact test ) when compared to KIF2a/MCAK-depleted cells ( Figure 6A ) . Strikingly KIF2a/MCAK-depleted 2:1 cells also underwent asymmetric cell divisions , yielding two daughter cells of different volumes ( Figure 6B and Videos 5 , 6 ) : while the two daughter cell volumes never differed by more than 20% in wild-type or KIF2a/MCAK depleted cells , 50% of KIF2a/MCAK/Sas-6-depleted 2:1 cells yielded one daughter cell that was at least 20% larger than its sister progeny ( Figure 6C ) . This suggested that an asymmetric metaphase plate position leads to asymmetric cell division . A recent study , however , demonstrated that asymmetric position of the entire spindle can also lead to asymmetric cell divisions in HeLa cells ( Kiyomitsu and Cheeseman , 2013 ) . To discriminate between the two possibilities , we quantified the position of the spindle center in relationship to the cell center at anaphase onset , which was often not centered in the middle of the cell at anaphase onset , as has been previously reported ( Figure 6D; Collins et al . , 2012; Kiyomitsu and Cheeseman , 2013 ) . The extent to which the spindle centers were offset was the same in KIF2a/MCAK-depleted and KIF2a/MCAK-depleted 2:1 cells , indicating that changes in spindle position were not at the origin of the asymmetric cell divisions seen in KIF2a/MCAK-depleted 2:1 cells . To independently confirm our hypothesis that an asymmetric metaphase plate position in anaphase leads to an asymmetric cell division , we acutely abrogated the spindle checkpoint with an Mps1 inhibitor in control-depleted cells , Sas-6-depleted 2:2 cells , Sas-6-depleted 2:1 cells in late prometaphase/early metaphase ( as visualized by live-cell imaging , where 43% of the cells have a metaphase plate not in the middle of the spindle , see Figure 1F ) or Sas-6-depleted 2:1 cells that were in late metaphase , which have symmetric spindles . Mps1 inhibition did not give rise to asymmetric cell division in control-depleted cells , 2:2 cells , or 2:1 cells treated in late metaphase; in contrast , in late prometaphase/early metaphase 2:1 cells with an asymmetric plate position , Mps1 inhibition resulted in an asymmetric cell division in 45% of the cases ( Figure 6E ) . The same result was also seen after acute Mps1 inhibition in laser-ablated cells: cells that were treated with a laser pulse in the cytoplasm divided in a symmetric manner; in contrast , 67% of the cells in which a single centriole was laser-ablated divided in an asymmetric manner when treated with an Mps1 inhibitor ( Figure 6F and supplementary Videos 7 , 8 ) . We conclude that entry into anaphase with an asymmetric position of the metaphase plate results in an asymmetric cell division . 10 . 7554/eLife . 05124 . 016Figure 6 . An asymmetric plate position at anaphase onset leads to segregation errors and asymmetric cell division . ( A ) Quantification of chromosome segregation errors in KIF2a/MCAK-depleted and KIF2a/MCAK/Sas-6-depleted 2:1 eGFP-centrin1/eGFP-CENPA cells based on time-lapse images . n = 12–21 cells in 2–4 experiments . Error bars indicate s . e . m . ( B , C ) Wild-type , KIF2a/MCAK-depleted or KIF2a/MCAK/Sas-6-depleted 2:1 eGFP-centrin1/CENPA cells were recorded by time-lapse imaging using the eGFP-centrin1 signal to count centrioles and phase contrast to detect the cell membrane as shown in B for a siKIF2a/MCAK/Sas-6 2:1 cell ( scale bar = 5 μm ) . Phase contrast images were used to quantify the ratio of the two daughter cell sizes , which was plotted as a histogram in C . Half the Kif2a/MCAK/Sas-6-depleted 2:1 cells had a ratio of over 1 . 2 , a ratio never observed in other conditions ( t-test with Welch's correction between KIF2a/MCAK and KIF2a/MCAK/Sas-6 2:1 , p = 3 . 1e-05; n = 20–45 cells in 2–4 experiments ) . ( D ) Quantification of spindle center position in relation to cell center in KIF2a/MCAK-depleted or KIF2a/MCAK/Sas-6 2:1 eGFP-centrin1/CENPA cells at anaphase onset . n = 20–45 cells in 2–4 experiments . ( E ) Wild-type HeLa eGFP-centrin1/eGFP-CENPA cells , 2:2 cells , 2:1 cells in late prometaphase ( still 1–2 chromosomes not perfectly aligned on the plate ) , or 2:1 cells in late metaphase ( plate perfectly in the middle ) were treated with an Mps1 inhibitor and recorded by time-lapse imaging using phase contrast to detect the cell membrane as shown in B . Shown is the ratio of the two daughter cell sizes; 45% of the 2:1 cells treated in late prometaphase/early metaphase had a ratio of over 1 . 2 , a ratio never observed in other conditions ( t-test with Welch's correction between 2:1 cells in late prometaphase and 2:1 cells in late metaphase , p = 0 . 0141; n = 11–17 cells in 3 experiments ) . See also Videos 5 , 6 . ( F ) HeLa eGFP-centrin1/CENPA cells were treated with a laser pulse in the cytoplasm ( control ) or ablation of a centriole ( 2:1 cells ) and acutely treated with an Mps1 inhibitor to force cells into anaphase . Shown is the ratio of the two daughter cell sizes . Note that 67% of the 2:1 cells treated in late prometaphase/early metaphase had a ratio of over 1 . 2 , a ratio never observed in other conditions ( t-test with Welch's correction between control and 2:1 cells p = 0 . 0451; n = 6 ) . See also Videos 7 , 8 . DOI: http://dx . doi . org/10 . 7554/eLife . 05124 . 01610 . 7554/eLife . 05124 . 017Video 5 . KIF2a/MCAK-depleted HeLa cell expressing eGFP-centrin1 ( centriole marker; green ) and eGFP-CENPA ( kinetochore marker; green ) entering anaphase and recorded with phase contrast microscopy . DOI: http://dx . doi . org/10 . 7554/eLife . 05124 . 01710 . 7554/eLife . 05124 . 018Video 6 . KIF2a/MCAK/Sas-6-depleted 2:1 HeLa cell expressing eGFP-centrin1 ( centriole marker; green ) and eGFP-CENPA ( kinetochore marker; green ) entering anaphase and recorded with phase contrast microscopy . Note the asymmetric cell division . DOI: http://dx . doi . org/10 . 7554/eLife . 05124 . 01810 . 7554/eLife . 05124 . 019Video 7 . Laser-ablated control ( ablation in the cytoplasm ) HeLa cell expressing eGFP-centrin1 ( centriole marker ) and eGFP-CENPA ( kinetochore marker ) treated in metaphase with an Mps1 inhibitor . Shown is the GFP-fluorescence channel ( left ) and the DIC channel ( right ) . Note how the cell divides in a symmetric manner . DOI: http://dx . doi . org/10 . 7554/eLife . 05124 . 01910 . 7554/eLife . 05124 . 020Video 8 . Laser-ablated 2:1 HeLa cell expressing eGFP-centrin1 ( centriole marker ) and eGFP-CENPA ( kinetochore marker ) treated in metaphase with an Mps1 inhibitor . Shown is the GFP-fluorescence channel ( left ) and the DIC channel ( right ) . Note how the cell divides in an asymmetric manner . DOI: http://dx . doi . org/10 . 7554/eLife . 05124 . 020 Here , we show that an equatorial position of the metaphase plate in the middle of the spindle is necessary for symmetric cell divisions and demonstrate that cells actively center the metaphase plate before anaphase onset . Metaphase plate centering requires the SAC , which provides cells with enough time to correct metaphase plate position . The SAC responds to subtle defects in kinetochore–microtubule stability that arise in cells with an asymmetric plate position and an imbalance of centrioles , implying that the SAC is more sensitive than previously assumed . Recent studies have shown that proper positioning of the spindle ensures symmetric cell divisions , and that , deviations from a symmetric position are corrected by dynein-dependent cortical forces and membrane elongation during anaphase ( Kiyomitsu and Cheeseman , 2013 ) . Here , we find that this external cortical correction mechanism in anaphase is complemented in metaphase by an internal centering mechanism that ensures a symmetric position of the metaphase plate within the spindle with the help of the SAC . This centering mechanism is particularly visible in cells with an asymmetric distribution of centrioles ( 2:1 cells ) , but it also acts in wild-type cells , indicating that it is active in every cell division . The centering of the metaphase plate is not just a consequence of establishing stable bipolar attachments at kinetochores , since KIF2a/MCAK/Sas-6-depleted 2:1 cells fail to center the plate , despite having reached stable bipolar attachments that satisfy the SAC; it is an active correction process , which in part depends on the regulation of microtubule dynamics , but whose precise molecular mechanisms will need to be uncovered . KIF2a/MCAK-depleted 2:1 cells fail to center the metaphase plate before anaphase and divide asymmetrically . These cells have no defect in spindle positioning when compared to the symmetrically-dividing KIF2a/MCAK-depleted cells , implying that the asymmetric position of the metaphase plate is the source of asymmetric cell division . This hypothesis is confirmed by our analysis of 2:1 cells treated with an Mps1 inhibitor: only 2:1 cells treated in late prometaphase/early metaphase , which still have asymmetrically positioned metaphase plates , give rise to asymmetric cell division , whereas the same 2:1 cell population treated in late metaphase , which have a centered metaphase plate position , carry out symmetric cell divisions . We thus postulate that a symmetric metaphase plate position is essential for symmetric cell divisions , explaining why it is conserved in all metazoans , plants , and many fungi . Control of this parameter is essential , since differences in cell size have been linked to cell fate ( Kiyomitsu and Cheeseman , 2013 ) . Metaphase plate position may also play a crucial role in asymmetric cell divisions that depend on asymmetric spindles in anaphase , such as in embryonic D . melanogaster neuroblasts . To form asymmetric spindles in a controlled and stereotypical manner , cells need an internal reference in space: breaking an existing symmetry , that is , a symmetric metaphase plate position , provides such a reference point . This is consistent with the progression of embryonic fly neuroblasts , which first align the metaphase plate in the middle of the spindle , before undergoing an asymmetric elongation of the spindle in anaphase . Our results also shed light on the mechanisms controlling the position of the cytokinetic furrow . Original studies in sand dollar eggs showed that the position of the centrosomes is a key determinant of the cytokinetic furrow position ( Rappaport , 1961 ) ; later studies in C . elegans found that a second signal emanating from the spindle midzone also contributes to the positioning of the cytokinetic furrow ( Dechant and Glotzer , 2003; Bringmann and Hyman , 2005 ) . A role for chromosomes was , however , discarded in these two organisms , since midzone formation and cytokinesis did not require them . In contrast , in human cells , chromosomes stabilize microtubules of the midzone and thus favor the formation of a cytokinetic furrow ( Canman et al . , 2003 ) . Here , we show that 2:1 cells only misplace the cytokinetic furrow in the presence of an asymmetric plate position in metaphase , implying that the position of the metaphase plate plays a crucial fine-tuning role in the positioning of the cytokinetic furrow . Future studies will have to test whether the metaphase plate acts via the microtubules of the midzone , or as recently postulated , by influencing the cortical populations of Anillin and Myosin in anaphase in a Ran-GTP-dependent manner ( Kiyomitsu and Cheeseman , 2013 ) . The depletion of KIF2a/MCAK satisfies the SAC in 2:1 cells , but not in cells with unattached or tension-free kinetochore-microtubule attachments , indicating that the SAC responds to kinetochore–microtubule attachments defects less severe than lack of attachment or a tension defect . What might be these defects ? Kinetochores in 2:1 cells bind a sufficient number of microtubules to form amphitelic attachments and stretch the two sister-kinetochores apart , but a number of kinetochores do not bind the full complement of stable microtubules required for SKAP loading . It is established that the SAC responds to detached kinetochores and is satisfied when kinetochores have bound the full set of microtubules . Based on our results , we postulate that the SAC also responds if a kinetochore is only bound by a fraction of the full set of microtubules . Such a lack of full occupancy would delay anaphase onset , such as in 2:1 cells; it would also explain the tendency for a higher rate of chromosome segregation errors in 2:1 cells treated with an Mps1 inhibitor or KIF2a/MCAK siRNAs . Stabilizing those kinetochore–microtubules by inhibiting Aurora-B or depleting KIF2a/MCAK establishes full binding of microtubules , SKAP loading , and satisfies the SAC , allowing cells to enter anaphase . This suggests a SAC that is more sensitive than a checkpoint that only senses detached kinetochores or kinetochores that become detached due to a tension defect . A SAC that detects such minor defects in kinetochore–microtubule occupancy caused by an imbalance of microtubule stability within the spindle would be able to indirectly probe for plate positioning , giving cells time to correct this imbalance and ensure a symmetric metaphase plate position . Such graded response to microtubule occupancy within a kinetochore complements studies showing that the SAC acts in a graded manner when it comes to the number of unattached kinetochores ( Collin et al . , 2013; Dick and Gerlich , 2013 ) . HeLa cells were grown in Dulbecco's modified medium containing 10% Fetal Calf Serum ( FCS ) , 100 U/ml penicillin , 100 mg/ml streptomycin , at 37°C with 5% CO2 in a humidified incubator . HeLa eGFP-centrin1/eGFP-CENPA , eGFP-Centrin1/mRFP-α-tubulin , and eGFP-Centrin2/H2B-mRFP ( kind gift of U Kutay , ETH ) cells were further maintained in 250 ng/ml puromycin and 250 μg/ml G418 . Live-cell imaging experiments were performed at 37°C in Lab-Tek II chambers ( Thermo Fischer , Switzerland ) with Leibovitz L-15 medium containing 10% FCS . SiRNA oligonucleotides ( Invitrogen and Thermo Fisher , Switzerland ) against control ( Scrambled ) , Sas-6 , Mad2 , Kid1 , KIF2a , and MCAK were transfected using Oligofectamine ( Invitrogen; Meraldi et al . , 2004; Ganem et al . , 2005; Leidel et al . , 2005; Wandke et al . , 2012; Mchedlishvili et al . , 2012 ) . Mad2 and Kid1 depletion had been previously validated in our laboratory ( Meraldi et al . , 2004; Wandke et al . , 2012 ) , Sas-6 depletion was validated by counting centrioles in all experiments; Mad2 and Sas-6 co-depletion was additionally validated by immunoblotting ( Figure 2—figure supplement 1 ) ; KIF2a/MCAK depletion in Sas-6-depleted cells was validated by immunoblotting ( Figure 2—figure supplement 1 ) . To exclude off-target effects , KIF2a and MCAK were also depleted with an alternative set of pooled siRNAs ( ON-TARGETplus Human KIF2C ( 11 , 004 ) and ON-TARGETplus Human KIF2a ( 3796 ) siRNA—SMARTpools; GE Healthcare , Switzerland ) . For drug treatments , cells were treated with 100 μM monastrol for 3 hr , 1 μM MG132 for 1 hr , or 10 nM taxol ( all Sigma-Aldrich , Switzerland ) for 15 min before fixation for immunofluorescence or live-cell imaging . The Mps1 inhibitors MPS1-IN-1 ( 10 μM; kind gift of NS Gray; [Kwiatkowski et al . , 2010] ) or Reversine ( 10 μM , Sigma–Aldrich ) , or the Aurora-B inhibitor ZM1 ( 2 μM , Tocris , United Kingdom ) were added to metaphase cells during live-cell imaging . To determine the response to spindle poisons , cells were treated for 16 hr with combinations of 100 ng/ml nocodazole , 100 μM monastrol , 10 μM MPS1-IN-1 or 2 μM ZM1 , and the percentage of mitotic cells determined by phase contrast microscopy . Cells were fixed with methanol at −20°C for 6 min , or with 20 mM PIPES ( pH 6 . 8 ) , 10 mM EGTA , 1 mM MgCl2 , 0 . 2% Triton X-100 , 4% formaldehyde for 7 min at room temperature . For the cold-stable assay , cells were incubated in cold medium whilst placed on ice for 7 min . The following primary antibodies were used: rabbit anti-Mad2 ( 1:1000; Bethyl ) ; mouse anti-α-tubulin ( 1:10 , 000 ) and rabbit anti-γ-tubulin ( 1:2000; both Sigma–Aldrich ) ; mouse anti-HEC1 ( 1:1000 ) , mouse anti-pericentrin ( 1:2000 , kind gift of U Kutay ) , rabbit anti-ninein ( 1:500 ) , mouse anti-TPX-2 ( 1:250 ) , and mouse anti-centrobin ( 1:1000; all Abcam , United Kingdom ) ; mouse anti-p150Glued ( 1:500; Becton Dickinson , Switzerland ) ; rabbit anti-phospho-Aurora-A ( 1:1000; Cell signalling , Danvers , MA ) ; rabbit anti-centrin ( 1:1000 ) and affinity-purified rabbit anti-SKAP ( 1 mg/ml; [Schmidt et al . , 2010]; both gifts of I Cheeseman ) . Cross-adsorbed secondary antibodies were used ( Invitrogen ) . Three-dimensional image stacks of mitotic cells were acquired in 0 . 2-μm steps using a 100× NA 1 . 4 objective on an Olympus DeltaVision microscope ( GE Healthcare ) equipped with a DAPI/FITC/TRITC/CY5 filter set ( Chroma , Bellow Falls , VT ) and a CoolSNAP HQ camera ( Roper Scientific , Tucson , AZ ) . For quantitative measurements , 3D image stacks were deconvolved with SoftWorx ( GE Healthcare ) and quantified with SoftWorx , Imaris ( Bitplane , Switzerland ) or ImageJ . Images were mounted as figures using Adobe Illustrator . Kinetochore protein intensities were measured as a ratio to the CREST signal as described ( McClelland et al . , 2007 ) . To monitor the polar ejection force , the distance between centrosomes and kinetochores was measured as described ( Wandke et al . , 2012 ) . For mitotic timing experiments , cells were recorded every 3 or 4 min as three-dimensional image stacks ( 12 × 1 μm steps using a 60× 1 . 4 NA objective , or 7 × 2 μm stacks using a 40× 1 . 3 NA objective ) on an Olympus DeltaVision microscope equipped with a GFP/mRFP filter set ( Chroma ) and a CoolSNAP HQ camera . To monitor cell contours , cells were illuminated with white light and recorded by phase-contrast microscopy . Time-lapse videos were visualized in Softworx to quantify mitotic timing and to detect rotating spindles . For kinetochore tracking , plate width and plate position experiments , fluorescence time-lapse imaging of metaphase HeLa eGFP–centrin1/eGFP-CENPA cells was recorded with a 100× 1 . 4 NA objective on an Olympus DeltaVision microscope . 35 Z-sections 0 . 5 μm apart were acquired with a sampling rate of 7 . 5 s for a total duration of 5 min . Three-dimensional image stacks were deconvolved with SoftWorx and subjected to the kinetochore tracking assay analysis run in MATLAB ( The Math Works , Inc , Natick , MA ) , to asses inter-kinetochore distances and kinetochore oscillations ( Jaqaman et al . , 2010 ) . The tracking assay was also used to quantify the length of the two half-spindles: the tracking assay estimates the metaphase plate by fitting a plane to the calculated kinetochore positions; metaphase plate position relative to the spindle poles was calculated using a custom MATLAB function that detects centrioles and calculates plate position as the intersection of the fitted plane with the spindle axis . The earliest time point data of each cell imaged was used for plate position and inter-kinetochore distance analysis to ensure that data come from early metaphase cells . To measure plate position at anaphase and to better visualize the centering mechanisms , we used a temporal resolution of 30 s and applied our combined kinetochore and centrosome tracking analysis . Videos were manually screened for the presence of chromosome segregation errors . To determine spindle positions within cells , we used the centrosome positions to determine the center of the spindle ( equidistant to both centrosomes ) and compared it to the cell center , which was determined using phase contrast images ( point on the spindle axis that is equidistant to both cell cortexes ) . To measure intra-kinetochore distances and SKAP signals , HeLa eGFP–centrin1/eGFP-CENPA cells were fixed and stained with anti-HEC1 and anti-SKAP antibodies , respectively . Three-dimensional image stacks of fixed cells were subjected to the kinetochore tracking assay for sister–kinetochore pair identification . Imaris was used in conjunction with a custom MATLAB function ( Source Code 1 ) to measure the HEC1 and SKAP signals of the detected sister-kinetochores . Centriole ablations were carried out by 2–4 series ( 10 Hz repetition rate ) of second-harmonic , single-mode , 532-nm pulses of an Nd:YAG laser ( ULTRA-CFR TEM00 Nd:YAG from Big Sky Laser , Quantel , United Kingdom ) . The pulse width was 8 ns and the pulse energy used was 1 . 5–2 μJ . A more detailed description of the laser-microsurgery unit can be found in ( Pereira et al . , 2009 ) . Imaging and laser focusing was performed using a 100× 1 . 4 NA plan-Apochromatic DIC objective on a Nikon TE2000U inverted microscope equipped with a Yokogawa CSU-X1 spinning-disk confocal head and an iXonEM+ Electron Multiplying CCD camera . Statistical analyses were performed in R 2 . 15 . 0 . Unpaired t-tests with Welch's correction and Mann–Whitney U tests ( against 2:2 cells ) were carried out to check for the statistical significance of normal and non-normal distributed data , respectively . Count data were analyzed using the Fisher's Exact test . Levene's test of equality of variance was used to check for equal variances , using the lawstat package ( α = 0 . 05 ) . Graphs were plotted in R using the ggplot2 package and mounted in Adobe Illustrator .	The genetic information of a cell is stored in the form of chromosomes . Before a cell divides , its entire set of chromosomes is duplicated so that the two newly formed daughter cells receive a full set . In animal cells , the chromosomes line up in the center of the cell . The two sets of chromosomes are then separated by a structure known as the spindle , which attaches long filaments of proteins to the chromosomes and pulls one set to either side of the cell . Birth defects and cancer can result from a cell ending up with too many , or too few , chromosomes . Therefore , a safety mechanism called the spindle assembly checkpoint ensures that all of the chromosomes have correctly attached to the spindle before chromosome separation begins . Although it has been known for over a hundred years that chromosomes line up precisely in the center of the cell before they are separated , the reason why this occurs has remained unknown . Tan et al . investigated this problem by altering human cells so that the chromosomes did not align in the middle of the cell , but instead lined up off-center . However , after a short delay the chromosomes relocated to the center . Further investigation revealed that the spindle assembly checkpoint gives cells the time required to re-position the chromosomes in the center of the cell . When the chromosomes are off-center , their connections to the spindle are altered . The spindle assembly checkpoint detects these changes and delays chromosome separation until these errors are corrected . When Tan et al . inactivated the spindle assembly checkpoint before the chromosomes had time to align at the center , the cells divided to produce two unequal cells . This study shows that the central position of chromosomes is essential for cells to divide into two equal cells . After understanding why this is , the next big challenge will be to find out how a cell re-positions chromosomes that are off-center and how it places them precisely in the middle of the cell .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2015	The equatorial position of the metaphase plate ensures symmetric cell divisions
Bridging brain-scale circuit dynamics and organism-scale behavior is a central challenge in neuroscience . It requires the concurrent development of minimal behavioral and neural circuit models that can quantitatively capture basic sensorimotor operations . Here , we focus on light-seeking navigation in zebrafish larvae . Using a virtual reality assay , we first characterize how motor and visual stimulation sequences govern the selection of discrete swim-bout events that subserve the fish navigation in the presence of a distant light source . These mechanisms are combined into a comprehensive Markov-chain model of navigation that quantitatively predicts the stationary distribution of the fish’s body orientation under any given illumination profile . We then map this behavioral description onto a neuronal model of the ARTR , a small neural circuit involved in the orientation-selection of swim bouts . We demonstrate that this visually-biased decision-making circuit can capture the statistics of both spontaneous and contrast-driven navigation . Animal behaviors are both stereotyped and variable: they are constrained at short time scale to a finite motor repertoire while the long-term sequence of successive motor actions displays apparent stochasticity . This dual characteristic is immediately visible in the locomotion of small animals such as Nematodes ( Stephens et al . , 2008 ) , Zebrafish ( Girdhar et al . , 2015 ) or Drosophila larvae ( Gomez-Marin and Louis , 2012 ) , which consists of just a few stereotyped maneuvers executed in a sequential way . In this case , behavior is best described as a set of statistical rules that defines how these elemental motor actions are chained . In the presence of sensory cues , two types of behavioral responses can be distinguished . If they signal an immediate threat or reward ( e . g . the presence of a predator or a prey ) , they may elicit a discrete behavioral switch as the animal engages in a specialized motor program ( e . g . escape or hunt , Budick and O’Malley , 2000; Fiser et al . , 2004; Bianco et al . , 2011; McClenahan et al . , 2012; Bianco and Engert , 2015 ) . However , most of the time , sensory cues merely reflect changes in external factors as the animal navigates through a complex environment . These weak motor-related cues interfere with the innate motor program to cumulatively promote the exploration of regions that are more favorable for the animal ( Tsodyks et al . , 1999; Fiser et al . , 2004 ) . A quantification of sensory-biased locomotion thus requires to first categorize the possible movements , and then to evaluate the statistical rules that relate the selection of these different actions to the sensory and motor history . Although the probabilistic nature of these rules generally precludes a deterministic prediction of the animal’s trajectory , they may still provide a quantification of the probability distribution of presence within a given environment after a given exploration time . In physics terms , the animal can thus be described as a random walker , whose transition probabilities are a function of the sensory inputs . This statistical approach was originally introduced to analyze bacteria chemotaxis ( Lovely and Dahlquist , 1975 ) . Motile bacteria navigate by alternating straight swimming and turning phases , so-called runs and tumbles , resulting in trajectories akin to random walks ( Berg and Brown , 1972 ) . Chemotaxis originates from a chemical-driven modulation of the transition probability from run to tumble: the transition rate is governed by the time-history of chemical sensing . How this dependency is optimized to enhance gradient-climbing has been the subject of extensive literature ( Macnab and Koshland , 1972; Adler and Tso , 1974; Mello and Tu , 2007; Yuan et al . , 2010; Celani and Vergassola , 2010 ) . More recently , similar descriptions have been successfully used to quantify chemotaxis and phototaxis in multicellular organisms such as Caenorhabditis elegans ( Ward , 1973; Miller et al . , 2005; Ward et al . , 2008 ) , Drosophila larva ( Sawin et al . , 1994; Kane et al . , 2013; Gomez-Marin et al . , 2011; Tastekin et al . , 2018 ) or different types of slugs ( Matsuo et al . , 2014; Marée et al . , 1999 ) . Although the sensorimotor apparatus of these animals are very different , the taxis strategies at play appear to be convergent and can be classified based on the gradient-sensing methods ( Fraenkel and Gunn , 1961; Gomez-Marin and Louis , 2012 ) . Tropotaxis refers to strategies in which the organism directly and instantaneously infers the stimulus direction by comparison between two spatially distinct sensory receptors . In contrast , during klinotaxis , the sensory gradient is inferred from successive samplings at different spatial positions . This second strategy is particularly adapted when the organism has only one receptor , or if the sensory gradient across the animal’s body is too small to be detected ( Humberg et al . , 2018 ) . It requires at least a basic form of memory , since the sensory information needs to be retained for some finite period of time . In the present work , we implement such a framework to produce a comprehensive statistical model of phototaxis in zebrafish larvae . Zebrafish larva is currently the only vertebrate system that allows in vivo whole-brain functional imaging at cellular resolution ( Panier et al . , 2013; Ahrens et al . , 2013 ) . It thus provides a unique opportunity to study how sensorimotor tasks , such as sensory-driven locomotion , are implemented at the brain-scale level . Although adult zebrafish are generally photophobic ( or scototactic , Serra et al . , 1999; Maximino et al . , 2007 ) , they display positive phototaxis at the larval stage , from 5 days post-fertilization ( dpf ) on ( Orger and Baier , 2005 ) . At this early stage , their locomotion consists of a series of discrete swimming events interspersed by ~1 s long periods of inactivity ( Girdhar et al . , 2015 ) . Previous studies have shown that , when exposed to a distant light source , the first bouts executed by the fish tend to be orientated in the direction of the source ( tropotaxis ) ( Burgess et al . , 2010 ) . Furthermore , Chen and Engert ( 2014 ) have shown , using a virtual reality assay , that zebrafish are able to confine their navigation within a bright region in an otherwise dark environment even when deprived from stereovisual contrast information . This latter study thus established that their phototactic behavior also involves a spatio-temporal integration mechanism ( klinotaxis ) . From a neuronal viewpoint , recent calcium imaging experiments identified a small circuit in the rostral hindbrain that plays a key role in phototaxis ( Ahrens et al . , 2013; Dunn et al . , 2016; Wolf et al . , 2017 ) . This region , called ARTR ( anterior rhombencephalic turning region ) or HBO ( hindbrain oscillator ) , displays pseudo-periodic antiphasic oscillations , such that the activity of the left and right subpopulations alternate with a ~20 s period . This alternation was shown to set the coordinated direction of the gaze and tail bout orientation , thus effectively organizing the temporal sequence of the successive reorientations . It was further shown that this circuit oscillation could be driven by whole-field illumination of the ipsilateral eye , such as to favor the animal’s orientation towards a light source ( Wolf et al . , 2017 ) . In the present study , we aim at quantifying the statistical rules that control the larva’s reorientation dynamics in the presence of a continuous angular gradient of illumination ( orientational phototaxis ) . Using a virtual-reality closed-loop assay , we quantify how swim bouts selection is statistically controlled by the light intensity received on both eyes prior to the bout initiation , or the change in illumination elicited by the previous swim bout . Our experimental configuration allows us to disentangle the contribution of the two aforementioned strategies: tropotaxis and klinotaxis . From the analysis of this short-term behavior , we built a minimal Markov model of phototaxis , from which we compute the long-term distribution of orientations for any angular profile of illumination . This model offers explicit predictions of the statistics of the fish orientation that quantitatively compare with the experimental observations . We further expand on a recent rate model of the ARTR circuit to propose a functional neuronal model of spontaneous navigation and contrast-biased orientation selection . We demonstrate that the statistics of turn orientation can be fully understood by assuming that this self-oscillating circuitry , that selects the orientation of turning bouts , integrates stereovisual contrast in the form of incoming currents proportional to the visual stimulus . Zebrafish larvae aged 5–7 dpf were placed one at a time in a Petri dish ( 14 cm in diameter ) . Their center-mass position and body axis orientation were tracked in real time at 35 frames/s ( Figure 1A–B ) . This information was used to deliver a body-centered visual stimulus using a video-projector directed onto a screen supporting the Petri dish . Prior to each phototactic assay , the larva was allowed an ≈8 min-long period of spontaneous exploration under uniform and constant illumination at maximum intensity Im⁢a⁢x=450⁢μ⁢W . c⁢m-2 . Such pre-conditioning phases were used to promote light-seeking behavior ( Burgess and Granato , 2007 ) , while enabling the quantification of the basal exploratory kinematics for each fish . Larval zebrafish navigation is comprised of discrete swim bouts lasting ≈100⁢m⁢s and interspersed with 1 to 2s-long inter-bout intervals ( τn ) during which the fish remains still ( Dunn et al . , 2016 ) . Each bout results in a translational motion of the animal and/or a change in its body axis orientation , and can thus be automatically detected from kinematic parameters . As we are mostly interested in the orientational dynamics , we extracted a discrete sequence of orientations αn measured just before each swimming event n ( Figure 1B–C ) from which we computed the bout-induced reorientation angles δ⁢αn=αn+1-αn . Although the complete swim bouts repertoire of zebrafish larvae is rich and complex ( Johnson et al . , 2019 ) , the statistical distribution of the reorientation angles P⁢ ( δ⁢αn ) in such unbiased conditions can be correctly captured by the weighted sum of two zero-mean normal distributions , P⁢ ( δ⁢αn ) =pt⁢u⁢r⁢n⁢𝒩⁢ ( 0 , σt⁢u⁢r⁢n2 ) +pf⁢w⁢d⁢𝒩⁢ ( 0 , σf⁢w⁢d2 ) , reflecting the predominance of only two distinct bouts types: turning bouts ( standard deviation σt⁢u⁢r⁢n=0 . 6 ) and forward scoots ( σf⁢w⁢d=0 . 1 ) ( Figure 1D ) . This bimodal distribution is consistent with the locomotor repertoire of larvae described by Marques et al . ( 2018 ) during spontaneous swimming and phototactic tasks . In the absence of a visual bias , the turning bouts and forward scoots were found to be nearly equiprobable , pt⁢u⁢r⁢n=1-pf⁢w⁢d=0 . 41 . Successive bouts were found to exhibit a slightly positive correlation in amplitude ( Figure 1F ) . This process can be captured by a two-state Markov-chain model that controls the alternation between forward and turning bouts , while the amplitude within each population is randomly sampled from the corresponding distribution ( Figure 1E ) . Within this scheme , we analytically derived the dependence in the amplitude of successive bouts and thus estimated the forward-to-turn and turn-to-forward transition rates , noted kf→t and kt→f ( all analytical derivations are detailed in Appendix From behavior to circuit modeling of light-seeking navigation in zebrafish larvae ) . We found that kf→t/pt⁢u⁢r⁢n=kt→f/pf⁢w⁢d≈0 . 8 . This indicates that the probability to trigger a turn ( resp . forward ) bout is decreased by only 20% if the previous bout is a forward ( resp . turn ) bout . For the sake of simplicity , we ignore in the following this modest bias in bout selection and assume that the chaining of forward and turning bout is memory-less by setting kf→t=pt⁢u⁢r⁢n and kt→f=pf⁢w⁢d . We checked , using numerical simulations , that this simplifying assumption has no significant impact on the long-term navigational dynamics: the results presented in the following , notably the diffusion coefficient , remain essentially unchanged when this small correlation in bout type selection is taken into account . In line with previous observations ( Chen and Engert , 2014; Dunn et al . , 2016 ) , we also noticed that successive turning bouts tended to be oriented in the same ( left or right ) direction ( Figure 1G ) . This orientational motor persistence was accounted for by a second Markov chain that set the orientation of turning bouts , and was controlled by the rate of flipping direction noted kf⁢l⁢i⁢p ( Figure 1E bottom ) . Notice that , in contrast with the model proposed by Dunn et al . ( 2016 ) , although the orientational state is updated at each bout , it only governs the direction of turning bouts . When a forward bout is triggered , its orientation is thus unbiased . This model provides an analytical prediction for the mean reorientation angle ⟨δ⁢αn⟩\|δαn-1 at bout n following a reorientation angle δ⁢αn-1 at bout n-1 . This expression was used to fit the experimental data ( Figure 1G ) and allowed us to estimate the flipping rate pf⁢l⁢i⁢p=0 . 19 ( 99% confidence bounds ±0 . 017 ) . We further computed the autocorrelation function of the reorientation angles and the Mean Square Reorientation ( MSR ) accumulated after n bouts ( Figure 1H–I ) . Both were consistent with their experimental counterparts . In particular , this model quantitatively captures the ballistic-to-diffusive transition that stems from the directional persistence of successive bouts ( Figure 1I ) . As a consequence , the effective rotational diffusivity at long time De⁢f⁢f=0 . 3⁢r⁢a⁢d2 is about twice as large than the value expected for a memory-less random walk ( i . e . with pf⁢l⁢i⁢p=0 . 5 , see dashed line in Figure 1I ) . In this discrete Markov-chain model , time is not measured in seconds but corresponds to the number of swim bouts . It thus implicitly ignores any dependence of the transition rates with the interbout interval . We examined this hypothesis by evaluating the correlation in bouts orientations as a function of the time elapsed between them . To do so , we first sorted the turning bouts by selecting the large amplitude events ( \|δ⁢α\|<0 . 22⁢r⁢a⁢d ) . We then binarized their values , based on their leftward or rightward orientation , yielding a discrete binary signal s⁢ ( tn ) =±1 . We finally computed the mean product ⟨s⁢ ( tn ) ⁢s⁢ ( tp ) ⟩ for various time intervals Δ⁢t=tp-tn . The resulting graph , shown in Figure 1J , demonstrates that the correlation in orientation of successive bouts decays quasi-exponentially with the inter-bout period . This mechanism can be captured by assuming that the orientation selection at each bout is governed by a hidden two-state continuous-time process . The simplest one compatible with our observations is the telegraph process , whose transition probability over a small interval d⁢t reads kf⁢l⁢i⁢p⁢d⁢t , and whose autocorrelation decays as exp⁡ ( -2⁢kf⁢l⁢i⁢p⁢t ) . Setting kf⁢l⁢i⁢p=pf⁢l⁢i⁢p/m⁢e⁢d⁢i⁢a⁢n⁢ ( τn ) =0 . 2⁢s-1 , this model correctly captures the τn-dependence of the orientational correlation of bouts . In the two following sections , we use the discrete version of the Markov-chain model to represent the fish navigation , and investigate how the model parameters are modulated in the presence of a virtual distant light source . We then go back to the underlying continuous-time process when introducing a neuronal rate model for the orientation selection process . We first examined the situation in which the perceived stereo-visual contrast is the only cue accessible to the animal to infer the direction of the light source ( tropotaxis regime ) . The visual stimulus consisted of two uniformly lit half-disks , each covering one visual hemifield . The intensity delivered to the left and right eyes , noted IL and IR respectively , were locked onto the fish’s orientation θ relative to the virtual light source ( Figure 2A ) : the total intensity ( IL+IR ) was maintained constant while the contrast c=IL-IR was varied linearly with θ , with a mirror symmetry at π/2 ( Figure 2B ) . This dependence was chosen to mimic the presence of a distant source located at θ=0 for which the contrast is null . The orientation of the virtual source in the laboratory frame of reference was randomly selected at initiation of each assay . After only a few bouts , the animal orientation was found to be statistically biased towards θ=0 , as shown in Figure 2C–D . This bias was quantified by computing the population resultant v defined as the vectorial mean of all orientations ( Figure 2E ) . Trajectories that are strongly biased towards the source tend to exit the ROI earlier than unbiased trajectories , which are more tortuous and thus more spatially confined . This generates a progressive selection bias as the number of bouts considered is increased , as revealed by the slow decay of the resultant vector length ( Figure 2—figure supplement 1 ) . In order to mitigate this selection bias , all analyses of stationary distributions were restricted to bout indices lower than the median number of bouts per trial ( N≤17 ) , and excluding the first bout . Under this condition , we found that ~77% of zebrafish larvae display a significant phototactic behavior ( Figure 2D–F , test of significance based on a combination of two circular statistic tests , see Materials and methods ) , a fraction consistent with values reported by Burgess et al . ( 2010 ) in actual phototactic assays . From these recordings , we could characterize how the contrast experienced during the inter-bout interval impacts the statistics of the forthcoming bout . Figure 2G displays the mean reorientation ⟨δ⁢θ⟩ as a function of the instantaneous contrast c . This graph reveals a quasi-linear dependence of the mean reorientation with c , directed toward the brighter side . Notice that the associated slope shows a significant decrease in the few first bouts , before reaching a quasi-constant value ( Figure 2—figure supplement 2 ) . This effect likely reflects a short term habituation mechanism as the overall intensity drops by a factor of 2 at the initiation of the assay . For a more thorough analysis of the bout selection mechanisms leading to the orientational bias , we examined , for all values of the contrast , the mean and variance of the two distributions associated with turning bouts and forward scoots , as well as the fraction of turning bouts pt⁢u⁢r⁢n ( Figure 2H–K ) . We found that the orientational drift solely results from a probabilistic bias in the selection of the turning bouts ( left vs right ) orientation: the mean orientation of the turning bouts varies linearly with the imposed contrast ( Figure 2I ) . Reversely , the ratio of turning bouts and the variance of the two distributions are insensitive to the contrast Figure 2J–K ) . These results indicate that the stereo-visual contrast has no impact neither on bout type selection nor on bout amplitude . As discussed in the preceding section , in the absence of visual cue , successive bouts tend to be oriented in the same direction . During phototaxis , the selection of the turning orientation is thus expected to reflect a competition between two distinct mechanisms: motor persistence , which favors the previous bout orientation , and stereo-visual bias , which favors the brighter side . To investigate how these two processes interfere , we sorted the bouts into two categories . In the first one , called 'reinforcement’ , the bright side is in the direction of the previous bout , such that both the motor and sensory cues act in concert . In the second one , called 'conflicting’ , the contrast tends to evoke a turning bout in a direction opposite to the previous one . For each category , we plotted the mean reorientation angle at bout n as a function of the reorientation angle at bout n-1 ( Figure 2L ) . We further estimated , for each condition and each value of the contrast , the probability of flipping orientation pf⁢l⁢i⁢p ( Figure 2M and Appendix 2 ) . These two graphs show that the stereo-visual contrast continuously modulates the innate motor program by increasing or decreasing the probability of flipping bout orientation from left to right and vice versa . Noticeably , in the conflicting situation at maximum contrast , the visual cue and motor persistence almost cancel each other out such that the mean orientation is close to ( pf⁢l⁢i⁢p∼0 . 4 ) . We now turn to the second paradigm , in which the stereo-visual contrast is null ( both eyes receive the same illumination at any time ) , but the total perceived illumination is orientation-dependent ( klinotaxis regime ) . We thus imposed a uniform illumination to the fish whose intensity I was locked onto the fish orientation θ relative to a virtual light source . We tested three different illumination profiles I⁢ ( θ ) as shown in Figure 3A: a sinusoidal and two exponential profiles with different maxima . Despite the absence of any direct orientational cue , a large majority of the larvae ( 78% ) displayed positive phototactic behavior: their orientational distribution showed a significant bias towards the virtual light source , that is the direction of maximum intensity ( Figure 3B–E ) . Although the efficiency of the phototactic behavior is comparable to the tropotaxis case previously examined , here we did not observe any systematic bias of the reorientation bouts towards the brighter direction ( Figure 3F ) . This indicates that the larvae do not use the change in intensity at a given bout to infer the orientation of the source in order to bias the orientation of the forthcoming turn . Instead , the phototactic process originates from a visually driven modulation of the orientational diffusivity , as measured by the variance of the bout angle distributions ( Figure 3G ) . The use of different profiles allowed us to identify which particular feature of the visual stimulus drives this modulation . Although the bout amplitude variance was dependent on the intensity I and intensity change δ⁢I experienced before the bout , these relationships were found to be inconsistent across the different imposed intensity profiles . In contrast , when plotted as a function of δ⁢I/I , all curves collapse ( Figure 3—figure supplement 1 ) . This observation is in line with the Weber-Fechner law ( Fechner , 1860 ) , which states that the perceived change scales with the relative change in the physical stimulus . One noticeable feature of this process is that the modulation of the turning amplitude is limited to illumination decrement ( i . e . negative values of δ⁢I/I ) . In the terminology of bacterial chemotaxis ( Oliveira et al . , 2016 ) , the zebrafish larva can thus be considered as a 'pessimistic’ phototactic animal: the orientational diffusivity increases in response to a decrease in illumination ( corresponding to a negative subjective value ) , whereas its exploratory kinematics remain unchanged upon an increase of illumination ( positive subjective value ) . Two kinematic parameters can possibly impact the orientational diffusivity: the fraction of turning bouts pt⁢u⁢r⁢n and their characteristic amplitude σt⁢u⁢r⁢n . We thus extracted these two quantities and plotted them as a function of δ⁢I/I ( Figure 3H–I ) . They appear to equally contribute to the observed modulation . To test whether this uniform phototactic process has a retinal origin , or whether it might be mediated by non-visual deep-brain phototreceptors ( Fernandes et al . , 2012 ) , we ran similar assays on bi-enucleated fish . In this condition , no orientational bias was observed , which indicates that the retinal pathway is involved in orientational klinotaxis Figure 3—figure supplement 2 , all p-values > 0 . 14 , pairwise T-tests ) . In the preceding sections , we quantified how visual stimuli stochastically modulate specific kinematic parameters of the subsequent bout . We used these relationships to build a biased random walk model of phototaxis . We then tested how such a model could reproduce the statistical orientational biases toward the directions of minimal contrast and maximal illumination . The phototactic model thus incorporates a visually-driven bias to the discrete Markov-chain model introduced to represent the spontaneous navigation ( Figure 4A ) . In line with the observation of Figure 2M , the rate of flipping orientational state ( left-to-right or right-to-left ) was a linear function of the imposed contrast: kR→L=kf⁢l⁢i⁢p+a⁢c and kl→r=kf⁢l⁢i⁢p-a⁢c . The value of a was set so as to capture the contrast-dependent orientational drift ( Figure 2G ) and was made dependent on bout index in order to account for the observed short-term habituation process ( Figure 2—figure supplement 2 ) . The selection of bout type was in turn linearly modulated by the relative change in intensity after negative rectification , [δ⁢I/I]-=m⁢i⁢n⁢ ( δ⁢I/I , 0 ) . Hence , the turn-to-forward and forward-to-turn transition rates read kt→f=kt⁢u⁢r⁢n+β⁢[δ⁢I/I]- and kf→t=kt⁢u⁢r⁢n-β⁢[δ⁢I/I]- , respectively . We also imposed a linear modulation of the turn amplitude variance σt⁢u⁢r⁢n=σt⁢u⁢r⁢ns⁢p⁢o⁢n⁢t-γ⁢[δ⁢I/I]- . The values of β and γ were adjusted to reproduce the observed dependence of the turn-vs-forward ratio and bout amplitude with δ⁢I/I ( Figure 3H–I ) . This stochastic model was tested under two conditions , tropo- and klino-phototaxis , similar to those probed in the experiments ( Figure 4B ) . In order to account for the sampling bias associated with the finite size of the experimental ROI , the particles in the simulations also progressed in a 2D arena . At each time step , a forward displacement was drawn from a gamma distribution adjusted on the experimental data ( Figure 5—figure supplement 1 ) . Statistical analysis was restricted to bouts executed within a circular ROI as in the experimental assay . The comparison of the data and numerical simulation is shown in Figure 4C for the tropotaxis protocol and in Figure 4D–F for the klinotaxis protocols . This minimal stochastic model quantitatively captures the distribution of orientations . It also reproduces the evolution of the orientational bias with the bout index as measured by the length of the resultant vector ( Figure 4G–J ) . The behavioral description proposed above indicates that larvae navigation can be correctly accounted for by two independent stochastic processes: one that organizes the sequence of successive bouts amplitude and in particular the selection of forward vs turning events , while a second one selects the left vs right orientation of the turning bouts . These two processes are independently modulated by two distinct features of the visual stimulus , namely the global intensity changes and the stereo-visual contrast , leading to the two phototactic strategies . This in turn suggests that , at the neuronal level , two independent circuits may control these characteristics of the executed swim bouts . As mentioned in the introduction , the ARTR is a natural candidate for the neuronal selection of bouts orientation . This small bilaterally-distributed circuit located in the anterior hindbrain displays antiphasic activity oscillation with ~ 20s period ( Ahrens et al . , 2013 ) . The currently active region ( left or right ) constitutes a strong predictor of the orientation of turning bouts ( Dunn et al . , 2016 ) . This circuit further integrates visual inputs as each ARTR subpopulation responds to the stimulation of the ipsilateral eye ( Wolf et al . , 2017 ) . Here , we adapted a minimal neuronal model of the ARTR , introduced in Wolf et al . ( 2017 ) to interpret the calcium recordings , and tested whether it could explain the observed statistics of exploration in both spontaneous and phototactic conditions . The architecture of the model is depicted in Figure 5A and the equations governing the network dynamics are provided in Appendix 2 . The network consists of two modules selective for leftward and rightward turning , respectively . Recurrent excitation ( wE ) drives self-sustained persistent activities over finite periods of time . Reciprocal inhibition ( wI ) between the left and right modules endows the circuit with an antiphasic dynamics . Finally , each ARTR module receives an input current from the visual system proportional to the illumination of the ipsilateral eye . Such architecture gives rise to a stimulus-selective attractor as described in Freeman ( 1995 ) and Wang ( 2002 ) . The various model parameters were adjusted in order to match the behavioral data ( see Appendix 2 ) . First , the self-excitatory and cross-inhibitory couplings were chosen such that the circuit displayed spontaneous oscillatory dynamics in the absence of sensory input . Figure 5B shows example traces of the two units’ activity in this particular regime . From these two traces , we extracted a binary 'orientational state’ signal by assigning to each time point a left or right value ( indicated in red and blue , respectively ) , based on the identity of the module with the largest activity . In the present approach , tail bouts are assumed to be triggered independently of the ARTR activity . The latter thus acts as mere orientational hub by selecting the orientation of the turning events: incoming bouts are oriented in the direction associated with the currently active module . In the absence of information regarding the circuit that organizes the swim bouts emission , their timing and absolute amplitude were drawn from the behavioral recordings of freely swimming larvae . Combined with the ARTR dynamics , this yielded a discrete sequence of simulated bouts ( leftward , rightward and forward , Figure 5B , inset ) . With adequate choice of parameters , this model captures the orientational persistence mechanism as quantified by the slow decay of the turning bout autocorrelation with the interbout interval ( Figure 5C and Figure 5—figure supplement 1 ) . In the presence of a lateralized visual stimulus , the oscillatory dynamics become biased towards the brighter direction ( Figure 5D–E ) . Hence , illuminating the right eye favors longer periods of activation of the rightward-selective ARTR unit . The mean reorientation displays a quasi-linear dependence with the imposed contrast ( Figure 5D ) consistent with the behavioral observations ( Figure 2G ) . At intermediate contrast values , the orientation of bouts remains stochastic; the effect of the contrast is to lengthen streaks of turning bouts toward the light ( Figure 5E ) . We also tested whether this model could capture the competition mechanism between stereovisual bias and motor persistence , in both conflicting and reinforcement conditions . We thus computed the dependence of the flipping probability pf⁢l⁢i⁢p as a function of the contrast in both conditions ( Figure 5F ) . The resulting graph is in quantitative agreement with its experimental counterparts ( Figure 2M ) . We finally used this model to emulate a simulated phototactic task . In order to do so , a virtual fish was submitted to a contrast whose amplitude varied linearly with the animal orientation , as in the lateralized assay . When a turning bout was triggered , its orientation was set by the ARTR instantaneous activity while its amplitude was drawn from the experimental distributions . After a few bouts , a stationary distribution of orientation was reached that was biased toward the virtual light source ( Figure 5G ) . Its profile was in quantitative agreement with its experimental counterpart ( mean resultant vector length v=0 . 23 in simulation for v=0 . 24 in experimental data for bouts 2 to 17 ) . Sensorimotor transformation can be viewed as an operation of massive dimensionality reduction , in which a continuous stream of sensory and motor-related signals is converted into a discrete series of stereotyped motor actions . The challenge in understanding this process is ( i ) to correctly categorize the motor events , that is to reveal the correct parametrization of the motor repertoire , and ( ii ) to unveil the statistical rules for action selection . Testing the validity of such description can be done by building a minimal model based on these rules . If the model is correct , the motor variability unaccounted for by the model should be entirely random , that is independent of the sensorimotor history . Here , we implemented a minimal model approach in order to unveil the basic rules underlying phototaxis . We showed that zebrafish light-driven orientational navigation can be quantitatively described by a stochastic model consisting of two independent Markov chains: one that selects forward scoots vs turning bouts and a second one that sets the orientation of the latter . We established that the stereo-visual contrast and global intensity modulation act separately on each of these selection processes . The contrast induces a directed bias of turning bouts toward the illuminated side , but does not impact the prevalence of turning bouts vs forward scoots . Reversely , a global decrement in illumination increases the ratio of turning bouts but does not favor any particular direction . For the contrast-driven configuration ( tropotaxis ) , the minimal model corresponds to an Ornstein-Uhlenbeck process ( Uhlenbeck and Ornstein , 1930 ) , which describes the dynamics of a diffusive brownian particle in a quadratic trap . In the klinotaxis configuration ( in the absence of stereo-visual contrast ) , the orientational bias solely results from a light-dependent modulation of the diffusivity , a mechanism reminiscent of bacterial chemotaxis . This stochastic minimal model is built on a simple decision tree ( Figure 4A ) with a set of binary choices . However , to fully capture the orientational dynamics , we had to incorporate the continuous increase in turning bout amplitude with the light decrement in an ad-hoc way . It is currently unclear whether all turn bouts in our experiments can be assigned to a single class of swim maneuvers that are modulated in amplitude , or whether these encompass distinct motor programs executed with varying frequencies . In the latter case , it might be possible to represent this amplitude modulation through an extension of the decision tree that would select between distinct turn bout categories . Compared to previous studies on phototaxis , for example ( Burgess et al . , 2010 ) , our approach allowed us to clearly disentangle the contributions of spatial ( stereovisual contrast ) and time-dependent ( motion-induced change in global illumination ) visual cues . Hence , the contrast-driven assays were performed under constant overall illumination intensity ( the sum of left and right intensities ) . This allowed us to establish that , rather surprisingly , the probability of triggering a turn ( vs a forward swim ) is insensitive to the imposed contrast . This possibility constitutes an important asset with respect to standard experimental configurations , such as the one examined by Burgess et al . ( 2010 ) , in which the animal is submitted to an actual light source . Although these configurations provide a more realistic context , the visual stimulus effectively perceived by each eye can not be quantitatively assessed , which precludes the design of predictive models . Conversely , once adjusted on well-controlled virtual assays , our model could be numerically implemented in realistic environments , and the trajectories could then be directly confronted with behavioral data . This would require to first infer how the intensity impinging on each eye depends on the source distance and orientation relative to the animal body axis . Another critical and distinct aspect of the present work is its focus on the steady-state dynamics . Our aim was to mimic the continuous exploration of an environment in which the brightness level displayed slowly varying angular modulations . The luminosity profiles were thus chosen to ensure that individual bouts elicited relatively mild changes in illumination . By doing so , we tried to mitigate visual startle responses that are known to be elicited upon sudden darkening ( Easter and Nicola , 1996 ) . Although we could not avoid the initial large drop in illumination at the onset of each trial , the associated short-term response ( i . e . the first bout ) was excluded from the analysis . In this respect , our experiment differs from the study of Chen and Engert ( 2014 ) in which a similar closed-loop setup was used to demonstrate the ability of larvae to confine their navigation within bright regions . This behavior was entirely controlled by the animal’s response to light-on or light-off stimuli as it crossed the virtual border between a bright central disk and the dark outer area . These sharp transitions resulted in clear-cut behavioral changes that lasted for a few bouts . In comparison , our experiment addresses a different regime in which subtle light-driven biases in the spontaneous exploration cumulatively drive the animal toward brightest regions . As we aimed to obtain a simple and tractable kinematic description , we ignored some other aspects of the navigation characteristics . First , we focused on the orientation of the animal and thus did not systematically investigate how the forward components of the swim bouts were impacted by visual stimuli . However , in the context of angle-dependent intensity profiles , this effect should not impact the observed orientational dynamics . More importantly , we ran most of our analysis using the bout number as a time-scale , and thus ignored possible light-driven modulations of the inter-bout intervals ( τn ) . We showed , however , that the orientational correlation is controlled by an actual time-scale . This result may have significant consequence on the fish exploration . In particular , we expect that changes in bout frequency , reflecting various levels of motor activity , may significantly affect the geometry of the trajectories ( and not only the speed at which they are explored ) . We illustrated this process by running numerical experiments at similar flipping rate kf⁢l⁢i⁢p but increasing bout frequencies . The trajectories , shown in Figure 5—figure supplement 2 , exhibit increasing complexity as measured by the fractal dimension . This mechanism may explain the changes in trajectories’ geometry observed by Horstick et al . ( 2017 ) in response to sudden light dimming . An important outcome of this study is to show that light-seeking navigation uses visual cues over relatively short time scales . The bouts statistics could be captured with a first-order autoregressive process , indicating that the stimulus perceived over one τn is sufficient to predict the forthcoming bout . However , one should be aware that such observation is only valid provided that the sensory context remains relatively stable . Hence for instance , a prolonged uniform drop in luminosity is known to enhance the overall motor activity ( generally estimated by the average displacement over a period of time ) for up to several tens of minutes ( Prober et al . , 2006; Emran et al . , 2007; Emran et al . , 2010; Liu et al . , 2015 ) . This long-term behavioral change , so-called photokinesis , might be regulated by deep brain photoreceptors ( Fernandes et al . , 2012; Horstick et al . , 2017 ) and thus constitutes a distinct mechanism . One particularly exciting prospect will be to understand how such behavioral plasticity may not only modulate the spontaneous activity ( Johnson et al . , 2019 ) but also affects the phototactic dynamics . One of the motivations of minimal behavioral models is to facilitate the functional identification and modeling of neural circuits that implement the identified sensorimotor operations in the brain . Here , we used the behavioral results to propose a neuronal model of the ARTR that quantitatively reproduces non-trivial aspects of the bout selection process . This recurrent neural circuit is a simplified version of working memory models developed by Brunel and Wang ( 2001 ) ; Wang ( 2001 ) ; Wang ( 2002 ) ; Wang ( 2008 ) and adapted in Wang ( 2002 ) for a decision task executed in the parietal cortex ( Shadlen and Newsome , 1996; Shadlen and Newsome , 2001 ) . In this class of models , the binary decision process reflects the competition between two cross-inhibitory neural populations . The circuit is endowed with two major functional capacities: ( 1 ) it can maintain mnemonic persistent activity over long periods of time , thanks to recurrent excitatory inputs; ( 2 ) it can integrate sensory signals in a graded fashion to continuously bias the statistics of the decision . This model thus naturally recapitulates the major functional features of the sensory-biased Markov side-chain - motor persistence and contrast-driven continuous bias - that organizes the orientation selection . It is tempting to generalize about this behavior-to-circuit approach , at least in small animals such as Zebrafish or Drosophila , by representing any behavior as a coordinated sequence of competing elemental actions biased by sensory feedback and organized within a hierarchical decision tree . The identification of such decision trees through quantitative behavioral analysis may provide a blueprint of the brain functional organization and significantly ease the development of circuit models of brain-scale sensorimotor computation . All experiments were performed on wild-type Zebrafish ( Danio Rerio ) larvae aged 5 to 8 days post-fertilization . Larvae were reared in Petri dishes in E3 solution on a 14/10 hr light/dark cycle at 28°C , and were fed powdered nursery food every day from 6 dpf . Experiments were conducted during daytime hours ( 10 am to 6 pm ) . The arena consisted of a 14 cm in diameter Petri dish containing E3 medium . It was placed on a screen illuminated from below by a projector ( ASUS S1 ) . Infrared illumination was provided by LEDs to enable video-monitoring and subsequent tracking of the fish . We used an IR-sensitive Flea3 USB3 camera ( FL3-U3-13Y13M-C , Point Grey Research , Richmond , BC , Canada ) with an adjustable macro lens ( Zoom 7000 , Navitar , USA ) equipped with an IR filter . The experimental setup was enclosed in a light-tight rig , which was maintained at 26°C using 'The Cube’ ( Life Imaging Services ) . For the stereovisual paradigm N = 47 larvae were tested , and N = 37 for the temporal paradigm [ ( uniform 1 ) : 12 , ( uniform 2 ) : 11 , ( uniform 3 ) : 14] . All fish ( N=75 ) that navigated in the ROI for a significant period of time during the habituation period were also used to assess spontaneous navigation statistics . Closed-loop tracking and visual stimulation were performed at a mean frequency of 35 Hz , with a custom-written software ( Karpenko , 2019b; copy archived at https://github . com/elifesciences-publications/Analysis_Behavioral_Phototaxis ) in Matlab ( The MathWorks ) , using the PsychToolBox ( PTB ) version 3 . 0 . 14 add-on . Positions and orientations ( heading direction ) of the fish , as well as bouts characteristics , were extracted online and the illumination pattern was updated accordingly , with a maximum latency of 34 ms . Heading direction was extracted with an accuracy of + /- 0 . 05 rad ( ∼3∘ ) . Behavioral monitoring was restricted to a circular central region of interest ( ROI ) of 8 . 2 cm diameter . When outside the ROI , the fish was actively brought back into the ROI through the opto-motor reflex ( OMR ) , using a concentrically moving circular pattern . One second after the fish re-entered the ROI , a new recording sequence was started . Prior to the phototactic assay , all tested fish were subjected to a period of at least 8 min of habituation under whole-field illumination at an intensity of Im⁢a⁢x=450⁢μ⁢W . c⁢m-2 . For both phototactic paradigms , the absolute orientation of the virtual source was randomly selected when initiating a new experimental sequence ( each time the animal would re-enter the ROI ) . The orientation of the fish relative to the light source θn was calculated online using the absolute orientation of the fish αn and the orientation of the virtual light source αsource:θn=αn−αsource . Lateralized paradigm . A circle of 6 cm in diameter was projected under and centered on the fish . The circle was divided into two parts , covering the left and right side of the fish . The separation between the two parts corresponded to the animal’s midline . A separation band ( 2 mm thick ) and an angular sector ( 30∘ ) in front of the animal were darkened to avoid interception of light coming from the right side of the fish by its left eye and vice-versa . The left and right intensities ( IL and IR ) were varied linearly as a function of θ , such that IL+IR=Im⁢a⁢x . Since during the habituation period , the whole arena was lit at maximum intensity Im⁢a⁢x , the total intensity received by the fish drops by a factor of ≈2 with the establishment of the circle , at the onset of the assay . Although our imposed contrast profile displays two angles for which the contrast is null , namely θ=0 and π , only does the first one correspond to a stable equilibrium point . When θ is close to zero , any excursion away from this particular direction results in a contrast that drives the animal back to the null angle . Conversely , when θ≈π , the contrast drives the animal away from π ( unstable equilibrium ) . Temporal paradigm . The whole arena was illuminated with an intensity locked onto the fish orientation θ relative to a virtual light source . The initial orientation was randomly chosen at the beginning of a recording sequence . Three different intensity angular profiles were implemented: ( uniform 1 ) a sinusoidal profile , with a maximum intensity of 60% of Im⁢a⁢x , ( uniform 2 ) an exponential profile , with a maximum intensity of 60% of Im⁢a⁢x and finally ( uniform 3 ) an exponential profile with a maximum intensity of 30% of Im⁢a⁢x . Data analysis was performed using a custom-written code in Matlab . All analysis programs and data are available at Karpenko ( 2019a ) . When representing the mean of one variable against another , bin edges were chosen such that each bin would encompass the same number of data points . Circular statistics analyses ( mean , variance , uniformity ) and circular statistics tests , namely the circular V-test of non-uniformity of data and the one-sample test for the mean angle of a circular distribution ( tested on the orientation of the light virtual light source ) were performed using CircStat toolbox for Matlab ( Berens , 2009 ) . Individual fish often exhibit a small yet consistent bias toward one direction ( either leftward or rightward ) . This bias was subtracted before performing the different analyses , in order to guarantee that ⟨α⟩=0 in the absence of a stimulus . The distribution of reorientation angles δ⁢θn during spontaneous swimming periods was fitted with a constrained double-Gaussian function . We imposed that both the mean absolute angle and variance of the fitting function be consistent with the experimental measurements . This yields an expression with only one independent fitting parameter pt⁢u⁢r⁢n in the form: ( 1 ) f⁢ ( x ) =12⁢π⁢ ( pt⁢u⁢r⁢nσt⁢u⁢r⁢n⁢e-12⁢ ( xσt⁢u⁢r⁢n ) 2+1-pt⁢u⁢r⁢nσf⁢e-12⁢ ( xσf⁢w⁢d ) 2 ) usingσt⁢u⁢r⁢n=μa⁢b⁢s⁢pt⁢u⁢r⁢n+μa⁢b⁢s2⁢pt⁢u⁢r⁢n2-[μa⁢b⁢s2-V⁢ ( 1-pt⁢u⁢r⁢n ) ]⁢[pt⁢u⁢r⁢n2+pt⁢u⁢r⁢n⁢ ( 1-pt⁢u⁢r⁢n ) ]pt⁢u⁢r⁢n2+pt⁢u⁢r⁢n⁢ ( 1-pt⁢u⁢r⁢n ) andσf⁢w⁢d=μa⁢b⁢s⁢π/2-pt⁢u⁢r⁢n⁢σt⁢u⁢r⁢n1-pt⁢u⁢r⁢nwithμa⁢b⁢s=⟨\|δ⁢θ\|⟩ , V=⟨δ⁢θ2⟩ To evaluate the mean and variance of the forward and turn bouts under various visual contexts , the distributions in different bins were also fitted with a constrained double-Gaussian model as in ( 1 ) . The stereovisual data distributions were fitted with two additional mean terms μt⁢u⁢r⁢n and μf⁢w⁢d ; and for the klinotaxis assay , with a constraint on σf⁢w⁢d and μf⁢w⁢d . The bins were constructed either on the contrast c experienced just before bout n or on the relative difference of intensity experienced at bout n−1:δI/I=2In−1−In−2In−1+In−2 . All distributions of θn and analyses of bias were computed using trajectories from bout index two to the median number of bouts per sequence in each type of experiment . The median number of bouts in each experiment was m⁢e⁢ds⁢t⁢e⁢r⁢e⁢o=17 for the tropotaxis experiment , and 27 , 15 , 17 for the klinotaxis assays for the 3 profiles 1–3 , respectively . The Markov-chain model simulations were performed using a custom-written code in MATLAB ( Karpenko , 2019b ) . Initial orientations and positions within the ROI were randomly sampled from , respectively , a uniform distribution and a normal distribution centered on a circle of radius 20 mm from the center of the ROI with a standard deviation of 1 . 3 mm ( mimicking the starting points of experimental data ) . At each step , an angular step-size is drawn from the data: either from the turning distribution with a probability pt⁢u⁢r⁢n or from the forward distribution with a probability 1-pt⁢u⁢r⁢n . Respective means are μt⁢u⁢r⁢n and μf⁢w⁢d and standard deviation σt⁢u⁢r⁢n and σf⁢w⁢d . The left-vs-right orientations of the turns is set by the probability of flipping sides pf⁢l⁢i⁢p . For the spatially constrained simulations , the walker also draws a distance step-size ( between two successive positions ) from two different gamma distributions: one for the turning bouts , a second one for the scoots . Under neutral conditions ( uniform illumination ) , all parameters are constant . For the simulation under stereovisual phototactic conditions , pf⁢l⁢i⁢p was varied linearly with the contrast ( based on the data represented in Figure 2M ) . When simulating temporal phototaxis , the parameters σt⁢u⁢r⁢n and pt⁢u⁢r⁢n were modulated by the relative illumination change δ⁢I/I experienced at the previous steps ( as represented in Figure 3H–I ) .	All animals with the ability to move use sensory signals to help them navigate towards areas that seem better than their current location . Such areas might contain desirable things like food and mates , or they might allow an animal to escape from threats such as predators . But how the brain gives rise to this navigation behavior is unclear . Karpenko et al . have now obtained insights into the underlying mechanism by studying a behavior in zebrafish larvae called phototaxis . Phototaxis is the tendency to move in response to light . The advantage of using zebrafish larvae to study this behavior is that their brains are small and semi-transparent . This makes it possible to record the activity of almost every neuron . As a result , an individual’s brain activity can be mapped on to their behavior more precisely than in most other species . To probe how visual cues influence fish behavior , Karpenko et al . exposed individual fish to a carefully controlled virtual light source and then tracked their movements with a camera . The fish used two strategies to move towards the light . They selected their next movement based partly on the difference in the amount of light reaching each of their eyes , and partly on the change in overall brightness with each swim movement . Karpenko et al . used this information to build a numerical model of fish phototaxis , and to show how a simple brain circuit could generate this behavior . Species whose brains differ in size and structure may nevertheless develop similar strategies to perform similar tasks . By quantifying a generic behavior in a simple animal model , this study could provide insights into comparable behaviors in other species . In addition , the study suggests a simple mechanism for how animals select actions on the basis of sensory signals , which may also be relevant to other species and other tasks .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2020	From behavior to circuit modeling of light-seeking navigation in zebrafish larvae
Starvation induces sustained increase in locomotion , which facilitates food localization and acquisition and hence composes an important aspect of food-seeking behavior . We investigated how nutritional states modulated starvation-induced hyperactivity in adult Drosophila . The receptor of the adipokinetic hormone ( AKHR ) , the insect analog of glucagon , was required for starvation-induced hyperactivity . AKHR was expressed in a small group of octopaminergic neurons in the brain . Silencing AKHR+ neurons and blocking octopamine signaling in these neurons eliminated starvation-induced hyperactivity , whereas activation of these neurons accelerated the onset of hyperactivity upon starvation . Neither AKHR nor AKHR+ neurons were involved in increased food consumption upon starvation , suggesting that starvation-induced hyperactivity and food consumption are independently regulated . Single cell analysis of AKHR+ neurons identified the co-expression of Drosophila insulin-like receptor ( dInR ) , which imposed suppressive effect on starvation-induced hyperactivity . Therefore , insulin and glucagon signaling exert opposite effects on starvation-induced hyperactivity via a common neural target in Drosophila . Energy homeostasis is vital for survival , growth , and reproduction of animal species ( Gautron et al . , 2015 ) . Energy deprivation drives a complex behavioral program to ensure adequate food intake ( Sternson et al . , 2013 ) . As an adaptive response to starvation , food intake is modulated by nutrient and hormonal cues . In mammals , the hypothalamic arcuate nucleus , especially the neurons expressing agouti-related protein ( AgRP ) /neuropeptide Y ( NPY ) , and those expressing pro-opiomelanocortin ( POMC ) , sense and integrate numerous nutritional cues such as circulating glucose , insulin , leptin , and ghrelin and modulate food intake in response ( Belgardt et al . , 2009 ) . In fruit flies , several hormonal signals modulate different aspects of food intake , these including Drosophila insulin-like peptides ( DILPs ) , the two homologs of mammalian NPY , neuropeptide F ( NPF ) and short neuropeptide F ( sNPF ) , and a handful of other neuropeptides such as allatostatin A ( AstA ) , leucokinin , and hugin ( Pool and Scott , 2014 ) . It is generally accepted that these hormonal cues represent the nutritional status of animals and translate them into physiological and behavioral responses ( Pool and Scott , 2014; Sternson et al . , 2013 ) . The search of appropriate food sources ( i . e . food seeking ) is often the first step of food intake , followed by the actual ingestion of food ( i . e . food consumption ) ( Stephens et al . , 2007 ) . Starvation enhances food-seeking behavior in two parallel ways . On the one hand , starvation modulates the perception of food associated sensory cues , increasing the likelihood to target desirable food sources . In fruit flies , starvation increases the sensitivity of Or42b+ olfactory receptor neurons ( ORNs ) that mediate odor attraction , and decreases the sensitivity of aversive Or85a+ ORNs , resulting in enhanced behavioral attraction to food odors ( Ko et al . , 2015; Root et al . , 2011 ) . Similarly , starvation also influences the sensitivity of sweet-sensing Gr5a+ gustatory receptor neurons ( GRNs ) and bitter-sensing Gr66a+ GRNs in opposite directions , resulting in enhanced attractiveness to food taste ( Inagaki et al . , 2012; Inagaki et al . , 2014; Marella et al . , 2012 ) . A number of neurohormonal cues , including DILPs , NPF , sNPF , dopamine , and tachykinin ( DTK ) , are involved in the sensory modulation by starvation ( Inagaki et al . , 2012; Inagaki et al . , 2014; Ko et al . , 2015; Marella et al . , 2012; Root et al . , 2011 ) . On the other hand , starvation also promotes locomotor activity in both rodents and fruit flies , which may facilitate the exploration of the environment and increase the possibility to locate potential food sources ( Dietrich et al . , 2015; Isabel et al . , 2005; Lee and Park , 2004; Yang et al . , 2015 ) . In fruit flies , starvation-induced hyperactivity requires the neuroendocrine cells producing AKH , the insect analog of glucagon ( Isabel et al . , 2005; Lee and Park , 2004 ) . However , AKH has a role in modulating lipid storage as well as starvation resistance and whether it imposes a direct effect on starvation-induced hyperactivity remains unclear ( Bharucha et al . , 2008; Grönke et al . , 2007 ) . We have also shown that octopamine , the insect analog of norepinephrine , is required for starvation induced hyperactivity ( Yang et al . , 2015 ) . But given the diverse anatomical distribution and physiological function of octopaminergic neurons in the fly brain , the underlying neural circuitry of starvation-induced hyperactivity and how it is modulated by hormonal cues remain largely unknown ( El-Kholy et al . , 2015 ) . Starvation-induced hyperactivity and food consumption are reciprocally inhibitory and the transition of these two behaviors relies on the detection of food cues ( Chen et al . , 2015; Yang et al . , 2015 ) . It is of great interest to understand how these two behaviors are dynamically regulated by the nutritional states and the availability of food sources . In rodents , acute activation of hypothalamic AgRP/NPY neurons induces both hyperactivity and food consumption , suggesting that these two behaviors are regulated by a same group of neurons and hormonal cues ( Aponte et al . , 2011; Atasoy et al . , 2012; Betley et al . , 2015; Dietrich et al . , 2015; Krashes et al . , 2011 ) . However , the detection of food sources rapidly suppresses starvation-induced hyperactivity along with the activity of AgRP/NPY neurons before actual food consumption , suggesting that these neurons may only be involved in the locomotor response upon starvation but not food consumption ( Betley et al . , 2015; Chen et al . , 2015 ) . Thus , it remains elusive whether starvation-induced hyperactivity and food consumption are modulated by the same set of hormones via a common neural target , or they are independently regulated ( Seeley and Berridge , 2015 ) . In this present study , we aimed to better understand how starvation-induced hyperactivity is regulated by hormonal signals , and whether starvation-induced hyperactivity and food consumption are interdependently or independently regulated . Through a neuronal-specific RNAi screen , we find that a small group of octopaminergic neurons located in the subesophageal zone ( SEZ ) of the fly brain are both necessary and sufficient for starvation-induced hyperactivity . These neurons co-express AKHR and dInR , the receptor of hunger hormone AKH and the receptor of satiety hormone DILPs , respectively , which modulate starvation-induced hyperactivity in opposite directions . Thus , these AKHR+dInR+ octopaminergic neurons represent a common neural target for two sets of functionally antagonizing hormones to modulate starvation-induced hyperactivity . Notably , manipulating starvation-induced hyperactivity does not interfere with starvation-induced food consumption , suggesting that starvation-induced changes in different food intake behaviors are independently regulated by different hormonal cues and neural circuitry in fruit flies . We have previously shown that starvation induces sustained increase in locomotor activity of adult flies ( Yang et al . , 2015 ) . To identify the hormonal cues that regulate this behavior , we performed a neuron-specific RNAi screen in adult flies and examined the influence on their locomotor activity , which was indirectly measured by their frequency to cross the midline of tubes in the Drosophila Activity Monitor System ( DAMS , Trikinetics ) ( Figure 1a–b ) . We crossed UAS-RNAi lines targeting 31 candidate neuropeptide receptors to a pan-neuronal GAL4 driver , elav-GAL4 , and assayed their female progeny for baseline locomotion under fed conditions ( Figure 1c ) . 16 out of the 31 lines exhibited significantly altered activity compared to the control , suggesting that the baseline locomotion may be a behavioral trait sensitive to multiple neuropeptidergic signaling systems ( Supplementary file 1 ) . 10 . 7554/eLife . 15693 . 003Figure 1 . A neuron-specific RNAi screen for starvation-induced hyperactivity in adult Drosophila . ( a–b ) The DAMS-based locomotion assay . ( a ) A single virgin female fly was hosted in a 5 × 65 mm polycarbonate tube , secured between 2% agar medium ( with or without 5% sucrose ) on one end and a piece of cotton on the other . The ruler was used for illustrating the size of the tube . ( b ) The tube was then inserted into the DAMS monitor ( Trikinetics ) . One DAM2 monitor can hold 32 tubes at a time . The passage of flies through the middle of the tube was counted by an infrared beam . The frequency of midline crossings therefore indirectly measured flies’ locomotor activity . ( c ) The workflow of our RNAi screen . Note that UAS-RNAi transgenes were integrated into either the second or the third chromosome . ( d ) Summary of our RNAi screen ( n = 29–48 ) . For each line , their average daily midline crossing activity when fed ad libitum with 5% sucrose ( 'Baseline locomotion' ) , their activity upon starvation ( 'Locomotion upon starvation' ) , and the relative increase in locomotion ( 'Increase in locomotion' ) were listed . 'Control' is the progeny of GAL4 driver line crossed to a wild type strain . The statistical difference between fed vs . starved conditions were listed for each RNAi line . Data are shown as means ( ± SEM ) . NS , p>0 . 05; p<0 . 05; p<0 . 01; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 15693 . 003 Besides the baseline activity , we also examined the changes in locomotor activity upon starvation ( Figure 1c ) . Consistent with our previous report ( Yang et al . , 2015 ) , control flies exhibited ~60% increase in their locomotor activity upon starvation ( Figure 1d , 'Control' ) . Amongst 31 RNAi lines we examined , only the neuronal knock-down of AKHR eliminated starvation-induced hyperactivity ( Figure 1d , 'AKHR' ) , whereas all other RNAi lines still exhibited significantly enhanced locomotion upon starvation ( Figure 1d ) . Notably , neuronal knock-down of AKHR did not interfere with the baseline locomotion under fed conditions ( Supplementary file 1 ) , suggesting that AKHR is not involved in general motor control , but specifically in the regulation of locomotion by the internal nutritional states . Consistent with the results from our RNAi screen , we found that AKHR-/- mutants exhibited no increase in locomotion upon starvation ( Figure 2a–c ) . It is worth noting that a previous report did not observe the behavioral difference between AKHR-/- mutants vs . the control under starvation conditions , which was likely due to the short time window of the behavioral assay ( Bharucha et al . , 2008 ) . 10 . 7554/eLife . 15693 . 004Figure 2 . AKH-AKHR signaling is required for starvation-induced hyperactivity . ( a–b ) Midline crossing activity of indicated genotypes assayed in the presence of 5% sucrose ( '+Sucrose' ) or 2% agar ( '+Sucrose' ) ( n = 61–63 ) . Yellow bars represent 12 hr light-on period in this and following figures . ( c ) Average daily midline crossing activity of flies assayed in a–b . ( d–e ) Midline crossing activity of indicated genotypes assayed in the presence of 5% sucrose ( '+Sucrose' ) or 2% agar ( '-Sucrose' ) ( n = 24–30 ) . ( f ) Average daily midline crossing activity of flies assayed in d–e . Error bars represent SEM . NS , p>0 . 05; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 15693 . 004 AKHR is the candidate receptor of AKH , the insect analog of mammalian glucagon . Similar to its mammalian counterpart , AKH secretion is induced by the reduction in circulating sugar levels , which in turn mobilizes lipid storage for energy supply ( Kim and Rulifson , 2004 ) . Previous reports have shown that genetic ablation of AKH-producing cells located in the corpora cardiaca led to increased fat storage and diminished starvation-induced hyperactivity ( Isabel et al . , 2005; Lee and Park , 2004 ) . Consistently , we also found that eliminating AKH expression abolished starvation-induced hyperactivity ( Figure 2d–f ) . Taken together , AKH-AKHR signaling is required for starvation-induced hyperactivity in adult flies . Previous reports have shown that AKHR is expressed in the fat body and regulates lipid storage ( Bharucha et al . , 2008; Grönke et al . , 2007 ) . We also confirmed that AKHR-/- mutant flies had excessive fat storage ( Figure 3—figure supplement 1 ) . Therefore , it was unclear whether AKHR signaling had a direct role in regulating starvation-induced hyperactivity in the nervous system , or merely delayed starvation-induced hyperactivity by enhancing lipid storage , or both . To further examine the function of neuronal AKHR , we used a different pan-neuronal driver neuronal synaptobrevin-GAL4 ( nSyb-GAL4 ) to knock down AKHR in the nervous system . Consistent with the results from our RNAi screen , pan-neuronal elimination of AKHR showed comparable locomotor activity under fed and starved conditions , whereas the two genetic controls exhibited significant increase in locomotion upon starvation ( Figure 3a–c ) . Notably , neuronal knock-down of AKHR did not affect lipid storage of flies ( Figure 3—figure supplement 1 ) , and fat body knock-down of AKHR did not eliminate starvation-induced hyperactivity ( Figure 3—figure supplement 2 ) , disassociating the behavioral and metabolic effects of AKHR signaling . These results confirm that only neuronal AKHR , but not fat body-expressed AKHR , is required for starvation-induced hyperactivity in flies . To further evaluate the function of neuronal AKHR in starvation-induced hyperactivity , we used a behavioral assay that directly quantified the walking velocity as well as the position of individual flies in the presence or absence of food cues ( Figure 3—figure supplement 3a ) . By using this behavioral assay , we confirmed that flies with neuronal knock-down of AKHR exhibited comparable walking velocity under fed vs . starved conditions , whereas the two control genotypes exhibited significantly increased walking velocity upon starvation ( Figure 3—figure supplement 3e ) . Besides increased locomotion , starved flies also exhibited increased interest to food and could rapidly locate and occupy food sources ( Ko et al . , 2015; Root et al . , 2011; Yang et al . , 2015 ) . Interestingly , starved flies from all three genotypes could locate and occupy food sources in the center of the behavioral chamber ( Figure 3—figure supplement 3b–d ) . Taken together , these data confirm that neuronal AKHR is required for starvation-induced hyperactivity without interfering their ability to locate and occupy food sources . Starvation promotes both locomotor activity and food consumption . But it remained unclear whether these two behaviors are regulated by the same set of hormonal cues or not . We thus asked whether neuronal AKHR was also required for starvation-induced increase in food consumption . We first examined food consumption during the course of a single meal , by using a capillary-based feeding assay named the MAFE ( MAnual FEeding ) assay ( Qi et al . , 2015 ) ( Figure 3d ) . We found that starvation could significantly increase food consumption , and that neuronal knock-down of AKHR did not affect starvation-induced increase in meal size ( Figure 3f ) . 10 . 7554/eLife . 15693 . 005Figure 3 . Neuronal AKHR is required for starvation-induced hyperactivity but not food consumption . ( a–b ) Midline crossing activity of indicated genotypes assayed in the presence of 5% sucrose ( '+Sucrose' ) or 2% agar ( '-Sucrose' ) ( n = 31–80 ) . ( c ) Average daily midline crossing activity of flies assayed in a–b . ( d–e ) Schematic illustration of the MAFE assay ( d ) or the FLIC assay ( e ) . ( f ) Volume of 800 mM sucrose consumed in a meal by indicated genotypes fed ad libitum with 5% sucrose , or starved for 36 hr using the MAFE assay . ( n = 30–39 ) . ( g–i ) Representative feeding plots of individual flies in the FLIC assay . Flies of the indicated genotypes were starved for 36 hr before the assays . The plots show the electrical current signals that reflected the physical contact between flies and the liquid food ( 800 mM sucrose ) . The a . u . higher than 120 ( red line ) was considered as feeding events ( arrows ) . Asterisks indicate possible 'tasting/licking' events . ( j–k ) Number of feeding bouts ( j ) and total duration of feeding time ( k ) during 1 hr recording in the FLIC assay ( n = 50–72 ) . Note that one control genotype ( green ) exhibited higher levels of feeding likely due to the genetic background of nSyb-GAL4 we used . Error bars represent SEM . NS , p>0 . 05; p<0 . 05; *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 15693 . 00510 . 7554/eLife . 15693 . 006Figure 3—figure supplement 1 . Neuronal AKHR is not involved in the regulation of fat storage . Average amount of triglyceride of indicated genotypes fed ad libitum ( n = 20–30 ) . NS , p>0 . 05; p<0 . 001 . For detailed protocols , please see Yang et al . ( 2015 ) . Octopamine mediates starvation-induced hyperactivity in adult Drosophila . Proceedings of the National Academy of Sciences 112 , 5219–5224 . DOI: http://dx . doi . org/10 . 7554/eLife . 15693 . 00610 . 7554/eLife . 15693 . 007Figure 3—figure supplement 2 . AKHR expressed in the fat body is not required for starvation-induced hyperactivity . ( a–b ) Midline crossing activity of indicated genotypes assayed in the presence of 5% sucrose ( '+Sucrose' ) or 2% agar ( '-Sucrose' ) ( n = 44–64 ) . ( c ) Average daily midline crossing activity of flies assayed in a–b . p<0 . 05; *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 15693 . 00710 . 7554/eLife . 15693 . 008Figure 3—figure supplement 3 . Neuronal AKHR is required for starvation-induced hyperactivity but not the localization and occupation of food sources . ( a ) Top view of a behavioral chamber ( 10 mm ( D ) × 4 mm ( H ) ) . The dashed circle outlines an agar patch ± 5% sucrose . ( b–d ) Spatial distribution of indicated genotypes assayed fed ad libitum in the presence of sucrose ( first row ) , and starved flies assayed in the absence or presence of sucrose ( second and third rows ) ( n = 10–20 ) . The average walking velocity for each treatment was listed at the upper left corner ( mm/s ) . Color temperature of the scale bar represents the percentage of time spent on each pixel for the duration of the assay ( 8 hr ) . ( e ) Walking velocity of flies assayed in b–d ( n = 10–20 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15693 . 008 We also assayed long-term food consumption using the FLIC ( Fly Liquid-food Interaction Counter ) assay that measured the frequency and duration of physical contacts between fly’s proboscis and the liquid food ( Ro et al . , 2014 ) ( Figure 3e , also see Figure 3g–i ) . We found that eliminating AKHR in neurons did not interfere with long-term feeding , as measured by both the number of feeding bouts and the total duration of feeding ( Figure 3j–k ) . Taken together , these data show that neuronal AKHR is specifically required for starvation-induced hyperactivity but not food consumption . It is likely that several other hormonal cues mediate the effect of starvation on food consumption ( Figure 7i ) ( Pool and Scott , 2014 ) . We next asked which AKHR+ neurons were required for starvation-induced hyperactivity . Previous reports have shown that besides the fat body , AKHR is also expressed in the sweet-sensing gustatory neurons ( Bharucha et al . , 2008 ) . However , we found that knocking down AKHR in Gr5a+ sweet-sensing gustatory neurons did not interfere with starvation-induced hyperactivity ( Figure 4—figure supplement 1 ) . Moreover , we have previously shown that starvation-induced hyperactivity did not require gustatory input ( Yang et al . , 2015 ) . These data suggest that AKHR is likely expressed in additional neurons other than those in the gustatory sensory organs , where it regulates starvation-induced hyperactivity . To specifically examine the neuronal expression of AKHR , we utilized the split-GAL4 system and generated transgenic flies carrying the GAL4 Activation Domain under the control of nSyb promoter ( nSyb:AD ) and the GAL4 DNA-binding Domain under the control of AKHR promoter ( AKHR:BD ) . Using the combinational nSyb:AD/AKHR:BD-GAL4 driver that induces gene expression only in AKHR+ neurons , we found that AKHR was expressed in a small group of neurons located in the SEZ ( 2–4 neurons per hemisphere ) but not in the ventral nerve cord ( data not shown ) ( Figure 4a–b ) . The cell bodies of these AKHR+ neurons were located in the ventrolateral region of the SEZ and sent Y-shaped neural projections into the SEZ ( Figure 4b ) . As previously reported ( Bharucha et al . , 2008 ) , we also found that AKHR was expressed in the sweet-sensing gustatory neurons and projected to the SEZ region of the brain ( Figure 4c–d ) . As discussed above , however , these gustatory neurons are not involved in starvation-induced hyperactivity ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 15693 . 009Figure 4 . AKHR+ neurons are required for starvation-induced hyperactivity . ( a , c ) The expression of membrane-bound GFP ( mCD8GFP ) in AKHR+neurons in the anterior ( a ) and the posterior ( c ) part of the fly brain . The dashed box outlines the region of SEZ . ( b , d ) An enlarged image of the SEZ region seen in a and c . Note the cell bodies located in the ventrolateral side of the SEZ ( arrowheads ) and their Y-shaped projections ( arrows ) in b . Green: GFP . Magenta: nc82 . Scale bars represent 20 μm in a–d . ( e–f ) Midline crossing activity of indicated genotypes assayed in the presence of 5% sucrose ( '+Sucrose' ) or 2% agar ( '-Sucrose' ) ( n = 60–75 ) . ( g ) Average daily midline crossing activity of flies assayed in e–f . ( h ) Volume of 800 mM sucrose consumed in a meal by indicated genotypes fed ad libitum with 5% sucrose , or starved for 24 hr using the MAFE assay ( n = 18–40 ) . ( i–j ) Number of feeding bouts ( i ) and total duration of feeding time ( j ) during 1 hr recording in the FLIC assay ( n = 20–24 ) . Error bars represent SEM . NS , p>0 . 05; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 15693 . 00910 . 7554/eLife . 15693 . 010Figure 4—figure supplement 1 . AKHR expressed in Gr5a+ gustatory sensory neurons is not required for starvation-induced hyperactivity . ( a–b ) Midline crossing activity of indicated genotypes assayed in the presence of 5% sucrose ( '+Sucrose' ) or 2% agar ( '-Sucrose' ) ( n = 46–64 ) . ( c ) Average daily midline crossing activity of flies assayed in a–b . *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 15693 . 010 To test whether AKHR+ neurons in the SEZ were required for starvation-induced hyperactivity , we expressed Shibirets1 , a temperature sensitive form of dynamin ( Kitamoto , 2001 ) , in AKHR+ neurons and acutely blocked neuronal transmission by transferring the flies to non-permissive temperature ( 30°C ) before the behavioral assays . We found that acute silencing of AKHR+ neurons eliminated starvation-induced hyperactivity ( Figure 4e–g ) . Therefore , AKHR+ neurons in the fly brain are required for increased locomotion upon starvation . We also examined whether AKHR+ neurons were required for starvation-induced increase in food consumption . Acute silencing of AKHR+ neurons did not block the increase in food consumption by starvation , as assayed by the changes in meal size upon starvation ( Figure 4h ) , and by the number of feeding bouts and total feeding duration ( Figure 4i–j ) . Therefore , AKHR and AKHR+ neurons specifically regulate starvation-induced hyperactivity , but not food consumption . We then sought to investigate whether AKHR+ neurons played an instructive or merely permissive role in regulating starvation-induced hyperactivity . To do so , we expressed NaChBac ( Nitabach et al . , 2006 ) , a bacterial sodium channel , in AKHR+ neurons . NaChBac expression increases the membrane excitability and facilitates the activation of these neurons ( Nitabach et al . , 2006 ) . We found that NaChBac expression in AKHR+ neurons accelerated the onset of hyperactivity upon starvation . While the two control lines only exhibited a significant increase in locomotion during Day two of the assay , NaChBac expression in AKHR+ neurons led to increased locomotor activity from Day one of the assay ( Figure 5a–d ) . Consistently , over-expression of AKHR in AKHR+ neurons also led to earlier onset of starvation-induced hyperactivity ( Figure 5—figure supplement 1 ) . We also examined whether activation of AKHR+ neurons influenced starvation-induced increase in food consumption . As shown in Figure 5e , NaChBac expression in AKHR+ neurons did not change the meal size of both fed and starved flies . Activating AKHR+ neurons did not alter the duration and frequency of feeding behavior , either ( Figure 5f–g ) . Taken together , the regulation of starvation-induced hyperactivity can therefore be disassociated from the regulation of starvation-induced food consumption , since the former is neither necessary nor sufficient for the latter ( Figures 4 and 5 ) . 10 . 7554/eLife . 15693 . 011Figure 5 . AKHR+ neurons promote starvation-induced hyperactivity . ( a , c ) Midline crossing activity of indicated genotypes assayed in the presence of 5% sucrose ( '+Sucrose' ) or 2% agar ( '-Sucrose' ) ( n = 48–80 ) . ( b , d ) Average daily midline crossing activity during Day one ( b ) and Day two ( d ) of flies assayed in a , c . ( e ) Volume of 800 mM sucrose consumed in a meal by indicated genotypes fed ad libitum with 5% sucrose , or starved for 18 and 36 hr using the MAFE assay ( n = 38–52 ) . ( f–g ) Number of feeding bouts ( f ) and total duration of feeding time ( g ) during 1 hr recording in the FLIC assay ( n = 54–60 ) . Error bars represent SEM . NS , p>0 . 05; p<0 . 05; p<0 . 01; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 15693 . 01110 . 7554/eLife . 15693 . 012Figure 5—figure supplement 1 . AKHR over-expression in AKHR+ neuron accelerates starvation-induced hyperactivity . ( a–b ) Midline crossing activity of indicated genotypes assayed in the presence of 5% sucrose ( '+Sucrose' ) or 2% agar ( '-Sucrose' ) ( n = 46–48 ) . ( c–e ) Average daily midline crossing activity of flies assayed in a–b , breaking down to three different time windows ( Day one-Daytime , Day one-Nighttime , Day two ) . NS , p>0 . 05; p<0 . 05; p<0 . 01; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 15693 . 012 We next sought to further investigate the mechanism underlying the regulation of starvation-induced hyperactivity by AKHR+ neurons . We hypothesized that besides AKH , AKHR+ neurons might also sense other hormonal cues that exerted modulatory effects on the activity of AKHR+ neurons and hence starvation-induced hyperactivity . To examine this hypothesis , we labeled AKHR+ neurons with mCD8GFP , extracted individual GFP+ neurons ( and randomly picked GFP- neurons as negative controls ) from fly brains , and performed single-cell RT-PCR analysis . All GFP+ neurons we examined were AKHR positive , confirming that the nSyb:AD/AKHR:BD-GAL4 driver reliably recapitulates the endogenous expression pattern of AKHR ( Figure 6a–b , 'GFP' and 'AKHR' ) . We also performed single-cell RNA-seq using the cells labeled by the nSyb:AD/AKHR:BD-GAL4 driver . Consistent with our single-cell RT-PCR experiments , all GFP+ cells we examined by RNA-seq showed considerable AKHR expression ( Figure 6c , 'AKHR' ) . In contrast , these GFP+ cells showed no or very low expression for genes specifically expressed in the peripheral nervous system , muscle , and the fat body ( Figure 6c , 'Or67d' , 'Gr5a' , 'MyoD' , and 'Slimfast' ) . 10 . 7554/eLife . 15693 . 013Figure 6 . Insulin signaling suppresses starvation-induced hyperactivity via AKHR+ neurons . ( a ) Heat map indicating the expression pattern of neuropeptide receptors we examined in individual AKHR+ ( green ) and AKHR- ( black ) neurons . Red and black blocks represent genes that could and could not be detected by RT-PCR , respectively . For a complete list of all 32 genes we examined see Supplementary file 1 . The first three genes , GFP , AKHR , and dInR are indicated in bold . ( b ) Representative RT-PCR bands for indicated genes shown in a . Note that the lower bands in the 'dInR' row are primer dimers . ( c ) Gene expression in five GFP+ cells assayed by RNA-seq ( shown in RPKM ) . Or67d and Gr5a are expressed in the primary sensory neurons; MyoD in muscle; and Slimfast in the fat body . ( d–e ) Midline crossing activity of indicated genotypes assayed in the presence of 5% sucrose ( '+Sucrose' ) or 2% agar ( '-Sucrose' ) ( n = 48–80 ) . ( f–g ) Average daily midline crossing activity during Day one and Day two of flies assayed in d–e . ( h–i ) Midline crossing activity of indicated genotypes assayed in the presence of 5% sucrose ( '+Sucrose' ) or 2% agar ( '-Sucrose' ) ( n = 48–80 ) . ( j–k ) Average daily midline crossing activity during Day one and Day two of flies assayed in h–i . Error bars represent SEM . NS , p>0 . 05; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 15693 . 01310 . 7554/eLife . 15693 . 014Figure 6—figure supplement 1 . Co-localization of AKHR and dInR in the fly brain . Two representative series are shown ( upper and lower rows ) . ( a , d ) Merged images , showing the co-localization of dInR ( red ) and AKHR ( green ) in the SEZ ( arrowheads ) . dInR expression was shown by antibody staining ( b , e ) . GFP expression was driven by the nSyb:AD/AKHR:BD-GAL4 driver ( c , f ) . Scale bars represent 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15693 . 01410 . 7554/eLife . 15693 . 015Figure 6—figure supplement 2 . dInR over-expression in AKHR+ neuron delays starvation-induced hyperactivity . ( a–b ) Midline crossing activity of indicated genotypes assayed in the presence of 5% sucrose ( '+Sucrose' ) or 2% agar ( '-Sucrose' ) ( n = 41–64 ) . ( c–d ) Average daily midline crossing activity during Day one and Day two of flies assayed in a–b . NS , p>0 . 05; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 15693 . 01510 . 7554/eLife . 15693 . 016Figure 6—figure supplement 3 . DMS-R2 in AKHR+ neurons is not required for starvation-induced hyperactivity . ( a–b ) Midline crossing activity of indicated genotypes assayed in the presence of 5% sucrose ( '+Sucrose' ) or 2% agar ( '-Sucrose' ) ( n = 48 ) . ( c ) Average daily midline crossing activity of flies assayed in a–b . p<0 . 01; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 15693 . 016 We then examined the expression of a collection of candidate neuropeptide receptors in AKHR+ neurons . Among all receptor genes we examined ( Supplementary file 1 ) , we found that the transcripts of dInR were steadily detected in GFP+ neurons by single-cell RT-PCR ( Figure 6a–b , 'dInR' ) . The co-localization of dInR and AKHR was also confirmed by RNA-seq ( Figure 6c , 'dInR' ) and antibody staining ( Figure 6—figure supplement 1 ) . dInR is the receptor of DILPs , the insect analogs of mammalian insulin . Similar to insulin , DILPs function as satiety signals . Dietary sugars and proteins induce the release of DILPs into the hemolymph , which in turn promotes growth and nutrient storage ( Buch et al . , 2008; Rulifson et al . , 2002 ) . Therefore , the expression of dInR in AKHR+ neurons suggests that insulin signaling may counteract the effect of AKH-AKHR signaling and suppress the activity of AKHR+ neurons hence starvation-induced hyperactivity . To test this hypothesis , we knocked down the expression of dInR in AKHR+ neurons and examined its effect on flies’ locomotion . We found that similar to the effect of activating AKHR+ neurons ( Figure 5a–d ) , eliminating dInR expression in AKHR+ neurons accelerated the onset of hyperactivity upon starvation ( Figure 6d–g ) . Conversely , over-expression of dInR in AKHR+ neurons delayed the onset of starvation-induced hyperactivity ( Figure 6—figure supplement 2 ) . Collectively , these data suggest that insulin signaling in AKHR+ neurons suppress the activity of these neurons and hence starvation-induced hyperactivity . To confirm this finding , we over-expressed Drosophila PTEN ( dPTEN ) , a negative regulator of insulin signaling pathway , in AKHR+ neurons ( Gao et al . , 2000 ) . Over-expression of dPTEN led to an earlier onset of starvation-induced hyperactivity , phenocopying the effect of dInR knock-down in AKHR+ neurons ( Figure 6h–k ) . Taken together , these data suggest that the activity of AKHR+ neurons is modulated by two sets of functionally antagonizing hormones , AKH and DILPs . The hunger hormone AKH activates these neurons via AKHR and promotes locomotor activity upon starvation , whereas the satiety hormones DILPs exert a suppressive effect on starvation-induced hyperactivity via dInR signaling . The interplay between these two sets of hormonal cues therefore modulates locomotor activity upon changes in flies’ internal nutritional states . Besides dInR , Myosuppressin receptor 2 ( DMS-R2 ) was also moderately enriched in AKHR+ neurons ( Figure 6a , Line 4 ) . Knocking down DMS-R2 in AKHR+ neurons , however , did not affect starvation-induced hyperactivity ( Figure 6—figure supplement 3 ) . We next asked how AKHR+ neurons connected the upstream hormonal signals to downstream neural circuitry . To do so , we aimed to understand which neurotransmitter ( s ) AKHR+ neurons utilized to modulate starvation-induced hyperactivity . We performed single-cell RT-PCR on AKHR+ neurons for genes critical for the biosynthesis of several neurotransmitters including dopamine , serotonin , octopamine , tyramine , acetylcholine , glutamate , GABA , and histamine ( Supplementary file 1 ) . Single-cell RT-PCR showed that the transcripts of tyramine beta-hydroxylase ( TβH ) were detected in GFP+ neurons ( Figure 7a–b ) . These results were also confirmed by single-cell RNA-seq ( Figure 7c ) . 10 . 7554/eLife . 15693 . 017Figure 7 . Octopamine mediates the effect of AKHR+ neurons on starvation-induced hyperactivity . ( a ) Heat map indicating the expression of neurotransmitter related genes in individual AKHR+ and AKHR- neurons . Red and black blocks represent genes that could and could not be detected by RT-PCR , respectively . For a complete list of all 10 genes we examined see Supplementary file 1 . ( b ) Representative RT-PCR bands for indicated genes shown in a . ( c ) Gene expression in five GFP+ cells assayed by RNA-seq ( shown in RPKM ) . ( d–e ) Midline crossing activity of indicated genotypes assayed in the presence of 5% sucrose ( '+Sucrose' ) or 2% agar ( '-Sucrose' ) ( n = 38–48 ) . ( f ) Average daily midline crossing activity of indicated genotypes in d–e . ( g–h ) Average daily midline crossing activity of indicated genotypes ( n = 43–65 ) . ( i ) A working model . Starvation promotes both food seeking and food consumption . Food seeking has two components: food targeting ( i . e . perception of food cues ) and environmental exploration ( i . e . hyperactivity ) . In this present study ( highlighted in red ) , we have shown that a small group of octopaminergic neurons located in the fly brain are both necessary and sufficient for starvation-induced hyperactivity , an important aspect of food seeking . These neurons express AKHR and dInR , the receptors of AKH and DILPs , respectively . These neurons are octopaminergic ( OA ) , and likely exert their behavioral effect via downstream neurons expressing certain octopamine receptor ( s ) ( OA-R ) . It is worth noting that the regulation of starvation-induced hyperactivity is independent from that of food consumption , and vice versa . NS , p>0 . 05; p<0 . 01; *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 15693 . 01710 . 7554/eLife . 15693 . 018Figure 7—figure supplement 1 . TβH is required for starvation-induced hyperactivity . ( a–b ) Midline crossing activity of indicated genotypes assayed in the presence of 5% sucrose ( '+Sucrose' ) or 2% agar ( '-Sucrose' ) ( n = 31–77 ) . ( c ) Average daily midline crossing activity of flies assayed in a–b . NS , p>0 . 05; p<0 . 01; ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 15693 . 018 TβH is critical for the biosynthesis of octopamine , the insect analog of vertebrate norepinephrine ( Roeder , 2004 ) . Octopamine is involved in the regulation of various behaviors , including sleep ( Crocker and Sehgal , 2008 ) , learning ( Burke et al . , 2012 ) , and aggression ( Hoyer et al . , 2008 ) . We have also shown previously that octopamine signaling was also required for starvation-induced hyperactivity ( Yang et al . , 2015 ) . We therefore asked whether octopamine mediated the effect of AKHR+ neurons on this behavior . We first showed that similar to the phenotype of TβHM18 null mutants ( Yang et al . , 2015 ) , pan-neuronal knock-down of TβH expression eliminated starvation-induced hyperactivity ( Figure 7—figure supplement 1 ) . Furthermore , knocking down the expression of TβH specificallyin AKHR+ neurons eliminated starvation-induced hyperactivity ( Figure 7d–f ) . Consistently , knocking down the expression of AKHR in octopaminergic neurons also blocked the increase in locomotion by starvation ( Figure 7g ) . Taken together , our data suggest that AKHR+ neurons in the fly brain are octopaminergic , and that octopamine signaling mediates the effect of these neurons on starvation-induced hyperactivity . Single-cell RT-PCR also identified that the transcripts of Dopa decarboxylase ( DDC ) , a key enzyme for the biosynthesis of both dopamine and serotonin ( Lundell and Hirsh , 1994 ) , was moderately enriched in AKHR+ neurons ( Figure 7a–b ) . We thus examined the potential role of DDC in AKHR+ neurons . Nevertheless , eliminating DDC expression in AKHR+ neurons exerted no effect on the increased locomotion by starvation ( Figure 7h ) . Food seeking and food consumption are essential for the acquisition of food sources , and hence survival , growth , and reproduction of animal species ( Gao and Horvath , 2007 ) . Starvation influences food-seeking behavior via both modulating the perception of food cues as well as enhancing flies’ locomotor activity ( Figure 7i ) . Accumulated evidence has suggested that starvation modulates the activity of ORNs via multiple neural and hormonal cues ( Beshel and Zhong , 2013; Ko et al . , 2015 ) , which in turn facilitates odor-driven food search ( Root et al . , 2011 ) and food consumption ( Farhadian et al . , 2012; Wang et al . , 2013 ) . Similarly , starvation also modulates the perception of food taste via the relative sensitivity of appetitive sweet-sensing and aversive bitter-sensing GRNs , which may in turn increase the attractiveness of food taste ( Inagaki et al . , 2012; Inagaki et al . , 2014; Marella et al . , 2012 ) . However , how starvation increases the locomotor activity of flies remains largely uncharacterized . Consistent with previous reports ( Isabel et al . , 2005; Lee and Park , 2004 ) , we have shown that starved fruit flies exhibit sustained increase in their locomotor activity , which can be suppressed by food consumption induced by both nutritive and non-nutritive food cues ( Yang et al . , 2015 ) . In the present study , we have shown that a small group of neurons located in the SEZ region of the fly brain are both necessary and sufficient for starvation-induced hyperactivity . These neurons sense the changes in flies’ internal nutritional states by directly responding to two sets of hormones , AKH and DILPs , and modulate locomotor activity in response . Single cell analysis has identified that these AKHR+dInR+ neurons are octopaminergic , which offers an entry point to trace the downstream neural circuitry that regulates starvation-induced hyperactivity . For example , there are 7 candidate octopamine receptors in fruit flies and it would be of interest to investigate whether any of these receptors and the receptor-expressing neurons are involved in locomotor regulation upon starvation ( El-Kholy et al . , 2015 ) . AKH and DILPs are two sets of functionally counteracting hormones in fruit flies . As its mammalian analog glucagon , the reduction in circulating sugars induces the release of AKH , which in turn mobilizes fat storage and provides energy supply for flies ( Bharucha et al . , 2008; Kim and Rulifson , 2004; Lee and Park , 2004 ) . In contrast , DILPs , the insect analog of mammalian insulin , function as satiety hormones . Dietary nutrient induces the release of DILPs into the hemolymph , which in turn promotes protein synthesis , body growth , and other anabolic processes ( Buch et al . , 2008; Rulifson et al . , 2002 ) . We have shown that these two hormonal signaling systems exert opposite effects on starvation-induced hyperactivity via a small group of AKHR+InR+ octopaminergic neurons . These results suggest that these AKHR+dInR+ neurons can integrate the inputs from the two hormonal signaling systems representing hunger and satiety at the same time , and modulate flies’ locomotor activity ( Figure 7i ) . This elegant yet concise design allows these neurons to be responsive to rapid changes in the internal nutritional states as well as food availability . Furthermore , it is possible that besides hunger and satiety , other physiological states such as wakefulness , stress , and emotions also influence flies’ locomotor activity . Notably , our single cell analysis has shown that these AKHR+dInR+ neurons also sparsely express other neuropeptide receptors , suggesting that at least small portions of these neurons may also receive input from other neuropeptidergic systems . Starved animals exhibited increased locomotion and food consumption , the transition of which relies on the detection of food cues ( Chen et al . , 2015; Yang et al . , 2015 ) . But whether these two behaviors are interdependently or independently regulated remains unclear . In this present study , we have shown that these two behaviors are dissociable from each other in fruit flies ( Figure 7i ) . On the one hand , although AKHR+ neurons exert a robust modulatory effect on starvation-induced hyperactivity , these neurons are neither necessary nor sufficient for starvation-induced food consumption ( Figures 4 and 5 ) . On the other hand , the regulation of food consumption is independent of starvation-induced hyperactivity as well . We have previously shown that a small subset of GABAergic neurons in the fly brain regulates food consumption but exerts no effect on starvation-induced hyperactivity ( Pool et al . , 2014 ) . In addition , several neuropeptides are known to regulate food consumption , such as Hugin , NPF , sNPF , Leucokinin , and AstA ( Al-Anzi et al . , 2010; Hergarden et al . , 2012; Lee et al . , 2004; Melcher and Pankratz , 2005; Wu et al . , 2005 ) . However , we found in our RNAi screen that the receptors of these neuropeptides were not involved in the regulation of starvation-induced hyperactivity ( Figure 1d ) . Taken together , it is likely that starvation-induced hyperactivity and food consumption are independently regulated by different sets of hormonal cues , and that AKHR+ neurons are only involved in the former but not the latter . Our results may shed light on the regulation of food intake in mammals , especially whether starvation-induced hyperactivity and food consumption are also independently regulated by different sets of hormones and distinct neural circuitry in mammals . Flies were kept on a standard fly medium made of yeast , corn , and agar at 25°C and 60% humidity on a 12 hr light: 12 hr dark cycle . Virgin female flies were collected shortly after eclosion and kept in groups ( 20 flies per vial ) for 5–6 d before experiments . For experiments involving Shibirets1 , flies were raised at 18°C for 8–9 d before transferring to 30°C right before the behavioral assays . All UAS-RNAi lines used in the screen ( #25832 , #25858 , #25925 , #25935 , #25936 , #25939 , #25940 , #26017 , #27275 , #27280 , #27494 , #27669 , #27506 , #27507 , #27509 , #27513 , #27529 , #27539 , #28580 , #28780 , #28781 , #28783 , #29414 , #29577 , #29624 , #31490 , #34947 , #31884 , #31958 , #33627 , #38347 ) , UAS-TβH RNAi ( #27667 ) , UAS-AKH RNAi ( #27031 ) , AKH-GAL4 ( #25683 ) , UAS-mCD8GFP ( #32186 ) , Tdc2-GAL4 ( #9313 ) , fat body specific ppl-GAL4 ( #58768 ) , and gustatory neuron specific Gr5a-GAL4 ( #57592 ) were obtained from the Bloomington Drosophila Stock Center at Indiana University . UAS-DDC RNAi ( #2197 . N ) and UAS-dInR RNAi ( #5713 ) were from the Tsinghua Fly Center . AKHR-/- and UAS-AKHR flies were from R . Kühnlein ( Grönke et al . , 2007 ) . Transgenic flies were generated using the methods described previously ( Pfeiffer et al . , 2010 ) . Briefly , the nSyb promoter ( −1826 to +78 ) and the AKHR promoter ( −2804 to +55 ) were generated by PCR from genomic DNA and cloned into pDONR221 vector ( Life Technologies ) , and subsequently cloned into pBPGUw ( Addgene #17575 ) , pBPp65ADZpUw ( Addgene #26234 ) , and pBPZpGAL4DBDUw ( Addgene #26233 ) to generate nSyb-GAL4 , AKHR-GAL4 , nSyb:AD and AKHR:BD constructs . These cloned DNA fragments were then integrated into AttP40 ( 25C6 ) , AttP2 ( 68A4 ) , and VK00031 ( 62E1 ) landing sites , respectively . The brains of ice anesthetized flies were dissected in PBS and then fixed in 4% PFA on ice for at least 1 hr . Fixed brains were incubated in Penetration/Blocking Buffer ( 2% Triton X-100 and 10% Calf Serum in PBS ) for 20–24 hr at 4°C and incubated in Dilution Buffer ( 0 . 25% Triton X-100 and 1% Calf Serum in PBS ) with primary antibodies at 4°C for 20–24 hr . The samples were then washed in Washing Buffer ( 3% NaCl and 1% Triton X-100 in PBS ) for 3 × 30 min at room temperature and subsequently incubated in secondary antibodies for 20–24 hr at 4°C . The brains were washed again in Washing Buffer before mounted in Fluoroshield with DAPI ( Sigma-Aldrich ) for confocal imaging ( Nikon 60 × A/1 . 20 WI ) . Antibodies were used at the following dilutions: mouse anti-nc82 ( 1:200 , DSHB ) , rabbit anti-GFP ( 1:500 , Life Technologies ) , mouse anti-GFP ( 1:500 , abcam ) , rabbit anti-dInR ( 1:100 , Cell Signaling Technology ) , Alexa Fluor 546 goat anti mouse ( 1:300 , Life Technologies ) , Alexa Fluor 488 goat anti rabbit ( 1:300 , Life Technologies ) , Alexa Fluor 546 goat anti rabbit ( 1:300 , Life Technologies ) , and Alexa Fluor 488 goat anti mouse ( 1:300 , Life Technologies ) . The method of single-cell cDNA preparation has been described previously ( Tang et al . , 2010 ) . In brief , individual GFP+ cells were picked with a glass micropipette in situ ( pipettes were pulled from thick-walled borosilicate capillaries ( BF120-69-10 , Sutter Instruments ) for initial tip opening of 1–2 μm under a dissecting microscope and transferred into lysate buffer ( 0 . 9 × PCR Reaction Buffer II , 1 . 35 mM MgCl2 , 0 . 45% NP40 , 4 . 5 mM DTT , 0 . 18 U/µl SUPERase-In , 0 . 36 U/µl RNase inhibitor , 12 . 5 nM UP1 primer , 0 . 045 mM dNTP mix ) immediately , followed by reverse transcription using oligo ( dT ) primers to generated first-strand cDNA . A poly ( A ) tail was then added to the 3′ end of the first-strand cDNA by terminal deoxynucleotidyl transferase . The cDNA was amplified by 29 cycles of PCR with universal oligo ( dT ) primers , and then tested by nested PCR with the primers listed in Supplementary file 1 . Individual GFP+ cells were harvested as described above , subjected to cDNA amplification ( SMARTer Ultra Low RNA Kit for Sequencing , Clonetech ) , library preparation ( NEBNext Ultra II DNA Library Prep Kit , NEB ) , and sequencing ( Illumina Hiseq2500/4000 platform ) . Reads were subsequently mapped to Drosophila genome and only uniquely mapped reads were kept for further analysis . Gene expression levels were quantified and compared by RPKM ( Reads Per Kilobase per Million mapped reads ) . Data presented in this study were verified for normal distribution by D’Agostino-Pearson omnibus test . Student’s t-test , one-way ANOVA , and two-way ANOVA were applied for pair-wise comparisons , comparisons among 3 or more data sets , and comparisons with more than one variant , respectively . The post hoc test with Bonferroni correction was performed for multiple comparisons following ANOVA .	Animals can be thought of as tightly controlled eating machines . An animal’s brain senses if it is hungry via signals from the nervous system or hormones , and then alters the animal’s behavior to obtain a supply of food . These behaviors include looking for food and eating it; and regulating both food seeking and food consumption behaviors is crucial for the animal’s chances of survival and reproduction . Studies that used fruit flies as a model have previously shown that flies walk more when they are hungry . This activity helped the flies to locate and occupy food sources , but it was not clear how this food seeking behavior was regulated . Now , Yu , Huang et al . find that a small group of neurons in the fly brain controls food seeking in starving flies . The neurons achieve this by sensing two groups of hormones with opposing activity . These hormones are the fly’s equivalents of glucagon and insulin , which are found in humans and other mammals . In humans , glucagon is released when blood sugar levels are low and stimulates hunger , while insulin is released when blood sugar is high and acts to suppress feelings of hunger . Therefore , food seeking in the flies is under the precise control of signals of hunger and satiety . Further experiments show that these fly neurons use a chemical messenger called octopamine to convey the hormone-based signals to other circuits of neurons . Notably , these downstream neurons are not involved in regulating the consumption of food . Therefore , food seeking and eating appear to be independently regulated in fruit flies . Further studies are now needed to dissect the downstream circuits of neurons that actually control the food seeking behavior . It will also be important to explore how this behavior is suppressed when a food source is detected .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Regulation of starvation-induced hyperactivity by insulin and glucagon signaling in adult Drosophila
PKMζ is a persistently active PKC isoform proposed to maintain late-LTP and long-term memory . But late-LTP and memory are maintained without PKMζ in PKMζ-null mice . Two hypotheses can account for these findings . First , PKMζ is unimportant for LTP or memory . Second , PKMζ is essential for late-LTP and long-term memory in wild-type mice , and PKMζ-null mice recruit compensatory mechanisms . We find that whereas PKMζ persistently increases in LTP maintenance in wild-type mice , PKCι/λ , a gene-product closely related to PKMζ , persistently increases in LTP maintenance in PKMζ-null mice . Using a pharmacogenetic approach , we find PKMζ-antisense in hippocampus blocks late-LTP and spatial long-term memory in wild-type mice , but not in PKMζ-null mice without the target mRNA . Conversely , a PKCι/λ-antagonist disrupts late-LTP and spatial memory in PKMζ-null mice but not in wild-type mice . Thus , whereas PKMζ is essential for wild-type LTP and long-term memory , persistent PKCι/λ activation compensates for PKMζ loss in PKMζ-null mice . LTP and long-term memory can be divided into two mechanistically distinct phases—a transient induction and a persistent maintenance ( Malinow et al . , 1988 ) . Induction is thought to rely solely on post-translational modifications . Maintenance requires new protein synthesis soon after strong synaptic stimulation or learning , and these newly synthesized proteins are believed to sustain the synaptic potentiation or behavioral modification ( Kandel and Schwartz , 1982 ) . Although numerous signal transduction molecules are important for LTP and long-term memory , most have been implicated in induction , with many participating in either the initial transient potentiation or the mechanisms for upregulating new protein synthesis ( Sanes and Lichtman , 1999 ) . In contrast , few molecules have been implicated in maintenance . Such a maintenance molecule would be both: 1 ) a product of de novo protein synthesis sufficient to enhance synaptic transmission and 2 ) a necessary component of the mechanism sustaining synaptic potentiation and long-term memory , as shown by inhibition of the molecule reversing sustained synaptic potentiation and erasing long-term memory . One hypothesis for a molecular mechanism of maintenance involves the persistent increase in an autonomously active PKC isoform , PKMζ ( Sacktor , 2011 ) . LTP and long-term memory maintenance may depend upon the function of PKMζ because data suggest the kinase possesses the two essential properties of a maintenance molecule . First , PKMζ is produced in LTP by new protein synthesis and is sufficient to potentiate synaptic transmission . PKMζ is generated from a PKMζ mRNA , but this mRNA is translationally repressed in the basal state of neurons ( Hernandez et al . , 2003 ) . During LTP , strong afferent synaptic stimulation derepresses the mRNA and rapidly increases the de novo synthesis of PKMζ ( Hernandez et al . , 2003 ) . The newly synthesized kinase , unlike most other protein kinases , is autonomously active without the requirement for second messenger stimulation ( Sacktor et al . , 1993; Hernandez et al . , 2003 ) , and the autonomous activity of PKMζ is sufficient to enhance synaptic transmission ( Ling et al . , 2002; 2006; Yao et al . , 2008 ) . Second , multiple inhibitors of PKMζ and a dominant-negative mutated form of PKMζ reverse established LTP maintenance or disrupt long-term memory storage ( Ling et al . , 2002; Serrano et al . , 2005; Pastalkova et al . , 2006; Shema et al . , 2011; Cai et al . , 2011 ) . Recently , this proposed function of PKMζ has been challenged by new findings that LTP and long-term memory appear normal in PKMζ-null mice ( Lee et al . , 2013; Volk et al . , 2013 ) . Moreover , the PKMζ-inhibitor ZIP , which disrupts the maintenance of LTP and long-term memory in wild-type animals , disrupts these same processes in PKMζ-null mice ( Lee et al . , 2013; Volk et al . , 2013 ) . Two hypotheses can account for these findings ( Frankland and Josselyn , 2013; Matt and Hell , 2013 ) . First , in a straightforward hypothesis , PKMζ is unnecessary for LTP or long-term memory , and therefore genetically deleting PKMζ has no effect on these processes ( Lee et al . , 2013; Volk et al . , 2013 ) . Second , PKMζ is essential for late-LTP and long-term memory in wild-type mice , and compensatory mechanisms emerge in the mutant mice to substitute for PKMζ , which are also inhibited by ZIP . Here , we used a pharmacogenetic approach to distinguish between the 'PKMζ is unnecessary hypothesis' and the 'PKMζ is compensated hypothesis' . We first confirmed the previously published findings that late-LTP appears similar in PKMζ-null and wild-type mice , and that ZIP ( 5 µM ) applied to the bath 3 hr after tetanization reverses late-LTP in both wild-type mice ( Serrano et al . , 2005 ) and PKMζ-null mice ( Volk et al . , 2013 , Figure 1A , B ) . In interface chambers in which maximal drug concentrations are achieved slowly , the reversal of late-LTP may be more rapid in wild-type mice than in PKMζ-null mice ( time to 50% of the pre-ZIP response is different: wild-type , 129 ± 28 min; PKMζ-null , 311 ± 72 min; t7 = 2 . 57 , p = 0 . 037 , d = 1 . 64 ) . 10 . 7554/eLife . 14846 . 003Figure 1 . ZIP reverses LTP maintenance in both wild-type mice and PKMζ-null mice and blocks synaptic potentiation mediated by PKMζ and PKCι/λ . Bath applications of ZIP ( 5 µM ) reverse ( A ) wild-type-LTP maintenance and ( B ) PKMζ-null-LTP maintenance ( filled circles ) . Above insets , numbered representative fEPSP traces correspond to time points noted below . Below , mean ± SEM . For ( A ) , wild-type , n = 5 , average response 5 min before ZIP compared to 3 . 5 hr post-ZIP , t4 = 4 . 83 , p = 0 . 0084 , d = 1 . 85; for ( B ) , PKMζ-null , n = 4 , t3 = 3 . 34 , p = 0 . 045 , d = 2 . 88 . Non-tetanized pathways are stable in the presence of ZIP ( open circles ) . For ( A ) , wild-type non-tetanized pathway: 5 min pre-ZIP vs . 3 . 5 hr post-ZIP; n = 5 , t4 = 1 . 73 , p = 0 . 16; d = 0 . 49; for ( B ) , PKMζ-null non-tetanized pathway: n = 4 , t3 = 0 . 82 , p = 0 . 47 , d = 0 . 058 . ( C ) ZIP inhibits both PKMζ and , at higher doses , the autonomous activity of PKCι/λ . The main effects and interactions are all significant ( kinase: F1 , 30 = 85 . 4 , p = 2 . 77 X 10–10 , η2 = 0 . 036; ZIP concentration: F4 , 30 = 200 . 56 , p = 3 . 48 X 10–21 , η2 = 0 . 34; interaction: F4 , 30 = 26 . 59 , p = 1 . 98 X 10–9 , η2 = 0 . 045 ) . Post-hoc tests show the kinases respond differently at 1 µM and 2 µM ZIP . ( D , E ) , ZIP blocks EPSC potentiation produced by postsynaptic dialysis of PKMζ or PKCι/λ in CA1 pyramidal cells in hippocampal slices . ZIP ( 5 µM ) is applied to the bath prior to obtaining whole-cell patch . Above insets , numbered representative EPSC traces correspond to time points noted below . Statistical comparisons are at 15 min after whole-cell patch . ( D ) PKMζ: n’s = 4 , F2 , 11 = 18 . 07 , p = 0 . 0007 , d = 1 . 78; post-hoc tests: PKMζ vs . baseline , p = 0 . 0012; PKMζ vs . PKMζ + ZIP , p = 0 . 0018; PKMζ + ZIP vs . baseline , p = 0 . 94 . ( E ) PKCι/λ: n’s = 4 , F2 , 11 = 35 . 2 , p = 1 . 66 X 10–5 , d = 1 . 79; post-hoc tests: PKCι/λ vs . baseline , p<0 . 0001; PKCι/λ vs . PKCι/λ + ZIP , p<0 . 0001; PKCι/λ + ZIP vs . baseline , p = 0 . 86 . DOI: http://dx . doi . org/10 . 7554/eLife . 14846 . 00310 . 7554/eLife . 14846 . 004Figure 1—figure supplement 1 . ZIP has no effect on synaptic potentiation induced by activation of conventional/novel PKCs and produces no perturbation of neuronal membrane conductance . ( A ) ZIP has no effect on fEPSP potentiation produced by phorbol esters ( phorbol 12 , 13-dibutyrate , PDBu , 1 µM ) , activators of conventional/novel PKCs . Synaptic potentiation induced by phorbol esters is blocked by the conventional/novel PKC inhibitor Gö 6983 ( 100 nmol ) , but not by ZIP ( 5 µM ) ( phorbol , n = 5 , phorbol + ZIP , n = 5 , phorbol + Gö 6983 , n = 6 , F2 , 13 = 10 . 9 , p = 0 . 0017 , d = 1 . 84; phorbol vs . phorbol + ZIP , p = 0 . 85; phorbol vs . phorbol + Gö 6983 , p = 0 . 0073; phorbol + ZIP vs . phorbol + Gö 6983 , p = 0 . 026 ) . ( B ) ZIP ( 5 µM ) produces no perturbation of membrane conductance of CA1 pyramidal cells , as expected for recordings of stable baseline fEPSP responses ( panel A , Figure 1A , B ) , as well as previous studies demonstrating membrane stability at this concentration of ZIP that reverses LTP ( Wang et al . , 2012 ) . Above , representative measurements of membrane conductance ( between arrows ) in response to current pulse , followed by evoked EPSC . Below , mean ± SEM ( n = 4 , t3 = 0 . 33 , p = 0 . 76 , d = 0 . 10 ) . We note that Volk et al . had reported ZIP induces decreases in both tetanized and untetanized pathways in slices ( Volk et al . , 2013 ) . These results conflict with evidence from a large number of studies showing exclusive actions of ZIP in tetanized or facilitated pathways and not in baseline pathways , including seven studies in brain slices of fEPSPs ( Ling et al . , 2002; Serrano et al . , 2005; Sajikumar et al . , 2005; Navakkode et al . , 2010; Lin et al . , 2012; Panaccione et al . , 2013; Chen et al . , 2015 ) , four studies of EPSCs ( Li et al . , 2010; Yao et al . , 2013; Velez-Hernandez et al . , 2013; Li et al . , 2014 ) , two studies in model systems ( Cai et al . , 2011; Balaban et al . , 2015 ) , and four studies in vivo of fEPSPs ( Pastalkova et al . , 2006; Madronal et al . , 2010; Dong et al . , 2015 ) and evoked responses ( Cooke and Bear , 2010 ) , as well as evidence of stable basal properties of hippocampal neurons following ZIP injections recorded in vivo ( Barry et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14846 . 00410 . 7554/eLife . 14846 . 005Figure 1—figure supplement 2 . Chelerythrine inhibits both PKMζ and PKCι/λ . ( A ) Chelerythrine inhibits PKMζ phosphorylation of the standard atypical PKC substrate , ε-peptide ( 25 µM , filled circles ) . In contrast , Lee et al . , who had reported no effect of chelerythrine on PKMζ activity ( Lee et al . , 2013 ) , had performed kinase assays using a substrate based upon the pseudosubstrate sequence of the conventional isoform PKCα ( TRF0108-D , PerkinElmer ) , which we find is a relatively poor substrate for PKMζ . We confirm that the inhibition of PKMζ by chelerythrine cannot be detected using this substrate at the concentration employed in Lee et al . ( 50 nM , open circles ) . We note that chelerythrine applied to the bath of hippocampal slices suppresses the synaptic potentiation produced by postsynaptic perfusion of PKMζ in CA1 pyramidal cells , indicating the drug inhibits PKMζ’s phosphorylation of physiologically relevant substrates in neurons ( Ling et al . , 2002; 2006 ) . ( B ) Under the appropriate assay conditions for observing inhibition using the standard ε-peptide substrate as described in ( A ) , chelerythrine inhibits both PKMζ and PKCι/λ . DOI: http://dx . doi . org/10 . 7554/eLife . 14846 . 005 Because these results clearly show that ZIP affects molecules that can maintain synaptic potentiation in addition to PKMζ , we examined whether ZIP blocks synaptic enhancement produced by other PKC isoforms . PKC is a gene family , consisting of conventional , novel , and atypical isoform classes ( Nishizuka , 1988 ) . ZIP inhibits the phosphotransferase activity of both PKMζ and , at higher doses , the other atypical PKC , PKCι/λ ( mouse PKCλ and human PKCι are orthologous genes ) ( Ren et al . , 2013 ) ( Figure 1C ) . We found that ZIP at the concentration standardly used to reverse late-LTP ( 5 µM ) ( Serrano et al . , 2005 ) blocks the synaptic potentiation produced by postsynaptic perfusion of both atypical PKCs ( Figure 1D , E ) . In contrast , this dose of ZIP does not block the synaptic potentiation produced by conventional/novel PKCs activated by phorbol esters ( Figure 1—figure supplement 1A ) . This dose of ZIP also has no effect on basal properties of neurons , such as field excitatory postsynaptic potentials ( fEPSPs ) in non-tetanized synaptic pathways recorded within the slices of the wild-type and PKMζ-null mice ( Figure 1A , B ) or on the membrane stability of CA1 pyramidal cells ( Figure 1—figure supplement 1B ) . Because ZIP blocks the synaptic potentiation produced by both PKMζ and PKCι/λ , it is unsuitable for evaluating PKMζ’s function in pharmacogenetic experiments . We therefore took advantage of the specific nucleotide sequence of the translation start site of the PKMζ mRNA that has no significant homology with sequences in any other known mRNA , except PKCζ mRNA that is not expressed in the hippocampus ( Hernandez et al . , 2003 ) . We hypothesized that if there were compensation for PKMζ’s function in PKMζ-null mice during LTP , a PKMζ-antisense oligodeoxynucleotide should prevent late-LTP in wild-type mice , but not in PKMζ-null mice that lack the antisense’s target mRNA ( Figure 2A , Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 14846 . 006Figure 2 . PKMζ is essential for late-LTP in wild-type mice , and compensation accounts for late-LTP in PKMζ-null mice . ( A ) Diagram illustrating the PKMζ-compensation hypothesis tested by pharmacogenetic analysis of LTP . The Prkcz gene consists of an autoinhibitory PKCζ regulatory domain exon region ( Reg , shown in red ) and a catalytic domain exon region ( Cat , green ) . In neurons in wild-type mice , PKMζ is produced by an internal promoter within the Prkcz gene , transcribing a PKMζ mRNA that expresses an independent ζ catalytic domain ( indicated as step [1] [Hernandez et al . , 2003] ) . The PKMζ mRNA is transported to dendrites ( Muslimov et al . , 2004 ) but is translationally repressed ( [2] [Hernandez et al . , 2003] ) . During tetanic stimulation , glutamate ( Glu ) activates the NMDAR to stimulate Ca2+-dependent induction mechanisms that release the translational block ( [3] [Hernandez et al . , 2003] ) , resulting in synthesis of PKMζ ( [4] [Hernandez et al . , 2003] ) , which potentiates postsynaptic AMPARs ( [5] [Ling et al . , 2002; Serrano et al . , 2005] ) . If wild-type mice express persistently enhanced AMPAR-mediated synaptic transmission through synthesis of PKMζ and PKMζ-null mice through compensatory mechanisms , then PKMζ-antisense will block LTP in wild-type mice ( left ) but have no effect in PKMζ-null mice ( right ) . ( B ) The PKMζ-antisense ( 20 µM ) blocks the new synthesis of PKMζ , but not PKCι/λ or the eukaryotic elongation factor 1A ( eEF1A ) that are also rapidly synthesized in LTP . In the presence of antisense or scrambled oligodeoxynucleotides , adjacent slices from the same hippocampus are either tetanized or untetanized , and 30-min post-tetanization CA1 regions are assayed by immunoblot . The levels of protein in the untetanized slices are set at 100% . Mean ± SEM; denotes significance; n . s . , no significance . PKMζ: scrambled , tetanized ( n = 17 ) vs . untetanized ( n = 19 ) , t34 = 3 . 81 , p = 0 . 00056 , d = 1 . 27; antisense , tetanized ( n = 12 ) vs . untetanized ( n = 18 ) , t28 = 1 . 35 , p = 0 . 19 , d = 0 . 50; antisense vs . scrambled , t27 = 2 . 12 , p = 0 . 043 , d = 0 . 80; PKCι/λ: scrambled , tetanized ( n = 17 ) vs . untetanized ( n = 18 ) , t33 = 3 . 72 , p = 0 . 00074 , d = 1 . 26; antisense , tetanized ( n = 12 ) vs . untetanized ( n = 17 ) , t27 = 3 . 59 , p = 0 . 0013 , d = 1 . 35; antisense vs . scrambled , t27 = 0 . 71 , p = 0 . 49 , d = 0 . 27; eEF1A: scrambled , tetanized ( n = 9 ) vs . untetanized ( n = 10 ) , t17 = 2 . 40 , p = 0 . 028 , d = 1 . 10; antisense , tetanized ( n = 12 ) vs . untetanized ( n = 18 ) , t28 = 2 . 07 , p = 0 . 048 , d = 0 . 77; antisense vs . scrambled , t19 = 0 . 47 , p = 0 . 64 , d = 0 . 21 . ( C ) PKMζ-antisense blocks late-LTP in wild-type mice but has no effect on LTP in PKMζ-null mice . Left , representative fEPSPs; numbers correspond to time points at right . Right , mean ± SEM . Scrambled version of the oligodeoxynucleotide has no effect on either genotype . Circles show tetanized pathways , and color-coded squares show control untetanized pathways within each slice that receive only test stimulation . Tetanization is at arrow . N’s are: wild-type + antisense: 7; PKMζ-null + antisense: 5; wild-type + scrambled: 8; PKMζ-null + scrambled: 6 . The genotype X drug X time ANOVA with repeated measures on time ( average of the 5 min ending at 30 min post-tetanization and 180 min post-tetanization ) confirmed a significant genotype X drug interaction , F1 , 22 = 4 . 7; p = 0 . 041 , η2 = 0 . 0026 . Post-hoc tests confirmed antisense on wild-type at 180 min post-tetanization is significantly less than all other responses . DOI: http://dx . doi . org/10 . 7554/eLife . 14846 . 00610 . 7554/eLife . 14846 . 007Figure 2—figure supplement 1 . The mutant PKMζ gene expresses neither PKMζ mRNA nor protein . ( A ) Strategy for the excision of the exon encoding the PKC/PKMζ catalytic domain ATP-binding site , as previously described ( Lee et al . , 2013 ) . DNA sequencing of the PKMζ locus in PKMζ-null mice shows upstream and downstream intronic sequences indicating the exon deletion . ( B ) Agarose gel shows the lowered molecular weight of the DNA PCR amplification product , following excision in PKMζ-null mouse , as compared to wild-type mouse . Primer locations are shown in panel ( A ) . ( C ) RT-PCR shows no amplification of PKMζ mRNA in the hippocampus of the PKMζ-null . ( D ) Immunoblot shows no expression of PKMζ protein in the hippocampus of the PKMζ-null . The full-length PKCζ is not expressed in wild-type hippocampus because its promoter is inactive ( Hernandez et al . , 2003 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14846 . 00710 . 7554/eLife . 14846 . 008Figure 2—figure supplement 2 . PKMζ-antisense bath-applied to hippocampal slices for 2 hr does not affect the amount of basal PKMζ or PKCι/λ . Left , representative immunoblots; right , mean ± SEM ( PKMζ: n’s = 8; t14 = 0 . 16 , p = 0 . 88 , d = 0 . 08; PKCι/λ: n’s = 8; t14 = 0 . 24 , p = 0 . 82 , d = 0 . 12; n . s . , no significance ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14846 . 008 We first validated that bath applications of PKMζ-antisense to hippocampal slices from wild-type mice block the new synthesis of PKMζ and not other gene products induced during LTP ( Figure 2B ) . PKMζ-antisense selectively blocks the increase of PKMζ during LTP , but not the activity-dependent increase of PKCι/λ , which occurs transiently in LTP ( Osten et al . , 1996; Kelly et al . , 2007 ) , or a third protein that is rapidly synthesized in LTP , eukaryotic elongation factor 1A ( eEF1A ) ( Tsokas et al . , 2005 ) ( Figure 2B ) . The brief application of the PKMζ-antisense has no effect on basal levels of PKMζ , as expected for the relatively long half-life of the kinase in untetanized hippocampal slices ( Osten et al . , 1996 ) , or on basal levels of PKCι/λ ( Figure 2—figure supplement 2 ) . We then tested the effect of PKMζ-antisense on late-LTP in wild-type and PKMζ-null mice . Whereas PKMζ-antisense blocks late-LTP in wild-type mice , the same antisense has no effect on late-LTP in PKMζ-null mice ( Figure 2C ) . These results are predicted by the PKMζ is compensated hypothesis but not by the PKMζ is unnecessary hypothesis . Control scrambled oligodeoxynucleotides have no effect on either wild-type- or PKMζ-null-LTP , further supporting the conclusion that the antisense effect is selective for PKMζ . What maintains LTP in PKMζ-null mice ? We first measured the basal levels of all eight remaining PKC isoforms expressed in the dorsal hippocampus of PKMζ-null mice and found increased basal amounts of PKCι/λ and the conventional PKC , PKCβI ( Figure 3—figure supplement 1 , 2A ) . We focused on PKCι/λ because it is the most closely related gene product to PKMζ and is blocked by ZIP ( Figure 1C , E ) . PKCι/λ , which is important for synaptic potentiation during an early phase of LTP through post-translational activation ( Kelly et al . , 2007; Ren et al . , 2013 ) , also has partial activity in the absence of lipid second messengers ( Akimoto et al . , 1994 ) ( Figure 3—figure supplement 2B ) and total levels of PKCι/λ increase transiently after tetanic stimulation in wild-type animals ( Osten et al . , 1996; Kelly et al . , 2007 ) . We therefore examined the possibility that the transient increase in PKCι/λ in wild-type-LTP changes in PKMζ-null-LTP ( Figure 3 ) . As expected , in wild-type mice , the increase of PKCι/λ observed 30 min post-tetanization ( Figure 2B ) returns to basal levels by 3 hr ( Osten et al . , 1996; Kelly et al . , 2007 ) ( Figure 3A ) . But in PKMζ-null mice , the increase in PKCι/λ during LTP lasts at least 3 hr , persisting like the increase of PKMζ in LTP maintenance in wild-type mice ( Figure 3A ) . These data suggest PKCι/λ as a candidate for functionally compensating for the loss of PKMζ in the PKMζ-null mice . 10 . 7554/eLife . 14846 . 009Figure 3 . PKCι/λ inhibitor reverses the maintenance of late-LTP in PKMζ-null mice , but not in wild-type mice . ( A ) Immunoblots show that in wild-type-LTP maintenance , PKMζ persistently increases for 3 hr , whereas PKCι/λ , as determined by both total and activation loop phosphorylated-PKCι/λ antisera , are at baseline . In PKMζ-null-LTP maintenance , both total and activation loop phosphorylated-PKCι/λ persistently increase for 3 hr . Wild-type: PKMζ , untetanized vs . tetanized ( n’s = 9 ) , t16 = 4 . 51 , p = 0 . 00036 , d = 2 . 13; PKCι/λ , untetanized ( n = 6 ) vs . tetanized ( n = 7 ) , t11 = 0 . 19 , p = 0 . 85 , d = 0 . 11; phospho-PKCι/λ , untetanized vs . tetanized ( n’s = 9 ) , t16 = 0 . 86 , p = 0 . 40 , d = 0 . 41 . PKMζ-null: PKCι/λ , untetanized vs . tetanized ( n’s = 7 ) , t12 = 2 . 41 , p = 0 . 033 , d = 1 . 29; phospho-PKCι/λ , untetanized ( n = 8 ) vs . tetanized ( n = 9 ) , t15 = 4 . 35 , p = 0 . 00058 , d = 2 . 11 . PKMζ-null: tetanized total PKCι/λ vs . phospho-PKCι/λ , t14 = 1 . 83 , p = 0 . 09 , d = 0 . 92 . Tetanized wild-type vs . PKMζ-null PKCι/λ , t12 = 2 . 28 , p = 0 . 042 , d = 1 . 22; tetanized wild-type vs . PKMζ-null phospho-PKCι/λ , t16 = 3 . 27 , p = 0 . 0048 , d = 1 . 54 . ( B ) PKMζ-null late-LTP maintenance ( filled circles ) is reversed by PKCι/λ-antagonist ICAP ( 10 µM ) applied 3 hr post-tetanization . Insert above , representative fEPSPs; numbers correspond to time points below . Below , mean ± SEM . Comparing average responses of the 5 min before drug and 3 . 5 hr after drug , n = 4 , t3 = 5 . 4 , p = 0 . 012 , d = 3 . 22 . ICAP has no effect on a second , independent synaptic pathway recorded within the slices that received no tetanization ( open circles ) . ( C ) ICAP has no effect on wild-type LTP maintenance ( filled circles; n = 7 , t6 = 1 . 88 , p = 0 . 11 , d = 0 . 67 ) , but blocks the initial potentiation following tetanization ( right arrow ) in the second synaptic pathway ( open circles ) . The effect of ICAP is different on LTP maintenance in the wild-type and PKMζ-null; t9 = 2 . 75 , p = 0 . 023 , d = 1 . 129 . Right insert above , inhibition of LTP induction is not due to prolonged perfusion in vitro because tetanization of a second pathway recorded for equivalent periods of time induces LTP . DOI: http://dx . doi . org/10 . 7554/eLife . 14846 . 00910 . 7554/eLife . 14846 . 010Figure 3—figure supplement 1 . Analysis of the complete PKC isoform family shows increases in basal expression of PKCι/λ and PKCβI in the dorsal hippocampus of PKMζ-null mice . ( A ) Total PKCι/λ- , activation loop-phosphorylated PKCι/λ- , and ζ-specific antisera show PKMζ-null mice with increased total and phospho-PKCι/λ in dorsal hippocampus , but not in the contralateral hemibrain . In previous studies , the amounts of basal PKCι/λ and phospho-PKCι/λ in the PKMζ-null and wild-type mice were reported to be indistinguishable on immunoblots of total brain homogenates ( Lee et al . , 2013; Volk et al . , 2013 ) . If the increases are regionally selective , however , immunoblots of total brain homogenates may not be sensitive enough to detect the basal increases in dorsal hippocampus ( dorsal and ventral hippocampus together constitute ~5% of total mouse brain [Kovacevic et al . , 2005] ) . Using a 'split-brain' preparation , we find the basal increases of PKCι/λ and phospho-PKCι/λ in dorsal hippocampus are below the level of detection if the entire contralateral hemibrain is homogenized and assayed . Dorsal hippocampus: denotes significance , total PKCι/λ , n’s = 8 , t14 = 4 . 38 , p = 0 . 00063 , d = 2 . 19; phospho-PKCι/λ , PKMζ-null , n = 11 , wild-type , n = 7 , t16 = 2 . 77 , p = 0 . 014 , d = 1 . 34; hemibrain: total PKCι/λ , PKMζ-null , n = 9 , wild-type , n = 8 , t15 = 0 . 73 , p = 0 . 47 , d = 0 . 36; phospho-PKCι/λ , PKMζ-null , n = 7 , wild-type , n = 6 , t11 = 0 . 014 , p = 0 . 89 , d = 0 . 077 . ( B ) The conventional PKCβI increases in dorsal hippocampus of PKMζ-null mice . Initial analysis of all seven conventional/novel PKC isoforms expressed in dorsal hippocampus with n’s of 3–4 revealed an increase in PKCβI ( t6 = 2 . 53 , p = 0 . 045 , d = 1 . 79 ) . To correct for multiple comparisons , the n’s were increased to wild-type , 10 , and PKMζ-null , 13 , resulting in t21 = 3 . 40 , p = 0 . 0026 , d = 1 . 43 , which reached significance with Bonferroni correction . Earlier studies of the PKMζ-null mice used antisera that do not distinguish between PKCβI and PKCβII isoforms ( Lee et al . , 2013; Volk et al . , 2013 ) . PKCβI is transiently activated by translocation to the cell membrane early in LTP induction ( Sacktor et al . , 1993 ) . The amount of the novel PKCθ isoform is below the level of detection by immunoblot in dorsal hippocampus ( Naik et al . , 2000 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14846 . 01010 . 7554/eLife . 14846 . 011Figure 3—figure supplement 2 . Characterization of PKCι/λ immunoreactivity , molecular weight , and kinase activity . ( A ) Left , immunoblot of extract of mouse hippocampus shows two different antisera to PKCι/λ detect a protein band with a molecular weight of 72 kDa ( yellow shows merged image of red and green ) . Right , the antisera to the activation loop of atypical PKC ( Cell Signaling ) recognizes both PKCι/λ ( comigrating with the band recognized by the BD Transduction antiserum ) and a broad band at 76–80 kDa , which are the molecular weights of conventional/novel PKCs ( c/nPKC ) ( Naik et al . , 2000 ) that have similar activation loop phosphorylation sequences . The 72 kDa molecular weight we observe for PKCι/λ is consistent with the molecular weight originally described for mouse PKCλ ( 74 kDa ) ( Akimoto et al . , 1994 ) and with data from Lee et al . ( 72 kDa ) ( Lee et al . , 2013 ) . We note that Volk et al . , who had not observed increases in PKCι/λ in hippocampus or changes in PKCι/λ during LTP in either wild-type or PKMζ-null mice , reported PKCι/λ’s molecular weight detected by a single antiserum as ~80–82 kDa ( Volk et al . , 2013 ) . The reason for the discrepancies between the data of Volk et al . on the molecular weight of PKCι/λ and its response to tetanization vs . the data from other laboratories is unclear . ( B ) Characterization of PKMζ and PKCι/λ enzymatic activity . Left , PKCι/λ is stimulated by lipids and has autonomous activity in the absence of lipids ( with and without phosphatidylserine , n's = 4 , t3 = 9 . 35 , p = 0 . 0026 , d = 4 . 68; baseline vs . without phosphatidylserine , t3 = 4 . 37 , p = 0 . 022 , d = 2 . 18 ) . PKMζ is entirely autonomously active , showing no significant difference in activity with or without lipids ( n’s = 4 , t3 = 1 . 33 , p = 0 . 28 , d = 0 . 67 ) . Right , phosphoinositide-dependent kinase-1 ( PDK1 ) phosphorylation of PKCι/λ’s activation loop , which is measured by phospho-specific antiserum in Figure 3A and Figure 3—figure supplement 1A , enhances PKCι/λ’s autonomous activity ( n’s = 8 , t7 = 4 . 88 , p = 0 . 0018 , d = 1 . 72 ) . PDK1 does not phosphorylate the standard ε-peptide substrate used to assay atypical PKCs . DOI: http://dx . doi . org/10 . 7554/eLife . 14846 . 011 We therefore tested the possible involvement of persistent increased activity of PKCι/λ in maintaining PKMζ-null-LTP . Inhibiting PKCι/λ blocks early-LTP and thus prevents the formation of late-LTP in wild-type animals ( Ren et al . , 2013 ) . Therefore , in order to test PKCι/λ’s potential function in maintaining LTP in PKMζ-null mice , we examined the effects of a PKCι/λ-antagonist applied only after late-LTP had been established . We used the cell-permeable antagonist [4- ( 5-amino-4-carbamoylimidazol-1-yl ) -2 , 3-dihydroxycyclopentyl] methyl dihydrogen ( ICAP ) that blocks PKCι/λ , and not ζ kinase activity ( Pillai et al . , 2011; Sajan et al . , 2013; 2014 ) . We first induced LTP in PKMζ-null and wild-type mice and then applied ICAP to the bath 3 hr later during late-LTP maintenance ( Figure 3B , C ) . The PKCι/λ-inhibitor reverses established PKMζ-null-LTP maintenance but has no effect on wild-type-LTP maintenance . To test the inhibitor’s efficacy in the wild-type slices in which no effect on late-LTP maintenance can be observed , we tetanized a separate synaptic pathway in the presence of the antagonist and found that it blocks early-LTP induction , as expected ( Ren et al . , 2013 ) ( Figure 3C ) . The selective effect of the PKCι/λ-inhibitor on only PKMζ-null-LTP maintenance is predicted by the PKMζ-compensation hypothesis , but not by the PKMζ-unnecessary hypothesis . We used the pharmacogenetic approach to re-examine PKMζ function in hippocampus-dependent spatial memory . General protein synthesis inhibitors are effective in blocking long-term memory in a time-window around the time of conditioning ( Davis and Squire , 1984 ) and intrahippocampally injected antisense lasts at least 2 hr ( Garcia-Osta et al . , 2006 ) . Therefore , for PKMζ-antisense to be present throughout spatial conditioning , we examined active place avoidance , a conditioned behavior that mice can rapidly learn and remember after three 30 min training sessions , spaced 2 hr apart ( Figure 4 ) . Following intrahippocampal injections of control scrambled oligodeoxynucleotides , 1-day memory retention in PKMζ-null and wild-type mice appears the same . In contrast , PKMζ-antisense disrupts 1-day long-term memory in wild-type mice but not in PKMζ-null mice ( Figure 4 , Figure 4—figure supplement 1 ) . These results are predicted by the PKMζ-compensation hypothesis but not the PKMζ-unnecessary hypothesis . 10 . 7554/eLife . 14846 . 012Figure 4 . PKMζ is essential for spatial long-term memory in wild-type mice , and compensation accounts for spatial long-term memory in PKMζ-null mice . PKMζ-antisense blocks spatial long-term memory in wild-type mice but has no effect on long-term memory in PKMζ-null mice . Inserts above , ( left ) schematic diagram of active place avoidance training apparatus and ( middle ) 1-day training protocol . Intrahippocampal injections were 1 nmol oligodeoxynucleotide in 0 . 5 µl vehicle/side , 20 min before each training session . Below left , representative paths during pretraining , the trial at the end of training , and during retention testing with the shock off 1 day after training . The shock zone is shown in red with shock on , and gray with shock off . Red circles denote where shocks are received , and gray circles where shocks would have been received if the shock were on . Right , time to first entry measure of active place avoidance memory ( mean ± SEM ) . There is a significant interaction between the effects of genotype and treatment ( scrambled , antisense ) ( F1 , 39 = 4 . 14 , p = 0 . 049 , η2 = 0 . 037 ) . The individual effects of genotype and treatment are F1 , 39 = 5 . 89 , p = 0 . 02 , η2 = 0 . 053 and F1 , 39 = 1 . 37 , p = 0 . 25 , η2 = 0 . 012 , respectively . Memory retention in the wild-type mice treated with PKMζ-antisense differs from the other groups ( * , significant post-hoc tests; wild-types , n’s = 12 , PKMζ-nulls , scrambled , n = 8 , antisense , 11 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14846 . 01210 . 7554/eLife . 14846 . 013Figure 4—figure supplement 1 . Fluorescence labeling of biotinylated PKMζ-antisense injected bilaterally in mouse hippocampi . Animal is sacrificed 50 min after the last of three 1 nmol in 0 . 5-µl vehicle injections per hippocampus , equivalent to the end of training in Figure 4 . DAPI is counterstain . Scale bar = 600 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 14846 . 013 We then examined whether the persistent action of PKCι/λ maintains spatial long-term memory in PKMζ-null mice . Analogous to the LTP experiments , we tested PKCι/λ’s function in spatial long-term memory maintenance by intracranially injecting the PKCι/λ-inhibitor ICAP in the dorsal hippocampus only after spatial long-term memory was established , using a protocol similar to that previously used for ZIP ( Pastalkova et al . , 2006 ) . We trained animals on active place avoidance and then 1 day later injected the PKCι/λ-inhibitor ICAP in the dorsal hippocampus , testing memory retention without shock 2 hr later ( Figure 5 ) . The PKCι/λ-inhibitor disrupts the retention of spatial long-term memory in PKMζ-null mice , but not in wild-type mice , as predicted by the PKMζ is compensated hypothesis . 10 . 7554/eLife . 14846 . 014Figure 5 . PKCι/λ inhibitor shows spatial long-term memory retention is mediated by distinct molecular mechanisms in PKMζ-null mice and wild-type mice . Insert above , schematic diagram of 1-day training protocol . Injections were 1 nmol ICAP in 0 . 5 µl vehicle/side , 2 hr before retention testing . Below left , representative paths during pretraining , the trial at the end of training , and during retention testing with the shock off 1 day after training . The shock zone is shown in red with shock on , and gray with shock off . Gray circles denote where shocks would have been received if the shock were on . Right , time to first entry measure of active place avoidance memory ( mean ± SEM ) . There is a significant difference in 1-day retention between genotype ( F1 , 13 = 14 . 12 , p = 0 . 0024 , d = 1 . 95 ) ; wild-type , n = 8; PKMζ-null , n = 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 14846 . 014 Because the data on spatial memory support the PKMζ-compensation hypothesis , we asked whether there might be differences in memory expression between wild-type mice that use PKMζ and mutant mice that rely on compensation . We examined this possibility in two ways . First , we made place avoidance more difficult to acquire by shortening the training sessions from 30 to 10 min , so that the mice require several training days for full memory expression ( Figure 6 ) . Clear differences in learning patterns between mutant and wild-type mice are revealed ( Figure 6A , B , Video 1 ) . In the first training trial , wild-type mice learn within minutes to move to the location opposite the shock zone . Mutant mice , although showing equivalent sensitivity to shock as wild-type mice ( Figure 6—figure supplement 1 ) , move to the least safe location of the rotating arena next to where it enters the shock zone . To determine whether the mutants eventually show the normal wild-type response , we extended the training over several more days . After 4 days of training , the PKMζ-null mice acquire the normal avoidance response moving to the quadrant furthest from the shock . 10 . 7554/eLife . 14846 . 015Figure 6 . With a weaker training protocol , PKMζ-null mice show inefficient spatial learning and deficits in spatial memory . ( A ) Above , schematic diagram of 5-day training protocol . Below , color-coded time-in-location maps for wild-type and PKMζ-null mice during pretraining , beginning of training , midpoint of training , asymptote levels of performance , and 1-day memory retention without shock . In the first training trial , wild-type mice move to areas of the arena opposing the shock zone , whereas PKMζ-null mice remain in the adjacent quadrant about to enter the shock zone . Only after multiple days of training do the PKMζ-null mice show the normal avoidance behavior . N’s = 10 . ( B ) Plotting the ratio of time spent in the adjacent quadrant about to enter the shock zone and the safe quadrant opposite the shock zone for each trial shows PKMζ-null mice remain in the adjacent quadrant more than wild-type mice for ~9 trials over 3 days of training . Mean ± SEM; data are analyzed by genotype x trial 2-way ANOVA followed by post-hoc tests as appropriate . The PKMζ-null mice prefer being in the least efficient place for avoiding shock ( genotype: F1 , 237 = 16 . 6 , p = 6 . 30 X 10–5 , η2 = 0 . 047 ) . Effects of trial and the genotype x trial interaction are not significant . Pretraining response is denoted as trial 0 . ( C ) Time to first entry into the shock zone increases to an asymptote over 3–4 days of training in both wild-type and PKMζ-null mice . Whereas the asymptote for wild-type mice is over 500 s , it is about half that for PKMζ-null mice . The main effects and interactions are all significant ( genotype: F1 , 255 = 22 . 4 , p = 3 . 67 X 10–6 , η2 = 0 . 053; trial: F16 , 255 = 7 . 3 , p = 2 . 35 X 10–14 , η2 = 0 . 28; interaction: F16 , 255 = 1 . 8 , p = 0 . 031 , η2 = 0 . 068 ) . Post-hoc tests do not distinguish PKMζ-null memory on any trial from the pretraining and first training trials when there is no avoidance memory , whereas wild-type memory is significantly better as early as day 3 , trial 3 , and is superior to pretraining and PKMζ-null estimates of 1-day memory from day 4 through the final retention test on day 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 14846 . 01510 . 7554/eLife . 14846 . 016Figure 6—figure supplement 1 . Wild-type and PKMζ-null mice are indistinguishable in their motivation to escape shock during active place avoidance training . The shock level is 0 . 2 mA for all animals , which is determined to be the minimum that elicited an escape response . After entering the shock zone , shocks are repeated every 1 . 5 s until the mouse leaves the zone . The ability to escape shock is estimated by counting the number of shocks the mouse received before it left the shock zone . Within the first training trial both the wild-type and PKMζ-null mice learn to leave the shock zone so they only rarely receive more than 1 shock . This indicates good and equivalent ability to escape in the two groups throughout the training . These observations are confirmed by a significant genotype X trial interaction ( F16 , 255 = 4 . 8 , p = 1 . 54 X 10–8 , η2 = 0 . 11 ) . Post-hoc test detects that for both genotypes during pretraining with shock off , the number of shocks the mice would have received for each shock zone entrance ( had the shock been on ) is greater than during all the training sessions with shock on , as expected . During retention with the shock off , post-hoc test detects that the PKMζ-null animals would have received more shocks for each entrance into the shock zone than the wild-type mice . This difference is consistent with worse retention of avoidance memory in the PKMζ-null mice and may indicate that the place avoidance memory is less persistent in the PKMζ-null mice because it rapidly extinguishes ( n’s = 10 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14846 . 01610 . 7554/eLife . 14846 . 017Video 1 . Inefficient place avoidance in the PKMζ-null mouse . These videos show place avoidance behavior during the first training trial . The video on the left shows a wild-type mouse and the video on the right a PKMζ-null mouse . The wild-type mouse rapidly learns to move opposite to the shock zone to better avoid shock , whereas the mutant mouse avoids shock inefficiently by remaining in the area adjacent to the shock zone , which is the most vulnerable place . DOI: http://dx . doi . org/10 . 7554/eLife . 14846 . 017 Like the stronger training protocol ( Figures 4 , 5 ) , the weaker training protocol produces measures of learning and memory in the PKMζ-null mice that are initially indistinguishable from wild-type mice ( Figure 6C ) . But after 3 days , the PKMζ-null mice begin to express more errors than the wild-type mice ( Figure 6A , C ) . The learning and memory deficits remain through the last training on day 5 with the two genotypes reaching distinct asymptotic levels of performance . The PKMζ-null mice also make more errors on memory testing without shock the following day . Second , we examined spatial memory in three unconditioned novelty-preference tests that vary in their cognitive demands ( Mumby et al . , 1999 ) , taking advantage of earlier findings showing that ZIP in hippocampus specifically disrupts information for place , but not environmental context ( Serrano et al . , 2008 , Figure 7 ) . As expected , when memory for discriminating the object-context associations or discriminating between a familiar and novel object-location association is tested ( Lee et al . , 2013 ) , PKMζ-null and wild-type mice remember equally well at 24 hr . But when memory for discriminating between two familiar and novel object-location associations ( object/place mismatch ) is tested , wild-type mice perform well , but PKMζ-null mice show no retention of discrimination memory at 24 hr . 10 . 7554/eLife . 14846 . 018Figure 7 . Long-term memory for where objects are encountered during exploration is tested in three unconditioned memory tests , revealing poor memory in PKMζ-null mice when the places of multiple objects are exchanged . ( A ) Mice explore distinctive boxes ( contexts ) during two 5 min trials per day for 3 days ( 4 objects ) and during three 5 min trials for 1 day ( 2 objects ) . Counterbalanced object/context mismatch and object/place mismatch tests , and a novel object location test are given a day after the last exploration trial , one test per day . For clarity , only a single version of the counterbalanced tests is shown . ( B ) Both wild-type ( n = 8 ) and PKMζ-null ( n = 8 ) mice express similar memory discrimination when two of four objects are encountered in the incorrect context , and when one of two objects is encountered in a novel location . However , despite similar levels of investigating the objects ( 32–36% of the time; effect of group F1 , 14 = 3 . 2; p = 0 . 09 , η2 = 0 . 18; effect of task F2 , 14 = 0 . 004; p = 0 . 95 , η2 = 0 . 00049; group X task interaction F1 , 14 = 0 . 18; p = 0 . 68 , η2 = 0 . 014 ) , only wild-type mice express discrimination memory when the places of two of four objects are exchanged . PKMζ-null mice do not discriminate between objects that are encountered in the familiar and unfamiliar places . The interaction between genotype and memory test is significant ( F2 , 13 = 4 . 49 , p = 0 . 03 , η2 = 0 . 06 ) . The individual effects of genotype and memory test are F1 , 14 = 2 . 23 , p = 0 . 16 , η2 = 0 . 0014 and F2 , 13 = 47 . 3 , p = 10–6 , η2 = 0 . 025 , respectively . Significant post-hoc tests distinguish wild-type and PKMζ-null memory on the object/place mismatch test . DOI: http://dx . doi . org/10 . 7554/eLife . 14846 . 018 PKMζ-null mice express LTP and long-term memory , and a straightforward hypothesis to explain these results is that PKMζ is unnecessary for synaptic plasticity and memory in wild-type mice , and therefore its genetic deletion has no effect ( Lee et al . , 2013; Volk et al . , 2013 ) . But a second hypothesis that could account for these data is that the mechanisms of LTP and long-term memory in wild-type and PKMζ-null mice are not the same , and that PKMζ is essential for these processes in wild-type mice and compensatory mechanisms are recruited in PKMζ-null mice . We used a pharmacogenetic approach to distinguish between these hypotheses . ZIP and chelerythrine are PKMζ-inhibitors but cannot easily be used in a pharmacogenetic approach because they can block the action of both PKMζ and PKCι/λ ( Figure 1C–E , Figure 1—figure supplement 2 ) . But when ζ- and ι/λ-selective inhibitors are examined , experiments reveal a double dissociation between the mechanisms of LTP and spatial long-term memory in PKMζ-null and wild-type mice ( Figures 2 , 3 , 4 , 5 ) . Thus , the data indicate that the mechanisms of LTP and spatial long-term memory in wild-type and PKMζ-null mice are not the same , and the PKMζ is unnecessary hypothesis by which the function of PKMζ in wild-type mice can be inferred from the results obtained from PKMζ-null mice is erroneous . To selectively block the action of PKMζ we took advantage of the specific nucleotide sequence of the PKMζ-mRNA translation start site to develop PKMζ-antisense oligodeoxynucleotides that suppress the activity-dependent de novo synthesis of PKMζ ( Figure 2B ) . In contrast to applications of inhibitors of PKMζ’s phosphotransferase activity that disrupt LTP and long-term memory after PKMζ has already been synthesized ( Ling et al . , 2002; Serrano et al . , 2005; Pastalkova et al . , 2006; Shema et al . , 2011; Cai et al . , 2011 ) , we applied the PKMζ-antisense during the critical temporal window of new protein synthesis during late-LTP and long-term memory formation when general protein synthesis inhibitors such as anisomycin are effective ( Frey and Morris , 1997; Davis and Squire , 1984 ) , and when PKMζ is formed ( Osten et al . , 1996 ) . Acute application of the PKMζ-antisense suppresses the new synthesis of PKMζ and blocks late-LTP without reducing basal levels of the kinase ( Figure 2B , Figure 2—figure supplement 2 ) . This suggests that the crucial pool of PKMζ protein sustaining synaptic potentiation in wild-type mice is synthesized de novo in response to tetanization , rather than through the recruitment of pre-existing PKMζ that had been synthesized before the tetanus . To examine the mechanism maintaining late-LTP in PKMζ-null mice , we focused on the kinase most closely related to PKMζ , the other ZIP-sensitive atypical PKC , PKCι/λ . We found that whereas the increase in PKCι/λ after tetanization is transient in wild-type mice ( Osten et al . , 1996; Kelly et al . , 2007 ) , the increase in PKCι/λ persists in PKMζ-null mice ( Figure 3A ) . In wild-type mice , PKCι/λ is important for early-LTP , and if PKCι/λ is inhibited at the time of afferent synaptic tetanization , neither early- nor late-LTP is formed ( Ren et al . , 2013 ) . In addition , PKCι/λ-null mice , which can only be partially compensated by PKCζ , are embryonically lethal ( Seidl et al . , 2013 ) , and therefore studies of LTP using null mutations in which both atypical PKCs are completely eliminated are not feasible . Therefore , to test for a possible role of PKCι/λ in LTP maintenance , we acutely suppressed the activity of PKCι/λ after late-LTP and spatial long-term memory had been established . We used ICAP that selectively blocks PKCι/λ and not ζ kinase activities and which was developed based upon data from the crystal structure of the PKCι/λ catalytic domain ( Pillai et al . , 2011; Sajan et al . , 2013; 2014 ) . ICAP is designed to convert in cells to a compound that targets and binds to a docking site present in the PKCι/λ catalytic domain , but not in the very closely related PKC/PKMζ catalytic domain , thus competing for binding of substrate proteins to ι/λ and not ζ ( Pillai et al . , 2011; Sajan et al . , 2013; 2014 ) . The PKCι/λ-inhibitor disrupts LTP maintenance and established long-term memory in the PKMζ-null mice but not in wild-type mice ( Figures 3B , C , 5 ) . But there is little information on ICAP on other kinases . Therefore , it is possible that ICAP may block the action of other molecules that may contribute to the maintenance of late-LTP and long-term memory in the PKMζ-null mice ( e . g . PKCβI , which also increases in PKMζ-null mice , Figure 3—figure supplement 1B ) . Although off-target effects of ICAP are thus possible , nonetheless , the specificity of the agent in reversing LTP and spatial memory maintenance in PKMζ-null mice and not in wild-type mice demonstrates that the mechanisms of maintenance in the two mouse genotypes are distinct and is predicted by the PKMζ-compensation hypothesis . The effects of ICAP on LTP maintenance only in PKMζ-null mice ( Figure 3B , C ) are thus consistent with the effects of ZIP on LTP in these mice ( Figure 1B , Volk et al . , 2013 ) , because ZIP blocks the synaptic potentiation produced by both atypical PKCs , PKMζ and PKCι/λ ( Figure 1D , E ) . ZIP does not block the ability of the conventional and novel PKCs to mediate synaptic potentiation , indicating selectivity of the inhibitor towards atypical PKC isoforms ( Figure 1—figure supplement 1A ) . Specificity of ZIP toward atypical PKCs is further supported by evidence that ZIP’s effects on LTP reversal and memory disruption in wild-type animals are completely blocked by preventing the reversal of PKMζ’s action on AMPAR-trafficking , indicating the absence of additional , off-target effects of ZIP in brain slices or in vivo ( Migues et al . , 2010; Pauli et al . , 2012; Evuarherhe et al . , 2014 ) . These results conflict with the non-specific toxic effects of ZIP reported on in vitro cultured neurons or at high concentrations in slices ( Sadeh et al . , 2015 ) . We observed , however , that ZIP causes no membrane perturbation at the lower doses used to block atypical PKCs , but not conventional/novel PKCs , and to specifically reverse potentiated , but not basal synaptic transmission ( Figure 1—figure supplement 1B , Wang et al . , 2012 ) . How PKCι/λ transforms from a transiently increasing kinase in wild-type mice into a persistently increasing kinase in PKMζ-null mice is not known , but likely involves additional compensatory mechanisms that sustain PKCι/λ’s synthesis or decrease its degradation during LTP . A possible role for new synthesis of PKCι/λ in PKMζ-null-LTP is consistent with data that a general protein synthesis inhibitor blocks the formation of late-LTP in PKMζ-null mice ( Volk et al . , 2013 ) . Our results suggest the possibility that in wild-type mice PKMζ suppresses the persistent increase of PKCι/λ and that in the PKMζ-null mice without this repression the increase of PKCι/λ is sustained . PKMζ and PKCι/λ compete for protein binding partners in developing neurons , suggesting that the loss of PKMζ may allow for increased binding of PKCι/λ to these proteins that might augment its stability and function ( Parker et al . , 2013 ) . Whereas PKMζ-null mice produce compensatory increases in PKCι/λ ( Figure 3—figure supplement 1A ) , the acute suppression of PKMζ synthesis by antisense does not ( Figure 2—figure supplement 2 ) . Conditional PKMζ knockdown mice , which show LTP like PKMζ-null mice ( Volk et al . , 2013 ) , have not yet been examined for compensation by PKCι/λ or other molecules . Like the antisense , shRNA knockdown of PKMζ has been found not to induce compensation by PKCι/λ , and to disrupt both late-LTP and long-term memory formation ( Dong et al . , 2015 ) . Therefore , further work will be required to determine when and how the loss of PKMζ induces compensation by PKCι/λ or other PKCs . Persistently increased atypical PKC activity by either PKMζ or PKCι/λ may thus be a common molecular mechanism for maintaining late-LTP and spatial long-term memory in wild-type and PKMζ-null mice , but when the cognitive demands of memory tasks increase , differences in performance emerge . When active place avoidance is made more difficult to acquire , for the first few days of conditioning wild-type mice explore the arena to find the safest location to avoid the shock , but PKMζ-null mice avoid in the least safe position adjacent to the shock zone ( Figure 6A , B , Video 1 ) . After several more days of training , the PKMζ-null mice switch to the wild-type strategy , but nonetheless perform poorly compared to wild-type mice ( Figure 6A , C ) . These differences in conditioned behavior may be a sign that the PKMζ-null mice are defective in learning and memory or in integrating information about the relative safety of multiple areas of the arena . Likewise , during unreinforced novel object placement tasks , both mice with and without PKMζ remember a single object-location association discrimination , but when the number of object location-association discriminations is increased to two , wild-type mice remember well , but PKMζ-null mice express no memory for the learned information ( Figure 7 ) . We speculate that some of the differences in memory expression between wild-type and PKMζ-null mice could be due to the fundamental molecular differences between PKMζ and PKCι/λ . In contrast to the second messenger-independent PKMζ , PKCι/λ can respond to second messengers that might be generated by short-term experiences ( Akimoto et al . , 1994 , Figure 3—figure supplement 2B ) . Thus , in tasks with high cognitive demand , wild-type animals that express both isoforms can use PKCι/λ and PKMζ for separate functions—PKCι/λ for encoding information about short-term experiences , and PKMζ for encoding information derived from these short-term experiences to be stored in long-term memory . But when PKCι/λ is used for both short-term memory and as a 'back-up' mechanism for long-term memory in the PKMζ-null mice , its continued responsiveness to second messengers produced by short-term experiences may interfere with its function in long-term memory . For example , during active place avoidance , when PKMζ-null mice attempt to find the safest location in an arena , information from one recently visited place might interfere with the integration of information from multiple places required for forming the optimal avoidance strategy . Likewise , in novel object placement , short-term memories of recently visited single object locations may suppress the encoding of multiple object-place associations in a single scheme . Further study of PKMζ-null mutant mice might thus reveal fundamental insights into how PKMζ encodes and stores information in long-term memory under physiological conditions in normal animals . Reagents were from Sigma unless specified otherwise . The ζ-specific rabbit polyclonal antiserum ( 1:20 , 000 for immunoblots ) was generated as previously described ( Hernandez et al . , 2003 ) . Total PKCι/λ antiserum was PKCλ mouse monoclonal antibody ( mAb , clone 41/PKCλ; 1:100 ) from BD Transduction Laboratories ( San Jose , CA ) . The identity of PKCι/λ was confirmed with PKCι ( C83H11 ) rabbit mAb #2998 ( 1:500 ) from Cell Signaling , Danvers , MA ( Figure 3—figure supplement 2A ) and by immunoprecipitation with PKCι/λ-specific antiserum ( H-76 , Santa Cruz Biotechnology , Dallas , TX ) ( data not shown ) . Phospho-atypical PKC activation loop antibody #9378 and phospho-PKC ( pan , 190D10 ) from Cell Signaling ( 1:50 ) were raised against the same epitope , the phosphorylated form of the atypical PKC activation loop phosphorylation site . The two antisera recognize the same set of bands and gave identical results ( data not shown ) . The eEF1A antiserum was mouse mAb , clone CBP-KK1 ( 1:5000 ) , from Upstate Biotechnology ( Lake Placid , NY ) , and the actin mouse mAb ( 1:5000 ) was from Sigma . ZIP was from Tocris Bioscience , Bristol , UK , and ICAP from United Chemical Resources , Birmingham , AL . Protein concentrations were determined by assay using bicinchoninic acid ( Pierce Biotechnology , ThermoFisher Scientific , Waltham , MA ) or for hippocampal extracts in reducing agents by the Bio-Rad RC-DC Protein Assay kit ( Hercules , CA ) , using bovine serum albumin as standard . Male mice from the PKMζ-null mouse line , previously described ( Lee et al . , 2013 ) and provided by Robert O Messing ( Univ Texas at Austin , TX ) , were at least 4-month old at testing , and wild-type and null alleles were genotyped using primer pairs ( forward: 5’-GGTATAGTAGGCAGCTATTGCG-3’ and reverse: 5’-TCCTGCCTCAGCCAGAAAACAAACCACACGG-3’ ) to identify homozygotes . All efforts were made to minimize animal suffering and to reduce the number of animals used . For Figure 2—figure supplement 1A , B , mouse genomic DNA was isolated from tail biopsies using DNA Extraction Kit from Agilent Technologies , Santa Clara , CA . The final volume of PCR was 25 µl , containing 0 . 15 ng of genomic DNA . PCR was performed using the primer pairs ( forward: 5’-GGTATAGTAGGCAGCTATTGCG-3’ and reverse: 5’-TGGTGGTAAGGACAGGCTTGAGTC-3’ ) . Amplification reactions were carried out under the following conditions: 10 mM Tris-HCl ( pH 8 . 8 ) , 50 mM KCL , 1 . 5 mM MgCl2 , 0 . 2 mM each dNTP , 0 . 2 µM primers , 0 . 06 ng/µl template , 0 . 04 U/µl Taq Polymerase . Genomic DNA ( 0 . 15 ng ) was used in a 25 µl final volume PCR , and then 15 µl of the PCR product was loaded on a 1% agarose gel . PCR conditions were as follows: initial denaturation at 95°C for 3 min was followed by 35 cycles of denaturation at 95°C for 30 s , annealing at 64°C for 30 s , and extension at 72°C for 3 min , with final extension at 72°C for 7 min . Temperature cycling was achieved with a DNA thermal cycler ( S1000 Thermal Cycler; Bio-Rad , Hercules , CA ) . For real-time qRT-PCR ( Figure 2—figure supplement 1C ) , total RNA was isolated from mouse hippocampi using the TRIzol reagent ( Life Technologies , Carlsbad , CA ) and reversed transcribed into cDNA using Superscript III ( Life Technologies , Carlsbad , CA ) , according to manufacturer’s instructions . The qPCR was performed using iQ SYBR Green Supermix Universal ( Bio-Rad , Hercules , CA ) . Ten nanogram of cDNA was used in a 20 µl final volume PCR . Amplification was for 40 cycles with 94°C for 30 s , 60°C for 30 s , and 72°C for 30 s as cycle parameters , with a final step of 72°C for 10 min . For amplification of PKMζ cDNAs , specific primers were: forward , 5’-GGCTGCAAGACTTCGACCTCATC-3’ and reverse , 5’-CTGGACGCCTGCTCAAACACATGT-3’ . Melting curve analysis was performed to confirm the specificity of PCR reactions . Relative expression of each gene was analyzed by ΔΔCT method . Data were normalized to a housekeeping gene , glyceraldehyde-3-phosphate dehydrogenase , using forward primer , 5’-TTGTGATGGGTGTGAACCACGAGA-3’ , and reverse primer , 5’-GAGCCCTTCCACAATGCCAAAGTT-3’ . Ten microliter of the qPCR product was loaded on a 2% agarose gel to test for PKMζ mRNA expression . For hippocampal slice experiments , methods were adapted from those previously described ( Hernandez et al . , 2003 ) . Briefly , slices removed from the recording chamber ( see below ) were immediately frozen on a glass slide on dry ice , or placed in appropriate volumes of RNAlater solution ( Ambion , ThermoFisher Scientific , Waltham , MA ) . The CA1 region was excised in a cold room ( 4°C ) and homogenized in 10 μl of ice-cold modified RIPA lysis buffer , consisting of the following ( in mM , unless indicated otherwise ) : 25 Tris-HCl ( pH 7 . 4 ) , 150 NaCl , 6 MgCl2 , 2 EDTA , 1 . 25% NP-40 , 0 . 125% SDS , 0 . 625% Na deoxycholate , 4 p-nitrophenyl phosphate , 25 Na fluoride , 2 Na pyrophosphate , 20 dithiothreitol ( DTT ) , 10 β-glycerophosphate , 1 μM okadaic acid , phosphatase inhibitor cocktail I & II ( 2% and 1% , respectively , Calbiochem ) , 1 phenylmethylsulfonyl fluoride , 20 μg/ml leupeptin , and 4 μg/ml aprotinin . For analysis of dorsal hippocampi and hemibrains , the tissue was dissected , snap-frozen , and stored at −80°C until lysis . Dorsal hippocampi were homogenized in 100 μl modified ice-cold RIPA buffer . Appropriate volumes of 4X NuPage LDS Sample Buffer ( Invitrogen , Carlsbad , CA ) and β-mercaptoethanol were added to the homogenates , and samples were boiled for 5 min followed by SDS-PAGE . Following transfer at 4°C , nitrocellulose membranes ( 0 . 2 μm pore size ) were blocked for at least 30 min at room temperature with LI-COR Odyssey Blocking Buffer ( LI-COR , Lincoln , NE ) , then probed overnight at 4°C using primary antibodies dissolved in LI-COR Odyssey Blocking Buffer with 0 . 1% Tween 20 and 0 . 01% SDS . After washing in phosphate-buffered saline ( PBS ) with 0 . 1% Tween 20 ( PBS-T; 3 washes , 5 min each ) , the membranes were incubated with IRDye ( LI-COR ) secondary antibodies . Proteins were visualized by the LI-COR Odyssey System . Densitometric analysis of the bands was performed using NIH ImageJ , and values were normalized to actin . For LTP experiments , acute mouse hippocampal slices ( 450 µm ) were prepared as previously described ( Serrano et al . , 2005 ) . Hippocampi were dissected , bathed in ice-cold dissection buffer , and sliced with a McIlwain tissue slicer in a cold room ( 4˚C ) . The dissection buffer contained ( in mM ) : 125 NaCl , 2 . 5 KCl , 1 . 25 NaH2PO4 , 26 NaHCO3 , 11 glucose , 10 MgCl2 , and 0 . 5 CaCl2 , and was bubbled with 95% O2/5% CO2 to maintain the pH at 7 . 4 . The slices were immediately transferred into an interface recording chamber ( 31 . 5 ± 1°C ) ( Serrano et al . , 2005 ) . The recording superfusate consisted of ( in mM ) : 118 NaCl , 3 . 5 KCl , 2 . 5 CaCl2 , 1 . 3 MgSO4 , 1 . 25 NaH2PO4 , 24 NaHCO3 , and 15 glucose , bubbled with 95% O2/5% CO2 , with a flow rate of 0 . 5 ml/min . In oligodeoxynucleotide experiments , the bath level was increased to fully submerge the slices , and the superfusate containing the oligodeoxynucleotide was recirculated ( 5 ml at 5 ml/min for 30 min ) , using a custom-made recirculation system employing piezoelectric pumps ( Bartels Mikrotechnik GmbH , Dortmund , Germany ) . Thereafter , the bath containing the oligodeoxynucleotide was lowered again to interface level , and the flow rate was returned to 0 . 5 ml/min for the remainder of the experiment . Field EPSPs were recorded with a glass extracellular recording electrode ( 2–5 MΩ ) placed in the CA1 stratum radiatum , and concentric bipolar stimulating electrodes were placed on either side within CA3 or CA1 . Hippocampal slices were excluded from study if initial analysis showed fEPSP spike threshold was <2 mV . Pathway independence was confirmed by the absence of paired-pulse facilitation between the two pathways . The high-frequency stimulation consisted of standard two 100 Hz-1 s tetanic trains , at 25% of spike threshold , spaced 20 s apart , which is optimized to produce a relatively rapid onset of protein synthesis-dependent late-LTP ( Tsokas et al . , 2005 ) . The maximum slope of the rise of the fEPSP is analyzed on a PC using the WinLTP data acquisition program ( Anderson and Collingridge , 2007 ) . For postsynaptic dialysis of PKMζ and PKCι/λ and activation of conventional/novel PKCs by bath applications of phorbol esters , hippocampal slices ( 400 µm ) were prepared from 19- to 30-day-old Sprague-Dawley rats , using a Vibratome tissue sectioner , as previously described ( Ling et al . , 2002 ) . The slices were placed in an incubation chamber at 31-33°C in oxygenated ( 95% O2 , 5% CO2 ) physiological saline consisting of ( in mM ) : 124 NaCl , 5 KCl , 26 NaHCO3 , 1 . 6 MgCl2 , 4 CaCl2 , 10 glucose for a minimum of 1 . 5 hr . Single slices were then transferred to a recording chamber ( 1 . 5 ml ) placed on the stage of an upright microscope ( Zeiss Axioskop 2; Carl Zeiss , Oberkochen , Germany ) and perfused with warm ( 31–33°C ) saline at ~4 . 5 ml/min . The recording pipettes had tip resistance of 2–4 MΩ and contained ( in mM ) : 130 Cs-MeSO4 , 10 NaCl , 2 EGTA , 10 HEPES , 1 CaCl2 , 2 Na-ATP , 0 . 5 Na-GTP . Purified PKMζ ( final concentration in the pipette , 7–20 nM , 0 . 5–0 . 9 pmol·min-1·μl-1 phosphotransferase activity [Ling et al . , 2002] ) or PKCι/λ ( final concentration , 7 . 4 ng/ml , 0 . 8 pmol·min-1·μl-1 [ProQinase GmbH , Breisgau , Germany] ) was added to the pipette solution prior to whole-cell patch . Whole-cell recordings were obtained from visualized CA1 pyramidal cells , and synaptic events were evoked by extracellular stimulation ( pulse width 0 . 1 ms ) every 15 s with bipolar electrodes placed in stratum radiatum . The cells were held at –75 mV , and EPSC was recorded under the voltage-clamp mode with a Warner Instruments PC-501A amplifier ( Hamden , CT ) and filtered at 2 kHz ( -3 dB , four-pole Bessel ) . Brief voltage steps ( −5 mV , 5 ms ) from holding potential were applied during the course of recording to monitor cell access resistance , input resistance , and capacitance . Only recordings with an initial input resistance of >100 MΩ and an initial access resistance of <10 MΩ with insignificant change ( <20% ) during the course of recordings were accepted for study . Signals were digitized with Digidata 1322A and acquired and analyzed with pClamp software ( Molecular Devices , Sunnyvale , CA ) running on a PC . The peak amplitude of EPSCs was further analyzed with Excel ( Microsoft , Redmond , WA ) . The means ± SEMs of 1 min bins of responses were plotted in the figures . The sequences of the single-stranded oligodeoxynucleotides were: PKMζ-antisense , ctcTTGGGAAGGCAtgaC; scrambled , aacAATGGGTCGTCtcgG , in which the lower case bases signifies phosphorothioate linkage 5'-3' . The PKMζ-antisense sequence is complementary to the translation start site in the PKMζ mRNA and shows no significant homology to any other sequence in the GenBank database , except PKCζ mRNA . Scrambled oligodeoxynucleotide also does not match any known sequence . Both oligodeoxynucleotides are phosphorothioated on the three terminal bases at each end to protect against nuclease degradation and were reverse phase cartridge-purified ( Gene Link , Hawthorne , NY ) ( Garcia-Osta et al . , 2006 ) . For LTP experiments , oligodeoxynucleotides ( 20 μM ) were applied to the bath after preparation of slices . A custom-made recirculation submersion system with piezoelectric pumps ( Bartels Mikrotechnik GmbH , Dortmund , Germany ) was used . The slices were perfused with a recirculating volume of 5 ml superfusate containing antisense- or scrambled-oligodeoxynucleotide for 1 . 5 hr before tetanization and for the duration of the experiment thereafter ( 30 min post-tetanus for immunoblots , Figure 2B; 3 hr for pharmacogenetic analysis , Figure 2C ) . For spatial long-term memory experiments , we adapted the approach used in Garcia-Osta et al . ( Garcia-Osta et al . , 2006 ) . In preparation for stereotaxic surgery to implant the injection cannula hardware , the mice were anesthetized by a mixture of dexmedetomidine ( 5 mg/kg i . p . ) and ketamine ( 28 mg/kg i . p . ) . The animals were mounted in a Kopf stereotaxic frame ( Tujunga , CA ) to implant a pair of guide cannulae with the tip above the injection target in the dorsal hippocampus ( AP −1 . 94 mm; L ±1 . 00 mm; DV −0 . 90 mm ) . The injection hardware was manufactured by Plastics One , Roanoke , VA ( Part Numbers: C235GS-5-2 . 0 , C235DCs-5 , 303DC/1 , C235IS-5; guide cannula , cannula dummy , cannula cap , injection needle , respectively ) . Antisedan ( 0 . 65 mg/kg i . p . ) was administered to reverse the sedation at the end of surgery . A week after surgery , the animals received active place avoidance training . Before testing the effect of the antisense injection on place avoidance , the animals received a bilateral injection of saline ( 1 µl/side ) and were left in the home cage to habituate to the procedure . The day after the initial pretraining exposure to the place avoidance apparatus , injections were 1 nmol oligodeoxynucleotide in 0 . 5 µl PBS/side , 20 min before each training session . The animals were restrained , the cannula cap and dummy removed , and the injection needle inserted into the guide cannula so that it protruded from the end of the guide by 0 . 5 mm . The other end of the needle was connected to a 1 µl Hamilton syringe via Tygon tubing . The oligodeoxynucleotide solution was infused for 1 min , and after the infusion the needle was left in place for 5 min before removal . The animals were returned to their home cage to recover from any acute effects of the injection and to allow diffusion of the antisense before training began . The data from 1 animal were excluded from behavioral analysis of the effects of antisense because histology revealed misplaced cannulae . The biotinylated PKMζ-antisense ( Gene Link , Hawthorne , NY ) was labeled by a 5'-biotin modification with a C6 spacer . To be equivalent to the training , three biotinylated PKMζ-antisense injections with 2 hr intervals ( 1 nmol in 0 . 5 PBS μl/side ) were given , and the brain fixed with 4% paraformaldehyde in PBS 50 min later . The 40 μm coronal sections were stained by immunocytochemistry using mouse anti-biotin antiserum followed by Cy3-conjugated secondary antiserum ( Jackson ImmunoResearch , West Grove , PA ) , counterstained with DAPI , and examined by confocal microscopy . PKMζ was recombinantly expressed and purified as previously described ( Ling et al . , 2002 ) . PKCι/λ was purchased from ProQinase GmbH ( Freiburg , Germany ) . The reaction mixture ( 50 μl final volume ) contained: 50 mM Tris-HCl ( pH 7 . 4 ) , 10 mM MgCl2 , 1 mM DTT , 25 μM ε-peptide substrate ( ERMRPRKRQGSVRRRV , AnaSpec , Freemont , CA ) , in the presence or absence of phosphatidylserine ( 5 μg/ml , Avanti Polar Lipids , Alabaster , AL ) , and PKCι/λ ( 184 ng , 0 . 2 pmol·min-1/assay ) or PKMζ ( 4 ng , 0 . 2 pmol·min-1/assay ) , in the presence or absence of ZIP or chelerythrine at concentrations given in the figures . The reaction , initiated with the addition of 50 μM ATP ( final concentration , ~1–3 μCi [γ-32P]/assay ) , was for 30 min at 30ºC , which is in the linear range for enzyme concentration ( data not shown ) . The reaction in which the substrate LANCE Ultra ULight-PKC substrate TRF0108-D ( 50 nM , PerkinElmer , Waltham , MA , Figure 1—figure supplement 2 ) was substituted for ε-peptide substrate was also in the linear range for enzyme concentration ( data not shown ) . The reaction was stopped by addition of 25 μl of 100 mM cold ATP and 100 mM EDTA , and 40 μl of the reaction mixture was spotted onto phosphocellulose paper and counted by liquid scintillation . Activity was measured as the difference between counts incorporated in the presence and absence of enzyme . Autonomous kinase activity is defined as activity in the absence of phosphatidylserine . To assay the effect of phosphoinositide-dependent kinase-1 ( PDK1 ) phosphorylation of PKCι/λ on PKCι/λ autonomous kinase activity , PDK1 ( 100 ng , 8 pmol·min-1·mg-1; ProQinase ) , either active or denatured by heating to 100°C , was added to the reaction mixture . The autonomous kinase activity of PKCι/λ was then measured as described above . PKCι/λ activity in the presence of active PDK1 was normalized to that in the presence of denatured PDK1 . A commercial computer-controlled active place avoidance system was used ( Bio-Signal Group , Acton , MA ) . The position of the mouse on a 40 cm diameter circular arena rotating at 1 rpm was determined 30 times per second by video tracking from an overhead camera ( Tracker , Bio-Signal Group ) . All experiments used the 'Room+Arena-' task variant that challenges the mouse on the rotating arena to avoid a shock zone that was a stationary 60° sector ( Pastalkova et al . , 2006 ) . A constant current foot-shock ( 60 Hz , 500 ms ) was delivered after entering the shock zone for 500 ms and was repeated each 1500 ms until the mouse left the shock zone . The arena rotation periodically transported the animal into the shock zone , forcing it to actively avoid the location of shock . The shock amplitude was 0 . 2 or 0 . 3 mA , which was determined for each animal in the first session to be the minimum that elicited flinch or escape responses . A clear wall made from Polyethylene Terephtalate Glycol-modified ( PET-G ) prevented the animal from jumping off the elevated arena surface . A 5-pole shock grid was placed on the rotating arena , the centroid of the mouse was tracked by the video tracker , and the shock was scrambled across the 5-poles when the mouse entered the shock zone . Every 33 ms , the software determined the mouse’s position , whether it was in the shock zone , and whether to deliver shock . The time series of the tracked positions was analyzed offline ( TrackAnalysis , Bio-Signal Group ) to extract a number of end point measures . The time to first enter the shock zone estimates ability to avoid shock and was taken as an index of between-session memory . A pretraining habituation period on the apparatus equivalent in time to a training session , but without shock , was provided . The training schedule for the pharmacogenetic analysis ( Figures 4 , 5 ) was as follows . The animals received three 30 min training trials , with an intertrial interval of 2 hr . Antisense or scrambled oligodeoxynucleotide ( Figure 4 ) was injected 20 min before each training trial , as described above . Retention testing was a 30 min trial without shock on the next day . ICAP ( Figure 5 ) was injected 2 hr before retention testing . The mouse trajectories depict the locations that were visited during the first 10 min , the time frame during which mice that learn the avoidance tend only rarely to enter the shock zone . The extended training schedule ( Figure 6 ) was as follows . Mice were trained across a 5-day period . The animals received three 10 min training trials per day , with an intertrial interval of 1 hr . On the first day of training , animals received an additional 10 min pretraining habituation session . Retention testing was a 10 min trial without shock on the 6th day . Unreinforced hippocampus-dependent long-term memory was tested using adaptations of the object/context mismatch , the object/place mismatch , and the novel object location tests of novelty-preference ( Mumby et al . , 1999 ) . The behavioral task , protocol , and analysis are described in more detail at Bio-protocol ( Lesburguères et al . , 2017 ) . Three open plastic boxes ( 42 x 42 x 20 cm ) were placed at the center of the experimental room . The visual appearance of each box was unique and customized with different patterns on three of the four walls to make distinct spatial contexts , denoted A , B , and C ( green , yellow , blue , respectively in Figure 7 ) . The fourth wall was transparent and faced south to provide orientation cues . Unique objects ( toys , flasks , jars ) were placed 5 cm away from the walls and were fixed to the floor . The apparatus , boxes , and objects were cleaned with 70% ethanol between subjects to eliminate odor cues . A video tracking system ( Tracker , Bio-Signal Group , Acton , MA ) monitored and recorded the exploratory activity of the animals for offline analysis . Each day , the mice were placed in the experimental room 30 min before beginning the behavioral experiments . Each mouse was trained and tested in the object/context mismatch and the object/place mismatch tasks first , followed by the object-location task . Each task had three phases: pretraining , training , and the retention test . Pretraining ( Day 1 ) : the animals were allowed to explore each environment for 10 min with no objects present . Training ( Days 2–4 ) : the mice were allowed to explore a pair of environments , Contexts A and B , each during two 5 min trials/day , separated by a 1 hr intertrial interval . This allowed the mice to learn the spatial arrangement of the four objects in each box . Retention testing ( Days 5 and 6 ) : retention of spatial memory was evaluated on subsequent days by the object/context mismatch and the object/place mismatch tests . In the object/context mismatch test , two of the four objects from one context replace two of the objects in the other context , whereas in the object/place mismatch test , the positions of two objects were exchanged within one context . Each mouse was allowed to explore the altered environment for 3 min , and the time spent exploring each object was recorded . On day 5 , an hour after the retention test , the mice received additional training ( 5 min exploration in contexts A and B with the original arrangement of objects ) . This additional training was intended to avoid extinction that could be induced by the retention test . The second retention test was on day 6 , to test the mismatch task that was not tested on day 5 . The order of the retention tests on days 5 and 6 , the order of exposure to the boxes during training , the objects and the places were all counterbalanced between the animals and groups . Novel location test pretraining ( Day 7 ) : the mice explored the third environment , Context C , for 10 min . Novel location test training ( Day 8 ) : two objects were placed in this environment , and the mice explored for 5 min . During three 5 min training sessions , with 1 hr intertrial intervals , the mice could learn the locations of the pair of objects . Novel location retention test ( Day 9 ) : one object was relocated to a novel unoccupied place , and the time spent exploring each object was recorded . The relocated object , the location , and the specific environment were counterbalanced between animals and genotypes . Offline analysis of the video identified exploration when the mouse’s nose was <2 cm away and oriented toward the object . Memory performance was quantified using a discrimination index calculated as the absolute difference in time spent exploring the changed ( i . e . incorrect , misplaced , or relocated ) and the unchanged objects divided by the total time spent exploring all the objects . Good memory retention corresponds to a positive discrimination index , which reflects that the animal spent more time exploring the incorrect ( object/context mismatch ) , displaced ( object/place mismatch ) , or relocated ( object-location ) objects compared to the objects that were not changed . Sample sizes vary for the different experimental approaches ( biochemistry , in vitro intracellular current and extracellular field potential physiology , and behavior ) . The PKMζ is unnecessary and the PKMζ is compensated hypotheses predict all-or-none effects in the experiments , and this provided a basis for sample size estimates . Power analyses were performed using GPower Version 3 . 0 . 6 with α = 0 . 05 and β = 0 . 8 and large effect sizes of 1 . 5–2 . 0 . The effect size estimates were based on prior studies that demonstrated essentially all-or-none effects of PKMζ inhibition on the biochemical , physiological , and behavioral assays used here ( Sacktor et al . , 1993; Osten et al . , 1996; Ling et al . , 2002; Kelly et al . , 2007; Pastalkova et al . , 2006 ) . Two-population Student’s t tests were performed to compare protein levels in the PKMζ-null and wild-type mice . For LTP experiments the responses to test stimuli were averaged across 5 min for statistical comparisons . Paired Student’s t tests were used to compare the change in the potentiated response at time points at the beginning and end of drug application . Multi-factor comparisons were performed using ANOVA with repeated measures , as appropriate . The degrees of freedom for the critical t values of the t tests and the F values of the ANOVAs are reported as subscripts . Post-hoc multiple comparisons were performed by Newman-Keuls tests as appropriate . Statistical significance was accepted at p<0 . 05 . Effect sizes for binary comparisons and one-way ANOVAs are reported as Cohen's d and as η2 for two-way ANOVA effects .	How are long-term memories stored in the brain ? The formation of memories is believed to depend on the strengthening of connections between neurons . During learning , neurons produce an enzyme called PKMzeta ( or PKMζ ) , which is thought to be responsible for maintaining the newly strengthened connections . Inhibitors of PKMzeta , such as a drug called ZIP , disrupt long-term memories . This suggests that the brain may be like a computer hard disc in that its stored information — its memories — could be erased . However , recent experiments on genetically engineered mice have thrown the role of PKMzeta into question . Knockout mice that lack the gene for PKMzeta can still strengthen connections between neurons and can still learn and remember . Moreover , ZIP still works to reverse the strengthening and to erase long-term memories . This indicates that ZIP can act on something other than the PKMzeta enzyme . These results have led many neuroscientists to doubt that PKMzeta has anything to do with memory . Yet there are two possible explanations for the normal memory in PKMzeta knockout mice . First , PKMzeta is not required for memory , so getting rid of it has no effect . Second , PKMzeta is essential for long-term memory in normal mice . However , knockout mice recruit a back-up mechanism for long-term memory storage , which is also sensitive to the effects of ZIP . To test these possibilities , Tsokas et al . used a modified piece of DNA that prevents neurons with the gene for PKMzeta from producing the enzyme . The DNA blocked memory formation in normal mice , consistent with a role for PKMzeta in memory . However , it had no effect in knockout mice — the DNA had nothing to work on . This suggests that another molecule does indeed act as a back-up for PKMzeta in these animals . Further experiments revealed that an enzyme closely related to PKMzeta , called PKCiota/lambda ( PKCι/λ ) , substitutes for PKMzeta during memory storage in the knockout mice . These findings restore PKMzeta to its early promise . They show that PKMzeta is crucial for long-term memory in normal mice , but that something as important as memory storage has a back-up mechanism should PKMzeta fail . Future work may reveal when and how this back-up becomes engaged .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Compensation for PKMζ in long-term potentiation and spatial long-term memory in mutant mice
Humans can run without falling down , usually despite uneven terrain or occasional pushes . Even without such external perturbations , intrinsic sources like sensorimotor noise perturb the running motion incessantly , making each step variable . Here , using simple and generalizable models , we show that even such small step-to-step variability contains considerable information about strategies used to run stably . Deviations in the center of mass motion predict the corrective strategies during the next stance , well in advance of foot touchdown . Horizontal motion is stabilized by total leg impulse modulations , whereas the vertical motion is stabilized by differentially modulating the impulse within stance . We implement these human-derived control strategies on a simple computational biped , showing that it runs stably for hundreds of steps despite incessant noise-like perturbations or larger discrete perturbations . This running controller derived from natural variability echoes behaviors observed in previous animal and robot studies . Human running is often modeled as being periodic ( Blickhan and Full , 1993; Seyfarth et al . , 2002; Srinivasan and Holmes , 2008 ) . But running is not exactly periodic , even on a treadmill at constant speed . Body motion during running varies from step to step ( Cavanagh et al . , 1977; Belli et al . , 1995; Jordan et al . , 2007; Jordan and Newell , 2008 ) . This step-to-step variability could be due to internal perturbative sources like muscle force noise and sensory noise ( Warren et al . , 1986; Harris and Wolpert , 1998; Osborne et al . , 2005 ) or small external perturbations ( e . g . visual field inhomogeneity , small ground imperfections ) . To run without falling , the body’s ‘running controller’ must prevent the effects of these small perturbations from growing too large . Here , we provide an experimentally derived low-dimensional characterization of this control that reveals how humans run without falling down . One classic modeling paradigm for running control assumes that the human leg behaves like a linear spring ( Blickhan , 1989; McMahon and Cheng , 1990; Blickhan and Full , 1993 ) . This paradigm has been used to argue how passive-elastic properties may reduce muscle work needed for locomotion ( Alexander and Vernon , 1975; Alexander , 1990 ) and has been useful in examining locomotion in a simplified setting . Variants of these spring-mass running models have demonstrated stable running ( Seyfarth et al . , 2002; Seipel and Holmes , 2005; Ghigliazza et al . , 2005; Geyer et al . , 2006; Srinivasan and Holmes , 2008; Englsberger et al . , 2016 ) . These models have been successful in fitting the average center of mass motion during running ( Blickhan and Full , 1993; Geyer et al . , 2006; Srinivasan and Holmes , 2008 ) . However , understanding running stability requires understanding how deviations from the average motion are controlled . It has been previously recognized that spring-like leg mechanics cannot explain how deviations from the average motion are controlled and eventually attenuated ( e . g . Ghigliazza et al . , 2005; Biewener and Daley , 2007; Maus et al . , 2015 ) . Here , we examine the role of active muscle control in running stability , using more general models of human locomotion rooted in Newtonian mechanics ( Srinivasan , 2011 ) . One way of characterizing the running controller is to apply perturbations ( for instance , pushes or pulls or sudden changes in terrain ) and examine how the body recovers from the perturbations ( Van Woensel and Cavanagh , 1992; Daley and Biewener , 2006; Qiao and Jindrich , 2014; Riddick and Kuo , 2016 ) . Instead of such external perturbations , here , we use the naturally occurring step-to-step variability ( Hurmuzlu and Basdogan , 1994; Maus et al . , 2015 ) to characterize the controller . Previous attempts at examining such variability for controller information focused only on walking ( Hurmuzlu and Basdogan , 1994; Wang and Srinivasan , 2012; Wang , 2013; Wang and Srinivasan , 2014 ) or considered variants of the spring-mass model ( Maus et al . , 2015 ) . Here , we directly characterize the control in terms of how humans modulate their leg force magnitude and direction during running . The only way to control the center of mass motion is for the leg to systematically change the forces and the impulses it applies on the ground . We uncover how such center of mass control is achieved . We then implement this human-derived controller on a simple mathematical model of a biped ( Srinivasan , 2011 ) , showing that this biped model runs without falling down , despite incessant noise-like perturbations , large external perturbations , and on uneven terrain . A human-derived controller such as the one proposed here could inform monitoring devices to quantify running stability or fall likelihood ( O'Loughlin et al . , 1993 ) , or could help understand running movement disorders . Further , implementing such controllers into robotic prostheses and exoskeletons ( Dollar and Herr , 2008; Shultz et al . , 2015 ) will allow the human body to interact more ‘naturally’ with the device , rather than having to compensate for an unnatural controller . Some running robots have demonstrated stable running , using a variety of control schemes ( Raibert , 1986; Chevallereau et al . , 2005; Tajima et al . , 2009; Nelson et al . , 2019 ) . But these robots fall short of human performance and versatility . Understanding human running may lead to better running robots . During flight phase , the body center of mass moves in a nearly parabolic trajectory and the runner has no control over this parabolic motion ( as the aerodynamic forces generated by the person are negligible , unlike birds ) . From Newton’s second law , it follows that the only way to control the center of mass motion is to modulate the total ground reaction force components during stance phase , when the leg is in contact with the ground . However , there are infinitely many ways to modulate the ground reaction forces to control the center of mass motion . Here , we examine how the ground reaction forces are modulated in response to center of mass state deviations during the previous flight apex ( Figure 1 ) . A flight apex is defined as when the center of mass height z is maximum . The center of mass position and velocity at flight apex are denoted by ( xa , ya , za ) and ( x˙a , y˙a , z˙a ) , respectively . Because the vertical velocity at flight apex z˙a=0 by definition , z˙a is not considered as an explanatory variable . The absolute horizontal position ( xa , ya ) on the treadmill changes with a much slower time-scale than other variables . Therefore , our default set of explanatory variables is ( x˙a , y˙a , za ) . We will include the horizontal position ( xa , ya ) when we comment later on ‘station keeping’ . The step-to-step variability in the center of mass state at flight apex over hundreds of steps is shown in Figure 2a . To be stable , the runner needs to prevent this motion variability from growing without bound . As noted , the only way to control this motion is by using the ground reaction forces ( GRFs ) . Consequently , the ground reaction force components over the stance phase are also variable ( Figure 2b ) . The net effect of the ground reaction forces on the center of mass velocity over a stance phase is captured by the force impulse , namely , the integral of the force . The variability in the sideways and fore-aft ground reaction impulses over a step ( Figure 2c ) are well-predicted by the variability in the center of mass state ( x˙a , y˙a , za ) at the previous flight apex ( Figure 3 ) . Moreover , the sideways impulse depends primarily on the sideways velocity x˙a and the fore-aft impulse depends primarily on the fore-aft velocity y˙a . Thus , it appears that the control in the fore-aft and sideways directions are independent or decoupled . Pooled over all subjects , the best-fit linear model for the sideways impulse Px is: ( 1 ) Left stance: ΔPx=−1 . 03Δx˙a , with R2=0 . 55 , andRight stance: ΔPx=−1 . 07Δx˙a , with R2=0 . 53 , and that for the fore-aft GRF impulse Py is: ( 2 ) Left stance: ΔPy=−0 . 72Δy˙a , with R2=0 . 32 , and Right stance: ΔPy=−0 . 72Δy˙a , with R2=0 . 33 , as in Figure 3 . All coefficients in Equations ( 1 ) and ( 2 ) are significant at p<10−4 . Both sideways and fore-aft impulses depend negligibly on vertical position deviations , so that including za in the regression increases the R2 values by less than 0 . 02 . The linear models for the fore-aft and sideways impulses in Equations ( 1 ) and ( 2 ) have a simple interpretation . The Δ⁢x˙a coefficient of about -1 in Equation ( 1 ) ( that is , Δ⁢Px≈-Δ⁢x˙a ) implies that sideways velocity deviations are completely corrected in one step , on average ( over all steps and all subjects ) . This correction could have been done over many steps , as would be the case if the coefficient were -0 . 5 , say . But humans seem to exhibit a ‘one-step dead-beat controller’ on average for sideways velocity deviations ( the term deadbeat control refers to when state deviations decay to zero in a finite amount of time ) . Of course , this single-step correction is not perfect . An R2 value of about 0 . 55 suggests that the system over-corrects or under-corrects deviations for any given step . Analogously , the coefficient of -0 . 72 in Equation ( 2 ) suggests that about 72% of a forward velocity deviation is corrected in a single step , on average . While this is not strictly ‘deadbeat control’ , it results in only ( 1-0 . 72 ) 2=0 . 08 of a perturbation remaining after two steps , and ( 1-0 . 72 ) 3=0 . 02 of a perturbation after three steps , indicating rapid control . We corroborate the above findings regarding perturbation decay with the ‘apex-to-apex maps’: that is , linear models that describe the relation between deviations in the state at one flight apex and those at the next flight apex . The right-to-left map from the state Sright=[x˙a , y˙a , za]right at an apex preceding a right stance to the state at the next flight apex ( preceding a left stance ) is , approximately: ( 3 ) [x˙ay˙aza]left=KR→L⋅[x˙ay˙aza] right whereKR→L=[−0 . 05∗−0 . 02∗+0 . 31∗−0 . 08∗+0 . 27−0 . 15∗+0 . 02∗+0 . 06+0 . 46] , where the superscript ∗ indicates that the coefficient is not significantly different from zero ( p>0 . 05 ) . The left-to-right matrix KL→R is similar to KR→L , except for the sign changes due to mirror-symmetry . The matrix product of KL→R and KR→L — Jacobians of the Poincare map ( Hurmuzlu and Basdogan , 1994; Guckenheimer and Holmes , 2013; Maus et al . , 2015 ) — quantify how apex state deviations grow or decay over one stride ( two steps ) . The eigenvalues of this matrix product were all less than one in absolute value , indicating a stable periodic motion . The largest eigenvalue was 0 . 14 , indicating that at most 14% of a perturbation remains after a stride on average in any direction . The low value of KR→L⁢ ( 1 , 1 ) , not significantly different from zero , suggests that a purely sideways velocity perturbation gets corrected essentially over one step on average , consistent with the sideways impulse control ( Equation 1 ) . Similarly , the value KR→L⁢ ( 2 , 2 ) =0 . 27 suggests that 73% of a forward velocity deviation is corrected in one step , consistent with the fore-aft impulse control ( Equation 2 ) . Finally , the ( 3 , 3 ) element of the step map ( Equation 3 ) suggests that less than 50% of a deviation in vertical position za remains after a step . See ( Maus et al . , 2015 ) for a detailed Floquet analysis of human running including more state variables , complementing the simplified version here . The control of vertical position is qualitatively different from that of control in the fore-aft and sideways directions , as we cannot use net vertical impulse for vertical position control due to the impulse-momentum considerations below . A flight apex occurs when the center of mass vertical velocity is zero . So , the net vertical impulse between two consecutive flight apexes is also zero ( as it equals the change in vertical momentum , according to the impulse-momentum equation ) . Therefore , changing the net vertical impulse over a stance phase will not accomplish any meaningful control in the vertical direction . However , we will show that by differentially modulating the vertical impulse within one stance phase , we can change the vertical position ( za ) from one flight apex to the next , without changing the net impulse . To show this most simply , consider infinitesimal flight phases and a stance phase from t=0 to t=Tstep . The total impulse Pz due to the vertical ground reaction force Fz⁢ ( t ) equals that due to gravity , which is given by , Pz=∫0TstepFz⁢ ( t ) ⁢𝑑t=∫0Tstepm⁢g⁢𝑑t=m⁢g⁢Tstep . For a triangular stance force ( Figure 4 ) with peak force Fpeak at tpeak , we get Fpeak=2⁢m⁢g . Then , by integrating the vertical acceleration ( Fz/m-g ) twice , the change in vertical position z⁢ ( Tstep ) -z⁢ ( 0 ) over a step is given by: ( 4 ) z⁢ ( Tstep ) -z⁢ ( 0 ) =g6⁢ ( ( Tstep-tpeak ) 2-tpeak2 ) . If the step was symmetric about mid-stance ( tpeak=Tstep/2 ) , there is no vertical position change over a step ( z⁢ ( Tstep ) =z⁢ ( 0 ) ) . The flight apex vertical position on the next step z⁢ ( Tstep ) can be changed by changing tpeak relative to Tstep/2 ( Figure 4 ) . For example , if z⁢ ( 0 ) at one flight phase was greater than its nominal value and the runner wishes to reduce it , this simple model predicts that the runner will decrease the first-half impulse and increase the second-half impulse; doing this is equivalent to delaying tpeak relative to Tstep/2 ( as in Figure 4 ) . This prediction is in agreement with the following experimentally-derived linear relations for the first half vertical impulse from t=0 to Tstep/2 , namely Δ⁢Pz\|0Tstep/2 , and the second half vertical impulse from t=Tstep/2 to Tstep , namely Δ⁢Pz\|Tstep/2Tstep: ( 5 ) Left stance: ΔPz\|0Tstep/2=−2 . 5Δza and ΔPz\|Tstep/2Tstep=+2 . 5Δza with R2=0 . 35 , and ( 6 ) Right stance: ΔPz\|0Tstep/2=−2 . 3Δza and ΔPz\|Tstep/2Tstep=+2 . 3Δza with R2=0 . 30 . We see that a positive Δza corresponds to a decrease in the first-half vertical impulse and an increase in the second half vertical impulse . In addition to the vertical impulse , the landing leg length is also modulated in response to vertical flight apex deviations . Regressing the leg length ℓ at the beginning of stance with the flight apex state , we found that this landing leg length is mostly a function of the vertical position at flight apex: ( 7 ) Δℓlanding=0 . 3Δza , with p<10−4 and R2=0 . 25 . Thus , a downward position deviation at flight apex would result in landing with a shorter leg length than nominal ( e . g . via flexed knee or ankle ) . A downward position deviation is analogous to a sudden step-up perturbation , so reducing the landing leg length reduces trip likelihood . The linear models above tell us how deviations from nominal motion at flight apex are corrected grossly over the next stance . But they do not tell us how the forces are modified continuously throughout a stance phase . The variability of the GRF components ( Fx , Fy , Fz ) depend on the ‘phase’ of the stride cycle , specifically , the time fraction ϕstance of stance ( Figure 2b ) . To explain this phase-dependent force variability within a single step , we compute the phase dependent sensitivity of ( Fx , Fy , Fz ) to the center of mass state as follows . For each output , say Fx , we divide the stance duration into 20 phases and compute a linear model for Fx at each of those phases , all with ( x˙a , y˙a , za ) as inputs . We refer to the coefficients in these linear models as a function of the phase ϕstance as the phase-dependent sensitivities of the GRFs ( Figure 5 ) to the corresponding predictor variable in ( x˙a , y˙a , za ) . The phase-dependent sensitivity of sideways GRF to x˙a shows that Fx is decreased over the whole step to correct a positive sideways velocity deviation at flight and that a majority of this correction occurs during the middle of stance ( Figure 5a ) . Similarly , in response to a positive fore-aft velocity perturbation , the fore-aft GRF is modulated so that there is a net negative force on the body over the next step ( Figure 5b ) . The sensitivity of the fore-aft force Fy is more in the first half of stance than during the second half of stance , being modulated more during the deceleration phase ( roughly ϕstance<0 . 5 ) than during the acceleration phase ( roughly ϕstance>0 . 5 ) . Placing the foot relative to the body allows a runner to modulate the leg force direction and thus the GRF components . The foot position ( xf , yf ) relative to center of mass position at the beginning of stance phase ( xs , ys ) is predicted by the previous flight apex state as described by the following equations . Specifically , sideways foot placement is described by the following equations: ( 8 ) Left stance: Δ ( xf−xs ) =0 . 95Δx˙a with R2=0 . 64 and Right stance: Δ ( xf−xs ) =1 . 00Δx˙a with R2=0 . 62 . The fore-aft foot placement is described by the following equations: ( 9 ) Left stance: Δ ( yf−ys ) =0 . 42Δy˙a−0 . 76Δzα with R2=0 . 45 and Right stance: Δ ( yf−ys ) =0 . 39Δy˙a−0 . 83Δza with R2=0 . 46 . That is , a sideways velocity perturbation to the body results in the foot being placed further along the direction of the perturbation . So , a rightward perturbation results in a more rightward step . Analogously , a forward velocity perturbation results in the foot being placed further forward relative to the body . As with the impulses , again , there is no significant coupling between sideways and fore-aft directions . Fore-aft foot placement modulation also depends on vertical position deviations , in a manner that the runner lands with a steeper leg when landing from a higher flight apex za . Such dependence of landing leg angle on vertical position is analogous to behavior in terrain-change experiments ( Daley and Biewener , 2006; Müller et al . , 2012; Qiao and Jindrich , 2012; Birn-Jeffery and Daley , 2012 ) , as discussed in detail later . We speculate that using foot placement based on center of mass state may be an efficient way to affect the center of mass motion , compared to , say , changing the leg force magnitudes and leg joint torques after the foot touches down ( Clark , 2018 ) . One possibility is that the foot placement deviations are achieved early on during the swing phase and this deviation is preserved during swing until the foot touchdown . However , this does not appear to be the case . Figure 6 shows the fraction of foot placement variance predicted by the swing foot state over the previous step . Less than 10% of the eventual foot placement is predicted by the swing foot at the beginning of flight phase ( Figure 6 ) . The explanatory power of the swing foot rises rapidly during the flight phase from less than 10% to a 100% when it becomes the next stance foot , suggesting that most swing foot re-positioning may happen during this flight phase . At the beginning of flight phase ( and earlier ) , the center of mass state is a vastly better predictor of the next foot placement than the swing foot itself ( Figure 6 ) . We can predict the foot placement using the center of mass state better than just the relative swing foot state until about 100 ms before foot touchdown . The explanatory power of the center of mass remains flat during flight . This flatness is likely because center of mass state follows a parabolic path during flight and thus accumulates no new variation . This lag between the explanatory power of the center of mass and the foot suggests that the error information in the center of mass state is yet to be fully incorporated into the swing foot re-positioning until the flight phase . During the brief flight phase , when the swing foot’s explanatory power increases , information from center of mass state is transferred to the foot , presumably via some mixture of feedback control and feedforward dynamics . As an alternative to control based on discrete monitoring of deviations at the previous flight apex state , we considered a ‘continuous control’ model . Specifically , we obtained linear models for the GRFs based on the current center of mass state during stance ( x˙ , y˙ , z ) . These linear models did not differ significantly in the fraction of GRF variance explained , compared to the apex-based control model ( p=0 . 94 ) . In the linear models above , adding the sideways and fore-aft apex body position ( xa , ya ) to the explanatory variables improves the R2 values by less than 1 . 5% . Thus , the runners did not prioritize controlling their position relative to the treadmill ( station-keeping ) . Further , the regression coefficients for ( x˙a , y˙a , za ) did not vary significantly across the three running speeds ( p>0 . 05 , paired t-test ) . The running control gains have approximate bilateral ( left-right ) symmetry . The gains that couple sideways direction variables and either fore-aft or vertical direction variables have mirror-symmetry ( see Equations 1 , 2 , 8 , 9 ) . That is , these gains for the left leg’s stance are the negatives of corresponding gains for the right leg’s stance . On the other hand , gains that couple one sideways variable with another sideways variable , or one fore-aft variable with another fore-aft variable , are the same for the left and right legs without any such sign changes . This mirror symmetry in running control likely follows from the approximate mirror symmetry in body physiology about the sagittal plane and was also found in walking ( Wang and Srinivasan , 2014; Ankaralı et al . , 2015 ) . This symmetry suggests the lack of a substantially dominant limb for running control , in contrast to the asymmetry and limb role differentiation that occurs in some other tasks ( Peters , 1988 ) . We now show that the experimentally derived control strategies described above are sufficient to control the running dynamics of a simple mathematical model of a biped . We consider a biped with point-mass upper body and massless telescoping legs capable of generating arbitrary force profiles ( unlike a spring ) . We considered two versions of this biped model ( Figure 7a ) , one with direct control of the leg force and another that produces leg forces via Hill-type muscles ( Figure 7b ) . See Materials and methods for how the nominal running motion and the feedback controllers are specified for the models . We find that the models’ ground reaction forces are similar to experimental data despite not explicitly matching the curves ( Figure 8a ) . Further , we find that the phase-dependent ground reaction force feedback gains for the models are qualitatively similar to the phase-dependent gains inferred from experiment ( Figure 8b ) , again , despite not explicitly fitting the shape of these phase-dependent gains . This shows that these simple models can not only capture the average motion during running , but also how the runner responds to deviations from the average motion . The simple models’ running motions are not stable without the controller: an arbitrarily small perturbation makes it diverge from the original running motion . With the foot placement and leg force controller turned on , the running motion is asymptotically stable . Figure 9a–d shows the model recovering from fore-aft , sideways , and vertical perturbations at flight apex . It is a mathematical theorem that a stable periodic motion that can reject perturbations at one phase ( say , flight apex ) can reject perturbations at any phase ( Guckenheimer and Holmes , 2013 ) . So , it follows that our model rejects perturbations at any phase . The inputs to the feedback controller ( x˙a , y˙a , za ) do not include the absolute sideways and fore-aft position ( xa , ya ) of the runner . Therefore , the controller does not correct position perturbations ( station-keeping ) . A sideways or fore-aft push to the model results in convergence to the nominal running motion , except for a sideways or fore-aft position offset ( Figure 9c ) . Figure 10 illustrates the leg work-loop for the unperturbed run ( net zero work ) and when positive perturbations are applied to sideways and fore-aft velocities , and vertical positions . All such positive perturbations result in net negative work on the first step after the perturbation , reflected in the work-loops with net negative area within them . Such net positive or negative leg work is clearly necessary to recover from perturbations that change the total mechanical energy of the runner , as was recognized in prior discussions of the energy-neutral spring-mass model of running ( Ghigliazza et al . , 2005; Biewener and Daley , 2007; Srinivasan and Holmes , 2008 ) . To simulate the step-to-step variability in real human running , we added ‘noise’ to our foot placement and leg forces ( for the direct force control model ) or muscle activations ( for the muscle-driven model ) and simulated the biped models for a few hundred steps . This noise is meant to model the phenomenon that intended muscle forces tend to deviate from actual muscle forces due to motor noise ( Harris and Wolpert , 1998 ) . We find that while the direct leg force control model falls down , the runner with muscles does not fall down for hundreds of steps despite the noise . The stable motion of the center of mass in the presence of noise-like perturbations is shown in Figure 11a . The variability in the center of mass state at flight apex for the model ( Figure 11b ) as a result of the simulated noisy control is qualitatively similar to the variability found in experiment ( Figure 2a ) . The model is also able to run without falling despite vertical position perturbations at flight apex , which are equivalent to uneven terrain . Thus , even though the model was derived using data on horizontal ground , it is capable of running robustly on uneven terrain . The muscle-driven model is robust to motor noise presumably because of the intrinsic stabilizing properties of force-length and force-velocity relations ( Hogan , 1984; Jindrich and Full , 2002 ) . We have mined the step-to-step variability in human running to show how humans modulate leg forces and foot placement to run stably . We then used these data-derived control strategies on a biped model , demonstrating robustness to discrete perturbations and persistent motor noise . We have shown that humans use foot placement or leg angle control in a manner that they step in the direction of the perturbation , thereby directing the leg force so as to oppose the perturbation . This result provides an empirical basis for ad hoc assumptions about leg angle control made in previous running models ( Seyfarth , 2003; Ghigliazza et al . , 2005; Peuker et al . , 2012 ) . The foot placement controller derived from running data is qualitatively similar to the classic Raibert-like controller used in early running robots ( Raibert , 1986 ) in that the foot placement opposes velocity deviations with no sideways-fore-aft coupling , but differs in that it has a dependence on vertical position perturbations . This makes such robotic controllers inadvertently biomimetic . Humans use similar foot placement control in walking , stepping in the direction of the perturbation ( Hof et al . , 2010; Wang and Srinivasan , 2014 ) . Previous work had shown that appropriate foot placement is used in running ostriches while turning ( Jindrich et al . , 2007 ) , running humans in cutting maneuvers ( Besier et al . , 2003 ) , and turning while walking ( Patla et al . , 1999 ) . Some past work on inferring stability from variability focused on kinematic measures of variability such as Floquet multipliers , finite-time Lyapunov exponents ( Dingwell et al . , 2001 ) and long term correlations in walking and running variability ( Hausdorff et al . , 1995; Jordan et al . , 2006; Kaipust et al . , 2012 ) . Such measures can provide discriminative diagnostic measures ( Kaipust et al . , 2012 ) , but do not attempt to provide a causal narrative about how locomotion is controlled . Our approach here , rooted in Newton-Euler mechanics , is able to discover potential causal strategies underlying locomotion stability , and by extension , could inform treatment of pathological unstable movements in addition to diagnosis . Other past studies have used variants of the principal component analysis ( Cappellini et al . , 2006; Maus et al . , 2015 ) to demonstrate that the intrinsic variability in human locomotion may reside in a lower dimensional manifold ( Cappellini et al . , 2006; Chang et al . , 2009; Yen et al . , 2009; Dingwell et al . , 2010; Maus et al . , 2015 ) . Here , by focusing on how the center of mass is controlled through forces , we have implicitly used a physics-based dimensionality reduction to examine the dominant control strategies . While our work relies on linear regressions from data , the basic physics relating the inputs and outputs in these models suggest a natural causal account . This causal account , based on simplifying modeling assumptions , ignores the effect of variables not considered here . Our goal here was to delineate the explanatory power of controller descriptions based on center of mass state . To identify the effect of perturbations of other possibly relevant state variables ( such as trunk attitude and angular velocity ) , we may need to either independently perturb these state variables or show that the natural variability in such variables is not significantly correlated with the center of mass state . The gain relating sideways foot placement and sideways velocity deviation was about 2 . 5 times greater than the gain relating fore-aft foot placement and fore-aft velocity deviation; a similar factor of 3 was found in walking ( Wang and Srinivasan , 2014 ) , perhaps reflecting the greater sideways instability of a biped without foot placement control ( Ghigliazza et al . , 2005 ) . Also consistent with lower control authority and a greater fall propensity in the sideways direction , we find that the recovery from a sideways perturbation is faster than from a fore-aft perturbation . While station-keeping was not prioritized over a single step , it may occur on a slower time-scale with a multi-step controller , not considered here . The results we have presented have been for data pooled over all subjects . Performing the regressions for data from individual subjects indicates that the dominant terms in the inferred controllers are similar for all subjects; the subject-to-subject variability in the estimated control gains are shown in Figure 12 . Figure 12 illustrates how the accuracy of an estimated control gain depends on the number of strides used for regression . For such linear regressions , the error estimate ( standard deviation ) is expected to decrease with Nstride like 1/Nstride2 , so that a factor of 10 decrease in error requires a 100-fold increase in sample size ( Wang and Srinivasan , 2012; Hamilton , 1994 ) . This dependence on Nstride may guide selection of sample sizes for future experimental designs . Our model predicts that when a runner starts at a higher-than-normal height at flight apex , or equivalently , encounters a step-down , the runner lands with a steeper leg angle ( Figure 9d ) . Such behavior has been observed in humans and bipedal running birds running with large unforeseen or visible step-downs ( Daley and Biewener , 2006; Grimmer et al . , 2008; Müller et al . , 2012; Qiao and Jindrich , 2012 ) . Conversely , step-ups decrease touch-down angle , as predicted ( Birn-Jeffery and Daley , 2012 ) . This behavior has been attributed to swing leg retraction just before foot contact ( Seyfarth , 2003 ) , but our foot placement controller captures this phenomenon despite not having explicit leg swing dynamics . While the terrain perturbations in the aforementioned experiments were large ( 5–20 cm ) , our model is based on data with tiny step-to-step deviations ( vertical position za s . d . 5 mm ) . This agreement indicates that humans may use qualitatively similar control strategies for large external perturbations and small intrinsic perturbations . Such foot placement control has also been used to control robots running on uneven terrain ( Hodgins and Raibert , 1991 ) . It is expected that any running controller that achieves asymptotic stability will need to perform net mechanical work in response to perturbations that decrease or increase the body’s mechanical energy ( Ghigliazza et al . , 2005; Srinivasan and Holmes , 2008; Maus et al . , 2015 ) . Our results are consistent with such expectation , as illustrated by the work-loops with net mechanical work in Figure 10 . Energy-conservative spring-like leg behavior does not allow such net mechanical work and can achieve only partial asymptotic stability , not being able to handle energy-changing perturbations ( as noted by ( Ghigliazza et al . , 2005 ) ) . Indeed , it is generally thought that even the spring-mass-like steady state center of mass motion in running is due to considerable muscle action and has been termed pseudo-elastic ( Ruina et al . , 2005 ) or pseudo-compliant ( McN . Alexander , 1997 ) . Remarkably , energy-optimal running movements in models with no leg springs produce similar spring-mass-like center of mass trajectories ( Srinivasan , 2011 ) , with leg muscles performing equal amounts of positive and negative work . A previous article ( Maus et al . , 2015 ) fit running data to variants of the spring-mass model , allowing the spring stiffness and spring length to change during stance , and showing that constant values for these parameters cannot fit running data . Here , we have used a simpler model to directly describe the control of stance leg force or activation ( Figure 8 ) . Such direct control of leg force or activation is perhaps more parsimonious than the simultaneous control of two variables , namely , spring stiffness and length . We have shown that humans modulate GRF continuously over the whole stance phase for control ( Figure 5 ) ; Maus and colleagues ( Maus et al . , 2015 ) assumed , for simplicity , an instantaneous finite energy input at mid-stance . The stabilizing responses we have characterized in this study are likely due to a mixture of feedforward dynamics and active neurally mediated feedback control . When we use the term "control" here , we implicitly refer to this mixture . It is hard to rigorously separate the roles of feedforward and feedback control without recording motor neuronal outputs and how these outputs interact with the properties of muscles . Nevertheless , we can determine the feasibility of feedback control by checking whether there is enough time for feedback control , given typical neuromuscular latencies . Our typical flight phase durations are greater than or about roughly equal to the typical short- or middle- latencies in reflex or feedback loops involving vestibular ( Fitzpatrick et al . , 1994; Iles et al . , 2007 ) or proprioceptive mechanisms ( Pearson and Collins , 1993; Sinkjær et al . , 1999 ) . This suggests feasibility of feedback based on flight phase or late stance phase information regarding center of mass state . While we have focused on the control of stance based on the previous flight apex , we have found that equivalent controllers based on the center of mass state at the end of previous stance have similar predictive ability ( Figure 6 ) , thus allowing more time for neural feedback . Specifically , the lag between the information in the center of mass state and the swing foot state regarding future foot placement is about 0 . 1 s for sideways placement and about twice that for fore-aft foot placement , suggesting sufficient time for neurally mediated feedback control of foot placement ( Figure 6 ) . Center of mass state or other body state information needed for feedback control could be estimated by the nervous system by integrating sensory signals from vision ( Patla , 1997 ) , proprioceptive sensors ( especially when the foot is on the ground ( Sainburg et al . , 1995 ) ) , and vestibular sensors ( Angelaki and Cullen , 2008 ) , potentially in combination with predictive internal models ( Wolpert et al . , 1995; Cullen , 2004 ) . In future work , repeating the calculations herein ( for instance , Figure 6 ) for experiments that systematically block or degrade ( say , by adding sensory noise ) one or more of these sensors may tell us the relative contributions of these sensors to running control . We speculate that most available relevant sensory information is used , perhaps analogous to an optimal state estimator ( Kuo , 2005; Srinivasan , 2009 ) , and degrading one sensor may result in sensory re-weighting on a slow time-scale ( Carver et al . , 2006; Assländer and Peterka , 2014 ) . Such experiments may also help explicitly distinguish the effects of sensory and motor noise , which we have implicitly combined here into a single residual term in the linear regressions . In this work , we have obtained a running controller with simplifying assumptions . Because humans have extended feet , non-point-mass upper bodies and legs with masses , the simple point-mass model may not capture all aspects of the running data ( Bullimore and Burn , 2006; Srinivasan and Holmes , 2008 ) . Further , we have made simplifying assumptions about muscle architecture , muscle properties ( linear force-velocity relations ) , and muscle activation , which are meant to capture the main qualitative dynamical features of muscles , rather than model them quantitatively precisely . For instance , we used a linear force-velocity relation , which may be sufficient to produce damping-like and stabilizing muscle behavior when activated , but this damping behavior may be accentuated by a more realistic nonlinear force-velocity relation . Future work will also involve obtaining controllers for more complex biped models and muscle models with different control architectures , which , for instance , might include feedback control based on not just the center of mass state , but the states of individual body segments . We have focused on linear relations between state deviations and control , as this is naturally suited for small deviations and perturbations that our data contains . In future work , we hope to examine the range of perturbation sizes for which this linear description is accurate by comparing this linear control to responses in experiments with larger perturbations , also inferring nonlinear descriptions should they improve predictive capability . We will also examine other control architectures , for instance , more explicitly incorporating state estimation and considering continuous control of motor outputs based on an estimated state , partly correcting for neural latencies using internal models . The methods used here are simple and non-invasive: they can be replicated to study running stability and control in other animals , or indeed , other approximately periodic tasks such as flapping flight and swimming . These methods are suitable for analyzing differences in different populations like athletes and non-athletes , the young and the elderly , and adults with and without movement disorders . Once such differences are well-characterized , this information could be used , say , in a rehabilitation setting to track progress from a controller in the presence of a movement disorder to a more healthy controller , and to design rehabilitation robots that assist in this progress . The protocols were approved by the Ohio State University Institution Review Board and subjects participated with informed consent . Eight subjects , three female and five male ( age 25 . 0 ±5 years , weight 66 . 8 ±7 kg , height 1 . 8 ±0 . 14 m , leg length 1 . 05 ±0 . 08 m , mean ± s . d . ) ran on a split-belt treadmill at three constant speeds: 2 . 5 , 2 . 7 and 2 . 9 m/s , presented in random order . Each speed had 2075 ± 67 strides across all subjects ( one stride = two steps ) with subjects running about 3 . 5 min on average . Subjects wore a loose safety harness that did not constrain their motion . Three-dimensional ground reaction forces and moments on each belt of the treadmill were recorded by separate six-axis load cells ( Bertec Inc , 1000 Hz ) . Body segment motion was measured using marker-based motion capture ( Vicon T20 , 100 Hz ) with four reflective markers on each foot and on the torso . We define flight apex as when the center of mass velocity reaches its peak height ( z˙=0 ) . The input to the running controller is drawn from the center of mass state at flight apex , namely position ( xa , ya , za ) and velocity ( x˙a , y˙a , z˙a ) . Unless otherwise specified , we use the flight apex state ( x˙a , y˙a , za ) as inputs in our linear models . The vertical velocity z˙a at flight apex is zero by definition and hence not included as an input . The center of mass velocities are obtained by integrating the center of mass accelerations , that is , the mass-normalized net force on the body: x¨=Fx/m , y¨=Fy/m , and z¨=Fz/m−g , where Fx , Fy and Fz are the measured ground reaction forces on the body . To obtain the integration constants , we assume that the mean velocity and acceleration over the whole trial are zero , because the person does not translate appreciably in the lab frame over a trial . To remove the slow integration drift in the center of mass velocity , we used a high-pass filter with a frequency cut-off equal to an eighth of mean step frequency ( Luinge and Veltink , 2005; Schepers et al . , 2009 ) . Changing this high-pass filter cut-off to a twentieth of the step frequency instead , or using a piecewise-linear de-trending over 20 steps , do not affect any of this article’s conclusions . This is because the stability-critical time-scales are much shorter . We ignored air drag here , because including it changed the velocities by less than 10-5 ms-1 , which is much smaller than the step-to-step variability . We use a weighted mean of four markers , roughly at the sacral level , as an approximation of the center of mass position ( Gard et al . , 2004; Wang and Srinivasan , 2014; Perry and Srinivasan , 2017 ) . We assume that the following variables are used to control the runner: GRFs , foot placement , and the landing leg length . Stance phases are identified as when the vertical GRF exceeds a threshold value to account for measurement noise ( Fz>30 N ) . The corresponding stance duration is Tstance . The GRF impulses ( Px , Py , Pz ) for each step are obtained by integrating the GRF components over the stance phase ( Px=∫0TstanceFx⁢𝑑t , etc ) . In addition to considering how GRF control occurs grossly over one step , we also consider GRF control throughout stance as a function of stance phase fraction ϕstance . Each stance phase is divided into n bins of duration Tstance/n . To approximate how the GRFs changes with the stance phase fraction ϕstance , we used the binned averages of the GRF in each of n=20 bins . We compute the mean values of the inputs over all steps in each trial and obtain deviations from these means ( Δ⁢x˙a , Δ⁢y˙a , Δ⁢za ) . Similarly , we compute the deviations from the means of the output variables Δ⁢F⁢ ( ϕstance ) , Δ⁢P , and Δ⁢ ( xf-xs , yf-ys ) . We use ordinary least squares regression to obtain linear models between the inputs and the outputs and report significant coefficients . Specifically , we have Δ⁢Output=J⋅Δ⁢Input , where the Jacobian matrix J represents the matrix of coefficients in the linear model . Each element of the matrix J quantifies the sensitivity of an output variable to small changes in a corresponding input variable , as inferred from the data and subject to the simplifying model assumptions . These sensitivity coefficients could be interpreted as partial derivatives , such as: ∂⁡Tstance/∂⁡x˙a , ∂⁡Fy⁢ ( ϕstance ) /∂⁡y˙a , and so on . Unless otherwise specified , the results presented are based on deviations of all subjects pooled together as one dataset , but we find that the models of individual subjects’ data are qualitatively similar ( as indicated in Figure 12 ) . The coefficients for the right leg and left leg are computed separately , to accommodate sign changes due to symmetry about the sagittal plane . In addition to the regressions described above using the flight apex state as the predictor , we used the center of mass state ( Δ⁢x˙a , Δ⁢y˙a , Δ⁢za ) at different phases over the previous step to predict each of the stance phase outputs . Specifically , for each stance phase output , we performed n=20 regressions , each using the center of mass state at one of the n=20 equally spaced gait phases over the previous step , where one full step is defined as starting and ending at a touch-down . This analysis allows us to investigate the predictive ability of the center of mass state at different phases . For these phase-dependent regressions , in addition to using the center of mass state as the predictor , we repeated the calculations using the swing foot state ( position and velocity relative to the center of mass ) , so as to compare the different predictive abilities as in Figure 6 . We consider two simple models of running , similar in spirit to previous models in terms of simplicity ( Blickhan and Full , 1993; Geyer et al . , 2006; Srinivasan and Holmes , 2008 ) , but generalized such that the leg forces are not constrained by ad hoc spring-like-leg assumptions ( Srinivasan , 2011 ) . Instead , the biped controller details are inferred from our experimentally obtained linear models . Both biped models have a point-mass upper body and massless legs ( Srinivasan and Ruina , 2006; Srinivasan , 2011 ) , that can change effective leg length during stance by modulating the leg force ( Figure 7a ) . During flight phase , the point-mass body undergoes parabolic free flight . The legs can apply forces on the upper body during stance phase . The two models , dubbed ‘direct force control model’ and ‘muscle control model’ differ in how the leg force is produced and controlled . For the muscle control model , we use a Hill muscle model with force-length and force-velocity relations ( Figure 7b , c and d ) . The 3D equations of motion of the point-mass biped are: mx¨=Fleg⋅ ( x−xfoot ) /ℓ , my¨=Fleg⋅ ( y−yfoot ) /ℓ , and mz¨=−mg+Fleg⋅ ( z−zfoot ) /ℓ , where Fleg is the scalar leg force , ( xfoot , yfoot , zfoot ) is the foot position with zfoot=0 on flat terrain and ℓ= ( x−xfoot ) 2+ ( y−yfoot ) 2+ ( z−zfoot ) 2 is the leg length from body to foot . In the ‘direct force control model’ , the object of control is the leg force Fleg during stance phase . In the ‘muscle control model’ , the object of control is the muscle activation amuscle , which is converted to muscle force via the force-length and force-velocity equations of Hill-type muscles ( Figure 7a–b ) . See ( Zajac , 1989; Srinivasan and Ruina , 2006; Srinivasan , 2011 ) for more detailed equations of motion and muscle model equations . Both models have two terms in their control: ( 1 ) a feedforward or ‘nominal’ term , that depends only on the average or desired periodic motion and ( 2 ) feedback modification of the control in response to state deviations at flight phase . The model’s leg force or muscle activation is modeled as a two-term sine series of the form A1sin⁡ ( 2πt/2Tstance ) +A2sin⁡ ( 2πt/Tstance ) , as shown in Figure 7e . By changing the relative weights of A1 and A2 , the shape of the leg force profile can be changed from being symmetric about the peak force to being asymmetric , with the peak force preceding or following mid-stance . We parameterize the running motion using stance duration Tstance , flight duration Tflight , 2D foot placement ( xfoot , yfoot ) , 3D initial conditions for stance ( x⁢ ( 0 ) , y⁢ ( 0 ) , z⁢ ( 0 ) ) , and the coefficients of the two-term sine series ( A1 and A2 ) . We solve for these variables to obtain a periodic running motion that accurately match the forward speed , step period , step width , and peak leg force from experimental data ( Figure 8a ) by using an optimization procedure ( Srinivasan and Holmes , 2008; Srinivasan , 2011 ) that enforces a constraint satisfaction tolerance of less than 10-6 . The runner leaves the ground when it reaches the maximum leg length , but the nominal leg length at landing is assumed to be shorter ( 95% ) than the maximum leg length , as seen in running data ( Voloshina and Ferris , 2015 ) . We enforce that left and right stances are mirror symmetric . Unlike previous simple running models , our model’s nominal periodic motion has non-zero step width and a stance phase that is asymmetric about mid-stance . This asymmetric stance is due to unequal landing and take-off leg lengths , and the asymmetry of the leg force or muscle activation about mid-stance . The foot placement control for the models are based on the experimentally derived control and given by the linear model in Equations 8 and 9 . The leg force feedback control based on apex body state , for the direct force control model , has gains as shown in Figure 8b . The muscle control model’s feedback control gains are also shown superimposed in Figure 8b . These control gains were derived for the two models by modifying the Fourier coefficients for the force and muscle activations respectively , so that the linear map from one apex to the next is best matched to that from data ( Equation 3 ) . While there are infinitely many controllers , even for this simple biped model , that can approximate the apex-to-apex map , our simplifying assumptions constrains the space of controllers to produce a unique fit . The leg forces and muscle activations are rectified , so that they never become negative despite feedback control ( Blum et al . , 2017 ) . The foot placement control and leg force feedback control are activated only when the apex state deviates from nominal . To obtain a running simulation over many steps , we break up each step into three phases: flight from apex to beginning of stance , the stance phase , and flight from the end of stance to flight apex . The control actions for the next stance are chosen at flight apex . As previously defined for the experimental data , the flight apex is when z˙ becomes zero . In some cases , if the vertical velocity is downward when a stance phase ends ( z˙<0 ) , there is no flight ‘apex’ and the controller uses the end of stance state instead of flight apex state as input . The end of flight and thus , the beginning of stance , are determined as the moment when the distance between the body and the target foot position is exactly equal to the landing leg length . The leg length at landing is also controlled based on flight apex state , based on the linear model in Equation 7 . At flight apex , if the distance to the next foot position is less than the target landing leg length , the runner immediately goes into stance . Such a simulation , when started from initial conditions exactly on the nominal periodic motion , results in a perfectly periodic motion when there are no further perturbations . We then re-simulated the two point-mass running models for hundreds of steps , in the presence of noisy foot placements and leg forces or muscle activations with step-to-step variability . To model the noise in foot placement , we computed the ‘desired’ foot placement based on the center of mass state at flight apex ( Equations 8-9 ) and then added a deviation drawn from a normal distribution , whose variance equals the foot placement variance not explained by Equation 8 . Similarly , we incorporated imprecise control of leg forces or muscle activation in the following manner: for each stance phase , once the leg force F⁢ ( t ) ( for model-1 ) or muscle activation a⁢ ( t ) ( for model-2 ) is determined based on the center of mass state at the previous flight apex , we ‘corrupt’ these functions by a multiplicative noise term , so that the actual leg force or muscle activation is F⁢ ( t ) ⁢ ( 1+ϵ ) or a⁢ ( t ) ⁢ ( 1+ϵ ) respectively , where ϵ is drawn from a normal distribution with variance equal to the unexplained step-to-step variability in leg force magnitude . Thus , we use the unexplained variance in the foot placement and leg forces from experimental regressions as a simple model of the intrinsic noise in active control .	Running at a constant speed seems like a series of repetitive , identical strides , but it is not . There are small variations in each stride . Self-inflicted errors in the forces generated by the muscles , or misperceptions from the senses , may cause these tiny imperfections . Uneven terrain or other outside forces , like a push , can also cause changes in a running stride . People must correct for these small changes as they run to avoid falling down . The only way to correct errors in a stride is by changing the force exerted on the ground by the leg . Now , Seethapathi and Srinivasan document step-by-step how people correct for small imperfections in their running stride to avoid falling . In the experiments , eight people ran on a treadmill at three different speeds while the motion of their torso and each foot was measured along with the forces of each foot on the treadmill . Seethapathi and Srinivasan found that these runners corrected for minor deviations by changing where each foot lands and how much force each leg applies to the treadmill . The runners placed their foot at a different position on each step and these varying foot positions could be predicted by the errors in the body movement between steps . These errors in body movement could also be used to predict how the runners would change the forces applied by their legs on each step . Imperfections in the stride were mostly corrected within the next step . Errors that would cause the runner to fall to the side were corrected more quickly than errors in forward or backward motion . Seethapathi and Srinivasan incorporated these corrective strategies into a computer simulation of a runner , resulting in a simulated runner that did not fall even when pushed . These findings may inform the design of robots that run more like humans . They may also help create robotic exoskeletons , prosthetic legs and other assistive devices that help people with disabilities move more fluidly and avoid falling .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology" ]	2019	Step-to-step variations in human running reveal how humans run without falling
Natural Killer ( NK ) cells confer protection from tumors and infections by releasing cytotoxic granules and pro-inflammatory cytokines upon recognition of diseased cells . The responsiveness of NK cells to acute stimulation is dynamically tuned by steady-state receptor-ligand interactions of an NK cell with its cellular environment . Here , we demonstrate that in healthy WT mice the NK activating receptor NKG2D is engaged in vivo by one of its ligands , RAE-1ε , which is expressed constitutively by lymph node endothelial cells and highly induced on tumor-associated endothelium . This interaction causes internalization of NKG2D from the NK cell surface and transmits an NK-intrinsic signal that desensitizes NK cell responses globally to acute stimulation , resulting in impaired NK antitumor responses in vivo . Natural Killer ( NK ) cells are key effectors in the immune response to pathogens and tumors ( Vivier et al . , 2008 ) . NK cells respond to infected or transformed cells by releasing cytotoxic granules and anti-tumor cytokines such as interferon-γ ( IFNγ ) ( Vivier et al . , 2008; Marcus et al . , 2014 ) . NK cells recognize unhealthy cells using an array of cell surface receptors ( Vivier et al . , 2011; Marcus et al . , 2014; Moretta et al . , 2014; Morvan and Lanier , 2016 ) . These receptors transmit activating or inhibitory signals upon binding cognate ligands on the target cell , and the net balance of these signals dictates whether the NK cell response is triggered . Tumors are often recognized and killed by NK cells in vitro and in vivo because cancer cells tend to upregulate ligands for activating receptors and downregulate ligands for inhibitory receptors ( Waldhauer and Steinle , 2008; Marcus et al . , 2014 ) . The responsiveness of NK cells to a given stimulus is dynamically tuned by the steady-state receptor-ligand interactions experienced by the NK cells ( Joncker and Raulet , 2008; Brodin et al . , 2009; Joncker et al . , 2009; Joncker et al . , 2010; Shifrin et al . , 2014 ) . Increases in steady-state stimulation cause NK cells to compensate by adopting a less responsive state ( Joncker et al . , 2010; Kadri et al . , 2016 ) – a process that will be referred to here as ‘desensitization’ – whereas NK cells receiving lower steady-state levels of stimulation exhibit a state of heightened responsiveness to acute activation . For example , the Ly49 family of inhibitory receptors on NK cells are known to engage host MHC I molecules at steady state , and this interaction is important for regulating NK responsiveness . Mice lacking MHC I molecules or inhibitory Ly49 receptors show dramatically weaker NK responses to a wide variety of acute stimulatory signals in vitro and in vivo ( Liao et al . , 1991; Fernandez et al . , 2005; Kim et al . , 2005; Anfossi et al . , 2006; Brodin et al . , 2009; Joncker et al . , 2010 ) . Desensitization may prevent NK cells from effecting autoreactivity and enable them to adjust to different tissue milieus , and mature NK cells can alter their responsiveness upon encountering a new MHC I environment ( Joncker and Raulet , 2008; Elliott et al . , 2010; Joncker et al . , 2010; Narni-Mancinelli et al . , 2013 ) . These dynamics are relevant for antitumor responses , as NK cells in WT mice become desensitized when they infiltrate MHC I-deficient tumors but not when they infiltrate matched MHC-I-positive tumors ( Ardolino et al . , 2014 ) . Similarly , humans receiving HLA-mismatched bone marrow show altered NK responses that match the trends described in mice ( Boudreau et al . , 2016 ) . It is presumed that steady-state interactions between MHC I and Ly49 receptors prevent NK desensitization by inhibiting steady-state signals from activating receptors . Indeed , transgenic overexpression of NK activating ligands causes NK desensitization ( Oppenheim et al . , 2005; Wiemann et al . , 2005; Sun and Lanier , 2008; Tripathy et al . , 2008 ) , but the endogenous receptor-ligand systems that transmit these activating signals to NK cells in healthy WT animals remain incompletely defined . In humans , activating KIR appear to be one such endogenous signal involved in steady-state NK cell tuning ( Fauriat et al . , 2010 ) . In mice , the activating receptor NKp46 may contribute to NK desensitization because NKp46-KO animals showed heightened NK responses to stimulation in one report ( Narni-Mancinelli et al . , 2012 ) , although not in another ( Sheppard et al . , 2013 ) . SLAM receptors are also reported to regulate NK responsiveness in some contexts ( Chen et al . , 2016 ) ( Veillette , 2010 ) . Very little is understood about which host cell types are responsible for engaging NK cells to regulate responsiveness . A recent study using β2M-KO bone marrow chimeras suggested that MHC-I-deficient nonhematopoietic cells may play a larger role than MHC-I-deficient hematopoietic cells in desensitizing NK cells , although both may participate ( Shifrin et al . , 2016 ) . In humans , different studies have implicated HLA molecules on hematopoietic cells ( Haas et al . , 2011 ) and nonhematopoietic cells ( Cooley et al . , 2011 ) as being critical for tuning . Clearly , much remains to be learned about these processes . Elucidating the receptor-ligand and cellular systems that regulate NK responses in homeostasis and cancer may suggest novel therapeutic strategies . NKG2D is a C-type lectin-like activating receptor expressed by all NK cells and subsets of T cells ( Raulet , 2003 ) . NKG2D binds a diverse array of MHC-like proteins . In mice , these include the RAE-1 family ( with α , β , γ , δ , and ε isoforms ) , the H60 family ( a , b , c ) , and MULT1 . Human NKG2D ligands include the ULBP family ( with isoforms 1–6 ) and the MICA and MICB proteins ( Raulet et al . , 2013 ) . Acute NKG2D engagement transmits powerful activating signals through the adaptor molecules DAP10 and DAP12 to drive cytotoxicity and cytokine production ( Raulet , 2003 ) . NKG2D ligands are thought to be absent from most healthy cells but can be induced consequent to DNA damage , oncogene signaling , and other stresses associated with cancer and infection ( Raulet et al . , 2013 ) . Many tumor cells express NKG2D ligands . In tumor transplant and spontaneous cancer models , expression of NKG2D ligand ( s ) on tumor cells triggers NK activation and protects the host from cancer ( Diefenbach et al . , 2001; Guerra et al . , 2008 ) . Interestingly , several recent studies have shown that NK cells in NKG2D-KO mice are hyper-responsive to stimulation when triggered through other activating receptors ( Zafirova et al . , 2009; Sheppard et al . , 2013 ) . Furthermore , tumor cells engineered to secrete soluble monomeric NKG2D ligands – which block but do not activate NKG2D – increase the responsiveness of tumor-infiltrating NK cells and enhance tumor rejection ( Deng et al . , 2015 ) . These data suggest that NKG2D may contribute to NK desensitization at steady state or in tumors . In this report , we provide important new findings concerning the cells and molecules that engage NK cells and regulate NK responsiveness , and we clarify the pleiotropic effect of NKG2D on NK activity . Unexpectedly , we show a steady-state interaction between NKG2D and one of its ligands , RAE-1ε , in healthy WT mice . Using bone marrow chimera experiments , we show that non-hematopoietic cells are the primary source of endogenous RAE-1ε . Endothelial cells in lymph nodes were found to be constitutively express RAE-1ε , and RAE-1ε was found to be super-induced on tumor-associated vasculature in transplant and autochthonous cancer models . Importantly , we demonstrate that this interaction between NKG2D and endogenous RAE-1ε desensitizes NK cells and impairs antitumor NK responses and tumor rejection . Cell surface NKG2D ligand expression is usually considered a hallmark of unhealthy cells , but expression on the surface of normal cells in healthy animals has not been exhaustively surveyed in vivo . NKG2D is known to be internalized upon ligand engagement ( Lanier , 2015 ) , so we reasoned that if NKG2D ligands are expressed and interact with NKG2D in healthy WT mice , antibody blockade of the relevant ligand ( s ) should result in increased levels of NKG2D on the surface of NK cells . Adult C57BL/6 ( B6 ) mice were injected with confirmed blocking antibodies ( Figure 1—figure supplement 1A ) specific for NKG2D ligands RAE-1δ , RAE-1ε , or MULT1 . NKG2D levels on NK cells were analyzed by flow cytometry 48 hr post-injection . In vivo blockade of RAE-1ε , but not RAE-1δ or MULT1 , substantially increased NKG2D surface levels on NK cells in blood ( Figure 1A ) , lymph nodes , and spleen ( Figure 1—figure supplement 1B ) . NKG2D elevation after RAE-1ε blockade occurred as early as 12 hr after antibody injection ( Figure 1B ) . We subsequently analyzed NKG2D surface levels in RAE-1-KO mice , which contain frameshift mutations ( induced by CRISPR/Cas9 ) in the genes for both RAE-1ε and RAE-1δ ( Deng et al . , 2015 ) . In healthy , unmanipulated animals , NK cells in RAE-1-KO mice showed substantially higher cell surface NKG2D levels than WT controls in all compartments tested , including blood , spleen , lymph nodes , and peritoneal wash ( Figure 1C ) . NK cells in bone marrow and liver also showed elevated NKG2D levels in RAE-1-KO mice ( Figure 1—figure supplement 2A ) . mRNA levels for Klrk1 ( the gene for NKG2D ) were identical in NK cells from WT and RAE-1-KO mice ( Figure 1D ) , consistent with the conclusion that host RAE-1ε causes internalization of NKG2D from the NK cell surface . Blocking RAE-1ε in WT mice increased NKG2D to levels comparable to RAE-1-KO mice at steady state , whereas anti-RAE-1ε had no effect on NKG2D levels in RAE-1-KO mice ( Figure 1—figure supplement 1C ) . Furthermore , blockade of RAE-1ε in combination with RAE-1δ in WT mice showed no additional effect on NKG2D levels compared with blocking RAE-1ε alone ( Figure 1—figure supplement 1D ) . To assess whether these phenotypes were intrinsic to NK cells , we transferred CFSE-labeled splenocytes from WT into RAE-1-KO mice and vice versa . When splenocytes were transferred from WT to RAE-1-KO mice , NKG2D levels on the transferred NK cells increased to match the RAE-1-KO mice ( Figure 1E ) . Reciprocally , NKG2D surface levels were reduced on NK cells transferred from RAE-1-KO into WT mice . Cumulatively , these data demonstrated that in healthy WT mice a subset of cells express RAE-1ε , which engages and downregulates NKG2D at steady state from the surface of NK cells . We next sought to understand the effect of host RAE-1ε on the function of NK cells . Splenic NK cell numbers and expression of CD11b and CD27 – cell surface markers associated with NK maturation ( Hayakawa and Smyth , 2006 ) – were similar in WT and RAE-1-KO mice ( Figure 2—figure supplement 1A ) . Release of cytotoxic granules and IFNγ are important NK cell functions ( Vivier et al . , 2008 ) , so we analyzed these responses in WT and RAE-1-KO NK cells after acute ex vivo activation through a variety of receptors . We used a standard 5 hr responsiveness assay in which cells were stimulated by plate-bound antibodies that crosslink activating NK receptors , followed by flow cytometry for degranulation ( marked by CD107a cell surface presentation ) and intracellular IFNγ ( Joncker et al . , 2009 , 2010 ) . As is typical with this assay , stimulation through the activating receptor NKp46 triggered robust NK cell degranulation and IFNγ production from WT splenic NK cells , and a significantly greater percentage of NK cells from RAE-1-KO mice responded to stimulation compared with WT NK cells ( Figure 2A and B ) . NK cells from RAE-1-KO mice also showed elevated responses when stimulated with platebound antibodies that ligate a distinct activating receptor , NK1 . 1 , or that ligate NKG2D itself ( Figure 2B ) . These data indicated that splenic NK cells from RAE-1-KO mice exhibit a hyper-responsive phenotype upon acute stimulation through a variety of activating receptors . In our experience , NK cells in the peritoneal cavity typically yield relatively low responses to ex vivo stimulation . We tested whether endogenous RAE-1ε regulated the responsiveness of these cells . Interestingly , peritoneal NK cells from RAE-1-KO mice showed markedly greater responses compared with their WT counterparts when stimulated through NKp46 , NK1 . 1 or NKG2D ( Figure 2C ) . This especially large increase gave us a greater window to examine the desensitization effect , so we next analyzed peritoneal NK responses after injecting WT mice i . p . with antibodies that block RAE-1ε . Similar to the RAE-1-KO mice , blockade of RAE-1ε caused a substantial increase in NK responses to stimulation through all receptors tested ( Figure 2D ) . The increased responses could be seen as early as 48 hr after antibody administration . To analyze killing of tumor cells , we performed a standard 4 hr 51Cr in vitro cytotoxicity assay , using YAC-1 cells as targets . Peritoneal wash cells from WT , RAE-1-KO , and NKG2D-KO mice were used as effectors . NKG2D-KO effectors were significantly less efficient at killing YAC-1 cells ( Figure 2—figure supplement 1B ) , consistent with published reports showing that NKG2D-mediated recognition is required for efficient YAC-1 killing ( Jamieson et al . , 2002; Guerra et al . , 2008 ) . In contrast , RAE-1-KO mice showed markedly enhanced NK killing of YAC-1 cells ( Figure 2—figure supplement 1B ) . Together , these data suggested that endogenous , steady-state RAE-1ε expression desensitizes NK responses to activation through multiple activating receptors and YAC-1 cells in vitro . RAE-1ε binds NKG2D , so we expected NKG2D-KO NK cells to be hyper-responsive to NKG2D-independent stimuli . Indeed , NKG2D-KO NK cells from spleen and peritoneal wash showed increased responses to stimulation compared with WT controls when stimulated through NKp46 and NK1 . 1 ( Figure 3A and B ) , as has also been previously reported ( Zafirova et al . , 2009; Sheppard et al . , 2013; Deng et al . , 2015 ) . We then directly compared the responses of peritoneal NK cells from matched WT , RAE-1-KO , and NKG2D-KO mice . When stimulated with platebound antibody ligating NKG2D , NK cells from RAE-1-KO showed elevated responses , whereas NKG2D-KO NK cells failed to respond , as expected ( Figure 3—figure supplement 1A ) . In contrast , stimulation through NKp46 resulted in elevated responses from both the RAE-1-KO and NKG2D-KO cohorts ( Figure 3C ) . Interestingly , NKG2D-KO NK cells were consistently even more responsive than the NK cells from RAE-1-KO mice ( Figure 3C ) . These data suggested that , in addition to RAE-1ε , other ligands may participate in NKG2D-mediated desensitization , or NKG2D may regulate NK responses partly through a ligand-independent mechanism in addition to the RAE-1ε-dependent mechanism documented herein . We considered that NKG2D-mediated desensitization could happen in a cell-intrinsic manner – that is , through a given NK cell’s interaction with ligand and consequent desensitization – or cell-extrinsically via a specific population of ‘suppressor’ cells . To discriminate between these hypotheses , we generated bone marrow chimeras containing NKG2D-WT and NKG2D-KO cells in the same animal , or singly reconstituted chimeras as controls . WT ( CD45 . 1 ) mice were lethally irradiated and reconstituted with bone marrow cells from WT ( CD45 . 1 ) mice , NKG2D-KO ( CD45 . 2 ) mice , or a 1:1 mixture of the two genotypes . Reconstitution efficiency was consistently greater than 99% , and the mixed chimeric mice contained similar numbers of WT and NKG2D-KO NK cells ( Figure 3—figure supplement 1B ) . We then tested the chimeras for NK cell responsiveness . Consistent with our earlier data , NK cells from mice reconstituted with NKG2D-KO bone marrow showed greater responses than NK cells from mice reconstituted with WT bone marrow . Interestingly , NK cells from the mixed chimeras recapitulated these responses , as NKG2D-KO NK cells were hyper-responsive compared with WT NK cells in the same animals ( Figure 3D ) . These data demonstrated that NKG2D desensitizes NK responses in a cell-intrinsic manner . We next sought to identify the cellular source of RAE-1ε responsible for engaging NKG2D and desensitizing NK cells . We used a bone marrow chimera approach to restrict RAE-1ε expression to hematopoietic or nonhematopoietic cells . We used a radiation dose ( 600 Gy + 500 Gy split dose ) that reliably led to replacement of >99% of cells in the hematopoietic compartment , although we cannot exclude the presence of some radio-resistant bone-marrow-derived cells in the chimeras . After irradiation , WT or RAE-1-KO mice were reconstituted with bone marrow from WT or RAE-1-KO mice , and NKG2D cell surface levels were analyzed on NK cells 8 weeks after reconstitution . As expected , KO → KO chimeras showed substantially higher NKG2D levels compared with WT → WT controls ( Figure 4A ) ( Figure 4-figure supplement 1A ) . Chimeric mice in which RAE-1ε was present only in hematopoietic cells ( WT → KO ) showed high NKG2D levels comparable to KO → KO chimeras , indicating that hematopoietic RAE-1ε does not play a major role in engaging NKG2D , although there was a reproducibly small effect in most experiments that failed to reach significance . In contrast , mice with RAE-1ε expression restricted to nonhematopoietic cells ( KO → WT ) completely recapitulated the low NKG2D levels seen in WT → WT animals ( Figure 4A ) ( Figure 4—figure supplement 1A ) . When we analyzed the functional responses of NK cells in these chimeras , a similar pattern emerged , with nonhematopoietic RAE-1 playing a dominant role in the desensitization of NK responses , although hematopoietic RAE-1 did show some effect ( Figure 4B ) . These data suggested that nonhematopoietic cells are the dominant source of RAE-1ε that engages NKG2D and regulates NK cell responsiveness . We then began a search for the nonhematopoietic source of RAE-1ε . Because RAE-1-KO mice had elevated NKG2D levels on NK cells in blood and other peripheral tissues , we reasoned that the cellular source of RAE-1ε must be accessible to these NK cells as part of their normal circulatory pattern . Therefore , we used flow cytometry to analyze RAE-1ε on nonhematopoietic cells in various organs encountered by circulating NK cells . Like other lymphocytes , circulating NK cells navigate to and from blood and secondary lymphoid organs . Lymph nodes are central hubs for circulating lymphocytes and have crucial regulatory roles . After gentle enzymatic dissociation of lymph nodes , four populations of nonhematopoietic ( CD45-neg ) lymph node cells can be delineated by expression of the adhesion molecule CD31 and the transmembrane protein Podoplanin ( PDPN ) ( Figure 4—figure supplement 1B ) ( Turley et al . , 2010 ) . Cells that are CD31+ PDPN-neg are blood endothelial cells ( BECs ) and CD31+ PDPN+ cells are lymphatic endothelial cells ( LECs ) . Lymphocytes intimately engage these endothelial cells to enter and exit lymph nodes ( Butcher et al . , 1986 ) . CD31-neg PDPN+ cells are fibroblastic reticular cells ( FRCs ) , which comprise a flexible cellular matrix that defines the lymph node architecture ( Turley et al . , 2010 ) . The CD31-neg PDPN-neg double negative ( DN ) population is poorly characterized . We isolated inguinal lymph nodes from naive B6 mice and used flow cytometry to analyze RAE-1ε on these four populations . Whereas DN cells and FRCs showed little to no RAE-1ε , we found substantial RAE-1ε expression on BECs and LECs ( Figure 4C ) . This was not due to promiscuous binding of the RAE-1ε antibody , because the staining completely disappeared in RAE-1-KO mice ( Figure 4—figure supplement 1C ) . Next , we examined whether RAE-1ε was expressed on endothelial cells in other tissues . Splenic CD31-hi endothelial cells did express low amounts of RAE-1ε , but endothelial cells in the lung , liver , and heart showed little to no RAE-1ε ( Figure 4D and Figure 4—figure supplement 2 ) ; all other nonhematopoietic cells in these cell preparations were also negative for RAE-1ε ( not shown ) . High Endothelial Venule ( HEV ) endothelial cells are a specialized subset of BECs that mediate lymphocyte entrance into lymph nodes ( Berg et al . , 1989 ) . HEV cells can be identified using the antibody MECA-79 , which recognizes a specific carbohydrate motif ( Figure 4—figure supplement 1D ) ( Streeter et al . , 1988 ) . Interestingly , RAE-1ε expression was substantially higher on HEV cells than the average expression on non-HEV BECs ( Figure 4—figure supplement 1E ) . In summary , these experiments showed that nonhematopoietic cells are the dominant compartment responsible for steady state RAE-1ε-mediated NKG2D engagement and NK desensitization , and our analysis of cellular RAE-1ε expression implicate endothelial cells in secondary lymphoid tissue as the relevant cellular source for RAE-1ε . These findings suggest a model in which NK cells , trafficking in and out of lymphoid tissue during homeostatic circulation , are continuously engaged and desensitized by RAE-1ε expressed on endothelial cells . NK cell responsiveness is controlled by systemic interactions and at local sites of inflammation such as the tumor microenvironment ( Joncker et al . , 2010; Ardolino et al . , 2014 ) . The powerful antitumor activity of NK cells often selects for tumor cells and microenvironments that can circumvent the NK response ( Marcus et al . , 2014 ) . To study the effects of endogenous RAE-1ε in tumor microenvironments , we used the B16-BL6 ( hereafter called B16 ) model of melanoma , a classic syngeneic tumor model that is sensitive to NK killing but lacks NKG2D ligand expression ( Lakshmikanth et al . , 2009 ) . WT mice were implanted subcutaneously with B16 cells . After establishment of tumors , mice were treated with antibodies against RAE-1δ , RAE-1ε , or both for 48 hr , after which the tumors were harvested , dissociated to single-cell suspensions , and NK cells infiltrating the tumors were analyzed for surface NKG2D levels . Blocking RAE-1ε but not RAE-1δ caused dramatic NKG2D upregulation on tumor-infiltrating NK cells ( Figure 5A ) . Because B16 tumors completely lack expression of NKG2D ligands , we suspected an endogenous source of RAE-1ε , so we analyzed NK cells infiltrating B16 tumors implanted in WT or RAE-1-KO mice . Tumor-infiltrating NK cells mice had elevated NKG2D in RAE-1-KO mice compared with WT controls ( Figure 5B ) . Similar results were obtained when tumor-infiltrating NK cells were examined in mice implanted subcutaneously with syngeneic RMA-S lymphoma cells , which also completely lack NKG2D ligand expression ( Figure 5B ) . We analyzed RAE-1 bone marrow chimeras to determine the cellular compartment of RAE-1ε in the tumor microenvironment . We found that nonhematopoietic RAE-1ε was dominant in downregulating NKG2D in B16 tumors ( Figure 5C ) , although in some experiments hematopoietic RAE-1 molecules seemed to show some variable effect , albeit not statistically significant ( Figure 5—figure supplement 1A ) . We analyzed single-cell suspensions from B16 , RMA-S , and TRAMP-C2 ( prostate adenocarcinoma model ) and found that tumor-associated endothelial cells expressed copious RAE-1ε ( Figure 5D ) in all tumor models tested . The staining specificity was confirmed using the RAE-1-KO mice as genetic controls ( Figure 5—figure supplement 1B ) . When we quantified RAE-1ε levels on tumor-associated endothelial cells compared with lymph node BECs , we saw a much greater expression of RAE-1ε on tumor-associated ECs ( Figure 5E ) , indicating that the tumor microenvironment is a substantial inducer of endothelial RAE-1ε . To address whether these findings applied to tumors that arise naturally , we explored endothelial RAE-1ε expression in the genetically engineered ‘KP’ cancer model ( DuPage et al . , 2012 ) . KP mice contain germline mutations targeting loxP sites to the Trp53 tumor suppressor gene and a lox-P-flanked STOP cassette preceding an oncogenic KrasG12D allele . Viral delivery of Cre recombinase results in deletion of p53 and expression of oncogenic KRAS in vivo , leading to tumorigenesis in the injected tissue ( DuPage et al . , 2009 ) . We injected lentivirus expressing Cre into the hind leg muscle of KP mice to generate autochthonous sarcomas . Established KP sarcomas or matched healthy tissue from the opposite leg were dissociated and analyzed for endothelial RAE-1ε . Healthy leg muscle lacked endothelial RAE-1ε expression , whereas endothelial cells in KP sarcomas showed robust RAE-1ε induction ( Figure 5F ) . Together , these data suggested that: ( 1 ) tumor-infiltrating NK cells are engaged by non-tumor RAE-1ε; ( 2 ) nonhematopoietic cells are the primary endogenous source of RAE-1ε responsible for NKG2D engagement; and ( 3 ) endothelial cells in transplanted tumors and autochthonous models of mouse cancer are induced to express especially high amounts of RAE-1ε . These data are consistent with a model in which NK cells recruited to tumors are engaged by RAE-1ε induced on endothelial cells in the tumor microenvironment . NK cells protect the host from tumors in vivo , and the strength of the NK antitumor response depends on the intrinsic responsiveness of NK cells ( Ardolino et al . , 2014 ) . Therefore , we hypothesized that disrupting interactions between NKG2D and host RAE-1ε would amplify the antitumor NK response in vivo . However , the situation is complicated by the fact that NKG2D ligands are often present on tumor cells , and this interaction provides an acute activation signal that kills tumor cells and protects the host . We undertook a series of experiments to clarify the effect of NKG2D ligands on host cells vs . tumor cells . First , we turned to the B16 model , which is sensitive to NK cell killing but does not express NKG2D ligands . B16 tumors can be injected intravenously ( metastasis model ) or subcutaneously ( solid tumor model ) . We reasoned that NKG2D-KO mice should show enhanced protection from B16 tumors given the observed NK hyper-responsiveness in these mice . WT and NKG2D-KO mice were injected intravenously with a limiting number of B16 cells and monitored for survival . Consistent with our hypothesis , NKG2D-KO mice showed reduced and delayed mortality compared with matched WT controls ( Figure 6A ) . Importantly , NK depletion resulted in dramatically accelerated mortality in WT and NKG2D-KO mice , eliminating the protective effect of NKG2D deficiency . When WT or NKG2D-KO mice were implanted with B16 cells subcutaneously , NKG2D-KO mice were more capable of controlling B16 tumor growth than WT counterparts ( Figure 6B ) . We also found that Rag2/NKG2D double knockout mice resisted B16 tumors better than Rag2-KO mice , indicating that T cells ( and B cells ) are not required for the protective effect of NKG2D deficiency ( Figure 6—figure supplement 1A ) . A subsequent experiment on the Rag2-KO background showed that treatment of NKG2D-WT mice with anti-NKG2D antibody in vivo starting the day before tumor cell implantation resulted in enhanced control of B16 tumors comparable to the NKG2D-KO mice ( Figure 6—figure supplement 1B ) . We then turned to the RMA-S lymphoma transplant tumor model , which is targeted in vivo by NK cells through ‘missing self’ recognition but also does not express NKG2D ligands . NKG2D-KO mice showed better control of subcutaneous RMA-S tumors than did their WT counterparts ( Figure 6C ) . These data indicate that NKG2D expression mitigates in vivo NK responses to tumors that lack NKG2D ligands , and that this desensitizing effect is reversed by acute antibody blockade of NKG2D . Because most tumors express NKG2D ligands , we analyzed tumor growth of B16 tumor cells transduced to express high surface levels of the NKG2D ligand MULT1 . MULT1 is a high-affinity NKG2D ligand that stimulates a strong acute NK cell response that enhances tumor rejection in WT mice ( Figure 6—figure supplement 1C ) . Despite the NK hyper-responsiveness in NKG2D-deficient animals , NKG2D-KO NK cells cannot recognize MULT1 , and the growth of B16-MULT1 tumors was accelerated in NKG2D-KO mice compared with WT controls ( Figure 6D ) . These data are consistent with previous studies showing a strong protective role for NKG2D interactions with tumor NKG2D ligands ( Cerwenka et al . , 2001; Diefenbach et al . , 2001; Guerra et al . , 2008 ) . Because we had found that RAE-1-KO mice have elevated NKG2D surface levels and enhanced NK responses , we hypothesized that RAE-1-KO mice would show enhanced protection from NKG2D ligand-positive tumors . Indeed , RAE-1-KO mice challenged with B16-MULT1 tumors exhibited superior tumor rejection compared with WT mice ( Figure 6E ) . We then turned to a transplant tumor model with endogenous NKG2D ligand expression , the TRAMP-C2 model of prostate adenocarcinoma . TRAMP-C2 cells express RAE-1δ and RAE-1ε and are sensitive to NK cells , and this sensitivity is partly dependent on NKG2D ( Jamieson et al . , 2002 ) . When injected subcutaneously into WT and RAE-1-KO mice , the RAE-1-KO mice showed enhanced control of TRAMP-C2 tumors compared with WT controls ( Figure 6F ) . Collectively , these data demonstrate that interactions between NKG2D and endogenous RAE-1ε desensitize NK responses to tumors in vivo , whereas NKG2D ligands expressed on tumor cells promote antitumor NK responses . The studies in this paper add considerably to our understanding of the receptors and ligands that regulate NK activity at steady state and in cancer . We show that the activating receptor NKG2D is engaged by endogenous RAE-1ε in healthy WT mice , causing constitutive downregulation of NKG2D from the NK cell surface , and leads to an intrinsic global desensitization of NK cells to acute stimulation . These effects were evidenced by an increase in the responsiveness of NK cells from NKG2D-KO or RAE-1-KO mice to stimulation through diverse activating receptors . Antibody blockade of RAE-1ε in normal mice also resulted in elevated NKG2D levels and increased NK cell responsiveness ( Figure 1A , B ) . Moreover , preventing NKG2D interactions with host RAE-1ε enhanced NK cell antitumor responses in vivo . Bone marrow chimera experiments identified nonhematopoietic cells as the dominant source of RAE-1ε , and endothelial cells were found to express RAE-1ε constitutively in lymph nodes . In tumors , RAE-1ε expression on tumor-associated endothelial cells was super-induced compared to endothelial cells in lymph nodes . These data support a model in which endothelial RAE-1ε interacts with NKG2D to desensitize NK cells and diminish antitumor NK responses . NK cell hyper-responsiveness in NKG2D-KO mice has been reported previously by several groups including ours ( Zafirova et al . , 2009; Sheppard et al . , 2013; Deng et al . , 2015 ) , although the trend of increased responsiveness did not reach statistical significance in our original characterization of the NKG2D-KO mouse ( Guerra et al . , 2008 ) . The data presented in this study build on those findings by showing that NKG2D engagement of endogenous ligands desensitizes NK cells , reducing antitumor responses . At the same time , our results support the classical view that NKG2D can participate in immune surveillance of cancer by recognizing NKG2D ligands on tumor cells to induce NK cell activation and tumor cell killing ( Diefenbach et al . , 2000; Cerwenka et al . , 2001; Diefenbach et al . , 2001; Jamieson et al . , 2002; Guerra et al . , 2008 ) . Hence , NKG2D confers opposing effects , both promoting killing of tumor cells that express NKG2D ligands , and desensitizing NK cells through interactions with host ligands . We propose that the net outcome of these opposing effects of NKG2D depends on the complement of NK-activating ligands that a given tumor cell expresses , among other factors . For tumor cells that express abundant NKG2D ligands , the ability of NKG2D to promote acute activation against the tumor can outweigh the opposing desensitizing interactions with host cells . In this circumstance , loss of NKG2D results in reduced tumor killing , as exemplified by the observation that NKG2D-deficient mice show reduced killing/rejection of B16-MULT1 tumors and YAC-1 cells ( Figure 6D , Figure 2—figure supplement 1B ) ( Jamieson et al . , 2002 ) . We speculate that this situation also applies to previous findings that NKG2D-KO mice were defective in immune surveillance of spontaneous tumors that express NKG2D ligands , such as prostate adenocarcinomas in TRAMP mice and lymphomas in Eu-Myc mice ( Guerra et al . , 2008 ) . In contrast , for tumor cells that express abundant other NK-activating ligands , and either do not express NKG2D ligands ( e . g . RMA-S and B16 cells ) , or depend little on NKG2D ligands for NK killing , the global desensitization of NK cells caused by NKG2D interactions with host ligands is predicted to exert a dominant effect . In this situation , NKG2D deficiency results in enhanced tumor killing ( Figure 6B , C ) . Interestingly , a recent study reported that in a spontaneous model of hepatocellular carcinoma , NKG2D-KO mice exhibited a reduced tumor incidence ( Sheppard et al . , 2017 ) . We speculate that NKG2D-mediated desensitization may be dominant in that model , though the authors offered an alternative explanation for their findings . It is also possible that some tumor environments contain signals that reverse NKG2D-mediated desensitization while preserving NKG2D-mediated activation , or vice versa . Thus , the net effect of NKG2D will likely depend on the activating and inhibitory ligands tumor cells express , as well as the context of the tumor microenvironment . Although endogenous RAE-1ε played a significant role in NK desensitization , it does not completely account for the desensitizing activity of NKG2D , based on the finding that NK cells attained higher responsiveness in NKG2D-KO mice than in RAE-1-KO mice ( Figure 3C ) . While our data clearly support a model in which endogenous interactions of NKG2D with RAE-1ε cause steady-state NK cell desensitization , we do not rule out other potential mechanisms contributing to hyper-responsiveness of NKG2D-KO cells -- such as interactions of NKG2D with other ligands at steady state , or ‘tonic’ signaling through NKG2D , at steady state or during NK development . Perhaps the most surprising finding in this report is the expression and functional relevance of RAE-1ε on endothelial cells . NKG2D ligand expression in adult mice has been thought largely restricted to cells undergoing certain stress responses associated with oncogenesis or infection , but here we show that blood endothelial cells and lymphoid endothelial cells in secondary lymphoid organs of healthy animals express RAE-1ε at steady-state . It was striking that , in the same mice , endothelium in several other tissues was completely negative for RAE-1ε . These findings suggest that the lymphoid tissue environment imparts specific signals that upregulate RAE-1ε on associated endothelial cells . Specific expression on endothelial cells in secondary lymphoid tissue further suggests that the system may be designed to desensitize NK cells that traverse this tissue . It is important to note that our data clearly identify non-hematopoietic cells as the dominant source of RAE-1ε-mediated NK desensitization , but we cannot conclusively establish endothelial cells as relevant desensitizing compartment without an endothelial-specific RAE-1ε-KO mouse . We were also surprised to find that resident liver NK cells ( CD49a-neg CD49b+ ) were engaged by host RAE-1ε . We do not know the cellular source of RAE-1ε responsible for engaging these NK cells , as we detected little to no RAE-1ε in cell suspensions from liver . We speculate that resident liver NK cells might traffic to distinct anatomical structures , perhaps liver-specific lymphoid tissue , that might contain RAE-1ε-expressing cells . We found especially high levels of RAE-1ε on endothelial cells in the tumor microenvironment . Notably , RAE-1ε was expressed on vasculature in all tumor models tested , including transplant models of melanoma , lymphoma , prostate cancer , and the genetically engineered KP model of autochthonous sarcoma . In all cases , levels of tumor-associated endothelial RAE-1ε were even greater than on endothelial cells in draining and non-draining lymph nodes ( Figure 5E ) . We can only speculate as to the mechanism of RAE-1ε induction on these cells . Vasculature in the tumor microenvironment is extremely dynamic , characterized by extensive angiogenesis . RAE-1 expression has been linked to high levels of cellular proliferation in development and wound healing ( Jung et al . , 2012 ) . Perhaps a similar mechanism is responsible for the super-induced expression of RAE-1ε on neovasculature in tumors . Steady-state RAE-1ε expression on lymph node endothelial cells , but not endothelium in normal non-lymphoid tissues , is more surprising . Clearly , more work is required to fully understand NKG2D ligand regulation in these cells . Endothelial cells are sometimes thought of as passive circulatory conduits . Our data and other studies clearly paint a more complex picture . Endothelial cells have recently been shown to produce immunomodulatory cytokines , and lymph node endothelial cells can tolerize peripheral T cells by presenting self MHC – peptide complexes ( Rouhani et al . , 2015 ) . The data in our study add to the emerging role of endothelial cells as active regulators in the immune response by suggesting that endothelial cells regulate NK cell activity . In a way , this seems logical , given the inevitable and intimate contact a lymphocyte must make with the vasculature upon entering and exiting lymph nodes and other organs during homeostatic circulation or inflammation . Because multiple NK cell receptors are known to regulate NK cell responsiveness , we are intrigued by the possibility that endothelial cells may also regulate NK cell activity through other receptor-ligand systems . In addition , it is interesting to speculate whether NK cells have effects on the endothelial cell biology , through NKG2D-RAE-1ε or other mechanisms . We wonder if endothelial-RAE-1ε-mediated NK desensitization is important in pathological contexts other than cancer . It is tempting to speculate that signals from vasculature may help inform infiltrating NK cells , or other lymphocytes , about the inflammatory state of the tissue . Perhaps NKG2D ligand expression on endothelial cells is advantageous for preventing autoreactivity and/or limiting inflammatory signals during the resolution of infection . Perhaps , tumor microenvironments mimic other physiological states , such as wound healing , for which RAE-1ε induction and NK desensitization is beneficial to the host . Endothelial RAE-1ε clearly regulates NK responses to tumors , but the relevant contribution of lymph node vs . tumor endothelium remains unclear . It is plausible that both environments control NK desensitization . We also speculate that physiological situations may arise in which NK cells transit from a desensitizing environment to one that reverses desensitization . Our data show that at steady state , NKG2D internalization is almost completely reversed as early as 12 hr after antibody blockade of RAE-1ε , and that NK function can be elevated at 48 hr after blockade . Further study is needed to understand how physiological changes in the host’s desensitizing environment might regulate the kinetics of the NK response . The differences in intracellular signaling mechanisms that culminate in NKG2D-mediated acute NK activation versus desensitization remain poorly understood . An attractive hypothesis is that receptor engagement in the absence of inflammatory signals results in initial activation followed by desensitization; if inflammatory cytokines are present , as in an infection , the activation response might be sustained . This model fits with evidence that inflammatory cytokines – such as IL-12/1 L-18 or IL-2/15 family cytokines – sustain or reinvigorate desensitized NK responses ( Ardolino et al . , 2014 ) . Similarly , activation vs . desensitization may be modulated by cell surface signals on the surface of the target cell . It might also be considered that different qualities of the encounter with ligand could result in divergent outcomes . For example , differences in the duration and/or frequency of the receptor-ligand interactions could contribute to opposing outcomes , or perhaps the relative affinity of a given ligand for its receptor , coupled with amount of ligand expressed , could potentially affect the signaling outcome . These and other molecular mechanisms might determine why interactions with some cells but not others results in desensitization . It is notable that a previous study did find a role for desensitization of NK cells by tumors lacking MHC I ( Ardolino et al . , 2014 ) , and tumor cells expressing NKG2D ligands can cause global NK cell desensitization in vitro ( Coudert et al . , 2008 ) . Furthermore , transgenic over-expression of NKG2D ligands is associated with NK cell desensitization as well ( Oppenheim et al . , 2005 ) . Thus , endothelial cells may not have unique desensitizing activity per se but rather could have an important role in NK desensitization by virtue of expressing NKG2D ligands ( which most healthy cells lack ) and/or their unique physiological role in extravasation , intravasation , and circulation . Furthermore , it is possible that the impact of endothelial cell RAE-1 on tumor rejection is to desensitize NK cells before they encounter tumor cells , whereas NKG2D ligands on tumor cells might affect NK cells differently after extravasation; these dynamics may also depend on the inflammatory state of the tumor microenvironment . Clearly , more work is needed to better understand the mechanisms that regulate NK cell activation and desensitization and its relation to tumor rejection . Our data may have translational implications . We hypothesize that treatments to disrupt interactions between NK cells and NKG2D ligands on vasculature , while preserving interactions with NKG2D ligands on tumor cells , may have powerful therapeutic benefits . This could be accomplished by antibody blockade of ligands expressed by endothelial cells but not tumors . Understanding the mechanisms supporting endothelial NKG2D ligand expression could also reveal targets for specific pharmacological inhibition . C57BL/6J mice were bred from mice obtained from The Jackson Laboratory ( Bar Harbor , ME ) . NKG2D-KO mice were previously generated in our lab as described ( Guerra et al . , 2008 ) . RAE-1-KO mice were previously generated in our lab using CRISPR-Cas9 and guide RNAs targeting the open-reading frames of the Raet1d and Raet1e genes , as described ( Deng et al . , 2015 ) . KP mice contain inducible mutations in the proto-oncogene Kras and the tumor suppresser gene Trp53 and were bred from mice obtained from The Jackson Laboratory . All mice were maintained at the University of California , Berkeley in accordance with guidelines from the Animal Care and Use Committee . Sex- and age-matched ( 8- to 12-week-old ) mice were used for the experiments . In most experiments , KO mice were compared with cousin WT controls ( i . e . derived from the same grandparents ) and were co-housed to minimize confounding genetic and environmental variables . In vivo blockade of NKG2D ligands was achieved by peritoneal injection of 100 μg of the indicated antibody . Antibodies against RAE-1δ ( clone 199205 ) and RAE-1ε ( clone 205001 ) were obtained from R&D Systems ( Bio-Techne Corp , Minneapolis , MN ) . Anti-MULT1 ( clone 1D6 ) was a kind gift from Stipan Jonjic . We confirmed the blocking efficacy of these reagents before in vivo use ( Figure 1—figure supplement 1A ) . All transplant tumor models were injected i . v . ( metastasis model ) or s . c . ( solid tumor model ) by injection with an insulin syringe ( BD Biosciences , San Jose , CA ) . For injection , tumor cells were suspended in 100 μl PBS and injected by the indicated route . In some groups , NK cells were depleted by twice weekly injections of 100 μg anti-NK1 . 1 antibody ( clone PK136 ) beginning the day before tumor injection . Tumor growth was measured by caliper , and tumor volume was calculated using the modified ellipsoid formula: V = 0 . 5 x [ ( length + width ) /2]3 . In some experiments , adult splenocytes were labeled with 1 μg/ml CFSE ( Biolegend , San Diego , CA ) according to the manufacturers instructions , and 5 × 107 labeled cells were transferred to recipient mice by i . v . injection . For bone marrow chimera experiments , recipient mice were lethally irradiated with 11 Gy ( 6 Gy +5 Gy split dose ) using an X-ray irradiator and then given 1 × 107 donor bone marrow cells by i . v . injection . Mice were allowed to recover and reconstitute for at least 8 weeks before analysis or tumor injection . Spleens were dissociated by mashing through a 70 μM filter into PBS . To dissociate lymph nodes , lungs , heart , liver , or tumors for flow cytometry , organs were gently dissociated with according to a published protocol optimized for stromal cell analysis ( Broggi et al . , 2014 ) . Briefly , organs were mechanically dissociated using a sharp blade , and then incubated in complete media with 3 . 5 mg/ml Collagenase D , 1 mg/ml Collagenase IV for 30 min at 37°C with rotation . Cells were then pipetted up and down rigorously 100 times to create a single cell suspension , with additional rounds of 10-min incubation followed by pipetting as needed . Total RNA was isolated from cells using the RNeasy kit ( Qiagen , Hilden , Germany ) and converted to cDNA using the iScript system ( Bio-Rad , Hercules , CA ) according to the manufacturer’s instructions . cDNA was subjected to real-time PCR using SsoFast EvaGreen supermix ( Bio-Rad ) in the presence of primers to amplify Klrk1 mRNA , or the transcripts of the housekeeping genes β-actin and Rpl19 , in a CFX96 RT-qPCR thermocycler ( BioRad ) . Relative mRNA values for Klrk1 were normalized to the levels of the housekeeping genes , using CFX96 software . All cell cultures were performed in a humidified 37°C incubator at 5% CO2 . Cells were cultured in DMEM or RPMI media ( Life Technologies , Carlsbad , CA ) supplemented with 5% fetal calf serum ( Omega Scientific , Tarzana , CA ) , 0 . 2 mg/ml glutamine , 100 U/ml penicillin , 100 μg/ml streptomycin ( Sigma–Aldrich , St . Louis , MO ) , 10 μg/ml gentamicin sulfate ( Lonza , Basel , Switzerland ) , and 20 mM HEPES ( Thermo Fisher Scientific , Waltham , MA ) . For all flow cytometry experiments , single-cell suspensions were generated and incubated for 20 min with supernatant from the 2 . 4G2 hybridoma to block FcγRII/III receptors , followed by incubation with fluorochrome- or biotin-conjugated specific antibodies for an additional 20 min . In some experiments , an additional incubation with fluorophore-conjugated streptavidin ( Biolegend ) was performed . For intracellular staining , cells were fixed and permeabilized using the Cytofix/Cytoperm kit ( BD Biosciences ) before incubation with intracellular antibodies . Samples were analyzed on a LSR Fortessa or LSR Fortessa X20 ( BD Biosciences ) and data were analyzed with FlowJo software ( Tree Star Inc . ) . Dead cells were excluded from analysis using DAPI ( Biolegend ) or Live-Dead fixable dead cell stain kits ( Molecular Probes ) following the manufacturer’s instructions . In some experiments , NK cells were sorted using the Influx Cell Sorter ( BD Biosciences ) . We used the following antibodies: from Biolegend; anti-CD3ε ( clone 145–2 C11 ) , anti-CD4 ( clone GK1 . 5 ) , anti-CD11b ( clone M1/70 ) , anti-CD19 ( clone 6D5 ) , anti–IFN-γ ( clone XMG1 . 2 ) , anti-NKp46 ( clone 29A1 . 4 ) , anti–NK1 . 1 ( clone PK136 ) , anti-Ter119 ( clone TER-119 ) , anti-CD31 ( clone 390 ) , anti-Podoplanin ( clone 8 . 1 . 1 ) , anti-HEV ( clone MECA-79 ) , mouse IgG2b isotype control , and rat IgG2b isotype control; from eBioscience; anti-CD27 ( clone 37 . 51 ) , anti-CD45 . 1 ( clone A20 ) , anti-CD45 . 2 ( clone 104 ) , anti-CD107a ( clone 1D4B ) , anti-NKG2D ( clone MI-6 ) ; from R and D Systems; mouse NKG2D-Fc fusion protein; from Jackson ImmunoResearch; goat anti-mouse IgG . For flow cytometry analysis of RAE-1ε , we used the EZ-Link-Sulfo-NHS-LC biotin kit ( Thermo Fisher ) to biotinylate clone 205001 mAb ( from R and D Systems ) , the same clone used for in vivo NKG2D ligand blockade . To analyze the responsiveness of NK cells ex vivo , 96-well high-binding flat-bottom plates ( Thermo Fisher ) were coated overnight with PBS plus the indicated antibody against NK activating receptors . Plate-bound antibodies were coated at the following concentrations: anti-NK1 . 1: 50 μg/ml; anti-NKG2D: 5 μg/ml; anti-NKp46: 5 μg/ml . Plates were washed three times with PBS before stimulation . Single-cell suspensions were generated from the indicated tissue and cultured in the coated plates for 5 hr in the presence of Golgi-Stop and Golgi-Plug ( 1:1000 each ) ( BD Biosciences ) , 1000 U/ml human IL-2 , and fluorophore-conjugated anti-CD107a ( 0 . 5 μg/ml ) ( Biolegend ) . After stimulation , cells were stained for extracellular markers to identify NK cells and then subjected to intracellular staining for IFN-γ , followed by flow cytometry analysis . To analyze NK killing of tumor cells in vitro , NK cells from mice of the indicated genotype were pre-activated by a single i . p . injection of 200 μg high-molecular weight Poly I:C ( Invivogen , San Diego , CA ) ; peritoneal wash cells were harvested 2 days later and pooled by genotype . YAC-1 target cells were labeled with 20 uCi 51Cr , washed three times , and plated in 100 μl medium at 1 × 104 cells per well in a 96-well round bottom plate . 100 μl of medium alone ( spontaneous release ) , 2% Triton-X-100 ( maximum release ) , or effector cells at the indicated effector:target ratios were added to targets , in quadruplicate . Cells were incubated for 4 hr at 37°C . Cells were then pelleted , and 100 μl of supernatant was analyzed by gamma counting . The spontaneous release was , in all cases , <20% of maximum release . The percent specific 51Cr release was calculated according to the following formula: % specific lysis = 100 x ( experimental release – spontaneous release ) / ( detergent release – spontaneous release ) . All statistical analyses were conducted using Prism software ( Graphpad , La Jolla , CA ) , as indicated in the figure legends . Statistical significance is indicated as follows: p<0 . 05 , p<0 . 01 , **p<0 . 001 . For most data sets , pilot experiments were performed with a small sample size ( usually n = 3 ) to determine approximate experimental variances and effect magnitudes , and this information was used to determine sample sizes for subsequent experiments .	White blood cells called “natural killer cells” are part of the first line of immune defense . Often called NK cells for short , one job of these cells is to help prevent cancer by killing tumor cells . If an NK cell spots a tumor cell , it must become energized so that it can deliver the killing blow , which comes in the form of a packet of cell-killing “cytotoxic” granules . Yet tumor cells look very similar to healthy cells , and NK cells must be able to tell the difference to be effective . Molecules on the outer surface of the NK cell control how the cell recognizes tumors , and deliver the signals the cell needs to become energized . One of these surface molecules is called NKG2D . It interacts with “partner” molecules found on the surface of cancer cells and tells the NK cell to attack . These partner molecules are not usually found on healthy cells , helping the immune system to tell the difference . After NKG2D interacts with its partner molecules , it moves inside the NK cell . This makes the cell less able to become energized . If the NK cells do not encounter any partner molecules in healthy mice , blocking the interactions should have no effect on NKG2D levels . But now , Thompson et al . find that blocking one of these interactions increased the levels of NKG2D on the surface of NK cells in healthy mice . Further experiments revealed that NK cells in mice constantly encounter an NKG2D partner molecule called RAE-1ε . A search for the source of RAE-1ε in healthy mice pointed to blood vessels inside the lymph nodes . NK cells pass through theses organs as part of their normal path around the body . Thompson et al . also saw that NK cells from healthy mice were less responsive than NK cells from mutant mice that lacked RAE-1ε . As a result of their encounters with RAE-1ε in healthy mice , the NK cells were less able to kill tumor cells . Blocking the interaction between NKG2D and RAE-1ε in mice re-energized their NK cells . More cells were able to enter tumors in these mice and the cells became better at killing tumors . Together these findings increase the current understanding of the biological processes that control NK cells . Further research may lead to new treatments for diseases like cancer . But first , scientists need to find out whether NK cells behave in the same way in humans as they do in mice . If so , developing ways to block the interaction could re-energize human NK cells to better kill cancer cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "immunology", "and", "inflammation", "cancer", "biology" ]	2017	Endothelial cells express NKG2D ligands and desensitize antitumor NK responses
Our ability to rapidly detect threats is thought to be subserved by a subcortical pathway that quickly conveys visual information to the amygdala . This neural shortcut has been demonstrated in animals but has rarely been shown in the human brain . Importantly , it remains unclear whether such a pathway might influence neural activity and behavior . We conducted a multimodal neuroimaging study of 622 participants from the Human Connectome Project . We applied probabilistic tractography to diffusion-weighted images , reconstructing a subcortical pathway to the amygdala from the superior colliculus via the pulvinar . We then computationally modeled the flow of haemodynamic activity during a face-viewing task and found evidence for a functionally afferent pulvinar-amygdala pathway . Critically , individuals with greater fibre density in this pathway also had stronger dynamic coupling and enhanced fearful face recognition . Our findings provide converging evidence for the recruitment of an afferent subcortical pulvinar connection to the amygdala that facilitates fear recognition . 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 minor issues remain unresolved ( see decision letter ) . Decades ago , rodent research uncovered a subcortical pathway to the amygdala that rapidly transmits auditory signals of threat even when the auditory cortex is destroyed ( Ledoux , 1998 ) . Since this discovery , researchers have sought an equivalent visual pathway that might explain how it is that people with a lesioned primary visual cortex can still respond to affective visual stimuli that they cannot consciously see ( Tamietto et al . , 2010 ) . The superior colliculus , pulvinar , and amygdala have been identified as nodes of a human subcortical route to the amygdala that bypasses the cortex ( Morris et al . , 1999 ) . These subcortical areas consistently coactivate in cortically blind patients ( Pegna et al . , 2005 ) – as well as in healthy adults ( Vuilleumier et al . , 2003; Morris et al . , 1999 ) – when they view emotional stimuli , such as angry or fearful faces . Magnetoencephalography studies using computational modelling have investigated whether the activation of these subcortical nodes is causally related . These studies have consistently found evidence for a forward connection between the pulvinar and amygdala ( McFadyen et al . , 2017; Garvert et al . , 2014; Rudrauf et al . , 2008 ) . The dynamic causal relationship between the superior colliculus and the pulvinar , however , remains unexplored in the human brain ( Soares et al . , 2017 ) . The pulvinar also has several functional and cytoarchitectural subregions ( Barron et al . , 2015 ) and it is unclear how these connect to the superior colliculus and the amygdala and what roles these subregions may play in mediating transmission along the subcortical route ( Koller et al . , 2018; Pessoa and Adolphs , 2010 ) . As such , the hypothesis that the subcortical route rapidly transfers information from the retina to the amygdala without interference has been heavily criticised ( Pessoa and Adolphs , 2010; Pessoa and Adolphs , 2011 ) . Furthermore , the pulvinar is highly connected with a widespread network of cortical regions that may contribute to transmission along the subcortical route ( Bridge et al . , 2016; Zhou et al . , 2016 ) . Hence , it remains unknown whether the functional activation of the human superior colliculus , pulvinar and amygdala during affective processing bears any relation to an underlying structural pathway ( Pessoa and Adolphs , 2010 ) . Recent animal research has revealed several potential direct subcortical pathways that have a causal relationship with fearful behaviour in response to visual threats ( Zhou et al . , 2017; Wei et al . , 2015; Shang et al . , 2015 ) . In the absence of relevant postmortem human research , however , our anatomical knowledge of the human subcortical route to the amygdala can only be derived from tractography of diffusion-weighted images ( DWI ) . Tamietto et al . , 2012 examined DWIs from a blindsight patient whose left primary visual cortex was destroyed . The white matter structure of the subcortical route was estimated for the patient and for ten healthy , age-matched controls . Critically , the subcortical route had greater fractional anisotropy in the patient’s damaged hemisphere , suggesting a neuroplastic increase in structural connectivity to compensate for the disrupted cortical pathways ( Tamietto et al . , 2012 ) . In a similar study , Rafal et al . ( 2015 ) used tractography to investigate the subcortical route in 20 healthy humans and eight macaques . The subcortical route was reconstructed in both hemispheres for 19 of the 20 human participants and 7 of the eight macaques ( Rafal et al . , 2015 ) . Notably , this sample of human participants was recently expanded and re-examined , further demonstrating that individuals with greater fractional anisotropy along the subcortical route also had a stronger bias toward threat when making saccades to scenes ( Koller et al . , 2018 ) . Diffusion tractography may grant insight into the strength of anatomical connectivity between regions , but it cannot reveal the direction of information transfer nor can it be used as direct evidence alone for the anatomical existence of a neural pathway . The anatomical presence and the direction-specific neural flow of emotional visual information along the subcortical route has never been concurrently investigated in humans to definitively show that the subcortical route is a direct , afferent pathway specifically associated with fear ( Pessoa and Adolphs , 2010; Pessoa and Adolphs , 2011 ) . Such a finding would have important implications for the very foundation of visual threat perception , given this pathway’s potential for rapid information transfer ( McFadyen et al . , 2017; Silverstein and Ingvar , 2015 ) and unconscious processing ( Tamietto et al . , 2010 ) . Here , we aimed to comprehensively investigate this putative amygdala pathway in a large sample of over 600 healthy human adults from the Human Connectome Project ( HCP ) dataset using a multimodal imaging approach to encompass structure , function and behaviour . First , we used DWI to reconstruct the subcortical route from the superior colliculus to the amygdala , via the pulvinar , and estimated its fibre density in a large sample . Next , we modelled the direction-specific flow of haemodynamic responses to faces , testing whether a functional subcortical route is recruited to transmit information toward the amygdala . Finally , we asked whether the fibre density of the subcortical route predicts both fearful face recognition as well as the strength of dynamic coupling between the superior colliculus , pulvinar , and amygdala . The first step in our investigation was to evaluate the evidence for an anatomical subcortical route to the amygdala in the healthy human brain . We exploited high-quality neuroimaging data from a large sample of 622 participants made available by the HCP ( Van Essen et al . , 2013 ) . We then reconstructed the white matter structure of the subcortical route using two complementary tractography methods for cross-validation . We began with global tractography , a Bayesian approach to reconstructing whole-brain fibre configurations that best explain DWI data ( see Materials and Methods for details ) . We discovered that the superior colliculus ( SC ) was connected to the pulvinar ( PUL; fibre counts for PUL; left: M = 13 . 23 , SD = 5 . 56 , right: M = 13 . 00 , SD = 5 . 59 , minimum of 2 fibres per participant ) . The pulvinar and the amygdala were also connected ( fibre counts for left: M = 5 . 33 , SD = 2 . 79 , and right: M = 6 . 75 , SD = 2 . 90 ) , with most participants having at least one connecting fibre ( zero fibres for left PUL-AMG for eight participants – only 1 . 28% of total sample ) . Thus , this relatively conservative method of fibre reconstruction ( as it takes into account the entire brain ) can reliably detect evidence for a subcortical route across a large sample of participants . We used the probabilistic JHU DTI-based white matter atlas ( Mori et al . , 2009 ) , implemented in FSL , to examine any overlap between 20 major fasciculi and the globally reconstructed fibres . After warping the tractograms into standard space and converting them into track density images , we calculated the total fibre density within each fasciculus . This revealed that up to 60% of the subcortical route overlapped with major fasciculi , mainly the anterior thalamic radiation and the corticospinal tract , as well as the inferior longitudinal and fronto-occipital fasciculi . For the SC-PUL pathway , the major overlap was found in the anterior thalamic radiation in the left ( M = 56 . 11% , SD = 15 . 56% , range = 9 . 46% to 100% ) and right ( M = 55 . 78% , SD = 16 . 97% , range = 4 . 81% to 96 . 55% ) hemispheres , followed by the corticospinal tract ( left: M = 4 . 82% , SD = 6 . 08% , range = 0% to 33%; right: M = 21 . 06% , SD = 12 . 75% , range = 0% to 69 . 09%; see Figure 1 ) . All other fasciculi had mean track densities less than 0 . 06% of the full SC-PUL pathway . Track density of the PUL-AMG pathway was mostly found in the corticospinal tract ( left: M = 33 . 69% , SD = 18 . 56% , range = 0% to 88 . 89%; right: M = 36 . 73% , SD = 17 . 80% , range = 0% to 82 . 35% ) , followed by the anterior thalamic radiation ( left: M = 20 . 58% , SD = 14 . 10% , range = 0% to 70%; right: M = 11 . 32% , SD = 8 . 88% , range = 0% to 52 . 75% ) . There was also some overlap with the inferior longitudinal fasciculus ( left: M = 5 . 44% , SD = 8 . 60% , range = 0% to 65 . 52%; right: M = 3 . 46% , SD = 5 . 63% , range = 0% to 40 . 68% ) and the inferior fronto-occipital fasciculus ( left: M = 1 . 49% , SD = 6 . 73% , range = 0% to 86 . 96%; right: M = 7 . 24% , SD = 10 . 33% , range = 0% to 75 . 72% ) . Mean track densities in all other fasciculi were less than 0 . 30% . After covarying out head motion and removing four participants with outlying standardised residuals ( z-score threshold ±3 ) , we established that there were significantly more fibres connecting the SC and PUL ( M = 13 . 119 , 95% CI = [12 . 738 , 13 . 500] ) than the PUL and AMG ( M = 6 . 040 , 95% CI = [5 . 867 , 6 . 214]; F ( 1 , 616 ) = 433 . 286 , p=2 . 842 × 10−73 , ηp2 = . 413 ) . We also found a hemispheric lateralisation , such that there were more reconstructed fibres for the right ( M = 9 . 879 , 95% CI = [9 . 624 , 10 . 135] ) than the left ( M = 9 . 280 , 95% CI = [9 . 023 , 9 . 537] ) hemisphere ( F ( 1 , 616 ) = 7 . 583 , p=0 . 006 , ηp2 = . 012 ) , specifically for the PUL-AMG pathway ( F ( 1 , 616 ) = 16 . 025 , p=7 . 000 × 10−5 , ηp2 = . 025; t ( 617 ) = −9 . 785 , p=4 . 070 × 10−18 , 95% CI [−1 . 714 , –1 . 141] ) . To uncover more anatomical features of the reconstructed fibres , we used subregion-specific masks of the amygdala ( basolateral , centromedial , and superficial ) and the pulvinar ( anterior , medial , superior , inferior , and lateral; see Materials and methods for ROI specification details ) to determine where the reconstructed fibres terminated . This masking approach revealed that the global tractography fibres present between the SC and PUL connected predominantly to the inferior and anterior pulvinar ( see Appendix 1—tables 3 to 5 for detailed statistics ) . Between the PUL and AMG , fibres terminated almost exclusively in the inferior PUL and then predominantly in the basolateral AMG . Hence , the inferior pulvinar served as the connecting node between the SC and the basolateral AMG for the globally reconstructed subcortical pathway . To assess the validity of our findings we used a second tractography method , namely ‘local’ probabilistic streamline tractography , as used by Rafal et al . ( 2015 ) to reconstruct the subcortical route to the amygdala ( Rafal et al . , 2015 ) . We generated streamlines between our regions of interest ( ROIs ) and found that the superior colliculus connected to the pulvinar ( streamline counts for left: M = 1403 . 32 , SD = 417 . 16 , right: M = 111 . 59 , SD = 358 . 60 , minimum of six streamlines per participant ) and the pulvinar connected to the amygdala ( left: M = 575 . 42 , SD = 203 . 03 , right: M = 575 . 42 , SD = 248 . 85 , minimum 66 streamlines per participant ) . To evaluate whether these streamlines counts were reconstructed significantly above chance , we compared the numbers with those produced by a null distribution algorithm ( Morris et al . , 2008 ) . We found that the number of streamlines was significantly different from chance for each connection , as determined by a series of paired two-sided t-tests ( see Appendix 1—table 9 ) , suggesting that the DWI data produced meaningful streamlines between our ROIs . We employed a recently developed method , SIFT2 , which estimates the apparent fibre density of the streamlines connecting two regions of interest . This method more accurately represents the true underlying white matter structure ( Smith et al . , 2015 ) . The apparent fibre density of the streamlines generated using local tractography followed the same pattern as the global tractography fibre counts , such that there was greater fibre density for the SC-PUL connection ( M = 5 . 793 , 95% CI = [5 . 663 , 5 . 923] ) than the PUL-AMG connection ( M = 4 . 461 , 95% CI = [4 . 368 , 4 . 554]; F ( 1 , 607 ) = 69 . 586 , p=4 . 930 × 10−16 , ηp2 = . 103 ) , after accounting for head motion and removing 13 outliers according to their residuals . Fibre density was also greater on the right ( M = 4 . 935 , 95% CI = [4 . 822 , 5 . 048] ) than the left ( M = 3 . 987 , 95% CI = [3 . 889 , 4 . 086] ) for the PUL-AMG connection ( t ( 608 ) = −18 . 205 , p=1 . 960 × 10−59 , 95% CI [−1 . 050 , –0 . 845] ) while , in contrast , there was greater fibre density for the left than right SC-PUL connection ( t ( 608 ) = 10 . 749 , p=8 . 600×10−25 , 95% CI = [0 . 742 , 1 . 073]; F ( 1 , 607 ) = 162 . 475 , p=3 . 828 × 10−33 , ηp2 = . 211 ) . Taken together , our tractography analyses provide strong convergent evidence for a subcortical white matter pathway to the amygdala in the human brain . Like in the global tractography , we investigated the overlap between the locally generated tracks and known white matter fasciculi . The pattern of results was the same , with up to 60% of fibres traversing the anterior thalamic radiation , corticospinal tract , and inferior longitudinal and fronto-occipital fasciculi . For the SC-PUL pathway , the majority of track density was found in the anterior thalamic radiation ( left: M = 52 . 85% , SD = 11 . 70% , range = 16 . 55% to 96 . 96%; right: M = 58 . 21% , SD = 16 . 24% , range = 5 . 49% to 89 . 30% ) and the corticospinal tract ( left: M = 12 . 89% , SD = 8 . 78% , range = 0 . 18% to 37 . 84%; right: M = 32 . 35% , SD = 12 . 49% , range = 0 . 56% to 63 . 64%; see Figure 1 ) . For the PUL-AMG pathway , the majority was found in the corticospinal tract ( left: M = 32 . 29% , SD = 15 . 58% , range = 0 . 59% to 65 . 25%; right: M = 37 . 38% , SD = 16 . 21% , range = 0 . 32% to 65 . 99% ) , followed by the anterior thalamic radiation ( left: M = 16 . 47% , SD = 9 . 20% , range = 5 . 24% to 59 . 52%; right: M = 7 . 00% , SD = 3 . 49% , range = 1 . 46% to 50 . 09% ) , and then the inferior longitudinal ( left: M = 5 . 69% , SD = 7 . 92% , range = 0% to 50 . 75%; right: M = 3 . 96% , SD = 6 . 33% , range = 0% to 47 . 51% ) and fronto-occipital ( left: M = 1 . 54% , SD = 4 . 21% , range = 0% to 72 . 79%; right: M = 7 . 96% , SD = 10 . 99% , range = 0% to 79 . 60% ) fasciculi . Mean track densities were lower than 0 . 20% and 0 . 01% in other fasciculi for SC-PUL and PUL-AMG , respectively . We also examined which subregions of the pulvinar and amygdala the seeded probabilistic tracks terminated in . For the SC-PUL pathway , the greatest number of streamlines terminated in the anterior PUL , followed by the inferior pulvinar ( see Appendix 1—tables 6 to 8 for detailed statistics ) , consistent with the global tractography . Also like the global tractography , the local tractography fibres between the PUL and AMG terminated almost exclusively in the inferior PUL . For the AMG , however , fibres terminated predominantly in the basolateral subregion in the left hemisphere ( consistent with the global tractography ) but in the centromedial amygdala on the right . We wanted to translate our work in humans to animal research that has demonstrated clear relationships between the anatomical presence of a subcortical route and fearful behaviour ( Zhou et al . , 2017; Wei et al . , 2015; Shang et al . , 2015 ) . To this end , we examined behavioural data from an out-of-scanner task , the Penn Emotion Recognition task , that assessed a different component of face processing than the in-scanner task ( analysed below ) . In the Penn Emotion Recognition task , participants were serially presented with 40 faces that were either happy , sad , angry , fearful , or neutral ( 8 faces presented in each category ) . Participants were most accurate with identifying the emotional expression of happy faces ( M = 7 . 96 , SD = 0 . 21 ) , followed by neutral ( M = 7 . 22 , SD = 1 . 18 ) , and then fearful faces ( M = 7 . 02 , SD = 1 . 03 ) . Recognition was poorest for angry ( M = 6 . 86 , SD = 0 . 98 ) and sad faces ( M = 6 . 82 , SD = 1 . 12 ) . We then investigated the association between these scores ( see Figure 2A ) with the fibre density of the subcortical route . We chose not to include happy or neutral expressions in our analysis because the data were substantially negatively skewed ( skewness for: happy = −5 . 821; neutral = −2 . 053; angry = −0 . 719; fearful = −1 . 188; sad = −1 . 090 ) . Thus , we entered fibre density measures for the SC-PUL and PUL-AMG pathways into two separate multivariate regressions ( one per tractography method , to reduce collinearity ) with recognition accuracy scores for fearful , angry , and sad faces as covariates , plus head motion as a control covariate . We removed outliers ( four for global tractography , 15 for local tractography ) with z-scored residuals ± 3 . While there were no significant multivariate relationships between global tractography and emotion recognition ( see Appendix 1—tables 10 and 11 for detailed statistics ) , there was a significant relationship between local tractography and recognition accuracy for fearful faces ( F ( 4 , 598 ) = 2 . 501 , p=0 . 042 , Wilk’s Λ = 0 . 984 , np2 = 0 . 016; see Figure 2B ) . This was driven predominantly by fibre density of the left ( β = 0 . 140 , p=0 . 004 ) and right ( β = 0 . 143 , p=0 . 012 ) PUL-AMG connections’ local fibre density . The local fibre density of the left and right SC-PUL did not contribute significantly to the model . These results suggest that the fibre density of the PUL-AMG half of the subcortical route is associated with fearful face recognition more so than with other negative ( sad ) or threatening ( angry ) emotional expressions . We used dynamic causal modelling to infer the dynamic ( or effective ) connectivity between each node of the subcortical route and determine the directionality of the functional interactions occurring along the anatomical pathway mapping described above . First , it was necessary to establish any differences in functional activation within these nodes . To do this , we used the ‘Emotion’ task from the HCP battery of fMRI experiments , in which participants performed a matching task using images of faces or shapes . We contrasted activation in face versus shape blocks and reviewed the results at the whole-brain group level across all 622 participants , p<0 . 05 FWE ( see Figure 3 ) . This revealed a network of significant BOLD clusters spread across occipital , temporal , frontal , parietal , and subcortical areas , replicating previous work with this dataset ( Barch et al . , 2013 ) . Critically , the most significant cluster included the left and right amygdala as well as the left and right fusiform gyri ( FG ) and inferior occipital gyri ( IOG ) . These latter two regions are key nodes in the cortical visual processing stream for faces , which may feed information forward to the amygdala ( Tamietto et al . , 2010 ) . We used the SPM Anatomy Toolbox to confirm the anatomical positions of our functional activations . In the absence of an anatomical template for the superior colliculus and the pulvinar , we masked the map of statistically significant voxels ( p < 0 . 05 , FWE ) with our a priori manual anatomically defined superior colliculi mask and functionally-defined pulvinar masks from Barron et al . ( 2015 ) ; see Materials and Methods for ROI generation ) . This revealed significant voxels in each area ( proportion of significant voxels within each mask: left SC = 65 . 08% , right SC = 73 . 55% , left PUL = 36 . 13% , right PUL = 51 . 49% ) . Therefore , the faces-vs-shapes fMRI HCP task established functional activation in the three subcortical nodes of interest , as well as two nodes of a potential cortical pathway to the amygdala for conveying information about emotional faces . After observing significant BOLD signal in regions along the subcortical route as well as in other visual cortical areas , we next asked whether these regions were dynamically connected . We designed a space of testable models that mapped onto the functional hypothesis of a subcortical route to the amygdala that operates alongside a cortical visual pathway and is modulated by faces . Due to the presence of the IOG and FG in the whole-brain corrected fMRI activation and their known roles in face processing ( Johnson , 2005 ) , we defined several plausible functional cortical connections to the amygdala . These consisted of reciprocal pathways between IOG and FG , FG and amygdala , IOG and amygdala , as well as pulvinar and IOG ( see Figure 4 ) . Note that , while previous research has defined motion-related area V5/MT as a significant component of the pulvinar’s subcortical visual network ( Zhou et al . , 2017 ) , we did not observe strong involvement of this area in the faces-vs-shapes fMRI task ( 12% probability in cluster 37 with two voxels ) . Hence , we omitted area V5/MT from our model space . We named models containing both a cortical and subcortical route to the amygdala as ‘Dual’ models , whereas models in which the subcortical route was absent were named ‘Cortical’ . Our model space also included different sources of visual input , namely to the superior colliculus , pulvinar , or both , given that the pulvinar also receives direct retinal input ( Cowey et al . , 1994 ) as well as input via the superior colliculus ( Berman and Wurtz , 2011 ) . This gave us six families of models: 1 ) Cortical with SC input , 2 ) Cortical with PUL input , 3 ) Cortical with SC and PUL input , 4 ) Dual with SC input , 5 ) Dual with PUL input , and 6 ) Dual with SC and PUL input . We considered all possible combinations of forwards and reciprocal ( forwards and backwards ) cortical and subcortical connections , giving us a comprehensive model space of 102 models . Of the available 622 participants , we conducted dynamic causal modelling on a subset of 237 participants who had sufficient above-threshold activation in all ROIs ( these were defined by the subcortical masks used thus far and by spheres surrounding the coordinate of peak group BOLD signal in the IOG and FG; see Materials and Methods for more details ) . We conducted Bayesian model selection on the model space ( grouped by families ) to estimate how well the models explained the data , taking into account model complexity . We used the random effects implementation to account for potential individual differences in the recruitment of a subcortical pathway for viewing faces ( Stephan et al . , 2009 ) . The winning family was the ‘Dual with SC and PUL input’ family ( expected probability = 67 . 34% , exceedance probability = 100% ) and the winning model across the entire model space was also within this family ( expected probability = 21 . 24% , exceedance probability = 98 . 01% , protected exceedance probability = 98 . 18%; see Figure 4 ) . This model included reciprocal cortical connections between IOG and FG and between FG and the amygdala . It also included a forwards-only subcortical route from the superior colliculus to the pulvinar to the amygdala , with input to both the superior colliculus and the pulvinar . The Bayesian Omnibus Risk score was p=1 . 78×10−124 , indicating a very small chance that the winning model was indistinguishable from all models tested ( Rigoux et al . , 2014 ) . We replicated this finding ( same winning family and winning model ) on a subsample consisting of only the unrelated ( i . e . non-sibling ) participants within this group ( 49 participants; see Appendix 1 ) . The winning model revealed that the functional network that best explained the BOLD responses in our sample of 237 participants included visual inputs to the superior colliculus and pulvinar , forward connections from superior colliculus to the amygdala via the pulvinar , and recurrent interactions between IOG and FG , as well as between FG and amygdala . To extrapolate this finding to the general population and assess the consistency of dynamic coupling at each individual connection , we performed inferential statistics ( t-tests ) on the parameter estimates of each connection within the winning model ( i . e . connectivity strength , in their natural space ) . We looked at the connectivity modulation parameters that represent the change in connection strength caused by the effect of faces . We removed extreme outliers ( >3 SDs from mean ) participants from each connection ( M = 5 . 25 , range = 3 to 8 participants excluded from sample of 237 ) and found that all connectivity modulations were significant ( one sample t-tests against a test value of 1; see Appendix 1—table 13 for detailed statistics ) except for the backward connection from left and right FG to left IOG ( see Figure 5 ) . These results suggest that the modulation of these connections by faces was consistently strong and so we can infer that a subcortical route for processing faces is likely present within the general population . Our findings from tractography , fMRI , and dynamic causal modelling provide convergent evidence for a subcortical route to the amygdala in humans . The final question we set out to answer was whether this converging evidence was correlated , such that participants with stronger structural connectivity also had stronger effective connectivity . In other words , we asked whether the structural connectivity along the subcortical amygdala route enables functional interactions amongst the nodes that lie within it . We computed eight partial correlations ( with head motion as a control covariate ) to examine the relationship between each parameter estimate and the corresponding global fibre count and local summed weights per connection ( left and right SC-PUL and PUL-AMG ) . After removing multivariate outliers ( leaving N = 213; see Appendix 1—table 12 for details ) , we discovered that participants with more global fibres also had greater modulatory activity for the right ( r = 0 . 180 , p = 0 . 004 , pbonf = 0 . 032; see Figure 6 ) but not the left ( r = 0 . 095 , p = 0 . 084 , pbonf = 0 . 672 ) PUL-AMG connection . The SC-PUL connection was not significantly related to its corresponding DCM parameters for global ( left: r = −0 . 022 , p = 0 . 627 , pbonf = 1 . 000; right: r = 0 . 002 , p = 0 . 488 , pbonf = 1 . 000 ) or local ( left: r = −0 . 101 , p = 0 . 928 , pbonf = 1 . 000; right: r = −0 . 028 , p = 0 . 659 , pbonf = 1 . 000 ) tractography . Note that we successfully replicated this finding within a subsample of unrelated participants ( 49 participants; see Appendix 1 ) . Thus , our study is the first to successfully harmonise functional and structural information about the subcortical pulvinar connection to the amygdala . The elusive subcortical route to the amygdala has posed a unique challenge in studies of the human brain , due to its depth and its fast activation . Evidence has accumulated over recent years across many studies using various neuroimaging modalities showing that this pathway may underlie primitive threat-related behaviour . These studies , however , often take a unimodal approach on typically small samples , making it difficult to relate the specific structural connections between superior colliculus , pulvinar , and amygdala to observed functional brain activity and behavioural output . Our study , which used a large sample of participants from the HCP , supports the existence of a subcortical pulvinar connection to the amygdala in the healthy human adult brain that facilitates dynamic coupling between these regions and also enhances fear recognition . We reconstructed the subcortical route to the amygdala using sophisticated tractography methods and found that the white matter fibre density of the pulvinar-amygdala connection significantly predicted individuals’ ability to recognise fearful faces . We then computationally modelled the functional neural networks along this structurally connected network that were engaged while people viewed emotional faces . We found that it was more likely for the network to include a subcortical visual route to the amygdala than a cortical route alone . Finally , we revealed converging evidence from structural and effectivity connectivity , such that the fibre density of the right pulvinar to amygdala pathway was positively correlated with the strength of the dynamic coupling ( i . e . effective connectivity ) between these regions . This study marks the first time that structural and effective connectivity have been concurrently investigated in the one large sample to address the controversy on the existence and functional role of the putative subcortical route to the amygdala . Up to 60% of its fibre density overlapped with major fasciculi , including the corticospinal tract , anterior thalamic radiation , inferior longitudinal fasiculus , and inferior fronto-occipital fasciculus . Tractography of diffusion images is susceptible to both false positives and false negatives and thus is seldom used in isolation to determine the existence of particular neuroanatomical pathways ( Jbabdi and Johansen-Berg , 2011 ) . We established the validity of our tractographically reconstructed subcortical route by directly relating our measures of fibre density to both behaviour and effective connectivity , as well as by using two different tractography methods . Had the fibre density measures been simply due to noise , we would not have expected these theoretically relevant relationships with fearful face processing to emerge within this large sample of individuals . Notably , these intermodal relationships were only found for the pulvinar-amygdala connection , despite there being greater fibre density between the superior colliculus and the pulvinar and this connection being present in the winning dynamic causal model . One explanation for this is that we had relatively less BOLD signal-to-noise ratio in the superior colliculus due to its small size and proximity to major blood vessels in the brain stem ( Wall et al . , 2009 ) , thus weakening the likelihood of finding consistent covariance of its functional coupling with fibre density . Another explanation , particularly regarding the behaviour-tractography relationship , is that the pulvinar plays a significant functional role in the subcortical route to the amygdala . Research on macaques has demonstrated the pulvinar’s response to emotional faces ( Soares et al . , 2017; Maior et al . , 2010 ) and its role in modulating attention ( Soares et al . , 2017 ) and so we would indeed expect the strength of the pulvinar-amygdala connection to be more predictive of fearful face recognition . Future research could more deeply investigate the relative contribution of each half of the subcortical route to emotional face processing by using an optimised fMRI approach ( Wall et al . , 2009 ) and contrasting different types of stimuli – for example , low vs . high spatial frequency ( Gomes et al . , 2017 ) or moving stimuli ( Berman and Wurtz , 2011 ) . Our decision to reconstruct the two halves of the subcortical route separately was motivated by our interest in the relative contribution of each connection to face-related processing ( as described above ) but was also a limitation imposed by anatomically-constrained tractography , where reconstructed fibres are terminated at boundaries between grey and white matter ( Smith et al . , 2012 ) . Given that the pulvinar is made up of thalamic cell bodies ( grey matter ) , the likelihood of reconstructing a continuous streamline of axon bundles traversing the pulvinar’s grey matter may have been restricted by these boundary constraints . Previous studies that have not imposed these constraints have successfully traced a continuous pathway from the superior colliculus to the amygdala via the pulvinar ( Rafal et al . , 2015; Tamietto et al . , 2012 ) , supporting animal research showing that inferior-lateral pulvinar neurons receiving superior colliculus afferents also have efferent connections to the lateral amygdala ( Day-Brown et al . , 2010 ) . Our investigation into pulvinar and amygdala subregions support these findings , such that we found the superior colliculus to project predominantly onto the inferior ( and anterior ) pulvinar , which was the same subregion to receive the vast majority of fibres from the amygdala ( see Figure 1 ) . Furthermore , pulvinar fibres terminated predominantly within the basolateral amygdala , which is known to process visual information about threat and faces ( Hortensius et al . , 2016 ) . Further studies could use both anatomically constrained tractography and this subregion-specific approach with ultra-high-resolution imaging to better differentiate grey-white matter boundaries and more accurately determine if and where a continuous , subcortical route might traverse the pulvinar . While our results suggest that the inferior pulvinar may serve as a disynaptic connection point between the superior colliculus and amygdala , the continuity of information flow along the subcortical route is still a disputed feature due to the strong cortical influences on the pulvinar ( Bridge et al . , 2016; Pessoa and Adolphs , 2011 ) . This dispute has also arisen from prior work investigating the spatial frequency content of information conveyed along the subcortical route . Research on blindsight patients has found evidence only for low spatial frequencies which suggests that such information originated from magnocellular cells in the superior colliculus ( Burra et al . , 2017; Méndez-Bértolo et al . , 2016 ) . On the other hand , work in healthy participants has found no such spatial frequency preference , which suggests that rapid pulvinar-amygdala transmission might include input from other parvocellular pathways ( McFadyen et al . , 2017 ) . We did not exhaustively explore the extent to which the cortex contributes information to the pulvinar-amygdala connection . The winning effective connectivity model , however , did not include cortical connections between the pulvinar and the inferior occipital gyrus . Hence , it is unlikely that the primary visual cortex contributed ( either via direct anatomical connections or functional coupling along the ventral visual stream; Pessoa and Adolphs , 2010 ) to the information transmitted along the subcortical route . The winning model did , however , include input to the superior colliculus as well as directly to the pulvinar , which could reflect direct retinal input or input from areas not explicitly included in the model , such as the parietal cortex , temporal cortex , or the LGN ( Bridge et al . , 2016 ) , that may transmit both low and high spatial frequency information . Furthermore , it remains to be shown how interactions between the pulvinar and other cortical areas , such as the inferotemporal cortex ( Zhou et al . , 2016 ) , may directly influence activity along the pulvinar-amygdala connection . Our findings open avenues for future studies on how this subcortical pathway might influence threat-related behaviour . While our findings demonstrated that greater pulvinar-amygdala fibre density related to better fearful face recognition , it remains to be seen how this might compare with structural connectivity of other cortical networks . In other words , would the fibre density of this subcortical connection explain fearful face recognition above and beyond , say , structural connections between the inferior temporal or orbitofrontal cortex and the amygdala ( Pessoa and Adolphs , 2011 ) or between the thalamus and the superior temporal sulcus ( Leppänen and Nelson , 2009 ) ? Evidence from blindsight patients suggests that this subcortical connection ensures redundancy and compensation , such that it strengthens when cortical connections are destroyed ( Tamietto et al . , 2012 ) . Taking this in conjunction with our findings , we might consider that the pulvinar-amygdala connection contributes to fear recognition in faces ( and effective connectivity underlying face perception ) in healthy participants but can increase or decrease its influence depending on the functioning of other networks . Such increases and decreases are already evident in certain clinical populations . For example , structural connectivity between the superior colliculus , pulvinar , and amygdala is weakened in individuals with autism compared to healthy controls ( Hu et al . , 2017 ) , and BOLD signal to fearful faces is reduced in these areas ( Kleinhans et al . , 2011; Green et al . , 2017 ) , unless participants are explicitly instructed to fixate on the eyes ( Hadjikhani et al . , 2017 ) . On the other hand , people who suffer from anxiety show hyperactive activity along the subcortical route compared to non-anxious individuals ( Hakamata et al . , 2016; Tadayonnejad et al . , 2016; Nakataki et al . , 2017 ) . How and why this subcortical visual pathway to the amygdala is altered in these clinical populations remains a significant and relatively unexplored avenue of research . We observed hemispheric lateralisation of the pulvinar-amygdala connection , such that both the local and global tractography showed greater fibre density along the right than the left , and there were stronger tractography-behaviour and tractography-connectivity relationships for the right than the left . Early studies on the subcortical route observed specifically right-sided BOLD responses during non-conscious fearful face viewing ( Morris et al . , 1999; Morris et al . , 1998 ) , and a previous tractography study has also found that only the fractional anisotropy of the right subcortical route was significantly related to threat-biased saccades ( Koller et al . , 2018 ) . There is mounting evidence for right-sided specialisation for ordered ( Wyczesany et al . , 2018 ) and disordered ( McDonald , 2017 ) emotion processing , particularly for non-conscious signals transmitted along the subcortical route ( Gainotti , 2012 ) . Thus , our results lend support to this theory by demonstrating evidence for the right pulvinar-amygdala connection’s stronger fibre density and its relationship to emotional face viewing and fearful face recognition . Our understanding of this lateralisation may be deepened by future exploration of left- vs . right-sided structural connectivity and function along the subcortical route during conscious vs . non-conscious emotion processing in healthy participants . One limitation of the present study is the discrepancy between how local and global measures of fibre density related to other measures; namely , that local tractography covaried with fearful face recognition scores while global tractography covaried with effective connectivity . While the reconstructed fibres shared many similarities ( e . g . the pattern of findings for each connection across hemispheres and subregions , as well as the overlap with major fasciculi; see Figure 1 ) even after accounting for head motion , it is possible that the local tractography’s relatively greater susceptibility to noise may have decreased its relationship to corresponding effective connectivity parameters . Indeed , global tractography has been shown to better reflect local connection architecture ( Jbabdi and Johansen-Berg , 2011 ) , such as the subcortical connections we have investigated . Such discrepancies between global and local tractography have been reported in other work ( Anastasopoulos et al . , 2014 ) and so further research ( particularly those that only recruit a single tractograpy method ) will benefit from specific investigations into why these discrepancies might arise . In conclusion , our study has made substantial progress towards settling the long-held debate over the existence and function of a subcortical route to the amygdala in the human brain . Our multimodal neuroimaging approach , leveraged by computational modelling , provides convergent evidence for a fundamental and conserved pulvinar-amygdala pathway that is specifically involved in fear . We demonstrate that the white matter tracts that form the subcortical structural pathway from the pulvinar to the amygdala enables functional , dynamic interactions involved in emotional face perception . Critically , we show that structural connectivity between the pulvinar and the amygdala leads to better recognition of fearful expressions . We used the data from the publicly available Human Connectome Project ( HCP ) S900 release , collected between 2012 and 2016 , containing data from 897 consenting adults ( Van Essen et al . , 2013 ) . Ethical permission to use this data and the associated restricted access data ( including variables such as specific age information ) was obtained from the University of Queensland Human Research Ethics Committee . Out of these participants , 730 young adults had complete MRI and dMRI data , as well as fMRI data for the faces-vs-shapes task ( Van Essen et al . , 2012 ) . Of these , we excluded 95 people due to positive drug/alcohol tests and an additional 13 for abnormal colour vision . This resulted in a final sample of 622 participants aged between 22 and 36 years ( M = 28 . 81 , SD = 3 . 68 years ) , 259 of whom were male and 363 female , with 569 right-handed and 53 left-handed . Within our sample , 495 participants were related to one or more other participants ( 328 families in total ) . This included 53 pairs of monozygotic twins , 50 pairs of dizygotic twins , and 289 participants with one or more non-twin siblings in the sample . The remaining 127 participants were unrelated . We acknowledged that the many siblings in the HCP sample might spuriously decrease the variance in our neural measures ( due to the structural and functional similarity between siblings , for example ) and thus influence our statistics . Because of this , we replicated some of the analyses from the full sample on the subsample of unrelated participants ( see Appendix 1 ) . We chose the superior colliculus , pulvinar , and amygdala as our ROIs . We created masks of these ROIs in standard MNI space using FSL . For the amygdala ( AMG ) binary mask , we used the probabilistic Harvard-Oxford Subcortical atlas at a threshold of at least 50% probability . For amygdala subregions , we used the basolateral , centromedial , and superficial amygdala regions in the Juelich Histological Atlas ( Amunts et al . , 2005 ) at a threshold of at least 40% probability . For the pulvinar ( PUL ) , we were interested in the structure as a whole , as well as its subregions ( results for the latter are detailed in Appendix 1 ) . To do this , we used the parcellated pulvinar mask generated by Barron et al . ( 2015 ) , who isolated five distinct pulvinar clusters based on functional co-activation profiles in fMRI data from 29 , 597 participants across 7772 experiments ( Barron et al . , 2015 ) . For the pulvinar as a whole ROI , we merged the five clusters together and used FSL to manually fill any holes in the resultant binary mask . Finally , we manually created binary masks for the left and right superior colliculi ( SC ) in the absence of an atlas-based mask by drawing the boundaries of the superior colliculus over the MNI152 single participant T1 template with reference to an anatomical atlas ( Tamraz and Comair , 2004 ) and filling the centre . We then used FSL to warp these masks into native diffusion space for each participant's tractography analysis . All our ROIs in MNI space are freely available online from the Open Science Framework: doi:10 . 17605/OSF . IO/KBPWM . In this study , we implemented two tractography methods that use different approaches to white matter reconstruction for cross-method validation . We first used the multi-tissue model of global tractography . This method takes a Bayesian approach to reconstructing a full-brain fibre configuration using a generative signal model to best explain the underlying data . It is less sensitive to noise that may accumulate for longer distance tracts in other ‘local’ tractography methods throughout their stepwise approach ( Christiaens et al . , 2015; Reisert et al . , 2011 ) . Hence , for comparison , we computed probabilistic ( ‘local’ ) tractography between our regions of interest ( Tournier et al . , 2010 ) . This method also uses a Bayesian approach to account for one or more distributions of fibre orientations within each voxel , thus incorporating uncertainty into the model ( Zhou et al . , 2017 ) . To acquire a biologically accurate measure of apparent fibre density ( Raffelt et al . , 2012 ) along the resultant streamlines , we used the Spherical-Deconvolution Informed Filtering of Tractograms version 2 ( SIFT2 ) method to weight each streamline by a cross-sectional area multiplier directly related to the underlying data ( Smith et al . , 2015; Raffelt et al . , 2012 ) . For both the global ( producing ‘fibre count’ as a variable ) and local tractography with SIFT2 ( producing ‘summed weights’ as a variable ) , we computed 2 ( hemisphere: left , right ) by 2 ( connection: SC-PUL , PUL-AMG ) repeated-measures ANOVAs to quantitatively examine the properties of these pathways . All computer codes that were used to produce the results ( from raw HCP data to track counts , fibre density , BOLD signal and DCM files ) is freely available online via GitHub ( McFadyen , 2018; copy archived at https://github . com/elifesciences-publications/hcp-diffusion-dcm ) and the Open Science Framework ( doi:10 . 17605/OSF . IO/KBPWM ) . The data analysed in this study came from the publicly-available Human Connectome Project S900 release: https://www . humanconnectome . org/study/hcp-young-adult/document/900-subjects-data-release . Restricted access was obtained through the HCP to acquire specific participant ages ( in years ) and drug/alcohol information . Ethical permission was granted by the University of Queensland Human Research Ethics Committee . No figures display raw data .	Being able to quickly detect and respond to potential threats is essential for survival . Fear and threat trigger a range of responses in the body , which are controlled by different regions in the brain . For example , a structure located deep within the brain called the amygdala is connected to other parts of the brain that regulate hormones , senses and muscles . The amygdala is highly responsive to signs of threat , and research in rodents has shown that it plays a role in transmitting sounds that indicate danger . However , so far it has remained unclear if this was also the case for visual information . This is particularly challenging to study in humans because it has been difficult to image the deeper regions in the human brain . Now , McFadyen et al . reconstructed the pathways between the deeper brain regions important for processing vision and the amygdala using the brain scans of 622 participants . Then , they tested whether there was any connection between these pathways and the ability to recognise emotional expressions . To do so , fMRI brain scanning was used to measure the blood flow in the brain of volunteers looking at 40 faces that were either happy , sad , angry , fearful or neutral . The results showed that when people were looking at pictures of fearful and angry faces , the blood flow between visual areas and the amygdala increased , especially in individuals with stronger connections , such denser nerve fibres , between the involved regions . The denser those fibres were , the better the people were at recognising when a face was fearful . These discoveries suggest that the amygdala also plays a role in transmitting signals from deep-brain visual areas indicating danger and is likely to be one of the first areas to trigger a fear response in the brain . People with autism respond less to fearful faces , while people with anxiety respond more . Future research could investigate if the pathways to the amygdala differ in these people .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "research", "communication", "neuroscience" ]	2019	An afferent white matter pathway from the pulvinar to the amygdala facilitates fear recognition
Faced with potential harm , individuals must estimate the probability of threat and initiate an appropriate fear response . In the prevailing view , threat probability estimates are relayed to the ventrolateral periaqueductal gray ( vlPAG ) to organize fear output . A straightforward prediction is that vlPAG single-unit activity reflects fear output , invariant of threat probability . We recorded vlPAG single-unit activity in male , Long Evans rats undergoing fear discrimination . Three 10 s auditory cues predicted unique foot shock probabilities: danger ( p=1 . 00 ) , uncertainty ( p=0 . 375 ) and safety ( p=0 . 00 ) . Fear output was measured by suppression of reward seeking over the entire cue and in one-second cue intervals . Cued fear non-linearly scaled to threat probability and cue-responsive vlPAG single-units scaled their firing on one of two timescales: at onset or ramping toward shock delivery . VlPAG onset activity reflected threat probability , invariant of fear output , while ramping activity reflected both signals with threat probability prioritized . When confronted with potential harm , an estimate of threat probability must be made and followed by an appropriate fear response . The ventrolateral periaqueductal gray ( vlPAG ) has long been implicated in this fear response ( Fanselow , 1993; Kim et al . , 1993; Liebman et al . , 1970; Carrive et al . , 1997; Bandler et al . , 1985; Vianna et al . , 2001; Arico et al . , 2017; Assareh et al . , 2017 ) . In the canonical fear circuit , threat probability estimates ( the stored associative strength of cue and foot shock , for example a danger cue in Pavlovian fear conditioning ) originate in amygdalar nuclei ( Duvarci and Pare , 2014; Fanselow and LeDoux , 1999; Davis , 2006; Maren et al . , 2013 ) . Amygdalar threat estimates are sent to the vlPAG , which in turn organizes behavioral components of fear output , most notably freezing ( Perusini and Fanselow , 2015; Tovote et al . , 2015; Dejean et al . , 2015; Koutsikou et al . , 2014; Walker et al . , 1997 ) . While the vlPAG contributes to diverse fear and pain-related processes ( Assareh et al . , 2017; Bellgowan and Helmstetter , 1998; Parsons et al . , 2010; McNally et al . , 2011; Johansen et al . , 2010 ) , the view of the vlPAG as a node for fear output is so prevalent , it is commonly presented in introductory neuroscience textbooks ( Bear et al . , 2016; Carlson and Birkett , 2017 ) . A series of studies have uncovered a vlPAG population showing short-latency increases in firing to a danger cue . This characteristic would be expected of neurons organizing fear output . Yet in these studies , robust relationships between vlPAG single-unit activity and freezing were observed in a minority of neurons ( Tovote et al . , 2016 ) , weakly observed at danger cue onset ( Watson et al . , 2016 ) , or mixed with activity that purely reflected the danger cue ( Ozawa et al . , 2017 ) . At best , freezing only partially accounted for vlPAG activity . If not freezing , then what aspect of fear do vlPAG neurons signal ? Here we test the hypothesis that vlPAG neurons signal threat probability . We drew from learning theory ( Rescorla , 1968 ) to devise a fear discrimination procedure in which three cues predict unique foot shock probabilities: danger ( p=1 . 00 ) , uncertainty ( p=0 . 375 ) and safety ( p=0 . 00 ) . This behavioral approach was essential as previous studies utilized procedures in which a single cue predicted foot shock with certainty ( Tovote et al . , 2016; Watson et al . , 2016; Ozawa et al . , 2017 ) , precluding the ability to observe neural activity reflecting threat probability . We measured fear output using conditioned suppression of reward seeking ( Rescorla , 1968; Bouton and Bolles , 1980 ) over the entire cue and in one-second cue intervals . Using this procedure , we have found that suppression non-linearly scales to shock probability . So while suppression is strong to danger , intermediate to uncertainty and weak to safety; uncertainty produces more suppression than would be expected given its shock probability ( Walker et al . , 2018; Ray et al . , 2018; DiLeo et al . , 2016; Wright et al . , 2015; Berg et al . , 2014 ) . Concurrent with fear discrimination , we recorded vlPAG single-unit activity . Indeed , the vlPAG could potentially signal fear output via conditioned suppression of reward seeking ( Arico et al . , 2017 ) , a long-established measure of fear ( Estes and Skinner , 1941 ) that correlates with freezing ( Bouton and Bolles , 1980 ) . The non-linear relationship between behavior and shock probability permitted us to determine whether vlPAG single-unit activity was better captured by fear output or threat probability . We recorded the activity of 245 neurons in six rats over 88 fear discrimination sessions . A previous study optogenetically identified vlPAG glutamate neurons ( vGluT2+ ) as having low baseline firing rates , compared to GABA ( gamma-aminobutyric acid ) neurons ( Gad1+ ) that exhibited high baseline firing rates ( Tovote et al . , 2016 ) . We performed k-means clustering for all 245 neurons using baseline firing rate and the waveform characteristics: amplitude ratio and half the duration ( Roesch et al . , 2007 ) . All neurons separated into one of two clusters purely on the basis of baseline firing rate ( Figure 1E ) , with majority of neurons falling into the low firing rate ( LFR ) cluster ( n = 199 ) and the remaining in the high firing rate ( HFR ) cluster ( n = 46 ) . Independent of cluster membership , we determined the cue-responsiveness of each neuron ( n = 245; cue firing of all neurons shown in Figure 1—figure supplement 3 ) . Previous studies have identified a population of vlPAG neurons showing short-latency firing increases to auditory cues paired with foot shock ( Tovote et al . , 2016; Watson et al . , 2016; Ozawa et al . , 2017 ) . We identified 29 neurons ( obtained from 5/6 rats , ~12% of all neurons recorded ) with phasic increases in firing to danger , uncertainty , or safety ( t-test for firing rate , baseline [2 s prior to cue onset] vs . first 1 s cue interval , p<0 . 017 , Bonferroni correction for three tests ) . All 29 neurons belonged to the LFR cluster and these neurons are referred to as the Onset population ( single-unit example , Figure 1F; information for each Onset neuron’s waveform , firing characteristics , subject and session can be found in Figure 1—figure supplement 4 ) . Consistent with the most recent report ( Ozawa et al . , 2017 ) , 17 neurons increased firing to at least one cue during the last 1 s cue interval ( interval > baseline , p<0 . 017 ) . Three neurons belonged to the high firing cluster and are likely a unique class of neurons ( see Figure 1—figure supplement 5 for a full analysis ) . The remaining 14 neurons ( obtained from 4/6 rats , ~6% of all neurons recorded ) belonged to the LFR cluster and are referred to as the Ramping population ( single-unit example , Figure 1G; information for each Ramping neuron’s waveform , firing characteristics , subject and session can be found in Figure 1—figure supplement 6 ) . All manuscript analyses were performed on these 29 Onset neurons and 14 Ramping neurons . Despite identifying neurons without regard for relative firing to the three cues , differential firing was observed in Onset neurons at the single-unit ( Figure 1F ) and population ( Figure 2A ) levels . Onset neurons ( n = 29 ) sharply increased activity during the first 1 s cue interval , with greatest firing to danger , lesser firing to uncertainty , and least firing to safety ( Figure 2B , Left ) . The differential firing pattern diminished as cue presentation proceeded and was absent by the last 1 s cue interval ( Figure 2B , Right ) . The temporal firing pattern ( onset → offset ) was observed for all trials in the session ( Figure 2—figure supplement 1A–D ) . ANOVA for normalized firing rate ( Z-score transformation ) for the 29 Onset neurons [data from Figure 2A; factors: cue ( danger vs . uncertainty vs . safety ) and bin ( 100 ms: 1 s prior to cue onset through 10 s cue ) ] revealed main effects of cue and bin ( Fs > 9 , ps<0 . 01 , ηp2 > 0 . 20 , op > 0 . 95 ) but most critically , a cue x bin interaction ( F218 , 6104 = 1 . 94 , p<0 . 01 , ηp2 = 0 . 06 , op = 1 . 00 ) . Consistent with the ANOVA interaction , Onset neurons showed significantly greater firing to danger compared to uncertainty in the first 1 s cue interval ( t28 = 4 . 54 , p=9 . 70×10−5 ) . While numerically greater firing to uncertainty over safety failed to reach significance in the first 1 s interval ( t28 = 1 . 37 , p=0 . 18 ) , ANOVA restricted to uncertainty and safety ( 100 ms: 1 s prior to cue onset through first 5 s of the cue ) revealed a significant cue x bin interaction ( F59 , 1652 = 1 . 50 , p=0 . 01 , ηp2 = 0 . 051 , op = 1 . 00 ) . Differential firing was not observed to danger vs . uncertainty ( t28 = 1 . 69 , p=0 . 10 ) or to uncertainty vs . safety ( t28 = 0 . 60 , p=0 . 55 ) in the last 1 s cue interval . Selective firing was observed in Ramping neurons at the single-unit ( Figure 1G ) and population ( Figure 2C ) levels . Ramping neurons ( n = 14 ) did not increase firing to any cue during the first 1 s cue interval ( Figure 2D , Left ) . Instead , activity ramped over cue presentation with greatest firing observed during the last 1 s cue interval ( Figure 2D , Right ) . Ramping activity was most apparent to danger , intermediate to uncertainty , and absent to safety . The temporal firing pattern ( onset → offset ) was consistent across trials ( Figure 2—figure supplement 1E–H ) . ANOVA for normalized firing rate for the 14 Ramping neurons [data from Figure 2C; factors: cue ( danger vs . uncertainty vs . safety ) and bin ( 100 ms: 1 s prior to cue onset through 10 s cue ) ] revealed main effects of cue and bin ( Fs >60 , ps <0 . 01 , ηp2 > 0 . 40 , op = 1 . 00 ) and a cue x bin interaction ( F218 , 2834 = 4 . 33 , p<0 . 01 , ηp2 = 0 . 24 , op = 1 . 00 ) . Illustrative of the ANOVA interaction , Ramping neurons showed no significant differences in firing to danger vs . uncertainty ( t13 = 0 . 62 , p=0 . 55 ) or uncertainty vs . safety ( t13 = 0 . 24 , p=0 . 81 ) in the first 1 s cue interval . However , differential firing to danger vs . uncertainty ( t13 = 3 . 17 , p=7 . 41×10−3 ) , and uncertainty vs . safety ( t13 = 8 . 26 , p=2 . 00×10−6 ) , was observed in the last 1 s cue interval . Ramping activity to danger and uncertainty could be the product of the time at which activity began to increase , or the rate of increase . We performed a two-tailed t-test for population firing to danger vs . safety ( Figure 2E , red line ) and uncertainty vs . safety ( Figure 2E , purple line ) in a 1 s window , starting with cue onset . We slid the 1 s window across the 10 s cue in 100 ms increments , to reveal the time in which danger and uncertainty population firing departed from safety . We then calculated the rate of firing increase from the departure window to the last 1 s interval for danger and uncertainty . Differential firing was determined by the time of departure from safety , as opposed to the rate of increase . Ramping activity to danger emerged earlier ( Figure 2E; 2 . 8 s following cue onset for p<0 . 05; 5 . 8 s for full Bonferroni correction , p<0 . 00025 [0 . 05/200] ) than ramping activity to uncertainty ( 5 . 7 s following cue onset; 6 . 5 s for full Bonferroni correction ) . Change in firing rate did not differ between danger and uncertainty ( Figure 2E , Inset ) . It is possible that while Onset and Ramping neurons are distinct , activity of one population drives the other . We identified four Onset-Ramping pairs recorded in the same session . We then asked if trial-by-trial variations in firing over each 1 s cue interval were negatively correlated for any of the four pairs . We failed to uncover such a relationship ( Figure 3 ) , demonstrating that Onset and Ramping populations are distinct and independent . To demonstrate that Onset population activity was the result of a consistent bias across neurons , we directly compared single-unit firing to cue pairs . Danger and uncertainty firing were correlated , and single-units were biased towards greater firing to danger ( Figure 4A ) . Uncertainty and safety firing were also correlated; however , the single-unit bias towards greater firing to uncertainty was not significant ( Figure 4B ) . Underscoring their specificity to cue onset , single-units were biased towards greater firing to danger in the first 1 s cue interval compared to the last interval , and there was no correlation between firing in the two epochs ( Figure 4C ) . Examining relative cue firing for each Onset neuron in the first 1 s interval revealed the most common pattern to be: danger > uncertainty > safety ( n = 14 ) . This was the only pattern to contain more units than would be expected than chance ( Figure 4D ) . We performed the same analysis for the Ramping population , only for the last 1 s interval . Ramping neurons showed a selective firing pattern . A significant correlation between firing to danger and uncertainty was observed , along with a single-unit bias towards greater firing to danger ( Figure 4E ) . Only now , there was no correlation between uncertainty and safety firing , but a consistent bias towards greater uncertainty firing ( Figure 4F ) . Ramping single-units were biased towards danger activity in the last 1 s cue interval , and there was no correlation between firing in the two epochs ( Figure 4G ) . The most common firing pattern in the last 1 s interval was: danger > uncertainty > safety ( n = 11 ) . This was the only pattern to contain more units than expected by chance ( Figure 4H ) . VlPAG activity is greatest to danger , the cue most strongly suppressing rewarded nose poking . It is therefore possible that Onset and Ramping neurons are simply responsive to nose poke cessation . To examine this possibility , we identified naturally occurring periods of nose poke cessation in inter-trial intervals , when no cues were presented . This analysis found no meaningful changes in Onset or Ramping activity during periods of nose poke cessation ( Figure 2—figure supplement 2 ) , demonstrating activity patterns are specific to cue-induced suppression of nose poking . At first glance , the firing patterns of Onset and Ramping neurons appear to support the prevailing hypothesis that vlPAG neurons signal fear output . Differential fear ( Figure 1C ) and differential firing ( Figure 2B , D ) show the same general pattern: danger > uncertainty > safety . However , closer inspection reveals that relative differences in fear do not match relative differences in firing . Rats showed robust discrimination between uncertainty and safety , regardless of the temporal resolution with which fear was measured ( Figure 1C , D ) . Yet , robust differential firing to uncertainty and safety was modest in the Onset population ( Figure 2B , left; Figure 4B ) . The Ramping population showed stronger differential firing between uncertainty and safety ( Figure 2D , right; Figure 4F ) , but this pattern did not emerge in until the end of the cue . Fear discrimination was reliably detected in the first 1 s interval ( Figure 1D ) , indicating that Ramping neurons cannot organize fear output early in cue presentation . While inconsistencies between fear output and neural activity are evident for the Onset and Ramping populations , the analyses conducted so far cannot conclusively test the relative contribution of threat probability and fear output to vlPAG single-unit activity . To formally test the degree to which vlPAG activity is captured by fear output and threat probability , we used simultaneous linear regression for single-unit firing ( Figure 5; see Materials and methods for example regression input ) . For each single-unit , we calculated the normalized firing rate for each trial ( 32 total: six danger , six uncertainty shock , 10 uncertainty omission , and 10 safety ) , in 1 s bins over the 10 s cue . For each trial , we calculated fear on two time scales: total fear ( suppression ratio for the entire 10 s cue ) and interval fear ( suppression ratio for the specific 1 s interval ) . The corresponding shock probability was assigned for each trial ( danger = 1 . 00 , uncertainty = 0 . 375 and safety = 0 . 00 ) . Total fear , interval fear and threat probability were used as regressors to explain trial-by-trial variance in single-unit firing . Statistical output was a beta coefficient quantifying the strength ( \|>0\| = stronger ) and direction ( >0 = positive ) of the predictive relationship between each regressor and single-unit firing . Beta coefficients for single-units comprising the Onset and Ramping populations were subjected to ANOVA with regressor ( total fear vs . interval fear vs . probability ) and interval ( 1 s intervals for 10 s cue ) as factors . This allowed us to determine the relative contribution of total fear , interval fear and probability to single-unit firing over the course of cue presentation . The results of primary interest for the Onset population came from the first 1 s cue interval , when activity was highest and differential firing was observed . Linear regression unequivocally revealed that Onset single-unit activity was captured by threat probability ( Figure 5A ) . The beta coefficient for the probability regressor was positive and significant , exceeding the beta coefficient for either measure of fear output – neither of which differed from zero . The population bias was observed across Onset neurons . Single-unit beta coefficients were positively biased for threat probability , but not for either measure of fear output ( Figure 5B , C ) . Examining the entirety of cue presentation , threat probability signaling was highest in the first interval , persisted several more seconds and diminished ( Figure 5D ) . Total fear or interval fear did not account for variance in single-unit firing at any interval . Consistent with this description , ANOVA for beta coefficient with factors of regressor and interval ( 10 total ) revealed a main effect of regressor ( F2 , 56 = 7 . 16 , p<0 . 01 , ηp2 = 0 . 20 , op = 0 . 92 ) and a regressor x interval interaction ( F18 , 504 = 2 . 38 , p<0 . 01 , ηp2 = 0 . 08 , op = 0 . 99 ) . Identical results were obtained when probability was compared to total fear or interval fear separately . In fact , significance for fear output could only be found if total fear was the only regressor used in the analysis – producing a result very similar to that of a previous study ( Watson et al . , 2016 ) . Even then , the predictive relationship was weaker than that of probability ( Figure 5—figure supplement 1 , A–E ) . The threat probability regressor in the above analyses utilized the actual shock probability assigned to each cue . Of course , the subjects had no a priori knowledge of shock probability assignments . It is then possible that vlPAG activity is ‘tuned’ to an alternative shock probability . To examine this , we performed single-unit linear regression for normalized firing in the first 1 s cue interval maintaining the probabilities for danger ( 1 . 00 ) and safety ( 0 . 00 ) , but incrementing the probability assigned to uncertainty from 0 to 1 in 0 . 125 steps ( 0 . 000 , 0 . 125 , 0 . 250 , 0 . 375 , 0 . 500 , 0 . 625 , 0 . 750 , 0 . 875 , and 1 . 000 ) . The mean beta coefficient for each of the nine increments is plotted as a threat-tuning curve for the Onset population ( Figure 5E ) . The beta coefficient resulting from regression using the actual shock probability ( uncertainty = 0 . 375 ) , was the ‘peak’ of the tuning curve . The probabilities with the next highest beta coefficients were those flanking 0 . 375 . Beta coefficients dropped off rapidly as the uncertainty assignment moved to the extremes . This result is particularly revealing for the analysis in which the uncertainty assignment was 0 . 000 ( first data point on the curve Figure 5E ) . Onset neurons showed high firing to danger but lower and more similar firing to uncertainty and safety , leaving open the possibility that Onset neurons signal a more binary output ( danger = 1 . 000 ) > ( uncertainty and safety = 0 . 000 ) . However , the actual uncertainty assignment ( 0 . 375 ) captured single-unit activity better than the binary assignment ( 0 . 000 ) in the first 1 s interval and across the remainder of cue presentation ( Figure 5—figure supplement 2 ) . Linear regression for the Ramping population in the last 1 s cue interval revealed that single-unit activity was captured by a mixture of threat probability and total fear output ( Figure 5F ) . Ramping single-units were biased towards positive beta coefficients for probability and total fear ( Figure 5G ) , but there was no correlation between these regressors . Ramping single-units were not biased towards positive beta coefficients for interval fear ( Figure 5H ) , but signaling of probability and interval fear were negatively correlated . Linear regression for all ten intervals revealed that threat probability signaling was prioritized over total fear and interval fear ( Figure 5I ) . ANOVA for beta coefficients with factors of regressor and interval revealed a main effect of regressor ( F2 , 26 = 8 . 96 , p<0 . 01 , ηp2 = 0 . 41 , op = 0 . 96 ) and regressor x interval interaction ( F18 , 234 = 2 . 16 , p<0 . 01 , ηp2 = 0 . 14 , op = 0 . 99 ) . Mean ± SEM beta coefficients for each regressor over all 10 intervals were: threat probability: 0 . 58 ± 0 . 14 , total fear: 0 . 12 ± 0 . 10 and interval fear: −0 . 03 ± 0 . 05 . Only the beta coefficient for probability differed significantly from zero ( t13 = 4 . 21 , p=1 . 02×10−3 ) and was significantly greater than either total fear ( t13 = 2 . 32 , p=0 . 04 ) or interval fear ( t13 = 4 . 32 , p=8 . 28×10−4 ) . This pattern of results held when probability was separately compared to total fear and interval fear , and when each regressor was considered in isolation ( Figure 5—figure supplement 1 , F–J ) . If Ramping neurons contain information about threat probability , as well as fear output , the tuning curve for ramping neurons ought to be shifted right of 0 . 375 . This is because the relative weighting of uncertainty for threat probability ( danger >> uncertainty > safety ) and average fear output ( danger > uncertainty >> safety ) differ . We constructed a population threat-tuning curve for normalized firing in the last 1 s interval ( Figure 5J , as in Figure 5E ) . Tuning was shifted right of the actual probability , with a ‘peak’ at 0 . 625 . This is consistent with mixed signaling of fear output and threat probability by Ramping neurons , rather than a pure threat probability signal . Onset neurons are tuned to initial cue presentation , providing a rapid estimate of threat probability . By contrast , Ramping neurons are initially unresponsive , but gradually increase activity over cue presentation . Ramping neurons may continue to signal threat probability through the delay period , up until foot shock receipt . We first examined population activity during the five seconds following cue offset ( 2 s delay , 0 . 5 s shock and 2 . 5 s post-shock ) . Activity during the 500 ms shock period was not analyzed because it may have been contaminated by electrical artifacts . Onset neurons showed negligible delay activity and little to no post-shock activity ( Figure 6A ) . Ramping neurons continued firing to danger and uncertainty throughout the delay period , but this firing diminished shortly after shock presentation ( Figure 6B ) . ANOVA for normalized firing rate [factors: trial type ( danger vs . uncertainty shock vs . uncertainty omission vs . safety ) and bin ( 100 ms: 5 s post cue ) ] revealed no main effects or interactions for Onset neurons ( Fs <2 . 3 , ps >0 . 09 ) . By contrast , identical ANOVA revealed main effects of trial and bin ( Fs >3 , ps <0 . 01 ) , as well as a trial x bin interaction for Ramping neurons ( F132 , 1716 = 3 . 14 , p<0 . 01 , ηp2 = 0 . 19 , op >0 . 99 ) . The Onset and Ramping firing patterns differed from one another . ANOVA with neuron type as a factor ( Onset vs . Ramping ) found significant interactions for trial type x neuron type , bin x neuron type , and trial type x bin x neuron type ( Fs >2 . 50 , ps <0 . 01 , ηp2 > 0 . 19 , op >0 . 94 ) . Observed differences in neural activity suggest that Ramping neurons maintain threat probability signaling throughout the delay period and this signal abruptly decreases following foot shock presentation . Single-unit regression was used to determine whether trial-by-trial , post-cue firing was best described by interval fear or probability ( total fear sampled only the prior 10 s cue period and was omitted ) . Regression was performed in 500 ms intervals for the 5 s post-cue period minus the shock interval ( nine total intervals ) . Onset neurons did not signal interval fear or probability at any time following cue offset ( Figure 6C ) . ANOVA revealed no main effect of regressor ( F1 , 28 = 1 . 02 , p=0 . 32 , ηp2 = 0 . 04 , op = 0 . 16 ) or regressor x interval interaction ( F8 , 224 = 0 . 53 , p=0 . 83 , ηp2 = 0 . 02 , op = 0 . 24 ) . Illustrative of the lack of information contained in Onset neurons , post hoc comparisons found that no beta coefficient , for any regressor differed from zero . By contrast , Ramping neurons signaled probability throughout the delay period ( Figure 6D , first four intervals ) , which diminished following shock delivery . In support , ANOVA revealed a significant regressor x interval interaction ( F8 , 104 = 5 . 20 , p<0 . 01 , ηp2 = 0 . 29 , op = 0 . 99 ) . Threat probability signaling prior to shock delivery was abruptly halted following shock delivery in Ramping neurons , but not Onset neurons . To reveal the degree to which this occurred , we constructed correlation matrices using the probability beta coefficient from each of the nine relevant intervals ( 4 delay and five post-shock ) . Of greatest interest were the twenty comparisons between the probability beta coefficient for the four delay and five post-shock intervals ( Figure 6E & F; bottom-left quadrant ) . Significant between-interval correlations for the probability beta coefficient were observed for 13/20 comparisons in Onset neurons ( Figure 6E ) , indicating that threat probability signaled during the delay tended to persist following foot shock . By contrast , significant correlations were found for only 4/20 comparisons in Ramping neurons ( Figure 6F ) ; and the proportion of intervals showing a significant correlation differed for the Onset and Ramping neurons ( χ2 = 8 . 08 , p=4 . 5×10−3 ) . Bonferroni correction ( 0 . 05/5 = 0 . 01 , five tests per interval ) found significant correlations for 6/20 Onset comparisons , 1/20 Ramping comparisons , and these proportions significantly differed ( χ2 = 4 . 22 , p=0 . 039 ) . We recorded vlPAG single-unit activity while rats discriminated between danger , uncertainty and safety . Consistent with previous reports ( Tovote et al . , 2016; Watson et al . , 2016; Ozawa et al . , 2017 ) , we found a population of Onset neurons with short-latency excitation to danger . Consistent with the most recent report ( Ozawa et al . , 2017 ) , we found a Ramping population that increased activity over danger presentation . Onset activity reflected an estimate of threat probability , invariant of fear output . Ramping activity reflected threat probability and fear output , though probability emerged earlier and was stronger overall . While vlPAG signals for fear output could potentially emerge at the ensemble level ( Jones et al . , 2007; Zhou et al . , 2018 ) , these multi-unit codes would be composed of single-units primarily signaling threat probability . Activity reflecting fear output may be found in other vlPAG populations , such as neurons showing inhibition of firing to cues ( Tovote et al . , 2015 ) , or in non-cue-responsive single-units ( Insanally et al . , 2019 ) . Yet , this would still mean that signals for threat probability and fear output co-exist in the vlPAG . It is important to note that these results are correlative and that Onset neuron activity may not play a causal role in fear output . Previous work has found that short-latency , excitatory responses to danger are largely observed in vlPAG VGlut2 +neurons , and that excitation of this population is sufficient to produce freezing ( Tovote et al . , 2016 ) . Though all Onset neurons fell into the low firing cluster , we cannot conclude these were VGlut2 +neurons . Moreover , inhibition of vlPAG GABA neurons also promotes freezing ( Tovote et al . , 2016 ) . These neurons have comparatively higher baseline firing rates and respond to danger cues through inhibition of neural activity . A causal , vlPAG signal for fear output may be observed in a GABAergic/cue-inhibited population , in concert with or independent of a glutamatergic/cue-excited population . At the same time , it is clear how the observed Onset signal could play a causal role in fear output . Blocking vlPAG Onset activity to danger and uncertainty would equate neural activity to that for safety , removing the threat impetus for suppression of reward seeking , freezing and related defensive behaviors . Before further discussing implications , we consider some alternative accounts for the observed firing patterns . Perhaps vlPAG neurons signaling fear output are anatomically distinct from those recorded here . We intentionally recorded from caudal vlPAG , the subregion preferentially activated by fearful contexts in rats ( Carrive et al . , 1997 ) . VlPAG manipulations that disrupt fear-related behaviors typically include this more caudal region ( De Oca et al . , 1998 ) , and high-resolution functional magnetic resonance imaging reveals caudal vlPAG activation specific to aversive stimuli in humans ( Satpute et al . , 2013 ) . We observed threat probability signaling in all recording locations ( bregma −7 . 62 → −7 . 98 ) . The vlPAG stretches ~0 . 7 mm beyond our most caudal recording site . It is therefore possible that neurons signaling fear output are restricted to the extreme caudal vlPAG . Maybe the vlPAG signals fear output , but we did not measure the relevant output . Previous studies have failed to find robust relationships between vlPAG activity and cued freezing . Here we used conditioned suppression of rewarded nose poking to provide an objective measure of fear on two timescales and to perhaps better capture vlPAG activity . This measure of fear did not capture Onset neuron firing , and only partially captured Ramping neuron firing at the end of cue presentation . Further , Onset and Ramping activity were not merely driven by nose poke cessation or withdrawal from the port . If not freezing , conditioned suppression , or nose poke cessation then perhaps another measure of fear ? Danger cues elicit active fear responses: escape-like behaviors such as darting ( Greiner et al . , 2018; Gruene et al . , 2015 ) . However , darting is prevalent in females , but less so in males . Further , the males used in this study had extensive experience with fear discrimination , and at no point was escape from the foot shock possible . Danger cues increase arterial blood pressure and reduce heart rate , however , neither of these abilities has been linked to vlPAG function ( Helmstetter and Tershner , 1994; LeDoux et al . , 1988; Wilson and Kapp , 1994 ) . Danger cues also enhance startle , perhaps more in line with Onset population firing . Yet , dorsal , rather than ventrolateral , PAG subregions have been implicated in startle behavior ( Walker et al . , 1997; Zhao et al . , 2009 ) . Of course , many other fear behaviors are possible: piloerection , hyperventilation , changes in body temperature , vocalization , etc . ( Kim et al . , 2010; Iwata and LeDoux , 1988; Gallego et al . , 2001; Vianna and Carrive , 2005 ) . Problematic is that most fear behaviors are initiated at cue onset and maintained until the aversive event occurs . We did not observe a substantial population of neurons with these temporal firing characteristics , making the vlPAG a poor candidate for sustained fear output . Above all , any potential behavior signaled by Onset neurons must closely match shock probability , confounding this behavior signal with probability itself . The present findings are best understood through comparison to the account of vlPAG function outlined in the predatory imminence continuum ( PIC ) , a highly influential theory of defensive behavior ( Fanselow and Lester , 1988 ) . Organizing features of the PIC are time and degree of threat . As predation becomes more imminent ( pre-encounter → post-encounter → circa-strike ) , the form and intensity of defensive behaviors change . Cued fear is argued to capture post-encounter defenses: immobility elicited when predators are nearby . In the neural instantiation of PIC , the amygdala integrates information about environmental stimuli ( auditory cues here ) , nociceptive information ( foot shock ) and time to produce a signal for degree of threat ( Fanselow and Lester , 1988 ) . This amygdala-derived signal is relayed to the vlPAG to organize fear output ( Fanselow , 1991; Fanselow , 1994 ) . Implicit in the PIC model , is that the vlPAG does not contain information about degree of threat – only the resultant fear output . Yet , we find that single vlPAG neurons contain detailed information about time and degree of threat . These results require more careful consideration of the role of the vlPAG in the fear circuit and the PIC . Rather than signaling fear output , vlPAG Onset neurons signal threat probability . This information could be used to organize a variety of fear responses , but these neurons do not intrinsically signal fear output . For example , vlPAG projections to the central amygdala ( CeA ) and rostral ventromedial medulla ( RVM ) could inform fear output via freezing ( Vianna et al . , 2008 ) , while projections to midline/intralaminar thalamus could rapidly relay threat probability estimates to a larger fear network ( basolateral amygdala , prelimibic cortex , infralimbic cortex , insular cortex , etc . ) ( Vertes et al . , 2015; Krout and Loewy , 2000; Buchanan and Thompson , 1994; Sengupta and McNally , 2014 ) , promoting a variety of threat-related processes ( Faull et al . , 2016 ) . In this way , rapid threat probability estimates generated by vlPAG Onset neurons may be more akin to brainstem-derived signals for rapid detection of visual threats , which are distributed to a wider brain network ( Liddell et al . , 2005 ) . Neurons responsive toward the end of cue presentation were more heterogeneous , in terms of their baseline firing rate and their signaling . Ramping neurons prioritized threat probability but also signaled fear output . However , Ramping activity could not drive fear output in full . Differential fear to safety , uncertainty and danger was observed even in the first second of cue presentation , when these neurons were unresponsive . Unlike Onset neurons , Ramping neurons showed selective firing throughout the delay period , only diminishing once foot shock had been delivered . Ramping neurons may provide a threat probability estimate that increases as threat draws nearer and peaks when threat is imminent . Ramping neurons may help sustain threat estimates in the absence of explicit stimuli , such as in trace conditioning ( McEchron et al . , 1998; Büchel et al . , 1999 ) , or estimate more precisely when the noxious event will occur . Shifts toward PAG-centric activity are apparent in humans , as capture becomes imminent ( Mobbs et al . , 2007 ) or natural threats draw closer Mobbs et al . , 2010; with the caveat that neither of these studies could specify the PAG subregion activated . Ramping neurons may also provide two forms of feedback: the estimated probability of foot shock and a readout of fear output on that trial . Either type of information could be compared to that trial’s outcome , particularly for uncertainty trials on which shock can be present or absent . This could be used to adjust estimates of threat probability and fear output on future encounters with the cue . This is broadly consistent with a central role for the vlPAG in feedback processes ( McNally et al . , 2011; Ozawa et al . , 2017; Yeh et al . , 2018 ) . Even more , this Ramping signal could alter or reduce foot shock processing through descending control of dorsal horn nociceptive inputs via endogenous opioid circuits . This behavioral phenomenon , conditioned analgesia ( MacLennan et al . , 1980; Fanselow and Bolles , 1979; Chance et al . , 1978 ) , may be related to a more general phenomenon of placebo analgesia . VlPAG activation is consistently observed in studies of placebo analgesia ( Tracey , 2010; Eippert et al . , 2009; Wager and Atlas , 2015; Petrovic et al . , 2002 ) and our observation of increasing neural activity toward presentation of a noxious event may provide a suitable neural substrate for this process . Perhaps most remarkable is that although independent , between the Onset and Ramping populations , the vlPAG contains an estimate of threat probability from the time of first encounter up through the noxious event itself . The vlPAG may not signal fear output per se , but is rich with information that would inform a variety of fear processes and behaviors . If fear output via conditioned suppression is non-linear , and vlPAG activity scales linearly to threat probability , how does the vlPAG fit into the fear circuit ? An ultimate explanation for non-linear fear output may be that threat systems evolved to avoid predation , not to precisely match degree of defensive behavior to threat probability . Erring on the side of greater fear to uncertain threats may promote survival . A proximate explanation may be that fear output is the summed product of multiple threat signals . VlPAG output to the RVM may instruct fear output to match threat probability . If this were the only threat signal , then vlPAG activity would in fact reflect fear output . We speculate that additional threat signals govern fear output . For example , neurons in the retrorubral field ( RRF ) project to the RVM , as well as the CeA ( Zahm and Trimble , 2008; Deutch et al . , 1988; Von Krosigk and Smith , 1991 ) . Preliminary data from our laboratory suggest that RRF neurons are primarily responsive to danger and uncertainty , but weight uncertainty similarly to danger ( danger = uncertainty > safety ) . RRF neurons may signal threat probability plus uncertainty-induced stress , or may favor cue-shock contiguity over contingency ( Rescorla , 1967 ) . Summation of vlPAG-derived and RRF-derived threat signals by RVM neurons would produce a non-linear fear output like that observed in our discrimination procedure . The results pose questions about the specific relationship between the vlPAG and the CeA . VlPAG threat probability signals may be trained up by the CeA , but become CeA-independent with sufficient training ( Ozawa et al . , 2017 ) . Consistent with this interpretation , the CeA is essential to the acquisition of conditioned suppression with limited training , but that more extended training mitigates the effects of CeA lesions ( Lee et al . , 2005; McDannald , 2010 ) . This is not to say that we would expect the CeA to become inessential following extensive fear discrimination training . Updating threat probability should occur when environmental stimuli become more or less predictive of noxious events . We anticipate the CeA is essential to updating vlPAG threat probability signaling ( McNally et al . , 2011; Ozawa et al . , 2017 ) . It is near universally accepted that the amygdala is a key node of dysfunction in stress ( Rauch et al . , 2000 ) and anxiety disorders ( Etkin and Wager , 2007 ) . This may be driven in part by technical considerations: whole-brain fMRI can detect amygdala BOLD signals ( Johnstone et al . , 2005 ) , while detecting subregion-specific PAG BOLD signals requires a more deliberate approach ( Satpute et al . , 2013 ) . Perhaps the primary intellectual driver is that the amygdala is theorized to be a privileged cite of integration/learning in the fear circuit ( Mahan and Ressler , 2012; Admon et al . , 2013 ) . The present findings illustrate that the amygdala is not privileged in this regard , and mark the vlPAG as likely node of dysfunction in psychiatric disorders of stress and anxiety . Appreciation for the vlPAG as a site of integration will hasten mapping of a more complete fear circuit . Deliberate study of vlPAG function ( Arico et al . , 2017; Assareh et al . , 2017; Rozeske et al . , 2018 ) and dysfunction in psychiatric disease ( Yeh et al . , 2018 ) , will be essential to developing effective therapies for disorders characterized by exaggerated threat estimation and aberrant fear . Ten adult male Long Evans rats ( RRID:RGD_2308852 ) weighing 241–268 g arrived from Charles River Laboratories , Raleigh , NC on postnatal day 55 . All rats were implanted with drivable microelectrode bundles . Data are reported from six rats; three rats did not yield units and one rat had incorrect electrode placement . All rats were single-housed throughout the duration of the experiment on a 12 hr light cycle ( lights off at 6:00pm ) and maintained at 85% of their free-feeding body weight with standard laboratory chow ( 18% Protein Rodent Diet #2018 , Harlan Teklad Global Diets , Madison , WI ) except during an 11 day , post-surgery recovery period where animals had ad libitum access to standard chow . Ad libitum access to water was always available in the home cage . All protocols were approved by the Boston College Animal Care and Use Committee and all experiments were carried out in accordance with the NIH guidelines regarding the care and use of rats for experimental procedures . Microelectrodes consisted of a drivable bundle of sixteen 25 . 4 µm diameter Formvar-Insulated Nichrome wires ( 761500 , A-M Systems , Carlsborg , WA ) within a 27-gauge cannula ( B000FN3M7K , Amazon Supply ) and two 127 µm diameter PFA-coated , annealed strength stainless-steel ground wires ( 791400 , A-M Systems , Carlsborg , WA ) . All wires were electrically connected to a nano-strip omnetics connector ( A79042-001 , Omnetics Connector Corp . , Minneapolis , MN ) on a custom 24-contact , individually routed and gold immersed circuit board ( San Francisco Circuits , San Mateo , CA ) . Stereotaxic surgery was performed aseptic conditions under isoflurane anesthesia ( 1–5% in oxygen ) . Carprofen ( i . p . , 5 mg/kg ) and lactated ringers solution ( ~2–5 mL ) were administered preoperatively . The skull was scoured in a crosshatch pattern with a scalpel blade to increase efficacy of implant adhesion . Five screws were installed in the skull to further stabilize the connection between the skull , electrode assembly and a protective head cap ( screw placements: two anterior to bregma , two between bregma and lambda about ~3 mm medial to the lateral ridges of the skull , and one on the midline ~5 mm posterior of lambda ) . A 1 . 4 mm diameter craniotomy was performed to remove a circular skull section centered on the implant site and the underlying dura was removed to expose the cortex . Nichrome recording wires were freshly cut with surgical scissors to extend ~2 . 0 mm beyond the cannula at a ~15° angle . Just before implant , current was delivered to each recording wire in a saline bath , stripping each tip of its formvar insulation . Current was supplied by a 12 V lantern battery and each Omnetics connector contact was stimulated for 2 s using a resistor-equipped lead . Machine grease was placed by the cannula and on the microdrive . For implantation dorsal to the vlPAG , the electrode assembly was slowly advanced at a 20° angle to the following coordinates from cortex ( anterior-posterior: −8 . 00 mm , medial-lateral: −2 . 45 mm and dorsal-ventral: −5 . 52 mm ) . Once in place , stripped ends of both ground wires were wrapped around a sixth screw inserted previously to ground the electrode ( anterior-posterior: −8 . 00 mm , medial-lateral: +2 . 45 mm ) . The microdrive base and a protective head cap surrounding the electrode assembly were cemented in place at the end of the procedure using orthodontic resin ( C 22-05-98 , Pearson Dental Supply , Sylmar , CA ) . The apparatus for Pavlovian fear conditioning consisted of two individual chambers with aluminum front and back walls retrofitted with clear plastic covers , clear acrylic sides and top , and a grid floor . Each grid floor bar was electrically connected to an aversive shock generator ( Med Associates , St . Albans , VT ) through a grounding device . This permitted the floor to be grounded at all times except during shock delivery . An external food cup and a central nose poke opening , equipped with infrared photocells were present on one wall . Auditory stimuli were presented through two speakers mounted on the ceiling . Prior to discrimination sessions , rats were food-deprived to 85% of their free-feeding body weight and were fed specifically to maintain this weight through the behavioral procedure . Starting on P59 , rats were shaped to nose poke for pellet delivery in the experimental chamber using a fixed ratio schedule in which one nose poke yielded one pellet . Shaping sessions lasted 30 min or until approximately 50 nose pokes were completed . Over the next 3 days , rats were placed on 5 days of variable interval ( VI ) schedules in which nose pokes were reinforced on average every 30 s ( day 1 ) , or 60 s ( days 2 through 5 ) . For the remainder of behavioral testing , nose pokes were reinforced on a VI-60 schedule independent of all Pavlovian contingencies . In two separate sessions , each rat was pre-exposed to the three cues to be used in Pavlovian discrimination . Auditory cues consisted of repeating motifs of broadband click , phaser or trumpet . These 42 min sessions consisted of four presentations of each cue ( 12 total presentations ) with a mean inter-trial interval ( ITI ) of 3 . 5 min . The order of trial type presentation was randomly determined by the behavioral program and differed for each rat during each session . Prior to recording , each rat received eight , 93 min sessions of fear discrimination . Each session consisted of 32 trials , with a mean ITI of 3 . 5 min . Auditory cues were 10 s in duration and consisted of repeating motifs of a broadband click , phaser , or trumpet . Each cue was associated with a unique probability of foot shock ( 0 . 5 mA , 0 . 5 s ) : danger , p=1 . 00; uncertainty , p=0 . 375; and safety , p=0 . 00 . Auditory identity was counterbalanced across rats . Foot shock was administered 2 s following the termination of the auditory cue on danger and uncertainty shock trials . This was done in order to observe possible neural activity during the delay period not driven by an explicit cue . A single session consisted of six danger trials , ten uncertainty no-shock trials , six uncertainty shock trials , and ten safety trials . The order of trial type presentation was randomly determined by the behavioral program , and differed for each rat , each session . After the eighth session , rats were removed from discrimination , given full food and received stereotaxic surgery . Following recovery , discrimination ( identical to that described above ) resumed with single-unit recording . Animals received discrimination every other day with recording . After each discrimination session with recording , electrodes were advanced either 0 . 042 mm or 0 . 084 mm to record from new units during the following session . Rats were deeply anesthetized using isoflurane and final electrode coordinates were marked by passing current from a 6 V battery through 4 of the 16 nichrome electrode wires . Rats were perfused with 0 . 9% biological saline and 4% paraformaldehyde in a 0 . 2 M Potassium Phosphate Buffered Solution . Brains were extracted and post-fixed in a 10% neutral-buffered formalin solution for 24 hr , stored in 10% sucrose/formalin and sectioned via microtome . All brains processed for light microscopy using anti-tryptophan hydroxylase immunohistochemistry ( T8575 , Sigma-Aldrich , St . Louis , MO ) and a NovaRed chromagen reaction ( SK-4800 , Vector Laboratories , Burlingame , CA ) Sections were mounted , imaged using a light microscope and electrode placement was confirmed ( Paxinos and Watson , 2007 ) . Sixteen individual recording wires were bundled and soldered to individual channels of an Omnetics connector . The bundle was integrated into a microdrive permitting advancement in ~0 . 042 mm increments . The microdrive was cemented on top of the skull and the Omnetics connector was affixed to the head cap . During recording sessions , a 1x amplifying head stage connected the Omnetics connector to the commutator via a shielded recording cable ( head stage: 40684–020 and Cable: 91809–017 , Plexon Inc , Dallas TX ) . Analog neural activity was digitized and high-pass filtered via amplifier to remove low-frequency artifacts and sent to the Ominplex D acquisition system ( Plexon Inc , Dallas TX ) . Behavioral events ( cues , shocks , nose pokes ) were controlled and recorded by a computer running Med Associates software . Timestamped events from Med Associates were sent to Ominplex D acquisition system via a dedicated interface module ( DIG-716B ) . The result was a single file ( . pl2 ) containing all time stamps for recording and behavior . Single-units were sorted offline with a template-based spike-sorting algorithm ( Offline Sorter V3 , Plexon Inc , Dallas TX ) . Timestamped spikes and events ( cues , shocks , nose pokes ) were extracted and analyzed with statistical routines in MATLAB ( Natick , MA ) . Neural activity was recorded throughout the 500 ms shock delivery period . However , we cannot be certain that shock artifact did not disrupt spike collection , so we do not present activity from this period . The following characteristics were determined for each neuron: baseline firing rate , half the duration of the mean waveform and amplitude ratio of the mean waveform . Duration was determined by measuring the time ( ms ) from peak depolarization to the trough of after-hyperpolarization and dividing by two . Amplitude ratio was calculated using ( n – p ) / ( n + p ) , in which p=initial hyperpolarization ( in mV ) and n = maximal depolarization ( in mV ) . This approach has been used to successfully separate neuron types in the ventral tegmental area ( Roesch et al . , 2007 ) . K-means clustering used these three firing characteristics to partition the 245 recorded neurons into two clusters ( k = 2 ) . Two clusters were chosen because previous studies have found that two neuron types , glutamatergic vGluT2 neurons and GABAergic Gad1 neurons , comprise the majority of vlPAG neurons , and these neurons can be differentiated by baseline firing rate ( Tovote et al . , 2016 ) . ANOVA for cluster results found that only baseline firing rate contributed to cluster membership ( F1 , 243 = 829 , p<0 . 001 ) . Neither amplitude ratio nor duration reached significance ( Fs <0 . 2 , ps >0 . 6 ) . All neurons were clustered , with the majority falling in the low firing rate cluster ( n = 199 ) and the remaining in the high firing rate cluster ( n = 46 ) . Independent of cluster analysis , all 245 neurons were screened for short-latency , excitatory firing to auditory cue onset ( Tovote et al . , 2016; Watson et al . , 2016; Ozawa et al . , 2017 ) . This was achieved using a paired , two-tailed t-test comparing raw firing rate ( spikes/s ) during a 2 s baseline period just prior to cue onset and during the first , 1 s cue interval . A t-test was performed for each of the three cues ( danger , uncertainty and safety ) , corrected for multiple comparisons ( p<0 . 017 ) . The remaining neurons were screened for longer-latency , excitatory firing to the later portion of auditory cues ( Ozawa et al . , 2017 ) , using an identical t-test , only now comparing firing rate during a 2 s baseline period just prior to cue onset and the last , 1 s cue interval . For each neuron , and for each trial type , firing rate ( spikes/s ) was calculated in 100 ms bins from 10 s prior to cue onset to 12 s following cue offset , for a total of 320 bins . Mean firing rate over the 320 bins was calculated by averaging all trials for each trial type . Mean differential firing was calculated for each of the 320 bins by subtracting mean baseline firing rate ( 2 s prior to cue onset ) , specific to that trial type , from each bin . Mean differential firing was Z-score normalized across all trial types within a single neuron , such that mean firing = 0 , and standard deviation in firing = 1 . Z-score normalization was applied to firing across the entirety of the recording epoch , as opposed to only the baseline period , in case neurons showed little/no baseline activity . As a result , periods of phasic , excitatory firing contributed to normalized mean firing rate ( 0 ) . For this reason , Z-score normalized baseline activity is below zero in Figure 2A & C . Z-score normalized firing during cue ( Figure 2A , C ) and post-cue periods ( Figure 6A , B ) , was analyzed with ANOVA using bin and trial-type as factors . F and p values are reported , as well as partial eta squared and observed power . For post hoc cue firing analyses ( Figure 2B , D ) , cue correlation analyses ( Figure 3 all ) , and cue regression analyses , it was necessary to calculate normalized firing in 1 s intervals . To do this , differential firing in the interval of interest ( for example , first cue 1 s interval ) was calculated for each individual of the 32 trials in a single session . Differential firing in this interval was then Z-score transformed . This process was repeated for each interval of interest . This done in order to maximize the distribution of firing within a single interval . Importantly , statistical outcomes were identical if a single Z-score transformation was applied to all intervals at once . An identical approach was used to Z-score normalize firing in 500 ms intervals for post-cue firing ( Figure 6 ) . The analysis for Onset neurons ( n = 29 ) utilized mean normalized firing to each cue ( danger , uncertainty and safety ) in the first 1 s interval; analysis for Ramping neurons ( n = 14 ) utilized firing in the last 1 s interval . Relative firing to the three cues was used to categorize each Onset and Ramping neuron: ( d > u > s ) , ( d > s > u ) , ( s > u > d ) or ( u > d > s ) . Counting the number observed in each category determined the actual number for each population . In order to determine the number in each category expected by chance , each neurons firing pattern was shuffled and the number of neurons in each category counted . Shuffling and counting neuron category was repeated 1000 times for each population . Box and whisker plots for each category/population were constructed showing the median , 25th percentile , 75th percentile and most extreme non-outliers . The actual number and expected statistics are reported plotted together in Figure 4D/H . Population firing was analyzed using analysis of variance ( ANOVA ) with trial type and bin ( 100 ms ) as factors . ANOVA for cue firing contained three trial types ( danger , uncertainty and safety ) . Uncertainty trial types were collapsed because they did not differ for either suppression ratio or firing analysis . This was expected , during cue presentation rats did not know the current uncertainty trial type . ANOVA for post-cue firing contained four trial types ( danger , uncertainty shock , uncertainty omission and safety ) . Uncertainty trial types were split because shock was delivered during the period . F statistic , p value , observed power and partial eta squared are reported for effects and interactions . Interval firing was compared within a population using a two-tailed , dependent samples t-test . Identical firing analyses were performed for the post-cue period , only now four trial types were used ( danger , uncertainty shock , uncertainty omission and safety ) . The sliding window analysis for the Ramping population ( Figure 2E ) employed a two-tailed , dependent samples t-test . Ramping population danger firing was compared to safety firing in 1 s intervals , with the window advancing in 100 ms steps across cue presentation . The p value for each comparison was recorded , and the first window with p<0 . 05 was termed the interval of departure ( when danger and safety firing significant differed ) . The same analysis was performed for uncertainty and safety . Change in firing from interval of departure to the final 1 s window was calculated by subtracting the departure firing rate from the last 1 s window firing rate and dividing by time between these two periods . Biases in single-unit firing to the three cues ( Figure 3 ) were determined using the sign test comparing z firing to danger vs . uncertainty , and uncertainty vs . safety . The relationship between firing was determined using the Pearson correlation coefficient . Single-unit , linear regression was used to determine the degree to which fear output and threat probability explained trial-by-trial variation in firing of single neurons in a specific time interval . The cue analysis used 1 s intervals , while the post-cue analyses used 500 ms intervals . Shorter intervals were used in the post-cue analysis to accommodate the 500 ms foot shock duration . For each regression , all 32 trials from a single session were ordered by type . Z-firing was specified for the interval of interest . The interval fear regressor was the suppression ratio for that specific interval/trial , while the total fear regressor was the suppression ratio for the entire cue , for that specific trial . The probability regressor was the foot shock probability associated with the specific cue . Regression ( using the regress function in MATLAB ) required a separate , constant input . To better visualize the organization of the regression input , the complete regression input for first interval firing of an Onset neuron is shown below . Identical regression analysis was performed for the post-cue period , only now 500 ms intervals were examined and the total fear regressor was removed . This interval was used to accommodate the 500 ms shock period and the total fear regressor was removed because it only sampled behavior from the 10 s cue period . trial #trial typeZ firingconstantinterval feartotal fearprobability1danger2 . 3211 . 001 . 0012danger0 . 5111 . 001 . 0013danger0 . 8111 . 001 . 0014danger1 . 5611 . 001 . 0015danger1 . 411-1 . 000 . 7316danger2 . 9211 . 001 . 0017unc-shock0 . 8111 . 001 . 000 . 3758unc-shock-0 . 401-0 . 330 . 470 . 3759unc-shock0 . 2111 . 000 . 640 . 37510unc-shock-0 . 7011 . 000 . 670 . 37511unc-shock-0 . 8511 . 001 . 000 . 37512unc-shock-0 . 701-1 . 000 . 640 . 37513unc-omission-0 . 7011 . 001 . 000 . 37514unc-omission-0 . 701-1 . 000 . 400 . 37515unc-omission-0 . 7010 . 000 . 230 . 37516unc-omission0 . 6611 . 001 . 000 . 37517unc-omission0 . 2111 . 001 . 000 . 37518unc-omission-0 . 7011 . 000 . 710 . 37519unc-omission-1 . 0011 . 000 . 430 . 37520unc-omission1 . 1111 . 001 . 000 . 37521unc-omission-0 . 091-0 . 600 . 560 . 37522unc-omission-0 . 8511 . 001 . 000 . 37523safety-0 . 701-0 . 330 . 06024safety-0 . 7010 . 000 . 27025safety-0 . 7011 . 00-0 . 04026safety-0 . 4011 . 000 . 14027safety-0 . 7011 . 000 . 54028safety-0 . 701-0 . 500 . 03029safety0 . 2111 . 000 . 47030safety-0 . 8511 . 001 . 00031safety0 . 0611 . 000 . 20032safety-0 . 7011 . 00-0 . 140beta coefficient:0 . 24-0 . 142 . 19 The regression output of greatest interest was the beta coefficient for each regressor ( interval fear , total fear and probability ) , quantifying the strength ( greater distance from zero = stronger ) and direction ( >0 = positive ) of the predictive relationship between each regressor and single-unit firing . ANOVA , two-tailed dependent samples t-test , sign test and The Pearson correlation coefficient was used to analyze beta coefficients , exactly as described for normalized firing . An identical approach was used for the pairs analysis ( Figure 3 ) only now firing of the Ramping neuron was used as a regressor to predict firing of the paired Onset neuron . Beta coefficient ( β ) and significance of the predictive relationship ( p ) are reported . Single-unit , linear regression was performed using the interval fear , total fear and probability regressor as above . Only now , nine separate regression analyses were performed in which the uncertainty component of the probability regressor was systematically varied from 0 to 1 in 0 . 125 increments ( 0 . 000 , 0 . 125 , 0 . 250 , 0 . 375 , 0 . 500 , 0 . 625 , 0 . 750 , 0 . 875 and 1 . 000 ) . The result of primary interest was the mean beta coefficient for the probability regressor from each variant of regression , as plotted in Figure 5E/J . A correlation matrix was created for the post-cue period for the Onset population ( n = 29 ) as well as the Ramping population ( n = 14 ) . The periods of interest were the four , 500 ms intervals prior to shock presentation and the five , 500 ms intervals following shock . Starting with interval 1 , The Pearson correlation coefficient ( R2 ) and associated p value was calculated by comparing the probability beta coefficient for each neuron to those for intervals 2–4 and 6–10 . Most critical were the comparison with intervals 6–10 , that permitted comparison of beta coefficients between pre-shock and post-shock periods . The same relationships were determined for each of the remaining intervals ( 2–4 , and 6–10 ) . The p value of R2 for each interval comparison is plotted . The proportion of significant intervals forming the pre-shock and post-shock comparison ( n = 20 ) were compared for the Onset and Ramping populations using the chi-square test . Full electrophysiology data set will be uploaded to http://crcns . org/ upon acceptance for publication . Med Associates programs used for behavior and MATLAB programs used for behavioral analyses are made freely available at our lab website: http://mcdannaldlab . org/resources .	The brain is hard-wired to detect and respond to threats . Catching sight of a spider or a snake , or hearing an unfamiliar sound in the house at night , can make you freeze momentarily . Multiple regions of the brain contribute to this process . According to the textbook view of threat detection , the amygdala and prefrontal cortex estimate the size of the threat . They then send this information to an area called the ventrolateral periaqueductal gray ( vlPAG ) . The vlPAG responds by triggering a fear response such as freezing . But some studies suggest that the vlPAG may do more than just trigger a fear response . To test this idea , Wright and McDannald trained rats to associate three different 10-second tones with different probabilities of receiving a mild electric shock to the foot . One of the tones was followed by a foot shock 100% of the time . Another tone was followed by a foot shock 37 . 5% of the time , while the third was never followed by a foot shock . The first tone thus signaled danger , the second uncertainty , and the third safety . Wright and McDannald recorded vlPAG activity as the rats heard each of the tones . The recordings revealed two distinct groups of vlPAG neurons . Both groups responded most to the tone that signaled danger , less to the tone that signaled uncertainty , and least to the tone that signaled safety . However , they responded at different times . One group of neurons was most active at the start of the tone . The activity of this group depended upon the degree of threat , and not upon whether the rat showed a fear response . The second group of neurons increased its activity over the course of each tone . The activity of this group mainly reflected the degree of threat , but also represented the rat's fear response to a lesser extent . The vlPAG thus helps to signal the size of a threat , rather than simply generating a fear response . This distinction is important because people with anxiety disorders tend to overestimate threats , and many treatments for anxiety target the brain regions involved in threat estimation . Future studies should examine how the vlPAG works together with other areas , including the amygdala and the prefrontal cortex , to evaluate threats . Understanding this circuit in full could ultimately lead to better treatments for phobias and anxiety .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2019	Ventrolateral periaqueductal gray neurons prioritize threat probability over fear output
The secretory pathway of eukaryotic cells packages cargo proteins into COPII-coated vesicles for transport from the endoplasmic reticulum ( ER ) to the Golgi . We now report that complete genetic deficiency for the COPII component SEC24A is compatible with normal survival and development in the mouse , despite the fundamental role of SEC24 in COPII vesicle formation and cargo recruitment . However , these animals exhibit markedly reduced plasma cholesterol , with mutations in Apoe and Ldlr epistatic to Sec24a , suggesting a receptor-mediated lipoprotein clearance mechanism . Consistent with these data , hepatic LDLR levels are up-regulated in SEC24A-deficient cells as a consequence of specific dependence of PCSK9 , a negative regulator of LDLR , on SEC24A for efficient exit from the ER . Our findings also identify partial overlap in cargo selectivity between SEC24A and SEC24B , suggesting a previously unappreciated heterogeneity in the recruitment of secretory proteins to the COPII vesicles that extends to soluble as well as trans-membrane cargoes . One-third of the vertebrate genome is predicted to encode proteins that are sorted into the secretory pathway en route to intracellular organelles , the cell surface , or the extracellular space ( Palade , 1975; Bonifacino and Glick , 2004 ) . Following synthesis in the endoplasmic reticulum , trans-membrane and soluble proteins co-translationally inserted into the ER are packaged into transport vesicles coated with COPII ( coat protein complex II ) for export from the ER and delivery to the Golgi for further processing ( Lee et al . , 2004 ) . The assembly of the COPII coat is initiated at ER exit sites upon activation of the GTPase SAR1 , which recruits the inner-coat heterodimer SEC23/SEC24 , followed by the assembly of the outer-coat heterotetramer SEC13/SEC31 to generate carrier vesicles destined for the Golgi ( Lee et al . , 2004; Gurkan et al . , 2006 ) . A key sorting event in ER-Golgi transport relies on recognition of specific signals within cargo molecules by the SEC24 subunit of the COPII complex ( Miller et al . , 2002 , 2003; Lee et al . , 2004 ) , though SEC23 may also contribute to cargo selection in some cases ( Kim et al . , 2012 ) . The SEC31/SEC13 outer-coat regulates the size and rigidity of COPII coats to package specialized cargos ( Copic et al . , 2012; Jin et al . , 2012 ) . Mammals express multiple paralogous forms of COPII , including two SAR1 GTPases ( SAR1A/B ) , two SEC23s ( SEC23A/B ) , four SEC24s ( SEC24A-D ) , as well as two SEC31s ( SEC31A/B ) , thus expanding the repertoire of potential COPII coat structures ( Wendeler et al . , 2007; Zanetti et al . , 2011 ) . Biochemical and structural studies of the COPII complex have identified multiple cargo recognition sites on SEC24 ( Bi et al . , 2002; Miller et al . , 2003; Bickford et al . , 2004; Gurkan et al . , 2006; Mancias and Goldberg , 2008 ) . The four mammalian SEC24 paralogs can be divided into two subfamilies , SEC24A/B and SEC24C/D , sharing ∼60% sequence identity within but only ∼25% identity across subfamilies ( Mancias and Goldberg , 2008; Zanetti et al . , 2011 ) . Mammalian SEC24A/B exhibit ∼30% sequence identity to yeast SEC24p , compared to ∼25% for SEC24C/D . The latter share ∼30% identity with LST1p or ISS1p , the non-essential SEC24 paralogs in yeast ( Roberg et al . , 1999; Peng et al . , 2000; Shimoni et al . , 2000 ) . Although deletions of SAR1 , SEC23 , or SEC24 are all lethal in yeast ( Lee et al . , 2004 ) , consistent with the broad function of ER-Golgi transport , mutations in the genes encoding several mammalian COPII components have been associated with remarkably limited phenotypes , often restricted to a specific cell type or tissue . Mutations in human SAR1B result in chylomicron retention disease ( Anderson Disease ) , a distinct defect in fat absorption due to reduced chylomicron assembly and secretion by intestinal enterocytes ( Jones et al . , 2003; Annesi et al . , 2007 ) . In contrast , missense mutations in human SEC23A result in a characteristic syndrome of malformations restricted to the craniofacial skeleton ( Boyadjiev et al . , 2006 , 2010 ) , with similar skeletal abnormalities observed in SEC23A-deficient zebrafish ( Lang et al . , 2006 ) . Mutations in human SEC23B result in the autosomal recessive disorder congenital dyserythropoietic anemia type II , with abnormalities restricted to the hematopoietic erythroid compartment ( Bianchi et al . , 2009; Schwarz et al . , 2009 ) . Surprisingly , although red blood cells appear grossly normal in SEC23B-deficient mice , these animals die at birth due to dramatic destruction of the pancreas ( Tao et al . , 2012 ) . Though human deficiency has not yet been reported for any of the four SEC24 paralogs , SEC24B deficiency in mice leads to failure in neural tube closure resulting from missorting of the signaling molecule VANGL2 ( Merte et al . , 2010 ) . In contrast , murine SEC24D deficiency results in very early embryonic lethality ( Baines et al . , In press ) . Here , we report that complete deficiency of SEC24A is compatible with normal development and survival in mice . However , these animals exhibit markedly reduced plasma cholesterol due to selective blockade in the secretion of PCSK9 , a circulating factor that negatively regulates cell surface LDL receptor expression . SEC24A-deficient mice were generated from an ES cell line carrying a gene trap insertion into intron two of Sec24a ( Figure 1A ) . The gene trap allele ( Sec24agt ) is predicted to direct expression of a hybrid mRNA fusing Sec24a exons 1 and 2 with the gene trap cassette , resulting in a chimeric protein missing the C-terminal ∼85% of SEC24A ( encoded by exons 3–23 ) . An intercross between Sec24a+/gt heterozygous mice produced offspring of all three genotypes at the expected Mendelian ratios ( Table 1 ) . RT-PCR analysis of liver RNA prepared from Sec24agt/gt mice showed a >1000 fold reduction in normal splicing from Sec24a exon 2 to exon 3 across the gene trap , compared to control wild type mice ( Figure 1B ) . Immunoblotting detected an ∼50% reduction in SEC24A protein in whole brain protein extracts from heterozygous Sec24a+/gt mice compared to wild type littermates , with no SEC24A detected in extracts from Sec24agt/gt mice ( Figure 1C ) . Of note , SEC24B protein levels were increased in brain lysates from Sec24agt/gt mice , with a potential slight increase in Sec24a+/gt heterozygous mice , though no differences were observed for SEC24C or SEC24D ( Figure 1C ) . 10 . 7554/eLife . 00444 . 003Figure 1 . SEC24A null mice are viable and exhibit normal survival and development . ( A ) Schematic of the first Sec24a mutant allele ( Sec24agt ) . Gray blocks represent exons with specific numbers indicated . SA , splice acceptor cassette; β-Geo , β-galactosidase-neo fusion; pA , poly-adenylation sequence . F , R , and V represent genotyping primers . Bottom , sequence of Sec24agt gene trap insertion junction; sequence of the gene trap cassette is underlined . ( B ) RT-PCR detection of splicing between exons 2 and 3 in Sec24agt/gt mice . Liver cDNA of wild type mice was serially diluted into liver cDNA of Sec24agt/gt mice as indicated and used as template for PCR with primers Sec24a-Exon2 and Sec24a-Exon3 ( see primer sequences ) . ( C ) Loss of SEC24A protein in Sec24agt/gt mice . Upper panel , PCR genotyping; lower panel , immunoblotting of brain protein extracts from mice with the genotypes indicated at the top , using the indicated SEC24A-D antibodies . ( D ) Body weights of SEC24A-deficient and wild type control mice . HF , high fat diet . Error bars represent SEM ( standard error of the mean ) . At least six mice were included in each group at each time point . ( E ) Kaplan Meier plot for survival of SEC24A-deficient mice ( n = 20 ) and littermate controls ( n = 15 ) . ( F ) Histology of several tissues from Sec24agt/gt mice . Li , liver; H , heart; K , kidney; Lu , lung . DOI: http://dx . doi . org/10 . 7554/eLife . 00444 . 00310 . 7554/eLife . 00444 . 004Table 1 . Distributions of offspring from intercrossDOI: http://dx . doi . org/10 . 7554/eLife . 00444 . 004CrossesGenotype distribution in %p value ( χ2 ) +/++/−−/−Expected %25%50%25%Sec24a+/gt X Sec24a+/gt25% ( 36 ) 48 . 6% ( 70 ) 26 . 4% ( 38 ) > 0 . 9Sec24a+/cgt X Sec24a+/cgt23 . 3% ( 7 ) 43 . 3% ( 13 ) 33 . 3% ( 10 ) > 0 . 5Sec24a+/gt2 X Sec24a+/gt227 . 9% ( 17 ) 49 . 2% ( 30 ) 23% ( 14 ) > 0 . 8Sec24a+/gtSec24b+/- X Sec24agt/gtSec24a+/gt 25%Sec24a+/gtSec24b+/- 25%Sec24agt/gt 25%Sec24agt/gtSec24b+/- 25%Observed20 . 7% ( 19 ) 32 . 6% ( 30 ) 25% ( 23 ) 21 . 7% ( 20 ) > 0 . 35Sec24a+/gtSec24d+/gt X Sec24agt/gtSec24a+/gt 25%Sec24a+/gtSec24d+/gt 25%Sec24agt/gt 25%Sec24agt/gtSec24d+/gt 25%Observed22 . 1% ( 21 ) 29 . 5% ( 28 ) 27 . 4% ( 26 ) 20 . 1% ( 20 ) > 0 . 6Observed numbers are listed in parentheses . Sec24agt/gt mice were grossly indistinguishable from their wild type littermates and developed normally to adulthood with no difference in body weight on a normal or high fat diet ( Figure 1D ) . Kaplan-Meier analysis showed no difference in the survival of wild type and Sec24agt/gt mice up to 12 months of age ( Figure 1E ) . Sec24agt/gt male and female mice both exhibited normal fertility at ∼8 weeks of age with normal litter size ( 7 . 7 ± 1 . 4 , n = 6 , compared to 7 . 1 ± 1 . 0 , n = 10 for congenic C57BL6/J controls , p>0 . 35 ) . Gross and routine microscopic survey of multiple tissues failed to identify any obvious morphologic abnormalities in adult Sec24agt/gt mice ( Figure 1F ) . Complete blood count ( CBC ) surveys were essentially normal in Sec24agt/gt mice compared to wild type controls , except for an ∼4% increase in mean corpuscular volume ( MCV ) and mean corpuscular hemoglobin ( MCH ) and a compensatory ∼6% decrease in red blood cell ( RBC ) number ( Table 2 ) . 10 . 7554/eLife . 00444 . 005Table 2 . Complete blood count surveyDOI: http://dx . doi . org/10 . 7554/eLife . 00444 . 005WT ( n = 10 ) Sec24agt/gt ( n = 6 ) p valueWBC ( X103 ) 11 . 77 ± 2 . 6711 . 53 ± 1 . 600 . 85RBC ( X106 ) 9 . 89 ± 0 . 419 . 28 ± 0 . 460 . 02 HGB ( g/dl ) 12 . 9 ± 0 . 712 . 7 ± 0 . 50 . 51HCT ( % ) 50 . 2 ± 1 . 549 . 2 ± 2 . 70 . 34MCV ( fl ) 50 . 78 ± 0 . 9752 . 72 ± 0 . 660 . 001 MCH ( pg ) 13 . 06 ±0 . 3513 . 48 ± 0 . 230 . 02 MCHC ( % ) 25 . 75 ± 0 . 5525 . 60 ± 0 . 280 . 55CHCM ( g/dl ) 26 . 39 ± 0 . 2126 . 28 ± 0 . 170 . 31RDW ( % ) 12 . 92 ± 0 . 6112 . 62 ± 0 . 700 . 38HDW ( % ) 1 . 692 ± 0 . 0861 . 630 ± 0 . 0710 . 16PLT ( X104 ) 123 . 6 ± 19 . 1118 . 8 ± 8 . 50 . 58MPV ( fl ) 5 . 2 ± 1 . 55 . 73 ± 0 . 20 . 46p < 0 . 05 by Student's t-test . To exclude a contribution from residual function of the Sec24agt allele or a potential ‘passenger' gene mutation as a result of gene targeting ( Westrick et al . , 2010 ) , we generated an additional series of Sec24a deficient mice from a second , independent Sec24a targeted allele ( Figure 2A ) . The parental allele ( Sec24acgt ) contains a conditional gene trap insertion in Sec24a intron 4 . Excision of the LoxP-flanked selection cassette and Sec24a exon 5 by Cre recombinase generated the Sec24agt2 allele , which results in a frame shift after the deleted exon 5 in addition to the gene trap . Removal of the FRT-flanked gene trap and selection cassette by Flpe recombinase generated the Sec24afl allele , which can be converted to the null Sec24a- allele by Cre recombinase . Sec24a+/cgt or Sec24a+/gt2 intercrosses both produced offspring with all three genotypes at the expected Mendelian ratios ( Table 1 ) . No SEC24A protein was detected in brain protein extracts from either Sec24acgt/cgt or Sec24agt2/gt2 mice ( Figure 2B ) . Taken together , these data demonstrated that SEC24A is not required for survival , development , or fertility in the mouse . 10 . 7554/eLife . 00444 . 006Figure 2 . Additional targeted alleles of Sec24a . ( A ) Schematic of additional Sec24a alleles , Sec24acgt , Sec24agt2 , Sec24afl and Sec24a- ( adapted from the Knockout Mouse Project; general conditional gene targeting scheme: https://www . komp . org/alleles . php#conditional-promoter-csd , Sec24a targeting: http://www . knockoutmouse . org/martsearch/project/24915 ) . Gray blocks represent exons . A , B , C , and D , genotyping primers . ( B ) PCR genotyping and immuno-blot analysis of brain extracts in tissues from Sec24acgt/cgt and Sec24agt2/gt2 mice . IB indicates immunoblotting antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 00444 . 006 SDS-PAGE analysis of non-reduced plasma samples from Sec24agt/gt mice and their wild type littermates demonstrated no changes in several abundant plasma proteins , including albumin and transferrin ( Figure 3A asterisks ) . However , a protein migrating at ∼25 kD consistently appeared under-represented by ∼50% in plasma samples from Sec24agt/gt mice ( Figure 3A arrow , p25 ) . Mass spectrometry identified this protein as apolipoprotein A-I ( APO-A1 ) ( Figure 3B ) , and quantification by spectral counts or mean peak area confirmed an ∼30–70% reduction of APO-A1 in Sec24agt/gt plasma compared to wild type . Immunoblotting confirmed an ∼40% decrease in APO-A1 levels in Sec24agt/gt plasma ( Figure 3C ) . 10 . 7554/eLife . 00444 . 007Figure 3 . SEC24A-deficient mice develop hypocholesterolemia . ( A ) Non-reduced plasma protein samples from four wild type and four Sec24agt/gt mice were separated on SDS-PAGE and stained with coomassie brilliant blue . The first lane contains size markers . Asterisks indicate transferrin and albumin . An ∼25kD protein ( p25 ) is under-represented in the plasma of Sec24agt/gt mice . Lower panel , quantification of the intensity for the band labeled ‘p25' in the upper panel; error bars represent SEM . Asterisk , p<0 . 01 by Student's t-test . ( B ) Identification of p25 as APO-A1 by mass spectrometry . Peptide sequences detected in HPLC-ESI-MS/MS analysis; all six peptides exhibit 100% match with mouse APO-A1 sequence . ( C ) Non-reduced plasma protein samples from four wild type or four SEC24A-deficient mice were analyzed by immunoblotting with antibodies to albumin or APO-A1 . ( D ) Pooled plasma samples from seven wild type and eight Sec24agt/gt mice were fractionated by FPLC and cholesterol in each fraction quantified with a colorimetric assay; total cholesterol for the fractions containing HDL , LDL , and VLDL/Chylomicrons are indicated in the table at the bottom . ( E ) Total plasma cholesterol in male wild type control ( n = 10 for 8 weeks of age , n = 4 for 6 months of age ) and Sec24agt/gt ( n = 5 for 8 weeks of age , n = 6 for 6 months of age ) . Error bars represent SEM . Asterisk , p<0 . 001 by Student's t-test . ( F ) Total plasma cholesterol in female wild type control ( n = 7 for 8 weeks of age , n = 4 for 6 months of age ) and Sec24agt/gt ( n = 5 for 8 weeks of age , n = 5 for 6 months of age ) . Error bars represent SEM . Asterisk , p<0 . 001 by Student's t-test . ( G ) Total plasma cholesterol in wild type control ( n = 7 for male , n = 5 for female ) and Sec24agt2/gt2 ( n = 4 for male , n = 4 for female ) . Error bars represent SEM . p<0 . 0001 by Student's t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 00444 . 007 APO-A1 is a core protein component of high-density cholesterol-containing lipoprotein particles and is also found in some other lipoprotein particle species in the circulation ( Hoofnagle and Heinecke , 2009 ) . For this reason , pooled plasma samples from overnight-fasted adult male Sec24agt/gt mice and their wild type littermates were fractionated by FPLC and cholesterol levels in each fraction were measured ( Figure 3D ) . Total cholesterol and High Density Lipoprotein ( HDL ) cholesterol were reduced by ∼40% in Sec24agt/gt plasma compared to wild type , and Low Density Lipoprotein ( LDL ) cholesterol by ∼60% . Similarly reduced plasma cholesterol levels were observed in both male and female SEC24A-deficient mice at different ages ( Figure 3E , F ) . The low cholesterol phenotype was also confirmed in mice harboring the second SEC24A-deficient allele , Sec24agt2/gt2 ( Figure 3G ) . To further exclude a potential contribution from a passenger gene to the hypocholesterolemia phenotype ( Westrick et al . , 2010 ) , we removed the conditional gene trap cassette from the Sec24acgt allele by germline Flpe-mediated excision ( see Experimental procedures ) to generate the Sec24afl allele . Sec24afl/fl mice exhibited full restoration of SEC24A expression in liver lysates ( Figure 4A ) , as well as normalization of total plasma cholesterol levels ( Figure 4B ) . 10 . 7554/eLife . 00444 . 008Figure 4 . Loss of hepatic SEC24A expression leads to hypocholesterolemia . ( A ) Immunoblotting of liver lysates from wild-type and Sec24acgt/cgt mice with an anti-SEC24A and control anti-LMAN1 antibody demonstrates loss of SEC24A expression in Sec24acgt/cgt mice . Removal of the gene trap from the Sec24acgt allele to generate Sec24afl restores wild type SEC24A expression . ( B ) Total plasma cholesterol levels from wild type , Sec24acgt/cgt , and Sec24afl/fl mice were quantified with a colorimetric assay; error bars represent SEM . Asterisk , p<0 . 001 by Student's t-test . ( C ) Protein extracts from the indicated tissues of a wild type mouse were subjected to immunoblotting with the indicated antibodies . ( D ) Hepatic inactivation of SEC24A was performed by intravenous injection of Sec24afl/fl mice with an adenovirus encoding Cre recombinase ( Adv-Cre ) . Liver protein extracts from these mice and control Adv-Cre-injected wild type mice were subjected to immunoblotting with antibodies to SEC24A or a control ( LMAN1 ) . Hepatic SEC24A is reduced by 70–80% in the Adv-Cre-treated Sec24afl/fl mice . ( E ) Total plasma cholesterol was quantified in Adv-Cre-treated mice; error bars represent SEM . n = 4 for wild type mice; n = 6 for Sec24afl/fl mice . p<0 . 001 by Student's t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 00444 . 008 Analysis of public mRNA expression data revealed that Sec24a is ubiquitously expressed in multiple tissues ( Rosenbloom et al . , 2012 ) . Immunoblotting confirmed that SEC24A was present in all tissues tested , although it was relatively enriched in the liver ( Figure 4C ) . Hepatic Sec24a was selectively inactivated in Sec24afl/fl mice by injection of a Cre recombinase-expressing Adenovirus ( Adv-Cre ) . Expression of SEC24A was reduced by 70–80% in the liver of Adv-Cre-treated Sec24afl/fl mice ( Figure 4D ) , resulting in an ∼25% decrease in total cholesterol levels compared to wild type control mice receiving the same adenovirus ( Figure 4E ) . Although this reduction in plasma cholesterol was less than observed in Sec24agt/gt mice , considering the incomplete deletion of hepatic Sec24a by Adv-Cre , these data suggest that loss of hepatic Sec24a expression is sufficient to explain the hypocholesterolemia observed in SEC24A-deficient mice . The SREBP1/2 transcription factors , key determinants of cholesterol metabolism ( Brown and Goldstein , 1997 , 2009 ) , undergo cleavage-dependent activation at the Golgi apparatus regulated by COPII-mediated transport from the ER through the chaperone protein SCAP ( Sun et al . , 2005 , 2007; Brown and Goldstein , 2009 ) . Thus , specific dependence of SREBP/SCAP on SEC24A for ER-Golgi transport could potentially account for the aberrant cholesterol metabolism in SEC24A-deficient mice . To test this hypothesis , we performed mRNA-Seq ( Mortazavi et al . , 2008 ) to examine the hepatic transcriptome in Sec24agt/gt mice ( Figure 5A , Supplementary file 1 ) . No significant differences were observed in the expression levels of the 20 most abundant hepatic transcripts between wild type and Sec24agt/gt mice ( <1 . 2 fold , False Discovery Rate [FDR] > 0 . 4 , Figure 5B ) . Among the ∼20 , 000 identified transcripts , only ∼30 showed significant down-regulation ( >2 fold , FDR < 0 . 05 , RPKM > 0 . 2 ) , with Sec24a as the most significantly down-regulated mRNA in Sec24agt/gt liver ( Figure 5C ) . Gene-ontology enrichment analysis failed to identify a consistent pattern in the changes in transcript levels , though modest shifts were noted in a few metabolic genes ( Supplementary file 2 ) . There was no significant correlation between the gene expression profile for Sec24agt/gt mice and previously reported profiles from mice with altered expression of SCAP or SREBP1/2 ( Horton et al . , 2003 ) . Specifically , transcripts for direct SREBP1/2 targets , including the Low Density Lipoprotein Receptor ( LDLR ) , HMG-CoA reductase , and PCSK9 , remained unchanged by mRNA-seq and qPCR , though SCD1 was decreased by ∼50% in Sec24agt/gt mice ( Figure 5D ) , as observed by mRNA-seq ( FDR < 3 × 10−13 ) . In addition , no detectable difference in the extent of SREBP1/2 cleavage was observed by immunoblotting of liver lysates from Sec24agt/gt mice and their wild type littermates ( Figure 5E ) . 10 . 7554/eLife . 00444 . 009Figure 5 . SEC24A deficiency does not alter SREBP signaling . ( A ) Volcano plot of liver transcriptome analysis by mRNA-Seq . X-axis , -Log ( p-value ) ; y-axis , Log2 ( fold difference WT/Sec24agt/gt ) . Significantly altered genes ( fold change > 2 , RPKM > 0 . 1 , and FDR < 0 . 05 ) are colored in red . ( B ) Twenty most abundant hepatic transcripts detected in wild type and Sec24agt/gt mice by mRNA-Seq . X-axis , gene name; y-axis: Log ( RPKM value ) . ( C ) Hepatic transcripts significantly down-regulated by SEC24A deficiency . X-axis , gene name; y-axis: Log2 ( RPKM value ) . ( D ) Liver mRNA samples from wild type ( n = 4 ) or SEC24A-deficient mice ( n = 4 ) were subjected to quantitative-PCR ( q-PCR ) with primers for the indicated SREBP-regulated transcripts . Error bars represent SEM . Asterisk , p<0 . 01 by Student's t-test . ( E ) Liver protein extracts from wild type and Sec24agt/gt mice were subjected to immunoblotting with the indicated antibodies . ( F ) Transcript abundance detected by mRNA-Seq for COPII genes in the liver of wild type and Sec24agt/gt mice . ( G ) SEC24A and SEC24C/D are expressed at comparable levels in the liver . Liver protein extracts from three wild type and three Sec24agt/gt mice and 293T cells expressing RFP-tagged SEC24A , SEC24C or SEC24D as references were analyzed with the indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 00444 . 009 Of note , RNA-seq detected expression of all four SEC24 paralogs in normal liver , with the number of transcripts for Sec24c and Sec24d–twofold greater than Sec24a and 5–10 fold greater than Sec24b ( Figure 5F ) . Quantitative immunoblotting using RFP-tagged SEC24A and SEC24C/D as references confirmed the presence of SEC24C/D proteins at similar levels to SEC24A in wild type liver , and no change in SEC24C/D levels in Sec24agt/gt liver ( Figure 5G ) . The reduction in circulating cholesterol in Sec24agt/gt mice could result from decreased hepatic output or increased clearance by the liver . To distinguish these two mechanisms , we crossed Sec24agt/gt mice with Apoe null mice , in which receptor-mediated clearance of cholesterol-rich lipoprotein remnant particles from the plasma is inhibited due to loss of the critical lipoprotein particle component APOE . Consistent with previous reports ( Piedrahita et al . , 1992; Zhang et al . , 1992 ) , Apoe-/- mice showed markedly elevated cholesterol levels ( 272 ± 13 mg/dl , n = 8 ) . However , in contrast to the ∼45% reduction in plasma cholesterol observed in mice singly deficient for Sec24a ( Figure 3D ) , the plasma cholesterol levels in Apoe-/-Sec24agt/gt mice ( 277 ± 14 mg/dl , n = 7 ) were indistinguishable ( p>0 . 8 ) from singly Apoe-/- littermates ( Figure 6A ) . Thus , Apoe is epistatic to Sec24a , suggesting that SEC24A may regulate cholesterol metabolism via the receptor-mediated clearance pathway . To further test this hypothesis , we also crossed Sec24agt/gt mice into the Ldlr null background . Consistent with previous reports ( Ishibashi et al . , 1993 ) , Ldlr-/- mice exhibited markedly elevated plasma cholesterol levels ( 211 ± 9 mg/dl , n = 14 ) . Doubly deficient Ldlr-/-Sec24agt/gt mice also displayed similarly elevated plasma cholesterol levels ( 197 ± 12 mg/dl , n = 11 ) , indistinguishable ( p>0 . 35 ) from their Ldlr-/- littermates ( Figure 6B ) . Taken together , these data suggest that reduction in plasma cholesterol in SEC24A-deficient mice could result from increased clearance of cholesterol-rich lipoproteins by the LDLR . Consistent with this hypothesis , immunoblotting demonstrated elevated LDLR levels in liver protein extracts or in proteins bound to a wheat germ agglutinin ( WGA ) matrix from Sec24agt/gt mice compared to littermate controls , whereas other proteins including APOE , APOB48/100 , or APO-A1 showed little alteration in the liver ( Figure 6C ) . Since hepatic Ldlr transcripts remained unaltered in Sec24agt/gt by both mRNA-Seq and qPCR ( Figure 5D ) , these data indicate that the reduction in LDLR protein levels in SEC24A-deficient mice was mediated via a translational or post-translational mechanism . 10 . 7554/eLife . 00444 . 010Figure 6 . SEC24A deficiency up-regulates LDLR protein levels by decreasing circulating PCSK9 . ( A ) SEC24A deficiency does not cause hypocholesterolemia in the setting of APOE deficiency . Total plasma cholesterols in Apoe-/- ( n = 8 ) and Apoe-/-Sec24agt/gt ( n = 7 ) mice . Error bars represent SEM . p=∼0 . 8 by Student's t-test . ( B ) . Total plasma cholesterol levels from Ldlr-/- ( n = 14 ) and Ldlr-/-Sec24agt/gt ( n = 11 ) mice . Error bars represent SEM . p=∼0 . 35 by Student's t-test . ( C ) Liver protein extracts from wild type and Sec24agt/gt mice were subjected to immunoblotting with the indicated antibodies . ( D ) Plasma PCSK9 levels from wild type ( n = 8 ) and Sec24agt/gt ( n = 8 ) mice were quantified by ELISA . Error bars represent SEM . p<0 . 001 by Student's t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 00444 . 010 LDLR levels are regulated by intracellular factors such as the E3 ligase IDOL ( Zelcer et al . , 2009 ) and circulating PCSK9 . The latter is secreted into the plasma primarily by hepatocytes , binding to LDLR and promoting its endocytosis and degradation ( Costet et al . , 2008; Horton et al . , 2009 ) . No difference in the level of IDOL was observed in immunoblotting of liver lysates prepared from SEC24A-deficient mice compared to their wild type littermates ( Figure 6C ) . To assess a potential role for PCSK9 in the low plasma cholesterol phenotype of SEC24A-deficient mice , we measured circulating PCSK9 levels by ELISA . An ∼55% reduction of plasma PCSK9 was observed in Sec24agt/gt mice ( Figure 6D ) , suggesting that SEC24A-deficiency may lower plasma cholesterol via reduced secretion of PCSK9 . No change in PCSK9 hepatic mRNA was observed between SEC24agt/gt mice and their wild type littermates by mRNA-seq or qPCR ( Figure 5D ) , suggesting a translational or post-translational mechanism for the reduction of circulating PCSK9 in SEC24A-deficient mice . Analysis of liver protein extracts by immunoblotting demonstrated significant accumulation of both mature and pro-PCSK9 in the liver of Sec24agt/gt mice compared to littermate controls ( Figure 7A ) . To determine the site of PCSK9 intracellular accumulation , we treated liver lysates with endoglycosidase H ( EndoH ) . Nearly all of the intracellularly-accumulated PCSK9 in Sec24agt/gt liver exhibited sensitivity to EndoH with faster eletrophoresis mobility ( Figure 7B ) , suggesting that SEC24A deficiency leads to accumulation of PCSK9 in the ER . 10 . 7554/eLife . 00444 . 011Figure 7 . PCSK9 is a soluble COPII cargo . ( A ) Liver protein extracts from wild type and Sec24agt/gt mice and cell lysates from 293T cells expressing PCSK9 were subjected to immunoblotting with two different anti-PCSK9 antibodies . Red asterisks indicate non-specific bands . ( B ) Liver protein extracts from SEC24A-deficient mice were subjected to Endo H treatment before SDS-PAGE and immunoblotting with an anti-PCSK9 antibody . n . s . , non-specific band . ( C ) Cell lysates and conditioned medium from 293T cells stably expressing PCSK9-FLAG were analyzed by immunoblotting with an anti-FLAG antibody following treatment with or without BFA . The arrows indicate un-cleaved ( upper ) and auto-cleaved ( lower ) forms of PCSK9 . ( D ) Permeabilized 293T cells stably expressing PCSK9-FLAG were employed in an in vitro COPII budding assay; the resulting vesicle fractions and permeabilized cell inputs were separated by SDS-PAGE and visualized by immunoblotting . The arrows indicate un-cleaved ( upper ) and auto-cleaved ( lower ) forms of PCSK9 . ( E ) Cell lysates and conditioned medium from 293T cells stably expressing PCSK9-FLAG transfected with a vector control or a plasmid expressing a dominant-negative mutant SAR1 ( H79G ) and analyzed by immunoblotting with an anti-FLAG antibody . The arrows indicate un-cleaved ( upper ) and auto-cleaved ( lower ) forms of PCSK9 . DOI: http://dx . doi . org/10 . 7554/eLife . 00444 . 011 These data suggest that COPII-dependent ER-Golgi transport plays a critical role in PCSK9 secretion from the cell . Belfeldin A ( BFA ) disrupts the Golgi apparatus and consequently the delivery of ER-derived COPII vesicles along the secretory pathway . When BFA was applied to a 293T cell line stably expressing PCSK9 fused with a C-terminal FLAG tag ( PCSK9-FLAG ) , PCSK9 secretion into the medium was abolished and the protein accumulated intracellularly ( Figure 7C ) . To directly test whether the COPII machinery plays a role in ER exiting of PCSK9 , we employed permeabilized 293T cells stably expressing PCSK9-FLAG in an in vitro COPII vesicle budding assay . In the presence of rat liver cytosol supplying COPII components , along with ATP and GTP , PCSK9 was packaged into COPII vesicles that sedimented in a high-speed pellet fraction , together with the classic COPII cargo SEC22B , but not the ER resident protein roboporin ( Figure 7D ) . ER budding of PCSK9 and SEC22B-containing vesicles were inhibited by GTPyS as well as constitutively active SAR1 ( SAR1A H79G ) , both of which prevent GTP hydrolysis and subsequently COPII vesicle budding from the ER membrane ( Figure 7D ) . Inhibition of COPII vesicle formation in cells by SAR1B H79G also prevented secretion of PCSK9 into the medium , with corresponding intracellular accumulation of the protein ( Figure 7E ) . These data demonstrate that PCSK9 secretion is dependent on ER exiting in COPII-coated vesicles . SEC24 plays a central role in the recruitment of cargos to the COPII vesicle ( Miller et al . , 2002 , 2003 ) , with different paralogs thought to provide specificity in cargo selection ( Mancias and Goldberg , 2008; Zanetti et al . , 2011 ) . To test whether SEC24 paralogs provide specificity in transporting PCSK9 , we introduced RFP-tagged SEC24A-D paired with GFP-tagged SEC23A into 293T cells stably expressing PCSK9-FLAG . SEC24A and SEC24B , but not SEC24C or SEC24D , associated with PCSK9 precipitated by anti-FLAG antibody in CHAPS-based buffer ( Figure 8A ) , suggesting that SEC24A and SEC24B may selectively mediate PCSK9 ER exiting and secretion . To directly test this hypothesis , we introduced RFP-tagged SEC24A-D paired with GFP-tagged SEC23A together with PCSK9-FLAG into 293T cells . Expression of SEC24A paired with SEC23A ( 5 . 4 ± 1 . 7 fold compared to RFP control , p<0 . 05 ) , and to a lesser extent SEC24B/SEC23A ( 2 . 1 ± 0 . 4 fold compared to RFP control , p<0 . 05 ) , reduced intracellular levels of PCSK9 and led to increased secretion of the protein into the medium , while SEC24C or SEC24D in complex with SEC23A had little effect ( Figure 8B , C ) . This effect was absent when ER-Golgi transport was blocked by BFA treatment , further suggesting that SEC24A , and to a lesser extent SEC24B , selectively mediate COPII-dependent secretion of PCSK9 . 10 . 7554/eLife . 00444 . 012Figure 8 . SEC24A regulates PCSK9 secretion . ( A ) Cell lysates from 293T cells stably expressing PCSK9-FLAG transfected with plasmids expressing RFP-tagged SEC24A-D or a control RFP vector together with GFP-tagged SEC23A , subjected to immune-precipitation; and immune-complexes and cell lysates then examined by immunoblotting with anti-RFP or anti-FLAG antibodies . The arrows indicate un-cleaved ( upper ) and auto-cleaved ( lower ) forms of PCSK9 . ( B ) Cell lysates and conditioned medium from 293T cells co-transfected with PCSK9-FLAG and SEC23/24 plasmids as in ( A ) were subjected to immunoblotting with the indicated antibodies . Cells treated with BFA were employed as controls . ( C ) Ratio of secreted PCSK9/intracellular PCSK9 for each transfected RFP-tagged SEC24 and control . Quantification was performed from five independent experiments . Error bars represent SEM . Asterisk , p<0 . 05 by Student's t-test . ( D ) Deficient McA-RH777 cells treated with the indicated siRNAs were subjected to immunoblotting to determine SEC24A and SEC24B levels . Numbers indicate the cells receiving different siRNAs . ( E ) Deficient McA-RH777 cells treated with the indicated siRNAs used as the source of cytosol for in vitro COPII budding assay as in ( 7d ) . ( F ) Quantification of PCSK9 packaging into COPII vesicles from four different experiments . Error bars represent standard deviation . p<0 . 01 by Student's t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 00444 . 012 To more directly examine the dependence of PCSK9 on SEC24A for efficient exit from the ER , the in vitro COPII vesicle budding assay was performed using cytosols from rat hepatoma McA-RH777 cells treated with control siRNA , or siRNAs against SEC24A , SEC24B , or both proteins . Depletion of SEC24A/B was confirmed by immunoblotting ( Figure 8D ) . Compared to control siRNA treatment , PCSK9 packaging into COPII vesicles was reduced upon depletion of SEC24A ( Figure 8E , F ) , and to a greater extent with depletion of both SEC24A and SEC24B . Similar reduction in COPII vesicle packing was observed for SEC22B , consistent with the previous report of its selectivity as a cargo for SEC24A/B ( Mancias and Goldberg , 2007 ) . In contrast , packaging of the non-selective cargo , LMAN1 , was not significantly altered by depletion of SEC24A , SEC24B , or both , consistent with previous reports ( Mancias and Goldberg , 2007; Wendeler et al . , 2007 ) . To test the selectivity of SEC24 paralogs in vivo , we performed crosses between SEC24B- and SEC24A-deficient mice ( Table 1 ) . Haplo-deficiency for both Sec24a and Sec24b resulted in no discernible phenotype and normal plasma cholesterol levels ( Figure 9A ) . However , Sec24agt/gt mice that were also haplo-deficient for Sec24b exhibited a further ∼25% reduction in plasma cholesterol compared to Sec24agt/gt littermates ( Figure 9A ) , although they appeared otherwise normal . These data suggest a partial overlap in function between SEC24A and SEC24B . 10 . 7554/eLife . 00444 . 013Figure 9 . SEC24B but not SEC24D exhibit partial overlap in function with SEC24A in vivo . ( A ) Total plasma cholesterol levels from mice generated from a Sec24a+/gtSec24b+/- X Sec24agt/gt cross . Error bars represent SEM . Asterisk , p<0 . 003 by Student's t-test . ( B ) Total plasma cholesterol levels from mice generated from a Sec24a+/gtSec24d+/gt X Sec24agt/gt cross . Error bars represent SEM . * p<0 . 003; n . s . , p=∼0 . 6 , by Student's t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 00444 . 013 The possibility of genetic interaction between Sec24a and Sec24d was also examined ( Table 1 ) . As for SEC24B above , Sec24a+/gtSec24d+/gt compound heterozygous mice also exhibited no discernible phenotype and had normal plasma cholesterol ( Figure 9B ) . Although SEC24D deficiency in mice results in early embryonic lethality ( Baines et al . , In press ) , Sec24agt/gtSec24d+/gt mice were also viable ( Table 1 ) and displayed no further reduction in plasma cholesterol compared to Sec24agt/gt littermates ( Figure 9B ) , indicating distinct functions between SEC24D and SEC24A/B . The finding that complete deficiency of the COPII subunit SEC24A is compatible with normal survival and development in the mouse is surprising , in light of its ubiquitous expression and presumed fundamental function in the secretory pathway . However , examination of these animals uncovered an unexpected reduction in plasma cholesterol due to a specific block in the secretion of PCSK9 , a circulating regulator of cell surface LDL receptor ( Costet et al . , 2008; Horton et al . , 2009 ) . Genetic crosses demonstrate that both Apoe and Ldlr are epistatic to Sec24a , suggesting that SEC24A primarily affects receptor-mediated cholesterol clearance of cholesterol-rich lipoproteins . Consistent with these genetic data , hepatic LDLR levels are up-regulated in SEC24A deficient mice as a consequence of a specific dependence of PCSK9 on SEC24A for efficient exit from the ER . Our findings also identify a partial overlap in cargo selectivity between SEC24A and B , and suggest a previously unappreciated heterogeneity in the recruitment of secretory proteins to the COPII vesicles that extends to soluble as well as trans-membrane cargos . Yeast SEC24p play a central role in the recruitment of cargo proteins into COPII vesicles to enter the secretory pathway ( Miller et al . , 2002 , 2003 ) , and the non-essential SEC24 homologs LST1p and ISS1p may co-operate with SEC24p for efficient sorting of a specialized set of cargos ( Roberg et al . , 1999; Peng et al . , 2000; Shimoni et al . , 2000 ) . The expansion of the SEC24 family in vertebrates may have facilitated the accommodation of a wider range of cargo proteins ( Jensen and Schekman , 2011; Zanetti et al . , 2011 ) . This notion is supported by the diverse phenotypes observed in vertebrates with selective deficiency of a single SEC24 paralog . The mammalian subfamily of SEC24A/B exhibits greater sequence identity than SEC24C/D to the essential yeast gene product SEC24p , with SEC24C/D closer to the non essential yeast gene products LST1p and ISS1p ( Peng et al . , 2000; Mancias and Goldberg , 2008; Zanetti et al . , 2011 ) . However , SEC24A-deficient mice are remarkably normal , aside from reduced plasma cholesterol . Even the more severe neural tube closure defect in SEC24B-deficient mice ( Merte et al . , 2010 ) stands in stark contrast to the very early embryonic lethality in SEC24D-deficient mice ( Baines et al . , In press ) . This wide range of SEC24-deficient phenotypes in vertebrates could be explained by either ( 1 ) paralog-specific sorting selectivity encoded by intrinsic cargo selectivity or ( 2 ) tissue/cell-type-specific differences in relative abundance . Our data , together with those of Merte et al . ( Merte et al . , 2010 ) , suggest that intrinsic cargo selectivity among SEC24 paralogs is a key determinant of SEC24 function . The latter authors demonstrated specific dependence of VANGL2 , a key component of the planar cell polarity signaling pathway , on the SEC24B paralog for ER exiting ( Merte et al . , 2010 ) . Our findings identify PCSK9 as a specific SEC24A-dependent COPII cargo . Although all four SEC24 paralogs are expressed in the liver , loss of SEC24A disrupts PCSK9 secretion without affecting other COPII-dependent processes such as SREBP activation . SEC24B exhibits limited compensation for SEC24A , with combined genetic deficiencies providing no evidence for functional overlap with SEC24D . These data suggest overlapping cargo specificities between the members of the SEC24A/B subfamily , yet distinct from SEC24C/D . Furthermore , steady state levels of SEC24B , but not SEC24C/D , are increased in the absence of SEC24A ( Figure 1C ) , although SEC24B mRNA levels remain constant . These data imply that SEC24A/B are balanced in a cytoplasmic pool distinct from SEC24C/D . Taken together , our data raise the possibility of extensive heterogeneity among COPII vesicles and their function in cargo selection and export , determined at least in part by SEC24 paralog composition . Intrinsic selectivity of cargos by SEC24s presumably relies on specific protein-protein interactions ( Lee et al . , 2004; Zanetti et al . , 2011 ) . Trans-membrane proteins including SEC22 , Syntaxin 5 , APP1 , and VANGL2 , could preferentially interact with some but not all SEC24 paralogs for differential recruitment to the COPII vesicle ( Mossessova et al . , 2003; Kim et al . , 2007; Mancias and Goldberg , 2008; Merte et al . , 2010 ) . In contrast , the cargo receptor LMAN1 interacts equally with all four SEC24 paralogs ( Mancias and Goldberg , 2007; Wendeler et al . , 2007 ) . Structural studies have revealed distinct cargo binding sites among SEC24 paralogs that recognize different sorting signals in specific trans-membrane cargos ( Mancias and Goldberg , 2008 ) . Nevertheless , whether and how soluble cargos are selectively transported via COPII vesicles remain to be fully defined ( Warren and Mellman , 1999; Lee et al . , 2004; Jensen and Schekman , 2011 ) . The ‘bulk flow' model proposes that soluble cargos exit the ER by default without selection ( Wieland et al . , 1987; Martinez-Menarguez et al . , 1999; Warren and Mellman , 1999; Thor et al . , 2009 ) , and would therefore predict that alterations of overall SEC24 abundance would proportionally affect secretion of all soluble cargos . The selective dependence of PCSK9 on SEC24A ( and to a lesser extent on SEC24B ) , is inconsistent with this model . Indeed , PCSK9 represents the first example of a soluble vertebrate cargo that is differentially regulated by specific interaction with selective components of the COPII machinery . This specificity might be conferred by a cargo receptor that recruits PCSK9 into COPII vesicles for efficient secretion , in contrast to the bulk flow of more abundant soluble secretory proteins ( Martinez-Menarguez et al . , 1999; Warren and Mellman , 1999 ) . Yeast α-factor is recruited to COPII vesicles and interacts with SEC24 ( Kuehn et al . , 1998; Shimoni et al . , 2000 ) , likely through the cargo receptor ERV29p ( Malkus et al . , 2002 ) . Genetic and biochemical studies have identified the LMAN1/MCFD2 complex as the cargo receptor for the soluble coagulation factors V and VIII in mammals ( Nichols et al . , 1998; Zhang et al . , 2003 , 2005; Khoriaty et al . , 2012 ) . Additionally , TANGO1 localized to ER exit sites facilitates loading of bulky cargos such as collagen VII ( Saito et al . , 2009 , 2011 ) . The existence of a PCSK9 ER cargo receptor has been proposed ( Nassoury et al . , 2007 ) , but its identity has been unclear . Alternatively , COPII vesicles coated by SEC24C/D might exclude selected soluble cargos such as PCSK9 . Additional cytosolic factor may also contribute to the selectivity among SEC24 paralogs as observed in the case of VANGL2 sorting ( Merte et al . , 2010 ) . Taken together with the broad range of specific phenotypes observed for human and murine mutations in the genes for individual COPII components , our findings suggest the potential for considerable heterogeneity in the COPII machinery , enabling the accommodation of diverse cargos . Although complete deficiency for SEC24B or SEC24A in mice is compatible with embryonic development or survival to adulthood , respectively , no human patients have yet been identified with genetic deficiencies in any of the four SEC24 paralogous genes . The SEC24A-deficient phenotype suggests a potential role for genetic variation at the SEC24A locus in the control of plasma cholesterol in humans , a key determinant of risk for myocardial infarction and stroke . Although genome-wide association studies for plasma lipid phenotypes have not identified a significant contribution from common genetic variants in the SEC24A gene ( Teslovich et al . , 2010 ) , a role for rare alleles at this locus cannot be excluded . Complete deficiency of SEC24A , even in the presence of haplo-insufficiency of SEC24B , is compatible with survival and normal development in the mouse , suggesting that pharmacologic inhibition of hepatic SEC24A expression/function to achieve reduction in plasma cholesterol may be well tolerated as a potential approach to inhibit PCSK9 scretion . ES cell clone XE182 ( 129 genetic background ) with a gene-trap insertion into intron 2 of the Sec24a gene , was obtained from the International Gene Trap Consortium ( IGTC ) . The XE182 ES cell clone was expanded and then injected into C57BL/6J blastocysts at the University of Michigan Transgenic Mouse Core . Germ-line transmission was achieved by mating chimeric founders with 129/SvImJ mice , and the resulting germ-line transmitted gene-trap allele was continuously backcrossed to C57BL/6J mice for at least seven generations prior to experimentation . Mice carrying a conditional Sec24a gene-trap allele in the C57BL6/J background were obtained from the Knockout Mouse Project ( KOMP ) . Sec24b and Sec24d mutant mice have been described previously ( Baines et al . , In press; Merte et al . , 2010 ) . Genotyping was performed with mouse tail clip DNA using Go-Taq Green Master MIX ( Promega , Madison , WI ) , and the resulting PCR products were resolved by 2% agarose gel electrophoresis . Primer sequences are listed under ‘Primer sequences' . Apoe ( stock no . 002052 ) , Apob ( stock no . 007682 ) , Ldlr ( stock no . 002207 ) , and Pcsk9 ( stock no . 005993 ) mutant mice were obtained from The Jackson Laboratory . Transgenic mice ( C57/BL6 background ) carrying Flpe recombinase driven by an actin promoter ( stock no . 005703 ) or Cre recombinase driven by an EIIA promoter ( stock no . 003724 ) were obtained from the University of Michigan Transgenic Animal Core . Complete Blood Counts ( CBC ) were measured in an Advia120 whole blood analyzer ( Bayer ) , according to the manufacturer's instructions . High fat diet ( 45% of Calories as fat ) was purchased from Research Diets . Chow diet ( TD 7001 ) supplemented with 60 mg/kg atorvastatin ( Lipitor ) was purchased from Harlan Teklad . Animals were housed according to the guidelines of the University of Michigan Unit of Laboratory Animal Medicine ( ULAM ) . Blood was collected using heparin-coated collection tubes ( Fisher , Pittsburgh , PA ) by retro-oribtal bleeding from mice anaesthetized with isoflurane . Fractionation of plasma samples and quantification of cholesterol were performed at the University of Cincinnati Mouse Metabolic Phenotyping Center . Adenovirus encoding Cre recombinase was purified and injected intravenously through mouse tail veins as previously described ( Li et al . , 2008 ) . Plasma samples were then collected by centrifugation of heparinized blood samples at 3 , 000 g for 5 min at 4°C . Plasma cholesterol levels were measured with a colorimetric assay using the LiquiColor Cholesterol test kit ( Stanbio , Boerne , TX ) . Plasma PCSK9 levels were determined using the mPCSK9 ELISA kit ( Circulex , Woburn , MA ) , according to the manufacturer's instructions . cDNAs of mouse Sec24 a/b/c/d were obtained from ATCC and sub-cloned into the pRFP-C1 vector ( Chen et al . , 2007 ) or the peGFP-C1 vector ( Clontech , Mountain View , CA ) . Mouse Sec23a cDNA in the peGFP-C1 vector has been described previously ( Tao et al . , 2012 ) . Mouse Pcsk9 cDNA was purchased from the ATCC and subcloned into the pLenti vector ( Chen et al . , 2007 ) , with or without a FLAG epitope fused at the C-terminus of PCSK9 . A mammalian expression construct of Sar1 H79G was kindly provided by B . Ye ( University of Michigan ) . Other constructs have been described previously ( Tao et al . , 2012 ) . All constructs were confirmed by complete DNA sequencing at the University of Michigan DNA Sequencing Core . Ten µl plasma samples were first diluted with 90 µl Phosphate Buffer Saline ( PBS; GIBCO , Grand Island , NY ) , and then mixed 1:1 with 2× SDS-PAGE sample buffer ( Invitrogen , Grand Island , NY ) without any reducing agent . 10 µl of each sample ( equal to 0 . 5 µl plasma ) was separated by 4–20% Tris-Glycine SDS-PAGE ( Invitrogen ) with the Seeblue Plus 2 protein size marker ( Invitrogen ) . Gels were first rinsed with de-ionized water for 10 min , and fixed with 20% methanol + 10% acetic acid for 30 min . Fixed gels were stained with 1% coomassie brilliant blue ( Sigma-Aldrich , St Louis , MO ) in 25% methanol for 60 min before de staining with 15% methanol for 15 min ( 3 times ) . HPLC-ESI-MS/MS was performed as described previously ( Yi et al . , 2006; Chao et al . , 2012 ) with instrument specific modifications . Briefly , the gel portions containing APOA1 were excised , destained , dehydrated , and subjected to trypsin digestion overnight . The resulting peptides were desalted and analyzed by on-line HPLC on a Linear Trap Quadrupole-Orbitrap Elite ( LTQ-Orbitrap Elite ) . An anti-SEC24A antibody was generated against a synthetic peptide ( NTYDEIEGGGFLATPQL-C ) in rabbits ( Pacific Immunology , Ramona , CA ) . Purification of the rabbit serum was performed with an affinity column conjugated with the antigenic peptide as previously described ( Chen et al . , 2011a ) . Rabbit anti-FLAG antibody , mouse anti-actin antibody , and rabbit anti-SEC23 antibody were purchased from Sigma-Aldrich . Rabbit anti-PCSK9 ( Ab-1 ) and rabbit anti-SREBP1 antibody were purchased from Santa Cruz Biotechnology . A second rabbit anti-PCSK9 antibody ( Ab-2 ) was kindly provided by J Horton ( UT Southwestern ) . Rabbit anti-APO-A1 antibody and rabbit anti-LMAN1 antibodies were purchased from Stressgen . Goat anti-albumin antibody , rabbit anti-SREBP2 antibody , rabbit anti-LDLR antibody , rabbit anti-Idol ( Mylip ) antibody , and HRP-conjugated anti-RFP antibody were purchased from Abcam ( Eugene , OR ) . A second rabbit anti-Idol ( Mylip ) antibody was purchased from Protein Tech Group . Mouse anti-RALA antibody was purchased from BD Biosciences ( Mountain View , CA ) . Preparation of anti-SAR1 , anti-riboporin , anti-SEC24B , and anti-SEC22B antibodies , and purification of recombinant COPII proteins have been described previously ( Kim et al . , 2007 ) . Tissues were dissected from euthanized animals in cold PBS and immediately transferred to RNAlater ( Ambion , Grand Island , NY ) for storage . Total RNA was isolated using an RNeasy Kit ( Qiagen ) . Whole transcriptome cDNA libraries were generated using TruSeq Stranded mRNA Sample Prep Kit ( Illumina ) , sequenced on a Genome Analyzer II ( Illumina , San Diego , CA ) , and sequencing data were mapped to the mouse reference genome mm9 using Bowtie ( Langmead et al . , 2009 ) . Using ERANGE software ( Mortazavi et al . , 2008 ) ( http://woldlab . caltech . edu/gitweb ) , the unique reads falling on the gene models and splice reads derived from splice junctions of genes were recorded . Following the author's recommendation , the unique reads were also re-evaluated by assessing a first-pass reads per KB per millions reads ( RPKM ) , and the unique reads were subsequently re-calculated with weights computed during the first pass . All candidate regions that were within a 20 kb radius of a gene , a default parameter recommend by the script author , were reported . Final exonic read density ( final RPKM ) was determined according to the expanded exonic read density . Differential expression analysis was performed to compare wild type vs Sec24agt/gt . To obtain robust variance estimates , we empirically estimated the gene variance levels as a function of the average normalized read count per gene . Specifically , the gene-wise variance levels were estimated by calculating the differences in expression between samples for each gene , using local regression to predict the absolute value of the log-difference based on log2-read counts , and then estimating the variance using the standard sample variance formula and Fisher's Z distribution as in the methods of the eBayes function in the limma R package and in the IBMT method ( Smyth , 2004; Sartor et al . , 2006 ) . This results in a moderated variance test that accounts for the dependence of variance on gene length and expression level . The variance estimates are expected to be slightly conservative , due to the assumption that relatively few genes are truly differentially expressed . Given the variance estimates , z-scores and p-values were calculated for each gene . Fold changes on the read counts ( normalized per million bp ) were calculated , and the p-values were adjusted for multiple testing using the Benjamini-Hochberg False Discovery Rate ( FDR ) approach ( Benjamini et al . , 2001 ) . FDR < 0 . 05 and a fold change > 2 cut-off were used to select final up- and down-regulated gene lists . In addition to the gene level analysis , read counts and fold changes were also calculated using ERANGE for individual exons . Based on the counts of reads of individual exons and samples for each gene , Fisher's exact tests were performed to identify significant changes in expression ratios across exons , indicating differential splicing between the samples . Fisher's exact p-values were also adjusted for multiple testing using the Benjamini-Hochberg FDR approach . Reverse transcription was performed with the Superscript II cDNA Synthesis Kit ( Invitrogen ) . The iQ SYBR Green Supermix ( ABI ) was used for quantitative RT-PCR as previously described ( Li et al . , 2008 ) . PCR primers are listed under ‘Primer sequences . ' 293T and rat hepatoma McA-RH777 cells were grown in Dulbecco's Modified Eagle Medium ( DMEM , GIBCO ) containing 10% FBS ( Sigma-Aldrich ) and 1% Pen-Strip ( GIBCO ) at 37°C in the presence of 5% CO2 . 293T cells were transfected with Lipofectamine 2000 ( Invitrogen ) according to the manufacturer's instruction . McA-RH777 cells grown in six wells were first transfected with 100 nM siRNA pools ( equal mixture of four different siRNAs; sequences are listed in ‘Primer Sequences' ) at ∼50% confluence . Cells were split 48 hr later to 60 mm dishes to ∼50% confluency for another round of siRNA treatment . Cytosol was prepared 48 hr after the third round of siRNA treatment . Production of lentivirus was carried out in 293T cells essentially as described ( Chen et al . , 2011a ) . Infection was performed by spinning viral particles onto cells at 1000 g for 45 min at room temperature in the presence of 8 μg/ml polybrene . For experiments involving conditioned medium collection , cells were washed once with serum free DMEM medium and then maintained in serum free DMEM for 4 hr before harvest . Conditioned medium was then centrifuged at 3 , 000 g for 5 min at 4°C to sediment floating cells or debris . Brefeldin A was purchased from Cell Signaling and dissolved in ethanol and used at a final concentration of 1 µg/ml . In vitro budding experiments using permeabilized cells were performed as previously described ( Kim et al . , 2007 ) with slight modification . Three plates of 293T cells stably expressing PCSK9-FLAG grown to ∼80–90% confluency were collected by trypsinization and sedimented at 750 g for 5 min at 4°C . Cells were resuspended in 6 ml of B88 buffer ( 20 mM HEPES pH 7 . 2 , 250 mM sorbitol , and 150 mM KOAc ) and permeabilized with 40 µg/ml digitonin on ice for 5 min . Permeabilization was stopped by adding ice-cold 8 ml of B88 buffer and permeabilized cells were sedimented at 750 g for 5 min at 4°C . After washing twice with 6 ml B88 buffer at 4°C , permeabilized cells were resuspended in 1 ml B88 buffer and centrifuged at 10 , 000 g for 1 min at 4°C . Permeabilized cells were then resuspended in 1 ml B88 buffer supplemented with 1 M MgCl2 for 10 min on ice to wash off cytosolic protein associated with cellular membranes . Permeabilized cells were then washed three times with 1 ml B88 buffer after being centrifuged at 10 , 000 g for 20 s , and finally resuspended in 0 . 2–0 . 3 ml B88 buffer . Budding reactions ( 100 µl ) were assembled in non-stick eppendorf tubes on ice with 20 µl ( OD600 = 0 . 1–0 . 2 ) semi-inact cells as donor membranes , 10 µl of 10× ATP regeneration system ( 10 mM ATP , 400 mM creatine phosphate , 2 mg/ml creatine phosphokinase , and 5 mM MgOAc in B88 buffer ) , 1 . 5 µl of 10 mM GTP , and rat liver cytosol at 4 mg/ml final concentration . For budding experiments with cytosols from McA-RH777 cells , the final concentration of cytosols was 2 mg/ml . Reactions were performed at 30°C for 60 min and stopped by centrifuging at 10 , 000 g for 10 min at 4°C . 75 µl supernatant was centrifuged in a TLA100 rotor at 100 , 000 g for 10 min at 4°C . The supernatants were discarded by pipetting with gel loading tips , and the high speed pellet fractions ( include COPII vesicles ) were thoroughly resuspended in 20 µl of 1× SDS sample buffer ( Invitrogen ) supplemented with 5% β-mercaptoethanol ( β-ME ) and heated at 55°C for 20 min before SDS-PAGE . Proteins were extracted from tissues or cells at 4°C for 60 min with buffer A ( 100 mM Tris pH 7 . 5 , 1% NP-40 , 10% glycerol , 130 mM sodium chloride , 5 mM magnesium chroride , 1 mM sodium vanadate , 1 mM sodium fluoride , and 1 mM EDTA ) supplemented with protease inhibitor tablets ( Roche ) , according to the manufacturer's instructions . Supernatants were collected after spinning at 13 , 000 g at 4°C for 10 min , and mixed 1:1 with 2× SDS sample buffer ( Invitrogen ) with 10% β-ME ( Sigma-Aldrich ) . For endoglycosidase H ( Endo H ) treatment , 20 µg of protein in 1× SDS sample buffer was incubated with 10U of Endo H ( NEB ) at 37°C for 2 hr . The enzyme was inactivated by heating at 70°C for 20 min . 10–20 µg of protein was loaded on 4–12% Tris-Glycine SDS-PAGE for separation . Protein extracts were incubated with 20 µl wheat germ agglutinin ( WGA ) beads ( EY Laboratories , San Mateo , CA ) for 4 hr at 4°C before washing four times with buffer A . Protein bound to WGA beads were solubilized with 1× SDS-PAGE sample buffer ( Invitrogen ) with 5% β-ME , and separated with 4–12% Tris-glycine SDS-PAGE ( Invitrogen ) . Immunoprecipitation was carried out with cellular proteins extracted at 4°C for 30 min with CHAPS based buffer B ( Kim et al . , 2002 ) ( 100 mM Tris pH 7 . 5 , 0 . 5% CHAPS , 10% glycerol , 130 mM sodium chloride , 10 mM sodium pyrophosphate , 5 mM magnesium chloride , 1 mM sodium fluoride , and 1 mM EDTA ) supplemented with protease inhibitor tablets ( Roche ) . Cell lysates were then incubated with 10 µl M2 agarose ( Sigma-Aldrich ) for 4 hr at 4°C before washing four times with buffer B . Immune-complexes were then solubilized with 1X SDS-PAGE sample buffer ( Invitrogen ) with 5% β-ME , and separated with 4–12% Tris-glycine SDS-PAGE ( Invitrogen ) . Following SDS-PAGE , proteins were transferred onto nitrocellulose membranes ( Bio-rad ) and immunoblotting was performed as previously described ( Chen et al . , 2011b ) . Sec24a-F ( 5′-GGGTAAGAGCAGCACCCGACTG ) Sec24a-R ( 5′-ATGTGCCCTAGGCATGAAAC ) Sec24a-V ( 5′-GGGTCTCAAAGTCAGGGTCA ) Sec24a-A ( 5′-CTGTCTTACAGGTTGTTCCGATGCACGCTG ) Sec24a-B ( 5′-CACAGCCAGCCCAGTGGTAT ) Sec24a-C ( 5′-AGGAAAAGAACCCTGTCATA ) Sec24a-C ( 5′-CACACCTCCCCCTGAACCTGAAAC ) Sec24a-Exon2 ( 5′-CAATATGTTTCTTCTGGAGACCC ) Sec24a-Exon3 ( 5′-AGCTGAGTTAAATGAAGGTGGGG ) Hmgcr ( 5′-ATGTTCACCGGCAACAACAA , 5′-GCGATGCACCGCGTTATC ) Ldlr ( 5′-CATCCGAGCCATTTTCACAGTC , 5′-CTGACTTGTCCTTGCAGTCTGC ) Scd1 ( 5′-GCTGGAGTACGTCTGGAGGAA , 5′-TCCCGAAGAGGCAGGTGTAG ) Mttp ( 5′-AGGCCGTCCAGAGCTTCCTG , 5′-GAGTCTGAGCAGAGGTGACG ) Apob ( 5′-GAGTTCCAGATGGTGTCTCCAAG , 5′-CTTGGAGTCTGACAAAGCTTAGC ) Apoe ( 5′-CAGACGCTGTCTGACCAGGTC , 5′-GTGTCTCCTCCGCCACTGGAC ) Pcsk9 ( 5′-AGTTGCCCCATGTGGAGTACA , 5′-TCTGGGCGAAGACAAAGGAGT ) Stealth RNAi against rat Sec24a/b sense sequences: rSec24a-1: UCGUUUCAGGUAAUCCUCAAAGAUU rSec24a-2: CCUAUCCCACCCGAUCGACUCUAAA rSec24a-3: CGGUCUGUCAAGAAGGUGACGUUCU rSec24a-4: GCUUUCCUGUUGGAGCUCUUAGGAU rSec24b-1: CAUGGAAUGACAUGUCACAAUCUAA rSec24b-2: CCAGAUAAAGCCAUCGCACAGUUAA rSec24b-3: CCACAGUACUCGUCUGUAUGAUUUA rSec24b-4: CAGUUGGUUUGGUGGUUCGUUUGUU	The endoplasmic reticulum ( ER ) is a structure that performs a variety of functions within eukaryotic cells . It can be divided into two regions: the surface of the rough ER is coated with ribosomes that manufacture various proteins , while the smooth ER is involved in activities such as lipid synthesis and carbohydrate metabolism . Proteins synthesized by the ribosomes attached to the rough ER are generally transferred to another structure within the cell , the Golgi apparatus , where they undergo further processing and packaging before being secreted or transported to another location within the cell . Proteins are shuttled from the ER to the Golgi apparatus by vesicles covered with coat protein complex II ( COPII ) . This complex is composed of an inner and outer coat , each of which is assembled primarily with two different SEC proteins: the SEC23/SEC24 protein heterodimer forms the inner coat of the COPII vesicle , and plays a key role in recruiting the appropriate protein cargos to the transport vesicle , while the SEC13/SEC31 protein heterotetramer forms the outer coat and is generally responsible for regulating vesicle size and rigidity . Previous work found that mammals , including humans and mice , harbor multiple copies of several SEC protein genes , including two copies of SEC23 and four copies of SEC24 . Both copies of SEC23 are derived from the same ancestral gene , and all four copies of SEC24 are derived from a different ancestral gene , and the availability of these copies potentially expands the range of properties that the vesicles can have . Insight into the roles of each SEC protein has come from work with SEC mutants . For example , a mutation in SEC23A was found to cause skeletal abnormalities in humans . Here , Chen et al . report the results of experiments which showed that mice with an inactive Sec24a gene could develop normally . However , these mice experienced a 45% reduction in their plasma cholesterol levels because they were not able to recruit and transport a secretory protein called PCSK9 , which is a critical regulator of blood cholesterol levels . The work of Chen et al . reveals a previously unappreciated complexity in the recruitment of secretory proteins to the COPII vesicle and suggests that the various combinations of SEC proteins influence the proteins selected for transport to the Golgi apparatus . The work also identifies Sec24a as a potential therapeutic target for the reduction of plasma cholesterol , a finding that could be of interest to researchers working on heart disease and other conditions exacerbated by high cholesterol .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2013	SEC24A deficiency lowers plasma cholesterol through reduced PCSK9 secretion

Subsets and Splits

No community queries yet

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